[AArch64][GlobalISel] Add some post-legalization cast combines. (#112509)

This helps clear up some of the legalization artefacts. Not all of the
cast_combines are added (notably select combines) as they currently have
questionable benefit in the test updates.

Author: davemgreen

llvm/include/llvm/Target/GlobalISel/Combine.td

@@ -1791,20 +1791,24 @@ class integer_of_opcode<Instruction castOpcode> : GICombineRule <
 
 def integer_of_truncate : integer_of_opcode<G_TRUNC>;
 
-def cast_combines: GICombineGroup<[
+def cast_of_cast_combines: GICombineGroup<[
   truncate_of_zext,
   truncate_of_sext,
   truncate_of_anyext,
-  select_of_zext,
-  select_of_anyext,
-  select_of_truncate,
   zext_of_zext,
   zext_of_anyext,
   sext_of_sext,
   sext_of_anyext,
   anyext_of_anyext,
   anyext_of_zext,
-  anyext_of_sext,
+  anyext_of_sext
+]>;
+
+def cast_combines: GICombineGroup<[
+  cast_of_cast_combines,
+  select_of_zext,
+  select_of_anyext,
+  select_of_truncate,
   buildvector_of_truncate,
   narrow_binop_add,
   narrow_binop_sub,

llvm/lib/Target/AArch64/AArch64Combine.td

@@ -314,9 +314,9 @@ def AArch64PostLegalizerLowering
 // Post-legalization combines which are primarily optimizations.
 def AArch64PostLegalizerCombiner
     : GICombiner<"AArch64PostLegalizerCombinerImpl",
-                       [copy_prop, combines_for_extload,
-                        combine_indexed_load_store,
-                        sext_trunc_sextload, mutate_anyext_to_zext,
+                       [copy_prop, cast_of_cast_combines, buildvector_of_truncate,
+                        integer_of_truncate, mutate_anyext_to_zext,
+                        combines_for_extload, combine_indexed_load_store, sext_trunc_sextload,
                         hoist_logic_op_with_same_opcode_hands,
                         redundant_and, xor_of_and_with_same_reg,
                         extractvecelt_pairwise_add, redundant_or,

llvm/test/CodeGen/AArch64/GlobalISel/combine-cast.mir

@@ -135,20 +135,13 @@ name:            test_combine_trunc_build_vector
 legalized: true
 body:             |
   bb.1:
-    ; CHECK-PRE-LABEL: name: test_combine_trunc_build_vector
-    ; CHECK-PRE: %arg1:_(s64) = COPY $x0
-    ; CHECK-PRE-NEXT: %arg2:_(s64) = COPY $x0
-    ; CHECK-PRE-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC %arg1(s64)
-    ; CHECK-PRE-NEXT: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC %arg2(s64)
-    ; CHECK-PRE-NEXT: %small:_(<2 x s32>) = G_BUILD_VECTOR [[TRUNC]](s32), [[TRUNC1]](s32)
-    ; CHECK-PRE-NEXT: $x0 = COPY %small(<2 x s32>)
-    ;
-    ; CHECK-POST-LABEL: name: test_combine_trunc_build_vector
-    ; CHECK-POST: %arg1:_(s64) = COPY $x0
-    ; CHECK-POST-NEXT: %arg2:_(s64) = COPY $x0
-    ; CHECK-POST-NEXT: %bv:_(<2 x s64>) = G_BUILD_VECTOR %arg1(s64), %arg2(s64)
-    ; CHECK-POST-NEXT: %small:_(<2 x s32>) = G_TRUNC %bv(<2 x s64>)
-    ; CHECK-POST-NEXT: $x0 = COPY %small(<2 x s32>)
+    ; CHECK-LABEL: name: test_combine_trunc_build_vector
+    ; CHECK: %arg1:_(s64) = COPY $x0
+    ; CHECK-NEXT: %arg2:_(s64) = COPY $x0
+    ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC %arg1(s64)
+    ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC %arg2(s64)
+    ; CHECK-NEXT: %small:_(<2 x s32>) = G_BUILD_VECTOR [[TRUNC]](s32), [[TRUNC1]](s32)
+    ; CHECK-NEXT: $x0 = COPY %small(<2 x s32>)
     %arg1:_(s64) = COPY $x0
     %arg2:_(s64) = COPY $x0
     %bv:_(<2 x s64>) = G_BUILD_VECTOR %arg1(s64), %arg2(s64)

