diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index 7198e134a2d26..26f4a6dfaa7bd 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -1186,8 +1186,10 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
 
       // Return the cost of multiple scalar invocation plus the cost of
       // inserting and extracting the values.
-      return getScalarizationOverhead(DstVTy, /*Insert*/ true, /*Extract*/ true,
-                                      CostKind) +
+      return getScalarizationOverhead(SrcVTy, /*Insert=*/ false,
+                                      /*Extract=*/ true, CostKind) +
+             getScalarizationOverhead(DstVTy, /*Insert=*/ true,
+                                      /*Extract=*/ false, CostKind) +
              Num * Cost;
     }
 
diff --git a/llvm/test/Analysis/CostModel/AArch64/cast.ll b/llvm/test/Analysis/CostModel/AArch64/cast.ll
index fa778864ae978..01e9d3483fc16 100644
--- a/llvm/test/Analysis/CostModel/AArch64/cast.ll
+++ b/llvm/test/Analysis/CostModel/AArch64/cast.ll
@@ -937,8 +937,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
@@ -947,8 +947,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r112 = fptoui <4 x float> undef to <4 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r113 = fptosi <4 x float> undef to <4 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r114 = fptoui <4 x float> undef to <4 x i16>
@@ -957,8 +957,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r117 = fptosi <4 x float> undef to <4 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
@@ -967,8 +967,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r132 = fptoui <8 x float> undef to <8 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r133 = fptosi <8 x float> undef to <8 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r134 = fptoui <8 x float> undef to <8 x i16>
@@ -977,8 +977,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r137 = fptosi <8 x float> undef to <8 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
@@ -987,8 +987,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 75 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 75 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %r152 = fptoui <16 x float> undef to <16 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %r153 = fptosi <16 x float> undef to <16 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r154 = fptoui <16 x float> undef to <16 x i16>
@@ -997,8 +997,8 @@ define i32 @casts_no_users() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r157 = fptosi <16 x float> undef to <16 x i32>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
@@ -1363,8 +1363,8 @@ define i32 @casts_no_users() {
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
@@ -1373,8 +1373,8 @@ define i32 @casts_no_users() {
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r112 = fptoui <4 x float> undef to <4 x i8>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r113 = fptosi <4 x float> undef to <4 x i8>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r114 = fptoui <4 x float> undef to <4 x i16>
@@ -1576,8 +1576,8 @@ define i32 @casts_no_users() {
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
@@ -1586,8 +1586,8 @@ define i32 @casts_no_users() {
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r112 = fptoui <4 x float> undef to <4 x i8>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r113 = fptosi <4 x float> undef to <4 x i8>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r114 = fptoui <4 x float> undef to <4 x i16>
@@ -3292,38 +3292,38 @@ define void @fp16cast() {
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r95 = fptosi <2 x half> undef to <2 x i16>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r96 = fptoui <2 x half> undef to <2 x i32>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x half> undef to <2 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r110 = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r111 = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r112 = fptoui <4 x half> undef to <4 x i8>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r113 = fptosi <4 x half> undef to <4 x i8>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r114 = fptoui <4 x half> undef to <4 x i16>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r115 = fptosi <4 x half> undef to <4 x i16>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r116 = fptoui <4 x half> undef to <4 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r117 = fptosi <4 x half> undef to <4 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %r118 = fptoui <4 x half> undef to <4 x i64>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %r119 = fptosi <4 x half> undef to <4 x i64>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r132 = fptoui <8 x half> undef to <8 x i8>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r133 = fptosi <8 x half> undef to <8 x i8>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r116 = fptoui <4 x half> undef to <4 x i32>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %r117 = fptosi <4 x half> undef to <4 x i32>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r118 = fptoui <4 x half> undef to <4 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r119 = fptosi <4 x half> undef to <4 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r132 = fptoui <8 x half> undef to <8 x i8>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r133 = fptosi <8 x half> undef to <8 x i8>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r134 = fptoui <8 x half> undef to <8 x i16>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r135 = fptosi <8 x half> undef to <8 x i16>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %r136 = fptoui <8 x half> undef to <8 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %r137 = fptosi <8 x half> undef to <8 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %r138 = fptoui <8 x half> undef to <8 x i64>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %r139 = fptosi <8 x half> undef to <8 x i64>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %r150 = fptoui <16 x half> undef to <16 x i1>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %r151 = fptosi <16 x half> undef to <16 x i1>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %r152 = fptoui <16 x half> undef to <16 x i8>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %r153 = fptosi <16 x half> undef to <16 x i8>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %r136 = fptoui <8 x half> undef to <8 x i32>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %r137 = fptosi <8 x half> undef to <8 x i32>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r138 = fptoui <8 x half> undef to <8 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r139 = fptosi <8 x half> undef to <8 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %r150 = fptoui <16 x half> undef to <16 x i1>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %r151 = fptosi <16 x half> undef to <16 x i1>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %r152 = fptoui <16 x half> undef to <16 x i8>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %r153 = fptosi <16 x half> undef to <16 x i8>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r154 = fptoui <16 x half> undef to <16 x i16>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r155 = fptosi <16 x half> undef to <16 x i16>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %r156 = fptoui <16 x half> undef to <16 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %r157 = fptosi <16 x half> undef to <16 x i32>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %r158 = fptoui <16 x half> undef to <16 x i64>
-; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %r159 = fptosi <16 x half> undef to <16 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %r156 = fptoui <16 x half> undef to <16 x i32>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %r157 = fptosi <16 x half> undef to <16 x i32>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %r158 = fptoui <16 x half> undef to <16 x i64>
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %r159 = fptosi <16 x half> undef to <16 x i64>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r250 = uitofp <8 x i1> undef to <8 x half>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r251 = sitofp <8 x i1> undef to <8 x half>
 ; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r252 = uitofp <8 x i8> undef to <8 x half>
@@ -3438,8 +3438,8 @@ define void @fp16cast() {
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r95 = fptosi <2 x half> undef to <2 x i16>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r96 = fptoui <2 x half> undef to <2 x i32>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x half> undef to <2 x i32>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r110 = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r111 = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r112 = fptoui <4 x half> undef to <4 x i8>
@@ -3448,28 +3448,28 @@ define void @fp16cast() {
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r115 = fptosi <4 x half> undef to <4 x i16>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r116 = fptoui <4 x half> undef to <4 x i32>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r117 = fptosi <4 x half> undef to <4 x i32>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %r118 = fptoui <4 x half> undef to <4 x i64>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %r119 = fptosi <4 x half> undef to <4 x i64>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r118 = fptoui <4 x half> undef to <4 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r119 = fptosi <4 x half> undef to <4 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r132 = fptoui <8 x half> undef to <8 x i8>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r133 = fptosi <8 x half> undef to <8 x i8>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r134 = fptoui <8 x half> undef to <8 x i16>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r135 = fptosi <8 x half> undef to <8 x i16>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r136 = fptoui <8 x half> undef to <8 x i32>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r137 = fptosi <8 x half> undef to <8 x i32>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %r138 = fptoui <8 x half> undef to <8 x i64>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %r139 = fptosi <8 x half> undef to <8 x i64>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %r150 = fptoui <16 x half> undef to <16 x i1>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %r151 = fptosi <16 x half> undef to <16 x i1>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r138 = fptoui <8 x half> undef to <8 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r139 = fptosi <8 x half> undef to <8 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %r150 = fptoui <16 x half> undef to <16 x i1>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %r151 = fptosi <16 x half> undef to <16 x i1>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r152 = fptoui <16 x half> undef to <16 x i8>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r153 = fptosi <16 x half> undef to <16 x i8>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r154 = fptoui <16 x half> undef to <16 x i16>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r155 = fptosi <16 x half> undef to <16 x i16>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r156 = fptoui <16 x half> undef to <16 x i32>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r157 = fptosi <16 x half> undef to <16 x i32>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %r158 = fptoui <16 x half> undef to <16 x i64>
-; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %r159 = fptosi <16 x half> undef to <16 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %r158 = fptoui <16 x half> undef to <16 x i64>
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %r159 = fptosi <16 x half> undef to <16 x i64>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r250 = uitofp <8 x i1> undef to <8 x half>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r251 = sitofp <8 x i1> undef to <8 x half>
 ; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r252 = uitofp <8 x i8> undef to <8 x half>
@@ -3511,8 +3511,8 @@ define void @fp16cast() {
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r95 = fptosi <2 x half> undef to <2 x i16>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r96 = fptoui <2 x half> undef to <2 x i32>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x half> undef to <2 x i32>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r110 = fptoui <4 x half> undef to <4 x i1>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r111 = fptosi <4 x half> undef to <4 x i1>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r112 = fptoui <4 x half> undef to <4 x i8>
@@ -3523,8 +3523,8 @@ define void @fp16cast() {
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r117 = fptosi <4 x half> undef to <4 x i32>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r118 = fptoui <4 x half> undef to <4 x i64>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r119 = fptosi <4 x half> undef to <4 x i64>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
-; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
+; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r132 = fptoui <8 x half> undef to <8 x i8>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r133 = fptosi <8 x half> undef to <8 x i8>
 ; FIXED-MIN-256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r134 = fptoui <8 x half> undef to <8 x i16>
@@ -3584,8 +3584,8 @@ define void @fp16cast() {
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r95 = fptosi <2 x half> undef to <2 x i16>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r96 = fptoui <2 x half> undef to <2 x i32>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x half> undef to <2 x i32>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98 = fptoui <2 x half> undef to <2 x i64>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99 = fptosi <2 x half> undef to <2 x i64>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r110 = fptoui <4 x half> undef to <4 x i1>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r111 = fptosi <4 x half> undef to <4 x i1>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r112 = fptoui <4 x half> undef to <4 x i8>
@@ -3596,8 +3596,8 @@ define void @fp16cast() {
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r117 = fptosi <4 x half> undef to <4 x i32>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r118 = fptoui <4 x half> undef to <4 x i64>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r119 = fptosi <4 x half> undef to <4 x i64>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
-; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r130 = fptoui <8 x half> undef to <8 x i1>
+; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %r131 = fptosi <8 x half> undef to <8 x i1>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r132 = fptoui <8 x half> undef to <8 x i8>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r133 = fptosi <8 x half> undef to <8 x i8>
 ; FIXED-MIN-2048-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r134 = fptoui <8 x half> undef to <8 x i16>
diff --git a/llvm/test/Analysis/CostModel/AArch64/fp-conversions-odd-vector-types.ll b/llvm/test/Analysis/CostModel/AArch64/fp-conversions-odd-vector-types.ll
index 51db7529386cd..414d78ed13d62 100644
--- a/llvm/test/Analysis/CostModel/AArch64/fp-conversions-odd-vector-types.ll
+++ b/llvm/test/Analysis/CostModel/AArch64/fp-conversions-odd-vector-types.ll
@@ -21,7 +21,7 @@ define <5 x i32> @fptoui_v5f32_to_v5i32(<5 x float> %in, ptr %dst) {
 
 define <3 x i16> @fptoui_v3f32_to_v3i16(<3 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptoui_v3f32_to_v3i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %conv = fptoui <3 x float> %in to <3 x i16>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %conv = fptoui <3 x float> %in to <3 x i16>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <3 x i16> %conv
 ;
   %conv = fptoui <3 x float> %in to <3 x i16>
@@ -30,7 +30,7 @@ define <3 x i16> @fptoui_v3f32_to_v3i16(<3 x float> %in, ptr %dst) {
 
 define <5 x i16> @fptoui_v5f32_to_v5i16(<5 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptoui_v5f32_to_v5i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %conv = fptoui <5 x float> %in to <5 x i16>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %conv = fptoui <5 x float> %in to <5 x i16>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <5 x i16> %conv
 ;
   %conv = fptoui <5 x float> %in to <5 x i16>
@@ -39,7 +39,7 @@ define <5 x i16> @fptoui_v5f32_to_v5i16(<5 x float> %in, ptr %dst) {
 
 define <3 x i8> @fptoui_v3f32_to_v3i8(<3 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptoui_v3f32_to_v3i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %conv = fptoui <3 x float> %in to <3 x i8>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %conv = fptoui <3 x float> %in to <3 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <3 x i8> %conv
 ;
   %conv = fptoui <3 x float> %in to <3 x i8>
@@ -48,7 +48,7 @@ define <3 x i8> @fptoui_v3f32_to_v3i8(<3 x float> %in, ptr %dst) {
 
 define <9 x i8> @fptoui_v9f32_to_v9i8(<9 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptoui_v9f32_to_v9i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %conv = fptoui <9 x float> %in to <9 x i8>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %conv = fptoui <9 x float> %in to <9 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <9 x i8> %conv
 ;
   %conv = fptoui <9 x float> %in to <9 x i8>
@@ -75,7 +75,7 @@ define <6 x i32> @fptosi_v6f32_to_v6i32(<6 x float> %in, ptr %dst) {
 
 define <3 x i16> @fptosi_v3f32_to_v3i16(<3 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptosi_v3f32_to_v3i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %conv = fptosi <3 x float> %in to <3 x i16>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %conv = fptosi <3 x float> %in to <3 x i16>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <3 x i16> %conv
 ;
   %conv = fptosi <3 x float> %in to <3 x i16>
@@ -84,7 +84,7 @@ define <3 x i16> @fptosi_v3f32_to_v3i16(<3 x float> %in, ptr %dst) {
 
 define <6 x i16> @fptosi_v6f32_to_v6i16(<6 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptosi_v6f32_to_v6i16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %conv = fptosi <6 x float> %in to <6 x i16>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %conv = fptosi <6 x float> %in to <6 x i16>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <6 x i16> %conv
 ;
   %conv = fptosi <6 x float> %in to <6 x i16>
@@ -93,7 +93,7 @@ define <6 x i16> @fptosi_v6f32_to_v6i16(<6 x float> %in, ptr %dst) {
 
 define <3 x i8> @fptosi_v3f32_to_v3i8(<3 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptosi_v3f32_to_v3i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %conv = fptosi <3 x float> %in to <3 x i8>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %conv = fptosi <3 x float> %in to <3 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <3 x i8> %conv
 ;
   %conv = fptosi <3 x float> %in to <3 x i8>
@@ -102,7 +102,7 @@ define <3 x i8> @fptosi_v3f32_to_v3i8(<3 x float> %in, ptr %dst) {
 
 define <5 x i8> @fptosi_v5f32_to_v5i8(<5 x float> %in, ptr %dst) {
 ; CHECK-LABEL: 'fptosi_v5f32_to_v5i8'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %conv = fptosi <5 x float> %in to <5 x i8>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %conv = fptosi <5 x float> %in to <5 x i8>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <5 x i8> %conv
 ;
   %conv = fptosi <5 x float> %in to <5 x i8>
@@ -138,7 +138,7 @@ define <3 x float> @uitofp_v3i16_to_v3f32(<3 x i16> %in, ptr %dst) {
 
 define <7 x float> @uitofp_v7i16_to_v7f32(<7 x i16> %in, ptr %dst) {
 ; CHECK-LABEL: 'uitofp_v7i16_to_v7f32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %conv = uitofp <7 x i16> %in to <7 x float>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %conv = uitofp <7 x i16> %in to <7 x float>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <7 x float> %conv
 ;
   %conv = uitofp <7 x i16> %in to <7 x float>
@@ -156,7 +156,7 @@ define <3 x float> @uitofp_v3i8_to_v3f32(<3 x i8> %in, ptr %dst) {
 