llvm/test/CodeGen/AArch64/GlobalISel/combine-trunc.mir

@@ -32,20 +32,12 @@ legalized: true
 body:             |
   bb.1:
   liveins: $h0
-    ; CHECK-PRE-LABEL: name: test_combine_trunc_anyext_s32_s16
-    ; CHECK-PRE: liveins: $h0
-    ; CHECK-PRE-NEXT: {{  $}}
-    ; CHECK-PRE-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
-    ; CHECK-PRE-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY]](s16)
-    ; CHECK-PRE-NEXT: $w0 = COPY [[ANYEXT]](s32)
-    ;
-    ; CHECK-POST-LABEL: name: test_combine_trunc_anyext_s32_s16
-    ; CHECK-POST: liveins: $h0
-    ; CHECK-POST-NEXT: {{  $}}
-    ; CHECK-POST-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
-    ; CHECK-POST-NEXT: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[COPY]](s16)
-    ; CHECK-POST-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[ANYEXT]](s64)
-    ; CHECK-POST-NEXT: $w0 = COPY [[TRUNC]](s32)
+    ; CHECK-LABEL: name: test_combine_trunc_anyext_s32_s16
+    ; CHECK: liveins: $h0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
+    ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[COPY]](s16)
+    ; CHECK-NEXT: $w0 = COPY [[ANYEXT]](s32)
     %0:_(s16) = COPY $h0
     %1:_(s64) = G_ANYEXT %0(s16)
     %2:_(s32) = G_TRUNC %1(s64)
@@ -82,20 +74,12 @@ legalized: true
 body:             |
   bb.1:
   liveins: $h0
-    ; CHECK-PRE-LABEL: name: test_combine_trunc_sext_s32_s16
-    ; CHECK-PRE: liveins: $h0
-    ; CHECK-PRE-NEXT: {{  $}}
-    ; CHECK-PRE-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
-    ; CHECK-PRE-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[COPY]](s16)
-    ; CHECK-PRE-NEXT: $w0 = COPY [[SEXT]](s32)
-    ;
-    ; CHECK-POST-LABEL: name: test_combine_trunc_sext_s32_s16
-    ; CHECK-POST: liveins: $h0
-    ; CHECK-POST-NEXT: {{  $}}
-    ; CHECK-POST-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
-    ; CHECK-POST-NEXT: [[SEXT:%[0-9]+]]:_(s64) = G_SEXT [[COPY]](s16)
-    ; CHECK-POST-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[SEXT]](s64)
-    ; CHECK-POST-NEXT: $w0 = COPY [[TRUNC]](s32)
+    ; CHECK-LABEL: name: test_combine_trunc_sext_s32_s16
+    ; CHECK: liveins: $h0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
+    ; CHECK-NEXT: [[SEXT:%[0-9]+]]:_(s32) = G_SEXT [[COPY]](s16)
+    ; CHECK-NEXT: $w0 = COPY [[SEXT]](s32)
     %0:_(s16) = COPY $h0
     %1:_(s64) = G_SEXT %0(s16)
     %2:_(s32) = G_TRUNC %1(s64)
@@ -107,20 +91,12 @@ legalized: true
 body:             |
   bb.1:
   liveins: $h0
-    ; CHECK-PRE-LABEL: name: test_combine_trunc_zext_s32_s16
-    ; CHECK-PRE: liveins: $h0
-    ; CHECK-PRE-NEXT: {{  $}}
-    ; CHECK-PRE-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
-    ; CHECK-PRE-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[COPY]](s16)
-    ; CHECK-PRE-NEXT: $w0 = COPY [[ZEXT]](s32)
-    ;
-    ; CHECK-POST-LABEL: name: test_combine_trunc_zext_s32_s16