 define <11 x float> @uitofp_v11i8_to_v11f32(<11 x i8> %in, ptr %dst) {
 ; CHECK-LABEL: 'uitofp_v11i8_to_v11f32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %conv = uitofp <11 x i8> %in to <11 x float>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %conv = uitofp <11 x i8> %in to <11 x float>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <11 x float> %conv
 ;
   %conv = uitofp <11 x i8> %in to <11 x float>
@@ -192,7 +192,7 @@ define <3 x float> @sitofp_v3i16_to_v3f32(<3 x i16> %in, ptr %dst) {
 
 define <6 x float> @sitofp_v6i16_to_v6f32(<6 x i16> %in, ptr %dst) {
 ; CHECK-LABEL: 'sitofp_v6i16_to_v6f32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %conv = sitofp <6 x i16> %in to <6 x float>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %conv = sitofp <6 x i16> %in to <6 x float>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <6 x float> %conv
 ;
   %conv = sitofp <6 x i16> %in to <6 x float>
@@ -219,7 +219,7 @@ define <8 x float> @sitofp_v8i8_to_v8f32(<8 x i8> %in, ptr %dst) {
 
 define <7 x float> @sitofp_v7i8_to_v7f32(<7 x i8> %in, ptr %dst) {
 ; CHECK-LABEL: 'sitofp_v7i8_to_v7f32'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %conv = sitofp <7 x i8> %in to <7 x float>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %conv = sitofp <7 x i8> %in to <7 x float>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <7 x float> %conv
 ;
   %conv = sitofp <7 x i8> %in to <7 x float>
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/cast.ll b/llvm/test/Analysis/CostModel/AMDGPU/cast.ll
index fc8f5fc393396..c6a38dd658e9d 100644
--- a/llvm/test/Analysis/CostModel/AMDGPU/cast.ll
+++ b/llvm/test/Analysis/CostModel/AMDGPU/cast.ll
@@ -56,13 +56,13 @@ define i32 @zext_sext(<8 x i1> %in) {
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %S = sext <8 x i1> %in to <8 x i32>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %A1 = zext <16 x i8> undef to <16 x i16>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %A2 = sext <16 x i8> undef to <16 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A = sext <8 x i16> undef to <8 x i32>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %A = sext <8 x i16> undef to <8 x i32>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %B = zext <8 x i16> undef to <8 x i32>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C = sext <4 x i32> undef to <4 x i64>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %C.v8i8.z = zext <8 x i8> undef to <8 x i32>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C.v8i8.s = sext <8 x i8> undef to <8 x i32>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %C.v4i16.z = zext <4 x i16> undef to <4 x i64>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C.v4i16.s = sext <4 x i16> undef to <4 x i64>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %C.v4i16.s = sext <4 x i16> undef to <4 x i64>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %C.v4i8.z = zext <4 x i8> undef to <4 x i64>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C.v4i8.s = sext <4 x i8> undef to <4 x i64>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %D = zext <4 x i32> undef to <4 x i64>
@@ -116,13 +116,13 @@ define i32 @zext_sext(<8 x i1> %in) {
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %S = sext <8 x i1> %in to <8 x i32>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %A1 = zext <16 x i8> undef to <16 x i16>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %A2 = sext <16 x i8> undef to <16 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A = sext <8 x i16> undef to <8 x i32>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %A = sext <8 x i16> undef to <8 x i32>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %B = zext <8 x i16> undef to <8 x i32>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C = sext <4 x i32> undef to <4 x i64>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %C.v8i8.z = zext <8 x i8> undef to <8 x i32>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C.v8i8.s = sext <8 x i8> undef to <8 x i32>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %C.v4i16.z = zext <4 x i16> undef to <4 x i64>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C.v4i16.s = sext <4 x i16> undef to <4 x i64>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %C.v4i16.s = sext <4 x i16> undef to <4 x i64>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %C.v4i8.z = zext <4 x i8> undef to <4 x i64>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C.v4i8.s = sext <4 x i8> undef to <4 x i64>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %D = zext <4 x i32> undef to <4 x i64>
@@ -249,7 +249,7 @@ define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B1 = sitofp <4 x i8> %b to <4 x float>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B2 = sitofp <4 x i8> %b to <4 x double>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %C1 = sitofp <4 x i16> %c to <4 x float>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C2 = sitofp <4 x i16> %c to <4 x double>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %C2 = sitofp <4 x i16> %c to <4 x double>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %D1 = sitofp <4 x i32> %d to <4 x float>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %D2 = sitofp <4 x i32> %d to <4 x double>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
@@ -271,7 +271,7 @@ define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B1 = sitofp <4 x i8> %b to <4 x float>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B2 = sitofp <4 x i8> %b to <4 x double>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %C1 = sitofp <4 x i16> %c to <4 x float>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C2 = sitofp <4 x i16> %c to <4 x double>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %C2 = sitofp <4 x i16> %c to <4 x double>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %D1 = sitofp <4 x i32> %d to <4 x float>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %D2 = sitofp <4 x i32> %d to <4 x double>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
@@ -299,19 +299,33 @@ define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 }
 
 define void @sitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
-; ALL-LABEL: 'sitofp8'
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = sitofp <8 x i1> %a to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = sitofp <8 x i8> %b to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = sitofp <8 x i16> %c to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = sitofp <8 x i32> %d to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
+; FAST-LABEL: 'sitofp8'
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = sitofp <8 x i1> %a to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = sitofp <8 x i8> %b to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %C1 = sitofp <8 x i16> %c to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = sitofp <8 x i32> %d to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
 ;
-; ALL-SIZE-LABEL: 'sitofp8'
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = sitofp <8 x i1> %a to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = sitofp <8 x i8> %b to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = sitofp <8 x i16> %c to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = sitofp <8 x i32> %d to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SLOW-LABEL: 'sitofp8'
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = sitofp <8 x i1> %a to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = sitofp <8 x i8> %b to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = sitofp <8 x i16> %c to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = sitofp <8 x i32> %d to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
+;
+; FAST-SIZE-LABEL: 'sitofp8'
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = sitofp <8 x i1> %a to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = sitofp <8 x i8> %b to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %C1 = sitofp <8 x i16> %c to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = sitofp <8 x i32> %d to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SLOW-SIZE-LABEL: 'sitofp8'
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = sitofp <8 x i1> %a to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = sitofp <8 x i8> %b to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = sitofp <8 x i16> %c to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = sitofp <8 x i32> %d to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %A1 = sitofp <8 x i1> %a to <8 x float>
   %B1 = sitofp <8 x i8> %b to <8 x float>
@@ -327,7 +341,7 @@ define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B1 = uitofp <4 x i8> %b to <4 x float>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B2 = uitofp <4 x i8> %b to <4 x double>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %C1 = uitofp <4 x i16> %c to <4 x float>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C2 = uitofp <4 x i16> %c to <4 x double>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %C2 = uitofp <4 x i16> %c to <4 x double>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %D1 = uitofp <4 x i32> %d to <4 x float>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %D2 = uitofp <4 x i32> %d to <4 x double>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
@@ -349,7 +363,7 @@ define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B1 = uitofp <4 x i8> %b to <4 x float>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %B2 = uitofp <4 x i8> %b to <4 x double>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %C1 = uitofp <4 x i16> %c to <4 x float>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %C2 = uitofp <4 x i16> %c to <4 x double>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %C2 = uitofp <4 x i16> %c to <4 x double>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %D1 = uitofp <4 x i32> %d to <4 x float>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %D2 = uitofp <4 x i32> %d to <4 x double>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
@@ -377,19 +391,33 @@ define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
 }
 