-    ; CHECK-POST: liveins: $h0
-    ; CHECK-POST-NEXT: {{  $}}
-    ; CHECK-POST-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
-    ; CHECK-POST-NEXT: [[ZEXT:%[0-9]+]]:_(s64) = G_ZEXT [[COPY]](s16)
-    ; CHECK-POST-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[ZEXT]](s64)
-    ; CHECK-POST-NEXT: $w0 = COPY [[TRUNC]](s32)
+    ; CHECK-LABEL: name: test_combine_trunc_zext_s32_s16
+    ; CHECK: liveins: $h0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $h0
+    ; CHECK-NEXT: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[COPY]](s16)
+    ; CHECK-NEXT: $w0 = COPY [[ZEXT]](s32)
     %0:_(s16) = COPY $h0
     %1:_(s64) = G_ZEXT %0(s16)
     %2:_(s32) = G_TRUNC %1(s64)
@@ -132,19 +108,11 @@ legalized: true
 body:             |
   bb.1:
   liveins: $w0
-    ; CHECK-PRE-LABEL: name: test_combine_trunc_anyext_s32_s32
-    ; CHECK-PRE: liveins: $w0
-    ; CHECK-PRE-NEXT: {{  $}}
-    ; CHECK-PRE-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
-    ; CHECK-PRE-NEXT: $w0 = COPY [[COPY]](s32)
-    ;
-    ; CHECK-POST-LABEL: name: test_combine_trunc_anyext_s32_s32
-    ; CHECK-POST: liveins: $w0
-    ; CHECK-POST-NEXT: {{  $}}
-    ; CHECK-POST-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
-    ; CHECK-POST-NEXT: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[COPY]](s32)
-    ; CHECK-POST-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[ANYEXT]](s64)
-    ; CHECK-POST-NEXT: $w0 = COPY [[TRUNC]](s32)
+    ; CHECK-LABEL: name: test_combine_trunc_anyext_s32_s32
+    ; CHECK: liveins: $w0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $w0
+    ; CHECK-NEXT: $w0 = COPY [[COPY]](s32)
     %0:_(s32) = COPY $w0
     %1:_(s64) = G_ANYEXT %0(s32)
     %2:_(s32) = G_TRUNC %1(s64)
@@ -156,20 +124,12 @@ legalized: true
 body:             |
   bb.1:
   liveins: $x0
-    ; CHECK-PRE-LABEL: name: test_combine_trunc_anyext_s32_s64
-    ; CHECK-PRE: liveins: $x0
-    ; CHECK-PRE-NEXT: {{  $}}
-    ; CHECK-PRE-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
-    ; CHECK-PRE-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
-    ; CHECK-PRE-NEXT: $w0 = COPY [[TRUNC]](s32)
-    ;
-    ; CHECK-POST-LABEL: name: test_combine_trunc_anyext_s32_s64
-    ; CHECK-POST: liveins: $x0
-    ; CHECK-POST-NEXT: {{  $}}
-    ; CHECK-POST-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
-    ; CHECK-POST-NEXT: [[ANYEXT:%[0-9]+]]:_(s128) = G_ANYEXT [[COPY]](s64)
-    ; CHECK-POST-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[ANYEXT]](s128)
-    ; CHECK-POST-NEXT: $w0 = COPY [[TRUNC]](s32)
+    ; CHECK-LABEL: name: test_combine_trunc_anyext_s32_s64
+    ; CHECK: liveins: $x0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+    ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+    ; CHECK-NEXT: $w0 = COPY [[TRUNC]](s32)
     %0:_(s64) = COPY $x0
     %1:_(s128) = G_ANYEXT %0(s64)
     %2:_(s32) = G_TRUNC %1(s128)