 define void @uitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
-; ALL-LABEL: 'uitofp8'
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = uitofp <8 x i1> %a to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = uitofp <8 x i8> %b to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = uitofp <8 x i16> %c to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = uitofp <8 x i32> %d to <8 x float>
-; ALL-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
+; FAST-LABEL: 'uitofp8'
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = uitofp <8 x i1> %a to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = uitofp <8 x i8> %b to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %C1 = uitofp <8 x i16> %c to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = uitofp <8 x i32> %d to <8 x float>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
 ;
-; ALL-SIZE-LABEL: 'uitofp8'
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = uitofp <8 x i1> %a to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = uitofp <8 x i8> %b to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = uitofp <8 x i16> %c to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = uitofp <8 x i32> %d to <8 x float>
-; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SLOW-LABEL: 'uitofp8'
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = uitofp <8 x i1> %a to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = uitofp <8 x i8> %b to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = uitofp <8 x i16> %c to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = uitofp <8 x i32> %d to <8 x float>
+; SLOW-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
+;
+; FAST-SIZE-LABEL: 'uitofp8'
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = uitofp <8 x i1> %a to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = uitofp <8 x i8> %b to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %C1 = uitofp <8 x i16> %c to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = uitofp <8 x i32> %d to <8 x float>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SLOW-SIZE-LABEL: 'uitofp8'
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %A1 = uitofp <8 x i1> %a to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %B1 = uitofp <8 x i8> %b to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %C1 = uitofp <8 x i16> %c to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %D1 = uitofp <8 x i32> %d to <8 x float>
+; SLOW-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %A1 = uitofp <8 x i1> %a to <8 x float>
   %B1 = uitofp <8 x i8> %b to <8 x float>
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/fptosi.ll b/llvm/test/Analysis/CostModel/AMDGPU/fptosi.ll
index b74d6a55da8e3..cd84194922d17 100644
--- a/llvm/test/Analysis/CostModel/AMDGPU/fptosi.ll
+++ b/llvm/test/Analysis/CostModel/AMDGPU/fptosi.ll
@@ -57,9 +57,9 @@ define i32 @fptosi_double_i32(i32 %arg) {
 define i32 @fptosi_double_i16(i32 %arg) {
 ; FAST-LABEL: 'fptosi_double_i16'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi double undef to i16
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptosi_double_i16'
@@ -71,9 +71,9 @@ define i32 @fptosi_double_i16(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptosi_double_i16'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi double undef to i16
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptosi <2 x double> undef to <2 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptosi <4 x double> undef to <4 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptosi <8 x double> undef to <8 x i16>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptosi_double_i16'
@@ -93,9 +93,9 @@ define i32 @fptosi_double_i16(i32 %arg) {
 define i32 @fptosi_double_i8(i32 %arg) {
 ; FAST-LABEL: 'fptosi_double_i8'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptosi double undef to i8
-; FAST-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptosi <2 x double> undef to <2 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptosi <4 x double> undef to <4 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptosi <8 x double> undef to <8 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptosi <2 x double> undef to <2 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptosi <4 x double> undef to <4 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptosi <8 x double> undef to <8 x i8>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptosi_double_i8'
@@ -107,9 +107,9 @@ define i32 @fptosi_double_i8(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptosi_double_i8'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptosi double undef to i8
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptosi <2 x double> undef to <2 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptosi <4 x double> undef to <4 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptosi <8 x double> undef to <8 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptosi <2 x double> undef to <2 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptosi <4 x double> undef to <4 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptosi <8 x double> undef to <8 x i8>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptosi_double_i8'
@@ -179,10 +179,10 @@ define i32 @fptosi_float_i32(i32 %arg) {
 define i32 @fptosi_float_i16(i32 %arg) {
 ; FAST-LABEL: 'fptosi_float_i16'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi float undef to i16
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptosi <2 x float> undef to <2 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptosi <2 x float> undef to <2 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptosi_float_i16'
@@ -195,10 +195,10 @@ define i32 @fptosi_float_i16(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptosi_float_i16'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptosi float undef to i16
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptosi <2 x float> undef to <2 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptosi <2 x float> undef to <2 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptosi <4 x float> undef to <4 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptosi <8 x float> undef to <8 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V16I16 = fptosi <16 x float> undef to <16 x i16>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptosi_float_i16'
@@ -220,10 +220,10 @@ define i32 @fptosi_float_i16(i32 %arg) {
 define i32 @fptosi_float_i8(i32 %arg) {
 ; FAST-LABEL: 'fptosi_float_i8'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptosi float undef to i8
-; FAST-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptosi <2 x float> undef to <2 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptosi <4 x float> undef to <4 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptosi <8 x float> undef to <8 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = fptosi <16 x float> undef to <16 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptosi <2 x float> undef to <2 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptosi <4 x float> undef to <4 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptosi <8 x float> undef to <8 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = fptosi <16 x float> undef to <16 x i8>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptosi_float_i8'
@@ -236,10 +236,10 @@ define i32 @fptosi_float_i8(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptosi_float_i8'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptosi float undef to i8
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptosi <2 x float> undef to <2 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptosi <4 x float> undef to <4 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptosi <8 x float> undef to <8 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = fptosi <16 x float> undef to <16 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptosi <2 x float> undef to <2 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptosi <4 x float> undef to <4 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptosi <8 x float> undef to <8 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = fptosi <16 x float> undef to <16 x i8>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptosi_float_i8'
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/fptoui.ll b/llvm/test/Analysis/CostModel/AMDGPU/fptoui.ll
index 381c2951dc1ef..8eb4540303188 100644
--- a/llvm/test/Analysis/CostModel/AMDGPU/fptoui.ll
+++ b/llvm/test/Analysis/CostModel/AMDGPU/fptoui.ll
@@ -57,9 +57,9 @@ define i32 @fptoui_double_i32(i32 %arg) {
 define i32 @fptoui_double_i16(i32 %arg) {
 ; FAST-LABEL: 'fptoui_double_i16'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptoui double undef to i16
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptoui <2 x double> undef to <2 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptoui <4 x double> undef to <4 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptoui <8 x double> undef to <8 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptoui <2 x double> undef to <2 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptoui <4 x double> undef to <4 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptoui <8 x double> undef to <8 x i16>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptoui_double_i16'
@@ -71,9 +71,9 @@ define i32 @fptoui_double_i16(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptoui_double_i16'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptoui double undef to i16
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptoui <2 x double> undef to <2 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptoui <4 x double> undef to <4 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptoui <8 x double> undef to <8 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptoui <2 x double> undef to <2 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptoui <4 x double> undef to <4 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptoui <8 x double> undef to <8 x i16>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptoui_double_i16'
@@ -93,9 +93,9 @@ define i32 @fptoui_double_i16(i32 %arg) {
 define i32 @fptoui_double_i8(i32 %arg) {
 ; FAST-LABEL: 'fptoui_double_i8'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptoui double undef to i8
-; FAST-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptoui <2 x double> undef to <2 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptoui <4 x double> undef to <4 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptoui <8 x double> undef to <8 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptoui <2 x double> undef to <2 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptoui <4 x double> undef to <4 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptoui <8 x double> undef to <8 x i8>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptoui_double_i8'
@@ -107,9 +107,9 @@ define i32 @fptoui_double_i8(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptoui_double_i8'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptoui double undef to i8
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptoui <2 x double> undef to <2 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptoui <4 x double> undef to <4 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptoui <8 x double> undef to <8 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptoui <2 x double> undef to <2 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptoui <4 x double> undef to <4 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptoui <8 x double> undef to <8 x i8>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptoui_double_i8'
@@ -179,10 +179,10 @@ define i32 @fptoui_float_i32(i32 %arg) {
 define i32 @fptoui_float_i16(i32 %arg) {
 ; FAST-LABEL: 'fptoui_float_i16'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptoui float undef to i16
-; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptoui <2 x float> undef to <2 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptoui <4 x float> undef to <4 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptoui <8 x float> undef to <8 x i16>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V16I16 = fptoui <16 x float> undef to <16 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptoui <2 x float> undef to <2 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptoui <4 x float> undef to <4 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptoui <8 x float> undef to <8 x i16>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V16I16 = fptoui <16 x float> undef to <16 x i16>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptoui_float_i16'
@@ -195,10 +195,10 @@ define i32 @fptoui_float_i16(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptoui_float_i16'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = fptoui float undef to i16
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I16 = fptoui <2 x float> undef to <2 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I16 = fptoui <4 x float> undef to <4 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = fptoui <8 x float> undef to <8 x i16>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V16I16 = fptoui <16 x float> undef to <16 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I16 = fptoui <2 x float> undef to <2 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V4I16 = fptoui <4 x float> undef to <4 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = fptoui <8 x float> undef to <8 x i16>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V16I16 = fptoui <16 x float> undef to <16 x i16>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptoui_float_i16'
@@ -220,10 +220,10 @@ define i32 @fptoui_float_i16(i32 %arg) {
 define i32 @fptoui_float_i8(i32 %arg) {
 ; FAST-LABEL: 'fptoui_float_i8'
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptoui float undef to i8
-; FAST-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptoui <2 x float> undef to <2 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptoui <4 x float> undef to <4 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptoui <8 x float> undef to <8 x i8>
-; FAST-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = fptoui <16 x float> undef to <16 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptoui <2 x float> undef to <2 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptoui <4 x float> undef to <4 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptoui <8 x float> undef to <8 x i8>
+; FAST-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = fptoui <16 x float> undef to <16 x i8>
 ; FAST-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret i32 undef
 ;
 ; SLOW-LABEL: 'fptoui_float_i8'
@@ -236,10 +236,10 @@ define i32 @fptoui_float_i8(i32 %arg) {
 ;
 ; FAST-SIZE-LABEL: 'fptoui_float_i8'
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I8 = fptoui float undef to i8
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I8 = fptoui <2 x float> undef to <2 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I8 = fptoui <4 x float> undef to <4 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I8 = fptoui <8 x float> undef to <8 x i8>
-; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = fptoui <16 x float> undef to <16 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V2I8 = fptoui <2 x float> undef to <2 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V4I8 = fptoui <4 x float> undef to <4 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V8I8 = fptoui <8 x float> undef to <8 x i8>
+; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V16I8 = fptoui <16 x float> undef to <16 x i8>
 ; FAST-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; SLOW-SIZE-LABEL: 'fptoui_float_i8'
diff --git a/llvm/test/Analysis/CostModel/ARM/arith-overflow.ll b/llvm/test/Analysis/CostModel/ARM/arith-overflow.ll
index 484cd201d14a0..6cf9ca8a38ca9 100644
--- a/llvm/test/Analysis/CostModel/ARM/arith-overflow.ll
+++ b/llvm/test/Analysis/CostModel/ARM/arith-overflow.ll
@@ -667,9 +667,9 @@ define i32 @smul(i32 %arg) {
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 164 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.smul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 328 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.smul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %I32 = call { i32, i1 } @llvm.smul.with.overflow.i32(i32 undef, i32 undef)
-; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
-; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
-; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
+; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
+; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.smul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
+; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 240 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.smul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %I16 = call { i16, i1 } @llvm.smul.with.overflow.i16(i16 undef, i16 undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.smul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.smul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
@@ -825,9 +825,9 @@ define i32 @umul(i32 %arg) {
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 156 for instruction: %V4I64 = call { <4 x i64>, <4 x i1> } @llvm.umul.with.overflow.v4i64(<4 x i64> undef, <4 x i64> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 312 for instruction: %V8I64 = call { <8 x i64>, <8 x i1> } @llvm.umul.with.overflow.v8i64(<8 x i64> undef, <8 x i64> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %I32 = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 undef, i32 undef)
-; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
-; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
-; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
+; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V4I32 = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> undef, <4 x i32> undef)
+; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V8I32 = call { <8 x i32>, <8 x i1> } @llvm.umul.with.overflow.v8i32(<8 x i32> undef, <8 x i32> undef)
+; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V16I32 = call { <16 x i32>, <16 x i1> } @llvm.umul.with.overflow.v16i32(<16 x i32> undef, <16 x i32> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %I16 = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 undef, i16 undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I16 = call { <8 x i16>, <8 x i1> } @llvm.umul.with.overflow.v8i16(<8 x i16> undef, <8 x i16> undef)
 ; V8M-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V16I16 = call { <16 x i16>, <16 x i1> } @llvm.umul.with.overflow.v16i16(<16 x i16> undef, <16 x i16> undef)
diff --git a/llvm/test/Analysis/CostModel/ARM/cast.ll b/llvm/test/Analysis/CostModel/ARM/cast.ll
index ae0d2347ec8be..38be8d90a3e38 100644
--- a/llvm/test/Analysis/CostModel/ARM/cast.ll
+++ b/llvm/test/Analysis/CostModel/ARM/cast.ll
@@ -173,8 +173,8 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r95 = fptosi <2 x float> undef to <2 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r96 = fptoui <2 x float> undef to <2 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r90h = fptoui <2 x half> undef to <2 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r91h = fptosi <2 x half> undef to <2 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r92h = fptoui <2 x half> undef to <2 x i8>
@@ -183,28 +183,28 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r95h = fptosi <2 x half> undef to <2 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r96h = fptoui <2 x half> undef to <2 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r97h = fptosi <2 x half> undef to <2 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r112 = fptoui <4 x float> undef to <4 x i8>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r113 = fptosi <4 x float> undef to <4 x i8>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r114 = fptoui <4 x float> undef to <4 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r115 = fptosi <4 x float> undef to <4 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r116 = fptoui <4 x float> undef to <4 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r117 = fptosi <4 x float> undef to <4 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r110h = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r111h = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r112h = fptoui <4 x half> undef to <4 x i8>
@@ -213,28 +213,28 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r115h = fptosi <4 x half> undef to <4 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r116h = fptoui <4 x half> undef to <4 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r117h = fptosi <4 x half> undef to <4 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r132 = fptoui <8 x float> undef to <8 x i8>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r133 = fptosi <8 x float> undef to <8 x i8>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r134 = fptoui <8 x float> undef to <8 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r135 = fptosi <8 x float> undef to <8 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r136 = fptoui <8 x float> undef to <8 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r137 = fptosi <8 x float> undef to <8 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r130h = fptoui <8 x half> undef to <8 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r131h = fptosi <8 x half> undef to <8 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r132h = fptoui <8 x half> undef to <8 x i8>
@@ -243,28 +243,28 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r135h = fptosi <8 x half> undef to <8 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r136h = fptoui <8 x half> undef to <8 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r137h = fptosi <8 x half> undef to <8 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 115 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 115 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r152 = fptoui <16 x float> undef to <16 x i8>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r153 = fptosi <16 x float> undef to <16 x i8>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r154 = fptoui <16 x float> undef to <16 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r155 = fptosi <16 x float> undef to <16 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r156 = fptoui <16 x float> undef to <16 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r157 = fptosi <16 x float> undef to <16 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r150h = fptoui <16 x half> undef to <16 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r151h = fptosi <16 x half> undef to <16 x i1>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r152h = fptoui <16 x half> undef to <16 x i8>
@@ -273,28 +273,28 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r155h = fptosi <16 x half> undef to <16 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r156h = fptoui <16 x half> undef to <16 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r157h = fptosi <16 x half> undef to <16 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r171 = sitofp <2 x i1> undef to <2 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r171 = sitofp <2 x i1> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r172 = uitofp <2 x i8> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r173 = sitofp <2 x i8> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r174 = uitofp <2 x i16> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r175 = sitofp <2 x i16> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r176 = uitofp <2 x i32> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r177 = sitofp <2 x i32> undef to <2 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r170h = uitofp <2 x i1> undef to <2 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r171h = sitofp <2 x i1> undef to <2 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r172h = uitofp <2 x i8> undef to <2 x half>
@@ -305,16 +305,16 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r177h = sitofp <2 x i32> undef to <2 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r190 = uitofp <4 x i1> undef to <4 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r191 = sitofp <4 x i1> undef to <4 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r192 = uitofp <4 x i8> undef to <4 x float>
@@ -323,8 +323,8 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r195 = sitofp <4 x i16> undef to <4 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r196 = uitofp <4 x i32> undef to <4 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r197 = sitofp <4 x i32> undef to <4 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r190h = uitofp <4 x i1> undef to <4 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r191h = sitofp <4 x i1> undef to <4 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r192h = uitofp <4 x i8> undef to <4 x half>
@@ -335,16 +335,16 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r197h = sitofp <4 x i32> undef to <4 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r210 = uitofp <8 x i1> undef to <8 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r211 = sitofp <8 x i1> undef to <8 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r212 = uitofp <8 x i8> undef to <8 x float>
@@ -353,8 +353,8 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r215 = sitofp <8 x i16> undef to <8 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r216 = uitofp <8 x i32> undef to <8 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r217 = sitofp <8 x i32> undef to <8 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r210h = uitofp <8 x i1> undef to <8 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r211h = sitofp <8 x i1> undef to <8 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r212h = uitofp <8 x i8> undef to <8 x half>
@@ -365,16 +365,16 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r217h = sitofp <8 x i32> undef to <8 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r230 = uitofp <16 x i1> undef to <16 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r231 = sitofp <16 x i1> undef to <16 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r232 = uitofp <16 x i8> undef to <16 x float>
@@ -383,8 +383,8 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r235 = sitofp <16 x i16> undef to <16 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r236 = uitofp <16 x i32> undef to <16 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r237 = sitofp <16 x i32> undef to <16 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r230h = uitofp <16 x i1> undef to <16 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r231h = sitofp <16 x i1> undef to <16 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r232h = uitofp <16 x i8> undef to <16 x half>
@@ -395,16 +395,16 @@ define i32 @casts() {
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r237h = sitofp <16 x i32> undef to <16 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
-; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
+; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
 ; CHECK-NEON-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; CHECK-MVE-RECIP-LABEL: 'casts'
@@ -559,36 +559,36 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r87hf = fpext <4 x half> undef to <4 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r88hf = fpext <8 x half> undef to <8 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r89hf = fpext <16 x half> undef to <16 x float>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r90 = fptoui <2 x float> undef to <2 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r91 = fptosi <2 x float> undef to <2 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r92 = fptoui <2 x float> undef to <2 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r93 = fptosi <2 x float> undef to <2 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r94 = fptoui <2 x float> undef to <2 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r95 = fptosi <2 x float> undef to <2 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r96 = fptoui <2 x float> undef to <2 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r90h = fptoui <2 x half> undef to <2 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r91h = fptosi <2 x half> undef to <2 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r92h = fptoui <2 x half> undef to <2 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r93h = fptosi <2 x half> undef to <2 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r94h = fptoui <2 x half> undef to <2 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r95h = fptosi <2 x half> undef to <2 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r96h = fptoui <2 x half> undef to <2 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r97h = fptosi <2 x half> undef to <2 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r90 = fptoui <2 x float> undef to <2 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r91 = fptosi <2 x float> undef to <2 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r92 = fptoui <2 x float> undef to <2 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r93 = fptosi <2 x float> undef to <2 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r94 = fptoui <2 x float> undef to <2 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r95 = fptosi <2 x float> undef to <2 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r96 = fptoui <2 x float> undef to <2 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r90h = fptoui <2 x half> undef to <2 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r91h = fptosi <2 x half> undef to <2 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r92h = fptoui <2 x half> undef to <2 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r93h = fptosi <2 x half> undef to <2 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r94h = fptoui <2 x half> undef to <2 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r95h = fptosi <2 x half> undef to <2 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r96h = fptoui <2 x half> undef to <2 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r97h = fptosi <2 x half> undef to <2 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r112 = fptoui <4 x float> undef to <4 x i8>
@@ -597,8 +597,8 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r115 = fptosi <4 x float> undef to <4 x i16>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r116 = fptoui <4 x float> undef to <4 x i32>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r117 = fptosi <4 x float> undef to <4 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 274 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 274 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r110h = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r111h = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r112h = fptoui <4 x half> undef to <4 x i8>
@@ -607,18 +607,18 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r115h = fptosi <4 x half> undef to <4 x i16>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r116h = fptoui <4 x half> undef to <4 x i32>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r117h = fptosi <4 x half> undef to <4 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 274 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 274 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 272 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r132 = fptoui <8 x float> undef to <8 x i8>
@@ -627,8 +627,8 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r135 = fptosi <8 x float> undef to <8 x i16>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r136 = fptoui <8 x float> undef to <8 x i32>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r137 = fptosi <8 x float> undef to <8 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1096 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1096 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 648 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 648 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r130h = fptoui <8 x half> undef to <8 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r131h = fptosi <8 x half> undef to <8 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r132h = fptoui <8 x half> undef to <8 x i8>
@@ -637,18 +637,18 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r135h = fptosi <8 x half> undef to <8 x i16>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r136h = fptoui <8 x half> undef to <8 x i32>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r137h = fptosi <8 x half> undef to <8 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1098 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1098 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 584 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 584 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1088 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1088 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 392 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 392 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r152 = fptoui <16 x float> undef to <16 x i8>
@@ -657,8 +657,8 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r155 = fptosi <16 x float> undef to <16 x i16>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r156 = fptoui <16 x float> undef to <16 x i32>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r157 = fptosi <16 x float> undef to <16 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4384 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4384 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2592 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2592 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r150h = fptoui <16 x half> undef to <16 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r151h = fptosi <16 x half> undef to <16 x i1>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r152h = fptoui <16 x half> undef to <16 x i8>
@@ -667,18 +667,18 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r155h = fptosi <16 x half> undef to <16 x i16>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r156h = fptoui <16 x half> undef to <16 x i32>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r157h = fptosi <16 x half> undef to <16 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4392 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4392 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2344 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2344 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2336 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2336 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4352 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4352 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2600 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2600 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1568 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1568 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2560 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2560 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r171 = sitofp <2 x i1> undef to <2 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r172 = uitofp <2 x i8> undef to <2 x float>
@@ -699,16 +699,16 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r177h = sitofp <2 x i32> undef to <2 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r190 = uitofp <4 x i1> undef to <4 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r191 = sitofp <4 x i1> undef to <4 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r192 = uitofp <4 x i8> undef to <4 x float>
@@ -729,16 +729,16 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r197h = sitofp <4 x i32> undef to <4 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 146 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r210 = uitofp <8 x i1> undef to <8 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r211 = sitofp <8 x i1> undef to <8 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r212 = uitofp <8 x i8> undef to <8 x float>
@@ -759,16 +759,16 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r217h = sitofp <8 x i32> undef to <8 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 394 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 586 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 832 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 832 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r230 = uitofp <16 x i1> undef to <16 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r231 = sitofp <16 x i1> undef to <16 x float>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %r232 = uitofp <16 x i8> undef to <16 x float>
@@ -789,16 +789,16 @@ define i32 @casts() {
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r237h = sitofp <16 x i32> undef to <16 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
-; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2346 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2344 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2344 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2344 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2344 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 3328 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
+; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 3328 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
 ; CHECK-MVE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; CHECK-V8M-MAIN-RECIP-LABEL: 'casts'
@@ -902,18 +902,18 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %s72 = zext <4 x i8> undef to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r72 = zext <8 x i8> undef to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r73 = zext <16 x i8> undef to <16 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_0 = sext <8 x i8> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_1 = zext <8 x i8> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %rext_2 = sext <8 x i16> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_3 = zext <8 x i16> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %rext_4 = sext <4 x i16> undef to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %rext_5 = zext <4 x i16> undef to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_6 = sext <2 x i8> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_7 = zext <2 x i8> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %rext_8 = sext <2 x i16> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_9 = zext <2 x i16> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_a = sext <2 x i32> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_b = zext <2 x i32> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %rext_0 = sext <8 x i8> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %rext_1 = zext <8 x i8> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_2 = sext <8 x i16> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %rext_3 = zext <8 x i16> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %rext_4 = sext <4 x i16> undef to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %rext_5 = zext <4 x i16> undef to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_6 = sext <2 x i8> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_7 = zext <2 x i8> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_8 = sext <2 x i16> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_9 = zext <2 x i16> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_a = sext <2 x i32> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_b = zext <2 x i32> undef to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %tv4i32i8 = trunc <4 x i32> undef to <4 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %tv4i32i16 = trunc <4 x i32> undef to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %tv4i16i8 = trunc <4 x i16> undef to <4 x i8>
@@ -961,8 +961,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r95 = fptosi <2 x float> undef to <2 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r96 = fptoui <2 x float> undef to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r90h = fptoui <2 x half> undef to <2 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r91h = fptosi <2 x half> undef to <2 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r92h = fptoui <2 x half> undef to <2 x i8>
@@ -971,16 +971,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r95h = fptosi <2 x half> undef to <2 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r96h = fptoui <2 x half> undef to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r97h = fptosi <2 x half> undef to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
@@ -991,8 +991,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r115 = fptosi <4 x float> undef to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r116 = fptoui <4 x float> undef to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r117 = fptosi <4 x float> undef to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r110h = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r111h = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r112h = fptoui <4 x half> undef to <4 x i8>
@@ -1001,16 +1001,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r115h = fptosi <4 x half> undef to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r116h = fptoui <4 x half> undef to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r117h = fptosi <4 x half> undef to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
@@ -1021,8 +1021,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r135 = fptosi <8 x float> undef to <8 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r136 = fptoui <8 x float> undef to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r137 = fptosi <8 x float> undef to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r130h = fptoui <8 x half> undef to <8 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r131h = fptosi <8 x half> undef to <8 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r132h = fptoui <8 x half> undef to <8 x i8>
@@ -1031,16 +1031,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r135h = fptosi <8 x half> undef to <8 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r136h = fptoui <8 x half> undef to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r137h = fptosi <8 x half> undef to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
@@ -1051,8 +1051,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r155 = fptosi <16 x float> undef to <16 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r156 = fptoui <16 x float> undef to <16 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r157 = fptosi <16 x float> undef to <16 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r150h = fptoui <16 x half> undef to <16 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r151h = fptosi <16 x half> undef to <16 x i1>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r152h = fptoui <16 x half> undef to <16 x i8>
@@ -1061,16 +1061,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r155h = fptosi <16 x half> undef to <16 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r156h = fptoui <16 x half> undef to <16 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r157h = fptosi <16 x half> undef to <16 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
@@ -1081,8 +1081,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r175 = sitofp <2 x i16> undef to <2 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r176 = uitofp <2 x i32> undef to <2 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r177 = sitofp <2 x i32> undef to <2 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r170h = uitofp <2 x i1> undef to <2 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r171h = sitofp <2 x i1> undef to <2 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r172h = uitofp <2 x i8> undef to <2 x half>
@@ -1091,16 +1091,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r175h = sitofp <2 x i16> undef to <2 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r176h = uitofp <2 x i32> undef to <2 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r177h = sitofp <2 x i32> undef to <2 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r190 = uitofp <4 x i1> undef to <4 x float>
@@ -1111,8 +1111,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r195 = sitofp <4 x i16> undef to <4 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r196 = uitofp <4 x i32> undef to <4 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r197 = sitofp <4 x i32> undef to <4 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r190h = uitofp <4 x i1> undef to <4 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r191h = sitofp <4 x i1> undef to <4 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r192h = uitofp <4 x i8> undef to <4 x half>
@@ -1121,16 +1121,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r195h = sitofp <4 x i16> undef to <4 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r196h = uitofp <4 x i32> undef to <4 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r197h = sitofp <4 x i32> undef to <4 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r210 = uitofp <8 x i1> undef to <8 x float>
@@ -1141,8 +1141,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r215 = sitofp <8 x i16> undef to <8 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r216 = uitofp <8 x i32> undef to <8 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r217 = sitofp <8 x i32> undef to <8 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r210h = uitofp <8 x i1> undef to <8 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r211h = sitofp <8 x i1> undef to <8 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r212h = uitofp <8 x i8> undef to <8 x half>
@@ -1151,16 +1151,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r215h = sitofp <8 x i16> undef to <8 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r216h = uitofp <8 x i32> undef to <8 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r217h = sitofp <8 x i32> undef to <8 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r230 = uitofp <16 x i1> undef to <16 x float>
@@ -1171,8 +1171,8 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r235 = sitofp <16 x i16> undef to <16 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r236 = uitofp <16 x i32> undef to <16 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r237 = sitofp <16 x i32> undef to <16 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r230h = uitofp <16 x i1> undef to <16 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r231h = sitofp <16 x i1> undef to <16 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r232h = uitofp <16 x i8> undef to <16 x half>
@@ -1181,16 +1181,16 @@ define i32 @casts() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r235h = sitofp <16 x i16> undef to <16 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r236h = uitofp <16 x i32> undef to <16 x half>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r237h = sitofp <16 x i32> undef to <16 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
@@ -1296,18 +1296,18 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %s72 = zext <4 x i8> undef to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r72 = zext <8 x i8> undef to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r73 = zext <16 x i8> undef to <16 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_0 = sext <8 x i8> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_1 = zext <8 x i8> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %rext_2 = sext <8 x i16> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_3 = zext <8 x i16> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %rext_4 = sext <4 x i16> undef to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %rext_5 = zext <4 x i16> undef to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_6 = sext <2 x i8> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_7 = zext <2 x i8> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %rext_8 = sext <2 x i16> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_9 = zext <2 x i16> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_a = sext <2 x i32> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_b = zext <2 x i32> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %rext_0 = sext <8 x i8> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %rext_1 = zext <8 x i8> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %rext_2 = sext <8 x i16> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %rext_3 = zext <8 x i16> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %rext_4 = sext <4 x i16> undef to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %rext_5 = zext <4 x i16> undef to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_6 = sext <2 x i8> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_7 = zext <2 x i8> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %rext_8 = sext <2 x i16> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_9 = zext <2 x i16> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_a = sext <2 x i32> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %rext_b = zext <2 x i32> undef to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %tv4i32i8 = trunc <4 x i32> undef to <4 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %tv4i32i16 = trunc <4 x i32> undef to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %tv4i16i8 = trunc <4 x i16> undef to <4 x i8>
@@ -1355,8 +1355,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r95 = fptosi <2 x float> undef to <2 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r96 = fptoui <2 x float> undef to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r90h = fptoui <2 x half> undef to <2 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r91h = fptosi <2 x half> undef to <2 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r92h = fptoui <2 x half> undef to <2 x i8>
@@ -1365,16 +1365,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r95h = fptosi <2 x half> undef to <2 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r96h = fptoui <2 x half> undef to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r97h = fptosi <2 x half> undef to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
@@ -1385,8 +1385,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r115 = fptosi <4 x float> undef to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r116 = fptoui <4 x float> undef to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r117 = fptosi <4 x float> undef to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r110h = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r111h = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r112h = fptoui <4 x half> undef to <4 x i8>
@@ -1395,16 +1395,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r115h = fptosi <4 x half> undef to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r116h = fptoui <4 x half> undef to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r117h = fptosi <4 x half> undef to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
@@ -1415,8 +1415,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r135 = fptosi <8 x float> undef to <8 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r136 = fptoui <8 x float> undef to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r137 = fptosi <8 x float> undef to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r130h = fptoui <8 x half> undef to <8 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r131h = fptosi <8 x half> undef to <8 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r132h = fptoui <8 x half> undef to <8 x i8>
@@ -1425,16 +1425,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r135h = fptosi <8 x half> undef to <8 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r136h = fptoui <8 x half> undef to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r137h = fptosi <8 x half> undef to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
@@ -1445,8 +1445,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r155 = fptosi <16 x float> undef to <16 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r156 = fptoui <16 x float> undef to <16 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r157 = fptosi <16 x float> undef to <16 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r150h = fptoui <16 x half> undef to <16 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r151h = fptosi <16 x half> undef to <16 x i1>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r152h = fptoui <16 x half> undef to <16 x i8>
@@ -1455,16 +1455,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r155h = fptosi <16 x half> undef to <16 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r156h = fptoui <16 x half> undef to <16 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r157h = fptosi <16 x half> undef to <16 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
@@ -1475,8 +1475,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r175 = sitofp <2 x i16> undef to <2 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r176 = uitofp <2 x i32> undef to <2 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r177 = sitofp <2 x i32> undef to <2 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r170h = uitofp <2 x i1> undef to <2 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r171h = sitofp <2 x i1> undef to <2 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r172h = uitofp <2 x i8> undef to <2 x half>
@@ -1485,16 +1485,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r175h = sitofp <2 x i16> undef to <2 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r176h = uitofp <2 x i32> undef to <2 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r177h = sitofp <2 x i32> undef to <2 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r190 = uitofp <4 x i1> undef to <4 x float>
@@ -1505,8 +1505,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r195 = sitofp <4 x i16> undef to <4 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r196 = uitofp <4 x i32> undef to <4 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r197 = sitofp <4 x i32> undef to <4 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r190h = uitofp <4 x i1> undef to <4 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r191h = sitofp <4 x i1> undef to <4 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r192h = uitofp <4 x i8> undef to <4 x half>
@@ -1515,16 +1515,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r195h = sitofp <4 x i16> undef to <4 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r196h = uitofp <4 x i32> undef to <4 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r197h = sitofp <4 x i32> undef to <4 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r210 = uitofp <8 x i1> undef to <8 x float>
@@ -1535,8 +1535,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r215 = sitofp <8 x i16> undef to <8 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r216 = uitofp <8 x i32> undef to <8 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r217 = sitofp <8 x i32> undef to <8 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r210h = uitofp <8 x i1> undef to <8 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r211h = sitofp <8 x i1> undef to <8 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r212h = uitofp <8 x i8> undef to <8 x half>
@@ -1545,16 +1545,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r215h = sitofp <8 x i16> undef to <8 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r216h = uitofp <8 x i32> undef to <8 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r217h = sitofp <8 x i32> undef to <8 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r230 = uitofp <16 x i1> undef to <16 x float>
@@ -1565,8 +1565,8 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r235 = sitofp <16 x i16> undef to <16 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r236 = uitofp <16 x i32> undef to <16 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r237 = sitofp <16 x i32> undef to <16 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r230h = uitofp <16 x i1> undef to <16 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r231h = sitofp <16 x i1> undef to <16 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r232h = uitofp <16 x i8> undef to <16 x half>
@@ -1575,16 +1575,16 @@ define i32 @casts() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r235h = sitofp <16 x i16> undef to <16 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r236h = uitofp <16 x i32> undef to <16 x half>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r237h = sitofp <16 x i32> undef to <16 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
@@ -1749,8 +1749,8 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r95 = fptosi <2 x float> undef to <2 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r96 = fptoui <2 x float> undef to <2 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r97 = fptosi <2 x float> undef to <2 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r98 = fptoui <2 x float> undef to <2 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r99 = fptosi <2 x float> undef to <2 x i64>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r90h = fptoui <2 x half> undef to <2 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r91h = fptosi <2 x half> undef to <2 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r92h = fptoui <2 x half> undef to <2 x i8>
@@ -1759,28 +1759,28 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r95h = fptosi <2 x half> undef to <2 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r96h = fptoui <2 x half> undef to <2 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r97h = fptosi <2 x half> undef to <2 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %r98h = fptoui <2 x half> undef to <2 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %r99h = fptosi <2 x half> undef to <2 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r100 = fptoui <2 x double> undef to <2 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r101 = fptosi <2 x double> undef to <2 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r102 = fptoui <2 x double> undef to <2 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r103 = fptosi <2 x double> undef to <2 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r104 = fptoui <2 x double> undef to <2 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r105 = fptosi <2 x double> undef to <2 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r106 = fptoui <2 x double> undef to <2 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r107 = fptosi <2 x double> undef to <2 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r108 = fptoui <2 x double> undef to <2 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r109 = fptosi <2 x double> undef to <2 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r110 = fptoui <4 x float> undef to <4 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r111 = fptosi <4 x float> undef to <4 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r112 = fptoui <4 x float> undef to <4 x i8>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r113 = fptosi <4 x float> undef to <4 x i8>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r114 = fptoui <4 x float> undef to <4 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r115 = fptosi <4 x float> undef to <4 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r116 = fptoui <4 x float> undef to <4 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r117 = fptosi <4 x float> undef to <4 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r118 = fptoui <4 x float> undef to <4 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r119 = fptosi <4 x float> undef to <4 x i64>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r110h = fptoui <4 x half> undef to <4 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r111h = fptosi <4 x half> undef to <4 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r112h = fptoui <4 x half> undef to <4 x i8>
@@ -1789,28 +1789,28 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r115h = fptosi <4 x half> undef to <4 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r116h = fptoui <4 x half> undef to <4 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r117h = fptosi <4 x half> undef to <4 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %r118h = fptoui <4 x half> undef to <4 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %r119h = fptosi <4 x half> undef to <4 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r120 = fptoui <4 x double> undef to <4 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r121 = fptosi <4 x double> undef to <4 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r122 = fptoui <4 x double> undef to <4 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r123 = fptosi <4 x double> undef to <4 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r124 = fptoui <4 x double> undef to <4 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r125 = fptosi <4 x double> undef to <4 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r126 = fptoui <4 x double> undef to <4 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r127 = fptosi <4 x double> undef to <4 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r128 = fptoui <4 x double> undef to <4 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r129 = fptosi <4 x double> undef to <4 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r130 = fptoui <8 x float> undef to <8 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r131 = fptosi <8 x float> undef to <8 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r132 = fptoui <8 x float> undef to <8 x i8>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r133 = fptosi <8 x float> undef to <8 x i8>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r134 = fptoui <8 x float> undef to <8 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r135 = fptosi <8 x float> undef to <8 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r136 = fptoui <8 x float> undef to <8 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r137 = fptosi <8 x float> undef to <8 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r138 = fptoui <8 x float> undef to <8 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r139 = fptosi <8 x float> undef to <8 x i64>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r130h = fptoui <8 x half> undef to <8 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r131h = fptosi <8 x half> undef to <8 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r132h = fptoui <8 x half> undef to <8 x i8>
@@ -1819,28 +1819,28 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r135h = fptosi <8 x half> undef to <8 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r136h = fptoui <8 x half> undef to <8 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r137h = fptosi <8 x half> undef to <8 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %r138h = fptoui <8 x half> undef to <8 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %r139h = fptosi <8 x half> undef to <8 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r140 = fptoui <8 x double> undef to <8 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r141 = fptosi <8 x double> undef to <8 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r142 = fptoui <8 x double> undef to <8 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r143 = fptosi <8 x double> undef to <8 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r144 = fptoui <8 x double> undef to <8 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r145 = fptosi <8 x double> undef to <8 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r146 = fptoui <8 x double> undef to <8 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %r147 = fptosi <8 x double> undef to <8 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r148 = fptoui <8 x double> undef to <8 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r149 = fptosi <8 x double> undef to <8 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 115 for instruction: %r150 = fptoui <16 x float> undef to <16 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 115 for instruction: %r151 = fptosi <16 x float> undef to <16 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r152 = fptoui <16 x float> undef to <16 x i8>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r153 = fptosi <16 x float> undef to <16 x i8>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r154 = fptoui <16 x float> undef to <16 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r155 = fptosi <16 x float> undef to <16 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r156 = fptoui <16 x float> undef to <16 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r157 = fptosi <16 x float> undef to <16 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 260 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r158 = fptoui <16 x float> undef to <16 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r159 = fptosi <16 x float> undef to <16 x i64>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r150h = fptoui <16 x half> undef to <16 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r151h = fptosi <16 x half> undef to <16 x i1>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r152h = fptoui <16 x half> undef to <16 x i8>