llvm/test/CodeGen/AArch64/add.ll

@@ -171,11 +171,7 @@ define void @v4i8(ptr %p1, ptr %p2) {
 ; CHECK-GI-NEXT:    ushll v0.8h, v3.8b, #0
 ; CHECK-GI-NEXT:    ushll v1.8h, v5.8b, #0
 ; CHECK-GI-NEXT:    add v0.4h, v0.4h, v1.4h
-; CHECK-GI-NEXT:    mov v1.h[0], v0.h[0]
-; CHECK-GI-NEXT:    mov v1.h[1], v0.h[1]
-; CHECK-GI-NEXT:    mov v1.h[2], v0.h[2]
-; CHECK-GI-NEXT:    mov v1.h[3], v0.h[3]
-; CHECK-GI-NEXT:    xtn v0.8b, v1.8h
+; CHECK-GI-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-GI-NEXT:    fmov w8, s0
 ; CHECK-GI-NEXT:    str w8, [x0]
 ; CHECK-GI-NEXT:    ret

llvm/test/CodeGen/AArch64/and-mask-removal.ll

@@ -530,10 +530,10 @@ define i64 @test_2_selects(i8 zeroext %a) {
 ; CHECK-LABEL: test_2_selects:
 ; CHECK:       ; %bb.0:
 ; CHECK-NEXT:    add w9, w0, #24
-; CHECK-NEXT:    mov w8, #131
+; CHECK-NEXT:    mov w8, #131 ; =0x83
 ; CHECK-NEXT:    and w9, w9, #0xff
 ; CHECK-NEXT:    cmp w9, #81
-; CHECK-NEXT:    mov w9, #57
+; CHECK-NEXT:    mov w9, #57 ; =0x39
 ; CHECK-NEXT:    csel x8, x8, xzr, lo
 ; CHECK-NEXT:    csel x9, xzr, x9, eq
 ; CHECK-NEXT:    add x0, x8, x9

llvm/test/CodeGen/AArch64/andorxor.ll

@@ -463,11 +463,7 @@ define void @and_v4i8(ptr %p1, ptr %p2) {
 ; CHECK-GI-NEXT:    ushll v0.8h, v3.8b, #0
 ; CHECK-GI-NEXT:    ushll v1.8h, v5.8b, #0
 ; CHECK-GI-NEXT:    and v0.8b, v0.8b, v1.8b
-; CHECK-GI-NEXT:    mov v1.h[0], v0.h[0]
-; CHECK-GI-NEXT:    mov v1.h[1], v0.h[1]
-; CHECK-GI-NEXT:    mov v1.h[2], v0.h[2]
-; CHECK-GI-NEXT:    mov v1.h[3], v0.h[3]
-; CHECK-GI-NEXT:    xtn v0.8b, v1.8h
+; CHECK-GI-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-GI-NEXT:    fmov w8, s0
 ; CHECK-GI-NEXT:    str w8, [x0]
 ; CHECK-GI-NEXT:    ret
@@ -514,11 +510,7 @@ define void @or_v4i8(ptr %p1, ptr %p2) {
 ; CHECK-GI-NEXT:    ushll v0.8h, v3.8b, #0
 ; CHECK-GI-NEXT:    ushll v1.8h, v5.8b, #0
 ; CHECK-GI-NEXT:    orr v0.8b, v0.8b, v1.8b
-; CHECK-GI-NEXT:    mov v1.h[0], v0.h[0]
-; CHECK-GI-NEXT:    mov v1.h[1], v0.h[1]
-; CHECK-GI-NEXT:    mov v1.h[2], v0.h[2]
-; CHECK-GI-NEXT:    mov v1.h[3], v0.h[3]
-; CHECK-GI-NEXT:    xtn v0.8b, v1.8h
+; CHECK-GI-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-GI-NEXT:    fmov w8, s0
 ; CHECK-GI-NEXT:    str w8, [x0]
 ; CHECK-GI-NEXT:    ret
@@ -565,11 +557,7 @@ define void @xor_v4i8(ptr %p1, ptr %p2) {
 ; CHECK-GI-NEXT:    ushll v0.8h, v3.8b, #0
 ; CHECK-GI-NEXT:    ushll v1.8h, v5.8b, #0
 ; CHECK-GI-NEXT:    eor v0.8b, v0.8b, v1.8b
-; CHECK-GI-NEXT:    mov v1.h[0], v0.h[0]
-; CHECK-GI-NEXT:    mov v1.h[1], v0.h[1]
-; CHECK-GI-NEXT:    mov v1.h[2], v0.h[2]
-; CHECK-GI-NEXT:    mov v1.h[3], v0.h[3]
-; CHECK-GI-NEXT:    xtn v0.8b, v1.8h
+; CHECK-GI-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-GI-NEXT:    fmov w8, s0
 ; CHECK-GI-NEXT:    str w8, [x0]
 ; CHECK-GI-NEXT:    ret