@@ -1849,28 +1849,28 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r155h = fptosi <16 x half> undef to <16 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r156h = fptoui <16 x half> undef to <16 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r157h = fptosi <16 x half> undef to <16 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 135 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %r158h = fptoui <16 x half> undef to <16 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %r159h = fptosi <16 x half> undef to <16 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r160 = fptoui <16 x double> undef to <16 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r161 = fptosi <16 x double> undef to <16 x i1>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r162 = fptoui <16 x double> undef to <16 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r163 = fptosi <16 x double> undef to <16 x i8>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %r164 = fptoui <16 x double> undef to <16 x i16>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %r165 = fptosi <16 x double> undef to <16 x i16>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r166 = fptoui <16 x double> undef to <16 x i32>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %r167 = fptosi <16 x double> undef to <16 x i32>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r171 = sitofp <2 x i1> undef to <2 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r168 = fptoui <16 x double> undef to <16 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r169 = fptosi <16 x double> undef to <16 x i64>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r170 = uitofp <2 x i1> undef to <2 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r171 = sitofp <2 x i1> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r172 = uitofp <2 x i8> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r173 = sitofp <2 x i8> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r174 = uitofp <2 x i16> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r175 = sitofp <2 x i16> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r176 = uitofp <2 x i32> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r177 = sitofp <2 x i32> undef to <2 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r178 = uitofp <2 x i64> undef to <2 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r179 = sitofp <2 x i64> undef to <2 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r170h = uitofp <2 x i1> undef to <2 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r171h = sitofp <2 x i1> undef to <2 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r172h = uitofp <2 x i8> undef to <2 x half>
@@ -1881,16 +1881,16 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r177h = sitofp <2 x i32> undef to <2 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r178h = uitofp <2 x i64> undef to <2 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r179h = sitofp <2 x i64> undef to <2 x half>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r180 = uitofp <2 x i1> undef to <2 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r181 = sitofp <2 x i1> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r182 = uitofp <2 x i8> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r183 = sitofp <2 x i8> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r184 = uitofp <2 x i16> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r185 = sitofp <2 x i16> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r186 = uitofp <2 x i32> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r187 = sitofp <2 x i32> undef to <2 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r188 = uitofp <2 x i64> undef to <2 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r189 = sitofp <2 x i64> undef to <2 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r190 = uitofp <4 x i1> undef to <4 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r191 = sitofp <4 x i1> undef to <4 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %r192 = uitofp <4 x i8> undef to <4 x float>
@@ -1899,8 +1899,8 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r195 = sitofp <4 x i16> undef to <4 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r196 = uitofp <4 x i32> undef to <4 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r197 = sitofp <4 x i32> undef to <4 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r198 = uitofp <4 x i64> undef to <4 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %r199 = sitofp <4 x i64> undef to <4 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r190h = uitofp <4 x i1> undef to <4 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r191h = sitofp <4 x i1> undef to <4 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r192h = uitofp <4 x i8> undef to <4 x half>
@@ -1911,16 +1911,16 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r197h = sitofp <4 x i32> undef to <4 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r198h = uitofp <4 x i64> undef to <4 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %r199h = sitofp <4 x i64> undef to <4 x half>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r200 = uitofp <4 x i1> undef to <4 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %r201 = sitofp <4 x i1> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %r202 = uitofp <4 x i8> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %r203 = sitofp <4 x i8> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r204 = uitofp <4 x i16> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r205 = sitofp <4 x i16> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r206 = uitofp <4 x i32> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %r207 = sitofp <4 x i32> undef to <4 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r208 = uitofp <4 x i64> undef to <4 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %r209 = sitofp <4 x i64> undef to <4 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r210 = uitofp <8 x i1> undef to <8 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r211 = sitofp <8 x i1> undef to <8 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %r212 = uitofp <8 x i8> undef to <8 x float>
@@ -1929,8 +1929,8 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r215 = sitofp <8 x i16> undef to <8 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r216 = uitofp <8 x i32> undef to <8 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %r217 = sitofp <8 x i32> undef to <8 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r218 = uitofp <8 x i64> undef to <8 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %r219 = sitofp <8 x i64> undef to <8 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r210h = uitofp <8 x i1> undef to <8 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r211h = sitofp <8 x i1> undef to <8 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r212h = uitofp <8 x i8> undef to <8 x half>
@@ -1941,16 +1941,16 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %r217h = sitofp <8 x i32> undef to <8 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r218h = uitofp <8 x i64> undef to <8 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %r219h = sitofp <8 x i64> undef to <8 x half>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r220 = uitofp <8 x i1> undef to <8 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %r221 = sitofp <8 x i1> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %r222 = uitofp <8 x i8> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %r223 = sitofp <8 x i8> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r224 = uitofp <8 x i16> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r225 = sitofp <8 x i16> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r226 = uitofp <8 x i16> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r227 = sitofp <8 x i16> undef to <8 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r228 = uitofp <8 x i64> undef to <8 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %r229 = sitofp <8 x i64> undef to <8 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r230 = uitofp <16 x i1> undef to <16 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r231 = sitofp <16 x i1> undef to <16 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %r232 = uitofp <16 x i8> undef to <16 x float>
@@ -1959,8 +1959,8 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %r235 = sitofp <16 x i16> undef to <16 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r236 = uitofp <16 x i32> undef to <16 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %r237 = sitofp <16 x i32> undef to <16 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r238 = uitofp <16 x i64> undef to <16 x float>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 244 for instruction: %r239 = sitofp <16 x i64> undef to <16 x float>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r230h = uitofp <16 x i1> undef to <16 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r231h = sitofp <16 x i1> undef to <16 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %r232h = uitofp <16 x i8> undef to <16 x half>
@@ -1971,16 +1971,16 @@ define i32 @casts() {
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %r237h = sitofp <16 x i32> undef to <16 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r238h = uitofp <16 x i64> undef to <16 x half>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %r239h = sitofp <16 x i64> undef to <16 x half>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r240 = uitofp <16 x i1> undef to <16 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %r241 = sitofp <16 x i1> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %r242 = uitofp <16 x i8> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %r243 = sitofp <16 x i8> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r244 = uitofp <16 x i16> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r245 = sitofp <16 x i16> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r246 = uitofp <16 x i16> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %r247 = sitofp <16 x i16> undef to <16 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
-; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r248 = uitofp <16 x i64> undef to <16 x double>
+; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %r249 = sitofp <16 x i64> undef to <16 x double>
 ; CHECK-V8R-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; CHECK-MVE-SIZE-LABEL: 'casts'
diff --git a/llvm/test/Analysis/CostModel/ARM/cast_ldst.ll b/llvm/test/Analysis/CostModel/ARM/cast_ldst.ll
index 4a2f9a25dc152..f09c12ad85e8e 100644
--- a/llvm/test/Analysis/CostModel/ARM/cast_ldst.ll
+++ b/llvm/test/Analysis/CostModel/ARM/cast_ldst.ll
@@ -175,42 +175,42 @@ define i32 @load_extends() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2816u = zext <2 x i8> %loadv2i8 to <2 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832s = sext <2 x i8> %loadv2i8 to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832u = zext <2 x i8> %loadv2i8 to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816s = sext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816u = zext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832s = sext <4 x i8> %loadv4i8 to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832u = zext <4 x i8> %loadv4i8 to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816s = sext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816u = zext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832s = sext <8 x i8> %loadv8i8 to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832u = zext <8 x i8> %loadv8i8 to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816s = sext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816u = zext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832s = sext <16 x i8> %loadv16i8 to <16 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832u = zext <16 x i8> %loadv16i8 to <16 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632s = sext <2 x i16> %loadv2i16 to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632u = zext <2 x i16> %loadv2i16 to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632s = sext <4 x i16> %loadv4i16 to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632u = zext <4 x i16> %loadv4i16 to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632s = sext <8 x i16> %loadv8i16 to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632u = zext <8 x i16> %loadv8i16 to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; CHECK-V8M-BASE-RECIP-LABEL: 'load_extends'
@@ -242,42 +242,42 @@ define i32 @load_extends() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2816u = zext <2 x i8> %loadv2i8 to <2 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832s = sext <2 x i8> %loadv2i8 to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832u = zext <2 x i8> %loadv2i8 to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816s = sext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816u = zext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832s = sext <4 x i8> %loadv4i8 to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832u = zext <4 x i8> %loadv4i8 to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816s = sext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816u = zext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832s = sext <8 x i8> %loadv8i8 to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832u = zext <8 x i8> %loadv8i8 to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816s = sext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816u = zext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832s = sext <16 x i8> %loadv16i8 to <16 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832u = zext <16 x i8> %loadv16i8 to <16 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632s = sext <2 x i16> %loadv2i16 to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632u = zext <2 x i16> %loadv2i16 to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632s = sext <4 x i16> %loadv4i16 to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632u = zext <4 x i16> %loadv4i16 to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632s = sext <8 x i16> %loadv8i16 to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632u = zext <8 x i16> %loadv8i16 to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; CHECK-V8R-RECIP-LABEL: 'load_extends'
@@ -807,7 +807,7 @@ define i32 @store_trunc() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i3264 = trunc i64 undef to i32
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2816 = trunc <2 x i16> undef to <2 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2832 = trunc <2 x i32> undef to <2 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4816 = trunc <4 x i16> undef to <4 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4832 = trunc <4 x i32> undef to <4 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4864 = trunc <4 x i64> undef to <4 x i8>
@@ -862,7 +862,7 @@ define i32 @store_trunc() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i3264 = trunc i64 undef to i32
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2816 = trunc <2 x i16> undef to <2 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2832 = trunc <2 x i32> undef to <2 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4816 = trunc <4 x i16> undef to <4 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4832 = trunc <4 x i32> undef to <4 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4864 = trunc <4 x i64> undef to <4 x i8>
@@ -1902,42 +1902,42 @@ define i32 @maskedload_extends() {
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2816u = zext <2 x i8> %loadv2i8 to <2 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832s = sext <2 x i8> %loadv2i8 to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832u = zext <2 x i8> %loadv2i8 to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816s = sext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816u = zext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832s = sext <4 x i8> %loadv4i8 to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832u = zext <4 x i8> %loadv4i8 to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816s = sext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816u = zext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832s = sext <8 x i8> %loadv8i8 to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832u = zext <8 x i8> %loadv8i8 to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816s = sext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816u = zext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832s = sext <16 x i8> %loadv16i8 to <16 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832u = zext <16 x i8> %loadv16i8 to <16 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632s = sext <2 x i16> %loadv2i16 to <2 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632u = zext <2 x i16> %loadv2i16 to <2 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632s = sext <4 x i16> %loadv4i16 to <4 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632u = zext <4 x i16> %loadv4i16 to <4 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632s = sext <8 x i16> %loadv8i16 to <8 x i32>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632u = zext <8 x i16> %loadv8i16 to <8 x i32>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; CHECK-V8M-BASE-RECIP-LABEL: 'maskedload_extends'
@@ -1954,42 +1954,42 @@ define i32 @maskedload_extends() {
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2816u = zext <2 x i8> %loadv2i8 to <2 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832s = sext <2 x i8> %loadv2i8 to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2832u = zext <2 x i8> %loadv2i8 to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864s = sext <2 x i8> %loadv2i8 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2864u = zext <2 x i8> %loadv2i8 to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816s = sext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4816u = zext <4 x i8> %loadv4i8 to <4 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832s = sext <4 x i8> %loadv4i8 to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4832u = zext <4 x i8> %loadv4i8 to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864s = sext <4 x i8> %loadv4i8 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4864u = zext <4 x i8> %loadv4i8 to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816s = sext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8816u = zext <8 x i8> %loadv8i8 to <8 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832s = sext <8 x i8> %loadv8i8 to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8832u = zext <8 x i8> %loadv8i8 to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864s = sext <8 x i8> %loadv8i8 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8864u = zext <8 x i8> %loadv8i8 to <8 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816s = sext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16816u = zext <16 x i8> %loadv16i8 to <16 x i16>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832s = sext <16 x i8> %loadv16i8 to <16 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16832u = zext <16 x i8> %loadv16i8 to <16 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864s = sext <16 x i8> %loadv16i8 to <16 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16864u = zext <16 x i8> %loadv16i8 to <16 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632s = sext <2 x i16> %loadv2i16 to <2 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v21632u = zext <2 x i16> %loadv2i16 to <2 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664s = sext <2 x i16> %loadv2i16 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v21664u = zext <2 x i16> %loadv2i16 to <2 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632s = sext <4 x i16> %loadv4i16 to <4 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v41632u = zext <4 x i16> %loadv4i16 to <4 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664s = sext <4 x i16> %loadv4i16 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v41664u = zext <4 x i16> %loadv4i16 to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632s = sext <8 x i16> %loadv8i16 to <8 x i32>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v81632u = zext <8 x i16> %loadv8i16 to <8 x i32>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664s = sext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v81664u = zext <8 x i16> %loadv8i16 to <8 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264s = sext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v23264u = zext <2 x i32> %loadv2i32 to <2 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264s = sext <4 x i32> %loadv4i32 to <4 x i64>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v43264u = zext <4 x i32> %loadv4i32 to <4 x i64>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
 ;
 ; CHECK-V8R-RECIP-LABEL: 'maskedload_extends'
@@ -2398,7 +2398,7 @@ define i32 @maskedstore_trunc() {
 ; CHECK-V8M-MAIN-RECIP-LABEL: 'maskedstore_trunc'
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2816 = trunc <2 x i16> undef to <2 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2832 = trunc <2 x i32> undef to <2 x i8>
-; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
+; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4816 = trunc <4 x i16> undef to <4 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4832 = trunc <4 x i32> undef to <4 x i8>
 ; CHECK-V8M-MAIN-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4864 = trunc <4 x i64> undef to <4 x i8>
@@ -2441,7 +2441,7 @@ define i32 @maskedstore_trunc() {
 ; CHECK-V8M-BASE-RECIP-LABEL: 'maskedstore_trunc'
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2816 = trunc <2 x i16> undef to <2 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2832 = trunc <2 x i32> undef to <2 x i8>
-; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
+; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2864 = trunc <2 x i64> undef to <2 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4816 = trunc <4 x i16> undef to <4 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4832 = trunc <4 x i32> undef to <4 x i8>
 ; CHECK-V8M-BASE-RECIP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4864 = trunc <4 x i64> undef to <4 x i8>
diff --git a/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll b/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
index aff7b19a9c87a..5b43e33aea9f4 100644
--- a/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
+++ b/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
@@ -24,86 +24,86 @@ define void @casts() {
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 240 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 406 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 378 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 314 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 276 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 276 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 276 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 404 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 376 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 626 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 548 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 548 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 548 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 496 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1360 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1304 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 922 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 844 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 844 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 844 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1352 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1296 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1834 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1676 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1676 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1672 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1568 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4912 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4800 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 3018 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2762 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2860 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2762 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2860 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2760 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2856 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2752 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4880 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4768 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 216 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 294 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 266 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 266 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 292 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 530 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 452 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 452 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 452 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 400 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 912 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 856 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 602 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 652 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 602 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 652 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 602 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 652 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 600 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 904 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 848 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1450 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1194 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1292 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1194 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1292 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1192 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1288 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1184 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 3120 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 3008 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2250 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1994 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2092 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1994 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2092 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1992 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2088 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1984 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 3088 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2976 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; CHECK-MVEFP-LABEL: 'casts'
@@ -135,18 +135,18 @@ define void @casts() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -155,18 +155,18 @@ define void @casts() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 278 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 234 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 324 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 304 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 166 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 186 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 212 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -175,18 +175,18 @@ define void @casts() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1148 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1104 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 762 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 684 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 684 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 684 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 648 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1192 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1152 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 656 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 570 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 458 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 492 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 458 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 492 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 458 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 492 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 456 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 744 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 704 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -195,18 +195,18 @@ define void @casts() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4488 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4400 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2698 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2474 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2540 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2474 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2540 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2472 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2536 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2464 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4560 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4480 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2696 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2608 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1930 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1706 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1772 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1706 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1772 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1704 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1768 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1696 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2768 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2688 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
@@ -334,46 +334,46 @@ define void @fp16() {
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 240 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 406 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 378 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 480 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 550 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1362 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1306 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1250 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 994 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1092 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 994 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1092 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 992 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1680 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4920 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4808 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 216 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 294 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 266 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 432 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 304 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 354 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 304 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 354 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 304 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 454 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 914 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 858 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1058 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 802 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 900 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 802 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 900 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 800 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1296 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1192 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 3128 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 3016 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; CHECK-MVEFP-LABEL: 'fp16'
@@ -393,10 +393,10 @@ define void @fp16() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
@@ -405,8 +405,8 @@ define void @fp16() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 278 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 166 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
@@ -415,8 +415,8 @@ define void @fp16() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1112 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1102 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 664 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 654 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
@@ -425,8 +425,8 @@ define void @fp16() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4484 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4400 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2692 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2608 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
diff --git a/llvm/test/Analysis/CostModel/ARM/mve-vecreduce-add.ll b/llvm/test/Analysis/CostModel/ARM/mve-vecreduce-add.ll
index 8edc2e6fc1568..e9a42e7179ed9 100644
--- a/llvm/test/Analysis/CostModel/ARM/mve-vecreduce-add.ll
+++ b/llvm/test/Analysis/CostModel/ARM/mve-vecreduce-add.ll
@@ -221,9 +221,9 @@ define void @add_i32() {
 
 define void @add_i64() {
 ; CHECK-LABEL: 'add_i64'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a0za = zext <1 x i8> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a0za = zext <1 x i8> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a0z = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a0za)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a0sa = sext <1 x i8> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a0sa = sext <1 x i8> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a0s = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a0sa)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a1za = zext <2 x i8> undef to <2 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %a1z = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> %a1za)
@@ -241,9 +241,9 @@ define void @add_i64() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 808 for instruction: %a4z = call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %a4za)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1322 for instruction: %a4sa = sext <16 x i8> undef to <16 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 808 for instruction: %a4s = call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %a4sa)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a5za = zext <1 x i16> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a5za = zext <1 x i16> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a5z = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a5za)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %a5sa = sext <1 x i16> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %a5sa = sext <1 x i16> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a5s = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a5sa)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a6za = zext <2 x i16> undef to <2 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %a6z = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> %a6za)
@@ -261,9 +261,9 @@ define void @add_i64() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 808 for instruction: %a9z = call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %a9za)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %a9sa = sext <16 x i16> undef to <16 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 808 for instruction: %a9s = call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %a9sa)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a10za = zext <1 x i32> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a10za = zext <1 x i32> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a10z = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a10za)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a10sa = sext <1 x i32> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a10sa = sext <1 x i32> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a10s = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a10sa)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a11za = zext <2 x i32> undef to <2 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %a11z = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> %a11za)
@@ -759,12 +759,12 @@ define void @mla_i32() {
 