llvm/test/CodeGen/AArch64/bitcast.ll

@@ -60,11 +60,7 @@ define i32 @bitcast_v4i8_i32(<4 x i8> %a, <4 x i8> %b){
 ; CHECK-GI-LABEL: bitcast_v4i8_i32:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.4h, v0.4h, v1.4h
-; CHECK-GI-NEXT:    mov v1.h[0], v0.h[0]
-; CHECK-GI-NEXT:    mov v1.h[1], v0.h[1]
-; CHECK-GI-NEXT:    mov v1.h[2], v0.h[2]
-; CHECK-GI-NEXT:    mov v1.h[3], v0.h[3]
-; CHECK-GI-NEXT:    xtn v0.8b, v1.8h
+; CHECK-GI-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-GI-NEXT:    fmov w0, s0
 ; CHECK-GI-NEXT:    ret
   %c = add <4 x i8> %a, %b
@@ -116,9 +112,7 @@ define i32 @bitcast_v2i16_i32(<2 x i16> %a, <2 x i16> %b){
 ; CHECK-GI-LABEL: bitcast_v2i16_i32:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.2s, v0.2s, v1.2s
-; CHECK-GI-NEXT:    mov v1.s[0], v0.s[0]
-; CHECK-GI-NEXT:    mov v1.s[1], v0.s[1]
-; CHECK-GI-NEXT:    xtn v0.4h, v1.4s
+; CHECK-GI-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
 ; CHECK-GI-NEXT:    fmov w0, s0
 ; CHECK-GI-NEXT:    ret
   %c = add <2 x i16> %a, %b
@@ -418,9 +412,7 @@ define <4 x i8> @bitcast_v2i16_v4i8(<2 x i16> %a, <2 x i16> %b){
 ; CHECK-GI-LABEL: bitcast_v2i16_v4i8:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.2s, v0.2s, v1.2s
-; CHECK-GI-NEXT:    mov v1.s[0], v0.s[0]
-; CHECK-GI-NEXT:    mov v1.s[1], v0.s[1]
-; CHECK-GI-NEXT:    xtn v0.4h, v1.4s
+; CHECK-GI-NEXT:    uzp1 v0.4h, v0.4h, v0.4h
 ; CHECK-GI-NEXT:    mov b1, v0.b[1]
 ; CHECK-GI-NEXT:    mov v2.b[0], v0.b[0]
 ; CHECK-GI-NEXT:    mov b3, v0.b[2]
@@ -455,11 +447,7 @@ define <2 x i16> @bitcast_v4i8_v2i16(<4 x i8> %a, <4 x i8> %b){
 ; CHECK-GI-LABEL: bitcast_v4i8_v2i16:
 ; CHECK-GI:       // %bb.0:
 ; CHECK-GI-NEXT:    add v0.4h, v0.4h, v1.4h
-; CHECK-GI-NEXT:    mov v1.h[0], v0.h[0]
-; CHECK-GI-NEXT:    mov v1.h[1], v0.h[1]
-; CHECK-GI-NEXT:    mov v1.h[2], v0.h[2]
-; CHECK-GI-NEXT:    mov v1.h[3], v0.h[3]
-; CHECK-GI-NEXT:    xtn v0.8b, v1.8h
+; CHECK-GI-NEXT:    uzp1 v0.8b, v0.8b, v0.8b
 ; CHECK-GI-NEXT:    mov h1, v0.h[1]
 ; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
 ; CHECK-GI-NEXT:    ushll v0.4s, v0.4h, #0