 define void @mla_i64() {
 ; CHECK-LABEL: 'mla_i64'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a0za = zext <1 x i8> undef to <1 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a0zb = zext <1 x i8> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a0za = zext <1 x i8> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a0zb = zext <1 x i8> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a0zm = mul <1 x i64> %a0za, %a0zb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a0z = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a0zm)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a0sa = sext <1 x i8> undef to <1 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a0sb = sext <1 x i8> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a0sa = sext <1 x i8> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a0sb = sext <1 x i8> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a0sm = mul <1 x i64> %a0sa, %a0sb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a0s = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a0sm)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a1za = zext <2 x i8> undef to <2 x i64>
@@ -799,12 +799,12 @@ define void @mla_i64() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1322 for instruction: %a4sb = sext <16 x i8> undef to <16 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 416 for instruction: %a4sm = mul <16 x i64> %a4sa, %a4sb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 808 for instruction: %a4s = call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %a4sm)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a5za = zext <1 x i16> undef to <1 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a5zb = zext <1 x i16> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a5za = zext <1 x i16> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a5zb = zext <1 x i16> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a5zm = mul <1 x i64> %a5za, %a5zb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a5z = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a5zm)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %a5sa = sext <1 x i16> undef to <1 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %a5sb = sext <1 x i16> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %a5sa = sext <1 x i16> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %a5sb = sext <1 x i16> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a5sm = mul <1 x i64> %a5sa, %a5sb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a5s = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a5sm)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a6za = zext <2 x i16> undef to <2 x i64>
@@ -839,12 +839,12 @@ define void @mla_i64() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %a9sb = sext <16 x i16> undef to <16 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 416 for instruction: %a9sm = mul <16 x i64> %a9sa, %a9sb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 808 for instruction: %a9s = call i64 @llvm.vector.reduce.add.v16i64(<16 x i64> %a9sm)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a10za = zext <1 x i32> undef to <1 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a10zb = zext <1 x i32> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a10za = zext <1 x i32> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a10zb = zext <1 x i32> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a10zm = mul <1 x i64> %a10za, %a10zb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a10z = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a10zm)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a10sa = sext <1 x i32> undef to <1 x i64>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %a10sb = sext <1 x i32> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a10sa = sext <1 x i32> undef to <1 x i64>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %a10sb = sext <1 x i32> undef to <1 x i64>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a10sm = mul <1 x i64> %a10sa, %a10sb
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %a10s = call i64 @llvm.vector.reduce.add.v1i64(<1 x i64> %a10sm)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %a11za = zext <2 x i32> undef to <2 x i64>
diff --git a/llvm/test/Analysis/CostModel/X86/arith-fp.ll b/llvm/test/Analysis/CostModel/X86/arith-fp.ll
index 90871e3a3831c..90f1d5f32bd26 100644
--- a/llvm/test/Analysis/CostModel/X86/arith-fp.ll
+++ b/llvm/test/Analysis/CostModel/X86/arith-fp.ll
@@ -1296,68 +1296,68 @@ define i32 @llrint(i32 %arg) {
 ;
 ; SSE2-LABEL: 'llrint'
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; SSE42-LABEL: 'llrint'
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX-LABEL: 'llrint'
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'llrint'
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; SLM-LABEL: 'llrint'
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; GLM-LABEL: 'llrint'
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = call i64 @llvm.llrint.i64.f32(float undef)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4F32 = call <4 x i64> @llvm.llrint.v4i64.v4f32(<4 x float> undef)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8F32 = call <8 x i64> @llvm.llrint.v8i64.v8f32(<8 x float> undef)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16F32 = call <16 x i64> @llvm.llrint.v16i64.v16f32(<16 x float> undef)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F64 = call i64 @llvm.llrint.i64.f64(double undef)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
-; GLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2F64 = call <2 x i64> @llvm.llrint.v2i64.v2f64(<2 x double> undef)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4F64 = call <4 x i64> @llvm.llrint.v4i64.v4f64(<4 x double> undef)
+; GLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8F64 = call <8 x i64> @llvm.llrint.v8i64.v8f64(<8 x double> undef)
 ; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
   %F32 = call i64 @llvm.llrint.i64.f32(float undef)
diff --git a/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll b/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
index 55b80350f595e..e4ce4a9a3d862 100644
--- a/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
+++ b/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
@@ -32,26 +32,26 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -60,8 +60,8 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
@@ -70,8 +70,8 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -80,8 +80,8 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
@@ -90,8 +90,8 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -100,8 +100,8 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 156 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
@@ -110,8 +110,8 @@ define void @casts() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 336 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SSE42-LABEL: 'casts'
@@ -135,26 +135,26 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -163,8 +163,8 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
@@ -173,8 +173,8 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -183,8 +183,8 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
@@ -193,8 +193,8 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -203,8 +203,8 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
@@ -213,8 +213,8 @@ define void @casts() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX1-LABEL: 'casts'
@@ -238,26 +238,26 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -266,8 +266,8 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
@@ -276,8 +276,8 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -286,8 +286,8 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
@@ -296,8 +296,8 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -306,8 +306,8 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
@@ -316,8 +316,8 @@ define void @casts() {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX2-LABEL: 'casts'
@@ -341,26 +341,26 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -369,8 +369,8 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
@@ -379,8 +379,8 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -389,8 +389,8 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
@@ -399,8 +399,8 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -409,8 +409,8 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
@@ -419,8 +419,8 @@ define void @casts() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512F-LABEL: 'casts'
@@ -444,26 +444,26 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -472,8 +472,8 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
@@ -482,8 +482,8 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -492,8 +492,8 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
@@ -502,8 +502,8 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -512,8 +512,8 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 69 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
@@ -522,8 +522,8 @@ define void @casts() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512DQ-LABEL: 'casts'
@@ -547,8 +547,8 @@ define void @casts() {
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
@@ -557,8 +557,8 @@ define void @casts() {
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
@@ -650,26 +650,26 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
@@ -678,8 +678,8 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
@@ -688,8 +688,8 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
@@ -698,8 +698,8 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
@@ -708,8 +708,8 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
@@ -718,8 +718,8 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
@@ -728,8 +728,8 @@ define void @casts() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
@@ -877,8 +877,8 @@ define void @fp16() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
@@ -887,8 +887,8 @@ define void @fp16() {
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 197 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 195 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 153 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 190 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
@@ -930,8 +930,8 @@ define void @fp16() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
@@ -940,8 +940,8 @@ define void @fp16() {
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 197 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 195 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 153 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 190 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
@@ -985,22 +985,22 @@ define void @fp16() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 203 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 212 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 179 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 212 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 179 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 202 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 169 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 206 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 171 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 153 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
@@ -1016,16 +1016,16 @@ define void @fp16() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
@@ -1034,28 +1034,28 @@ define void @fp16() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 186 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 216 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 206 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 206 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512DQ-LABEL: 'fp16'
@@ -1069,8 +1069,8 @@ define void @fp16() {
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
@@ -1087,28 +1087,28 @@ define void @fp16() {
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 186 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 216 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 206 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 206 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'fp16'
@@ -1142,8 +1142,8 @@ define void @fp16() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
@@ -1152,8 +1152,8 @@ define void @fp16() {
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 71 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 199 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 197 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 187 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 153 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
diff --git a/llvm/test/Analysis/CostModel/X86/fptosi.ll b/llvm/test/Analysis/CostModel/X86/fptosi.ll
index 44a4401f3f713..e840ff2bae48b 100644
--- a/llvm/test/Analysis/CostModel/X86/fptosi.ll
+++ b/llvm/test/Analysis/CostModel/X86/fptosi.ll
@@ -13,30 +13,30 @@
 define i32 @fptosi_double_i64(i32 %arg) {
 ; SSE2-LABEL: 'fptosi_double_i64'
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptosi double undef to i64
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; SSE42-LABEL: 'fptosi_double_i64'
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi double undef to i64
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX-LABEL: 'fptosi_double_i64'
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi double undef to i64
-; AVX-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512F-LABEL: 'fptosi_double_i64'
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi double undef to i64
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512DQ-LABEL: 'fptosi_double_i64'
@@ -48,9 +48,9 @@ define i32 @fptosi_double_i64(i32 %arg) {
 ;
 ; SLM-LABEL: 'fptosi_double_i64'
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi double undef to i64
-; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x double> undef to <2 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = fptosi <4 x double> undef to <4 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = fptosi <8 x double> undef to <8 x i64>
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
   %I64 = fptosi double undef to i64
@@ -199,34 +199,34 @@ define i32 @fptosi_double_i8(i32 %arg) {
 define i32 @fptosi_float_i64(i32 %arg) {
 ; SSE2-LABEL: 'fptosi_float_i64'
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptosi float undef to i64
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; SSE42-LABEL: 'fptosi_float_i64'
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi float undef to i64
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX-LABEL: 'fptosi_float_i64'
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi float undef to i64
-; AVX-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512F-LABEL: 'fptosi_float_i64'
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi float undef to i64
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512DQ-LABEL: 'fptosi_float_i64'
@@ -239,10 +239,10 @@ define i32 @fptosi_float_i64(i32 %arg) {
 ;
 ; SLM-LABEL: 'fptosi_float_i64'
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptosi float undef to i64
-; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptosi <2 x float> undef to <2 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = fptosi <4 x float> undef to <4 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V8I64 = fptosi <8 x float> undef to <8 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V16I64 = fptosi <16 x float> undef to <16 x i64>
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
   %I64 = fptosi float undef to i64
diff --git a/llvm/test/Analysis/CostModel/X86/fptoui.ll b/llvm/test/Analysis/CostModel/X86/fptoui.ll
index c31151750540a..683bd03d3d6fc 100644
--- a/llvm/test/Analysis/CostModel/X86/fptoui.ll
+++ b/llvm/test/Analysis/CostModel/X86/fptoui.ll
@@ -13,37 +13,37 @@
 define i32 @fptoui_double_i64(i32 %arg) {
 ; SSE2-LABEL: 'fptoui_double_i64'
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %I64 = fptoui double undef to i64
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 136 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; SSE42-LABEL: 'fptoui_double_i64'
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui double undef to i64
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX1-LABEL: 'fptoui_double_i64'
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui double undef to i64
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX2-LABEL: 'fptoui_double_i64'
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %I64 = fptoui double undef to i64
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512F-LABEL: 'fptoui_double_i64'
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptoui double undef to i64
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512DQ-LABEL: 'fptoui_double_i64'
@@ -55,9 +55,9 @@ define i32 @fptoui_double_i64(i32 %arg) {
 ;
 ; SLM-LABEL: 'fptoui_double_i64'
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui double undef to i64
-; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = fptoui <2 x double> undef to <2 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V4I64 = fptoui <4 x double> undef to <4 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V8I64 = fptoui <8 x double> undef to <8 x i64>
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
   %I64 = fptoui double undef to i64
@@ -206,42 +206,42 @@ define i32 @fptoui_double_i8(i32 %arg) {
 define i32 @fptoui_float_i64(i32 %arg) {
 ; SSE2-LABEL: 'fptoui_float_i64'
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui float undef to i64
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; SSE42-LABEL: 'fptoui_float_i64'
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui float undef to i64
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX1-LABEL: 'fptoui_float_i64'
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui float undef to i64
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX2-LABEL: 'fptoui_float_i64'
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %I64 = fptoui float undef to i64
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512F-LABEL: 'fptoui_float_i64'
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = fptoui float undef to i64
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512DQ-LABEL: 'fptoui_float_i64'
@@ -254,10 +254,10 @@ define i32 @fptoui_float_i64(i32 %arg) {
 ;
 ; SLM-LABEL: 'fptoui_float_i64'
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %I64 = fptoui float undef to i64
-; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
-; SLM-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V2I64 = fptoui <2 x float> undef to <2 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V4I64 = fptoui <4 x float> undef to <4 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %V8I64 = fptoui <8 x float> undef to <8 x i64>
+; SLM-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V16I64 = fptoui <16 x float> undef to <16 x i64>
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
   %I64 = fptoui float undef to i64
diff --git a/llvm/test/Analysis/CostModel/X86/shuffle-replication-i16.ll b/llvm/test/Analysis/CostModel/X86/shuffle-replication-i16.ll
index aecb60e441374..3318b0e4e8e8e 100644
--- a/llvm/test/Analysis/CostModel/X86/shuffle-replication-i16.ll
+++ b/llvm/test/Analysis/CostModel/X86/shuffle-replication-i16.ll
@@ -197,10 +197,10 @@ define void @replication_i16_stride3() nounwind "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %vf1 = shufflevector <1 x i16> undef, <1 x i16> poison, <3 x i32> zeroinitializer
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %vf2 = shufflevector <2 x i16> undef, <2 x i16> poison, <6 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <12 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 456 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512BWVEC512-LABEL: 'replication_i16_stride3'
@@ -418,11 +418,11 @@ define void @replication_i16_stride5() nounwind "min-legal-vector-width"="256" {
 ; AVX512FVEC256-LABEL: 'replication_i16_stride5'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %vf1 = shufflevector <1 x i16> undef, <1 x i16> poison, <5 x i32> zeroinitializer
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %vf2 = shufflevector <2 x i16> undef, <2 x i16> poison, <10 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <20 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 91 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <40 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <80 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 366 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <160 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 732 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <320 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <20 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <40 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 188 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <80 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 376 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <160 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 752 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <320 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512BWVEC512-LABEL: 'replication_i16_stride5'
@@ -529,11 +529,11 @@ define void @replication_i16_stride6() nounwind "min-legal-vector-width"="256" {
 ; AVX512FVEC256-LABEL: 'replication_i16_stride6'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %vf1 = shufflevector <1 x i16> undef, <1 x i16> poison, <6 x i32> zeroinitializer
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %vf2 = shufflevector <2 x i16> undef, <2 x i16> poison, <12 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 109 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <384 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 225 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 450 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 900 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <384 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512BWVEC512-LABEL: 'replication_i16_stride6'
@@ -639,12 +639,12 @@ define void @replication_i16_stride7() nounwind "min-legal-vector-width"="256" {
 ;
 ; AVX512FVEC256-LABEL: 'replication_i16_stride7'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %vf1 = shufflevector <1 x i16> undef, <1 x i16> poison, <7 x i32> zeroinitializer
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %vf2 = shufflevector <2 x i16> undef, <2 x i16> poison, <14 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <28 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 127 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <56 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 255 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <112 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 510 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <224 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1020 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <448 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %vf2 = shufflevector <2 x i16> undef, <2 x i16> poison, <14 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %vf4 = shufflevector <4 x i16> undef, <4 x i16> poison, <28 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %vf8 = shufflevector <8 x i16> undef, <8 x i16> poison, <56 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 262 for instruction: %vf16 = shufflevector <16 x i16> undef, <16 x i16> poison, <112 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 524 for instruction: %vf32 = shufflevector <32 x i16> undef, <32 x i16> poison, <224 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1048 for instruction: %vf64 = shufflevector <64 x i16> undef, <64 x i16> poison, <448 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512BWVEC512-LABEL: 'replication_i16_stride7'
diff --git a/llvm/test/Analysis/CostModel/X86/shuffle-replication-i8.ll b/llvm/test/Analysis/CostModel/X86/shuffle-replication-i8.ll
index e82ea6871c7ec..5e5bc7b3e8ad2 100644
--- a/llvm/test/Analysis/CostModel/X86/shuffle-replication-i8.ll
+++ b/llvm/test/Analysis/CostModel/X86/shuffle-replication-i8.ll
@@ -196,11 +196,11 @@ define void @replication_i8_stride3() nounwind "min-legal-vector-width"="256" {
 ; AVX512FVEC256-LABEL: 'replication_i8_stride3'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %vf2 = shufflevector <2 x i8> undef, <2 x i8> poison, <6 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <12 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 107 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 215 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 430 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 860 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <384 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <384 x i32> <i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512VBMIVEC512-LABEL: 'replication_i8_stride3'
@@ -417,12 +417,12 @@ define void @replication_i8_stride5() nounwind "min-legal-vector-width"="256" {
 ;
 ; AVX512FVEC256-LABEL: 'replication_i8_stride5'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %vf2 = shufflevector <2 x i8> undef, <2 x i8> poison, <10 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <20 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <40 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 177 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <80 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 355 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <160 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 710 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <320 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1420 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <640 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127, i32 127, i32 127>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <20 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <40 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <80 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <160 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <320 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1480 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <640 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127, i32 127, i32 127>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512VBMIVEC512-LABEL: 'replication_i8_stride5'
@@ -528,12 +528,12 @@ define void @replication_i8_stride6() nounwind "min-legal-vector-width"="256" {
 ;
 ; AVX512FVEC256-LABEL: 'replication_i8_stride6'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %vf2 = shufflevector <2 x i8> undef, <2 x i8> poison, <12 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 212 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 425 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 850 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <384 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1700 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <768 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <24 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <48 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 221 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <96 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 443 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <192 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 886 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <384 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1772 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <768 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512VBMIVEC512-LABEL: 'replication_i8_stride6'
@@ -639,12 +639,12 @@ define void @replication_i8_stride7() nounwind "min-legal-vector-width"="256" {
 ;
 ; AVX512FVEC256-LABEL: 'replication_i8_stride7'
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %vf2 = shufflevector <2 x i8> undef, <2 x i8> poison, <14 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <28 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 125 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <56 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 247 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <112 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 495 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <224 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 990 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <448 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1980 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <896 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 65, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 66, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 67, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 68, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 69, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 70, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 71, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 72, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 73, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 74, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 75, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 76, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 77, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 78, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 79, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 80, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 81, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 82, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 83, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 84, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 85, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 86, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 87, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 88, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 89, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 90, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 91, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 92, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 93, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 94, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 95, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 96, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 97, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 98, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 99, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 100, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 101, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 102, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 103, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 104, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 105, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 106, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 107, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 108, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 109, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 110, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 111, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 112, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 113, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 114, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 115, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 116, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 117, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 118, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 119, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 120, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 121, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 122, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 123, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 124, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 125, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 126, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %vf4 = shufflevector <4 x i8> undef, <4 x i8> poison, <28 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %vf8 = shufflevector <8 x i8> undef, <8 x i8> poison, <56 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 258 for instruction: %vf16 = shufflevector <16 x i8> undef, <16 x i8> poison, <112 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 516 for instruction: %vf32 = shufflevector <32 x i8> undef, <32 x i8> poison, <224 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1032 for instruction: %vf64 = shufflevector <64 x i8> undef, <64 x i8> poison, <448 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2064 for instruction: %vf128 = shufflevector <128 x i8> undef, <128 x i8> poison, <896 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 3, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 7, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 9, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 10, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 11, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 12, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 13, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 14, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 15, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 17, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 18, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 19, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 20, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 21, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 22, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 23, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 24, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 25, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 26, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 27, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 29, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 30, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 33, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 34, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 35, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 36, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 37, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 38, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 39, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 40, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 41, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 42, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 43, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 44, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 45, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 46, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 47, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 48, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 49, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 50, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 51, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 52, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 53, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 54, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 55, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 56, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 57, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 58, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 59, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 60, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 61, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 62, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 63, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 64, i32 65, i32 65, i32 65, i32 65, i32 65, i32 65, i32 65, i32 66, i32 66, i32 66, i32 66, i32 66, i32 66, i32 66, i32 67, i32 67, i32 67, i32 67, i32 67, i32 67, i32 67, i32 68, i32 68, i32 68, i32 68, i32 68, i32 68, i32 68, i32 69, i32 69, i32 69, i32 69, i32 69, i32 69, i32 69, i32 70, i32 70, i32 70, i32 70, i32 70, i32 70, i32 70, i32 71, i32 71, i32 71, i32 71, i32 71, i32 71, i32 71, i32 72, i32 72, i32 72, i32 72, i32 72, i32 72, i32 72, i32 73, i32 73, i32 73, i32 73, i32 73, i32 73, i32 73, i32 74, i32 74, i32 74, i32 74, i32 74, i32 74, i32 74, i32 75, i32 75, i32 75, i32 75, i32 75, i32 75, i32 75, i32 76, i32 76, i32 76, i32 76, i32 76, i32 76, i32 76, i32 77, i32 77, i32 77, i32 77, i32 77, i32 77, i32 77, i32 78, i32 78, i32 78, i32 78, i32 78, i32 78, i32 78, i32 79, i32 79, i32 79, i32 79, i32 79, i32 79, i32 79, i32 80, i32 80, i32 80, i32 80, i32 80, i32 80, i32 80, i32 81, i32 81, i32 81, i32 81, i32 81, i32 81, i32 81, i32 82, i32 82, i32 82, i32 82, i32 82, i32 82, i32 82, i32 83, i32 83, i32 83, i32 83, i32 83, i32 83, i32 83, i32 84, i32 84, i32 84, i32 84, i32 84, i32 84, i32 84, i32 85, i32 85, i32 85, i32 85, i32 85, i32 85, i32 85, i32 86, i32 86, i32 86, i32 86, i32 86, i32 86, i32 86, i32 87, i32 87, i32 87, i32 87, i32 87, i32 87, i32 87, i32 88, i32 88, i32 88, i32 88, i32 88, i32 88, i32 88, i32 89, i32 89, i32 89, i32 89, i32 89, i32 89, i32 89, i32 90, i32 90, i32 90, i32 90, i32 90, i32 90, i32 90, i32 91, i32 91, i32 91, i32 91, i32 91, i32 91, i32 91, i32 92, i32 92, i32 92, i32 92, i32 92, i32 92, i32 92, i32 93, i32 93, i32 93, i32 93, i32 93, i32 93, i32 93, i32 94, i32 94, i32 94, i32 94, i32 94, i32 94, i32 94, i32 95, i32 95, i32 95, i32 95, i32 95, i32 95, i32 95, i32 96, i32 96, i32 96, i32 96, i32 96, i32 96, i32 96, i32 97, i32 97, i32 97, i32 97, i32 97, i32 97, i32 97, i32 98, i32 98, i32 98, i32 98, i32 98, i32 98, i32 98, i32 99, i32 99, i32 99, i32 99, i32 99, i32 99, i32 99, i32 100, i32 100, i32 100, i32 100, i32 100, i32 100, i32 100, i32 101, i32 101, i32 101, i32 101, i32 101, i32 101, i32 101, i32 102, i32 102, i32 102, i32 102, i32 102, i32 102, i32 102, i32 103, i32 103, i32 103, i32 103, i32 103, i32 103, i32 103, i32 104, i32 104, i32 104, i32 104, i32 104, i32 104, i32 104, i32 105, i32 105, i32 105, i32 105, i32 105, i32 105, i32 105, i32 106, i32 106, i32 106, i32 106, i32 106, i32 106, i32 106, i32 107, i32 107, i32 107, i32 107, i32 107, i32 107, i32 107, i32 108, i32 108, i32 108, i32 108, i32 108, i32 108, i32 108, i32 109, i32 109, i32 109, i32 109, i32 109, i32 109, i32 109, i32 110, i32 110, i32 110, i32 110, i32 110, i32 110, i32 110, i32 111, i32 111, i32 111, i32 111, i32 111, i32 111, i32 111, i32 112, i32 112, i32 112, i32 112, i32 112, i32 112, i32 112, i32 113, i32 113, i32 113, i32 113, i32 113, i32 113, i32 113, i32 114, i32 114, i32 114, i32 114, i32 114, i32 114, i32 114, i32 115, i32 115, i32 115, i32 115, i32 115, i32 115, i32 115, i32 116, i32 116, i32 116, i32 116, i32 116, i32 116, i32 116, i32 117, i32 117, i32 117, i32 117, i32 117, i32 117, i32 117, i32 118, i32 118, i32 118, i32 118, i32 118, i32 118, i32 118, i32 119, i32 119, i32 119, i32 119, i32 119, i32 119, i32 119, i32 120, i32 120, i32 120, i32 120, i32 120, i32 120, i32 120, i32 121, i32 121, i32 121, i32 121, i32 121, i32 121, i32 121, i32 122, i32 122, i32 122, i32 122, i32 122, i32 122, i32 122, i32 123, i32 123, i32 123, i32 123, i32 123, i32 123, i32 123, i32 124, i32 124, i32 124, i32 124, i32 124, i32 124, i32 124, i32 125, i32 125, i32 125, i32 125, i32 125, i32 125, i32 125, i32 126, i32 126, i32 126, i32 126, i32 126, i32 126, i32 126, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127, i32 127>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512VBMIVEC512-LABEL: 'replication_i8_stride7'
diff --git a/llvm/test/Analysis/CostModel/X86/sitofp.ll b/llvm/test/Analysis/CostModel/X86/sitofp.ll
index fff1ffe856571..ac0a11d5b5e1e 100644
--- a/llvm/test/Analysis/CostModel/X86/sitofp.ll
+++ b/llvm/test/Analysis/CostModel/X86/sitofp.ll
@@ -157,15 +157,15 @@ define i32 @sitofp_i64_double() {
 ; AVX-LABEL: 'sitofp_i64_double'
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i64_f64 = sitofp i64 undef to double
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v2i64_v2f64 = sitofp <2 x i64> undef to <2 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %cvt_v4i64_v4f64 = sitofp <4 x i64> undef to <4 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %cvt_v8i64_v8f64 = sitofp <8 x i64> undef to <8 x double>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v4i64_v4f64 = sitofp <4 x i64> undef to <4 x double>
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %cvt_v8i64_v8f64 = sitofp <8 x i64> undef to <8 x double>
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512F-LABEL: 'sitofp_i64_double'
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i64_f64 = sitofp i64 undef to double
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v2i64_v2f64 = sitofp <2 x i64> undef to <2 x double>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %cvt_v4i64_v4f64 = sitofp <4 x i64> undef to <4 x double>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %cvt_v8i64_v8f64 = sitofp <8 x i64> undef to <8 x double>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v4i64_v4f64 = sitofp <4 x i64> undef to <4 x double>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %cvt_v8i64_v8f64 = sitofp <8 x i64> undef to <8 x double>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512DQ-LABEL: 'sitofp_i64_double'
@@ -358,8 +358,8 @@ define i32 @sitofp_i64_float() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i64_f32 = sitofp i64 undef to float
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v2i64_v2f32 = sitofp <2 x i64> undef to <2 x float>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i64_v4f32 = sitofp <4 x i64> undef to <4 x float>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %cvt_v8i64_v8f32 = sitofp <8 x i64> undef to <8 x float>
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %cvt_v16i64_v16f32 = sitofp <16 x i64> undef to <16 x float>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %cvt_v8i64_v8f32 = sitofp <8 x i64> undef to <8 x float>
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %cvt_v16i64_v16f32 = sitofp <16 x i64> undef to <16 x float>
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512DQ-LABEL: 'sitofp_i64_float'
diff --git a/llvm/test/Analysis/CostModel/X86/trunc.ll b/llvm/test/Analysis/CostModel/X86/trunc.ll
index 2d062742ef685..300a36c332c3b 100644
--- a/llvm/test/Analysis/CostModel/X86/trunc.ll
+++ b/llvm/test/Analysis/CostModel/X86/trunc.ll
@@ -315,29 +315,29 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 518 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 888 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1036 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 416 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
@@ -349,27 +349,27 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i32 = trunc <10 x i32> undef to <10 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i16>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -382,29 +382,29 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 518 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 336 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 888 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1036 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 672 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
@@ -416,27 +416,27 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i32 = trunc <10 x i32> undef to <10 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i16>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -516,29 +516,29 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 518 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 336 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 888 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1036 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 672 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
@@ -550,27 +550,27 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i32 = trunc <10 x i32> undef to <10 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i16>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -650,29 +650,29 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 518 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 336 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 888 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1036 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 672 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
@@ -684,27 +684,27 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i32 = trunc <10 x i32> undef to <10 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i16>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -784,29 +784,29 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 518 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 336 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 888 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1036 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 672 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
@@ -818,27 +818,27 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i32 = trunc <10 x i32> undef to <10 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i16>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -851,29 +851,29 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 93 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 111 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 185 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 222 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 370 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 444 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 518 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 740 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 888 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1036 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 416 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i16
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i16>
@@ -885,27 +885,27 @@ define i32 @trunc_vXi16() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = trunc <8 x i32> undef to <8 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i32 = trunc <10 x i32> undef to <10 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 123 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i16>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i16>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -1295,29 +1295,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 164 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 328 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 656 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1312 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i8
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>
@@ -1328,29 +1328,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 207 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 416 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i8
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i8>
@@ -1363,27 +1363,27 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i8>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -1397,29 +1397,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 528 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1056 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i8
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>
@@ -1430,29 +1430,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 207 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 288 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i8
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i8>
@@ -1465,27 +1465,27 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i8>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -1601,29 +1601,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 528 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1056 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i8
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>
@@ -1634,29 +1634,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 207 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 288 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i8
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i8>
@@ -1669,27 +1669,27 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i8>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -1805,29 +1805,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 528 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1056 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i8
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>
@@ -1838,29 +1838,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 207 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 288 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i8
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i8>
@@ -1873,27 +1873,27 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i8>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -2009,29 +2009,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 264 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 528 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1056 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i8
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>
@@ -2042,29 +2042,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 207 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 288 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i8
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i8>
@@ -2077,27 +2077,27 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i8>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -2111,29 +2111,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 219 for instruction: %V96i64 = trunc <96 x i64> undef to <96 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 164 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 328 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 656 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1312 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i8
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i8>
@@ -2144,29 +2144,29 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 207 for instruction: %V96i32 = trunc <96 x i32> undef to <96 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 416 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i8
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i8>
@@ -2179,27 +2179,27 @@ define i32 @trunc_vXi8() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %V96i16 = trunc <96 x i16> undef to <96 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i8>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i8>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -2751,28 +2751,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -2786,28 +2786,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i1
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i1>
@@ -2823,26 +2823,26 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i1>
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i8 = trunc i8 undef to i1
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i8 = trunc <2 x i8> undef to <2 x i1>
@@ -2894,28 +2894,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -2929,28 +2929,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i1
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i1>
@@ -2966,26 +2966,26 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i1>
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i8 = trunc i8 undef to i1
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i8 = trunc <2 x i8> undef to <2 x i1>
@@ -3033,32 +3033,32 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 133 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 190 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 266 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 380 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 456 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 532 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 760 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 912 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1064 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1520 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1824 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 2128 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -3107,28 +3107,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V12i16 = trunc <12 x i16> undef to <12 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V14i16 = trunc <14 x i16> undef to <14 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 167 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 234 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i8 = trunc i8 undef to i1
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2i8 = trunc <2 x i8> undef to <2 x i1>
@@ -3142,28 +3142,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V12i8 = trunc <12 x i8> undef to <12 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V14i8 = trunc <14 x i8> undef to <14 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16i8 = trunc <16 x i8> undef to <16 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i8 = trunc <32 x i8> undef to <32 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 115 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i8 = trunc <64 x i8> undef to <64 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 229 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i8 = trunc <128 x i8> undef to <128 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 327 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 393 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 458 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i8 = trunc <256 x i8> undef to <256 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 655 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 786 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 917 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i8 = trunc <512 x i8> undef to <512 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
-; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1310 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1572 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
+; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 1834 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i8 = trunc <1024 x i8> undef to <1024 x i1>
 ; AVX512FVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -3172,36 +3172,36 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i64 = trunc <2 x i64> undef to <2 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3i64 = trunc <3 x i64> undef to <3 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4i64 = trunc <4 x i64> undef to <4 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V5i64 = trunc <5 x i64> undef to <5 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V5i64 = trunc <5 x i64> undef to <5 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 252 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 360 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 432 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 504 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 720 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 864 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1008 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1440 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1728 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2016 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -3285,28 +3285,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i8 = trunc <12 x i8> undef to <12 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i8 = trunc <14 x i8> undef to <14 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = trunc <16 x i8> undef to <16 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i8 = trunc <32 x i8> undef to <32 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i8 = trunc <64 x i8> undef to <64 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 227 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i8 = trunc <128 x i8> undef to <128 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 325 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 455 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i8 = trunc <256 x i8> undef to <256 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 780 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 910 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i8 = trunc <512 x i8> undef to <512 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
-; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1300 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1560 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
+; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 1820 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i8 = trunc <1024 x i8> undef to <1024 x i1>
 ; AVX512FVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -3393,28 +3393,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V12i16 = trunc <12 x i16> undef to <12 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V14i16 = trunc <14 x i16> undef to <14 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 167 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 234 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i8 = trunc i8 undef to i1
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2i8 = trunc <2 x i8> undef to <2 x i1>
@@ -3428,28 +3428,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V12i8 = trunc <12 x i8> undef to <12 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V14i8 = trunc <14 x i8> undef to <14 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16i8 = trunc <16 x i8> undef to <16 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i8 = trunc <32 x i8> undef to <32 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 115 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i8 = trunc <64 x i8> undef to <64 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 229 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i8 = trunc <128 x i8> undef to <128 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 327 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 393 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 458 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i8 = trunc <256 x i8> undef to <256 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 655 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 786 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 917 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i8 = trunc <512 x i8> undef to <512 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
-; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1310 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1572 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
+; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 1834 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i8 = trunc <1024 x i8> undef to <1024 x i1>
 ; AVX512DQVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -3571,28 +3571,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V12i8 = trunc <12 x i8> undef to <12 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V14i8 = trunc <14 x i8> undef to <14 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = trunc <16 x i8> undef to <16 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %V20i8 = trunc <20 x i8> undef to <20 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %V24i8 = trunc <24 x i8> undef to <24 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V28i8 = trunc <28 x i8> undef to <28 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i8 = trunc <32 x i8> undef to <32 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 81 for instruction: %V40i8 = trunc <40 x i8> undef to <40 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %V48i8 = trunc <48 x i8> undef to <48 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %V56i8 = trunc <56 x i8> undef to <56 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i8 = trunc <64 x i8> undef to <64 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 162 for instruction: %V80i8 = trunc <80 x i8> undef to <80 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 227 for instruction: %V112i8 = trunc <112 x i8> undef to <112 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i8 = trunc <128 x i8> undef to <128 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 325 for instruction: %V160i8 = trunc <160 x i8> undef to <160 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V192i8 = trunc <192 x i8> undef to <192 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 455 for instruction: %V224i8 = trunc <224 x i8> undef to <224 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i8 = trunc <256 x i8> undef to <256 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 650 for instruction: %V320i8 = trunc <320 x i8> undef to <320 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 780 for instruction: %V384i8 = trunc <384 x i8> undef to <384 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 910 for instruction: %V448i8 = trunc <448 x i8> undef to <448 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i8 = trunc <512 x i8> undef to <512 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
-; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1300 for instruction: %V640i8 = trunc <640 x i8> undef to <640 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1560 for instruction: %V768i8 = trunc <768 x i8> undef to <768 x i1>
+; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 1820 for instruction: %V896i8 = trunc <896 x i8> undef to <896 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i8 = trunc <1024 x i8> undef to <1024 x i1>
 ; AVX512DQVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
@@ -3605,32 +3605,32 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 133 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 190 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 266 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 380 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 456 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 532 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 760 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 912 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1064 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1520 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1824 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2128 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -3644,28 +3644,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 245 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 350 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 420 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 490 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 700 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 840 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 980 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
-; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1400 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1680 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
+; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 1960 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i1>
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i1
 ; AVX512BWVEC512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i1>
@@ -3744,36 +3744,36 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i64 = trunc <2 x i64> undef to <2 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3i64 = trunc <3 x i64> undef to <3 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4i64 = trunc <4 x i64> undef to <4 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V5i64 = trunc <5 x i64> undef to <5 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V5i64 = trunc <5 x i64> undef to <5 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %V6i64 = trunc <6 x i64> undef to <6 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V7i64 = trunc <7 x i64> undef to <7 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8i64 = trunc <8 x i64> undef to <8 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %V10i64 = trunc <10 x i64> undef to <10 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 252 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 360 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 432 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 504 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 720 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 864 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1008 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1440 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1728 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2016 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -3787,28 +3787,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 85 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 170 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 238 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 320 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 340 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 408 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 476 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 640 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 768 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 896 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 680 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 816 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 952 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1280 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1536 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
-; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1792 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1360 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1632 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
+; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 1904 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i1>
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i1
 ; AVX512BWVEC256-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i1>
@@ -3895,28 +3895,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V12i64 = trunc <12 x i64> undef to <12 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %V14i64 = trunc <14 x i64> undef to <14 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %V16i64 = trunc <16 x i64> undef to <16 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %V20i64 = trunc <20 x i64> undef to <20 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %V24i64 = trunc <24 x i64> undef to <24 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V28i64 = trunc <28 x i64> undef to <28 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i64 = trunc <32 x i64> undef to <32 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %V40i64 = trunc <40 x i64> undef to <40 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %V48i64 = trunc <48 x i64> undef to <48 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 129 for instruction: %V56i64 = trunc <56 x i64> undef to <56 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i64 = trunc <64 x i64> undef to <64 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 183 for instruction: %V80i64 = trunc <80 x i64> undef to <80 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 256 for instruction: %V112i64 = trunc <112 x i64> undef to <112 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i64 = trunc <128 x i64> undef to <128 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 365 for instruction: %V160i64 = trunc <160 x i64> undef to <160 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 438 for instruction: %V192i64 = trunc <192 x i64> undef to <192 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 511 for instruction: %V224i64 = trunc <224 x i64> undef to <224 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i64 = trunc <256 x i64> undef to <256 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 730 for instruction: %V320i64 = trunc <320 x i64> undef to <320 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 876 for instruction: %V384i64 = trunc <384 x i64> undef to <384 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1022 for instruction: %V448i64 = trunc <448 x i64> undef to <448 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i64 = trunc <512 x i64> undef to <512 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1460 for instruction: %V640i64 = trunc <640 x i64> undef to <640 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1752 for instruction: %V768i64 = trunc <768 x i64> undef to <768 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 2044 for instruction: %V896i64 = trunc <896 x i64> undef to <896 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i64 = trunc <1024 x i64> undef to <1024 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i32 = trunc i32 undef to i1
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i32 = trunc <2 x i32> undef to <2 x i1>
@@ -3930,28 +3930,28 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V12i32 = trunc <12 x i32> undef to <12 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V14i32 = trunc <14 x i32> undef to <14 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %V16i32 = trunc <16 x i32> undef to <16 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %V20i32 = trunc <20 x i32> undef to <20 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %V24i32 = trunc <24 x i32> undef to <24 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %V28i32 = trunc <28 x i32> undef to <28 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i32 = trunc <32 x i32> undef to <32 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %V40i32 = trunc <40 x i32> undef to <40 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %V48i32 = trunc <48 x i32> undef to <48 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %V56i32 = trunc <56 x i32> undef to <56 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i32 = trunc <64 x i32> undef to <64 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 173 for instruction: %V80i32 = trunc <80 x i32> undef to <80 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 242 for instruction: %V112i32 = trunc <112 x i32> undef to <112 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i32 = trunc <128 x i32> undef to <128 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 345 for instruction: %V160i32 = trunc <160 x i32> undef to <160 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 414 for instruction: %V192i32 = trunc <192 x i32> undef to <192 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 483 for instruction: %V224i32 = trunc <224 x i32> undef to <224 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i32 = trunc <256 x i32> undef to <256 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %V320i32 = trunc <320 x i32> undef to <320 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %V384i32 = trunc <384 x i32> undef to <384 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 966 for instruction: %V448i32 = trunc <448 x i32> undef to <448 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i32 = trunc <512 x i32> undef to <512 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1380 for instruction: %V640i32 = trunc <640 x i32> undef to <640 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1656 for instruction: %V768i32 = trunc <768 x i32> undef to <768 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1932 for instruction: %V896i32 = trunc <896 x i32> undef to <896 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i32 = trunc <1024 x i32> undef to <1024 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i16 = trunc i16 undef to i1
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i16 = trunc <2 x i16> undef to <2 x i1>
@@ -3967,26 +3967,26 @@ define i32 @trunc_vXi1() "min-legal-vector-width"="256" {
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = trunc <16 x i16> undef to <16 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %V20i16 = trunc <20 x i16> undef to <20 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %V24i16 = trunc <24 x i16> undef to <24 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %V28i16 = trunc <28 x i16> undef to <28 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32i16 = trunc <32 x i16> undef to <32 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %V40i16 = trunc <40 x i16> undef to <40 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %V48i16 = trunc <48 x i16> undef to <48 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %V56i16 = trunc <56 x i16> undef to <56 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64i16 = trunc <64 x i16> undef to <64 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 165 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 231 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 168 for instruction: %V80i16 = trunc <80 x i16> undef to <80 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 235 for instruction: %V112i16 = trunc <112 x i16> undef to <112 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128i16 = trunc <128 x i16> undef to <128 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 330 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 462 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 335 for instruction: %V160i16 = trunc <160 x i16> undef to <160 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 402 for instruction: %V192i16 = trunc <192 x i16> undef to <192 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 469 for instruction: %V224i16 = trunc <224 x i16> undef to <224 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256i16 = trunc <256 x i16> undef to <256 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 660 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 924 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 670 for instruction: %V320i16 = trunc <320 x i16> undef to <320 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 804 for instruction: %V384i16 = trunc <384 x i16> undef to <384 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 938 for instruction: %V448i16 = trunc <448 x i16> undef to <448 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512i16 = trunc <512 x i16> undef to <512 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1320 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1584 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
-; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1848 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1340 for instruction: %V640i16 = trunc <640 x i16> undef to <640 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1608 for instruction: %V768i16 = trunc <768 x i16> undef to <768 x i1>
+; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1876 for instruction: %V896i16 = trunc <896 x i16> undef to <896 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V1024i16 = trunc <1024 x i16> undef to <1024 x i1>
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %i8 = trunc i8 undef to i1
 ; BTVER2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2i8 = trunc <2 x i8> undef to <2 x i1>