[AArch64] Add a subvector extract cost. (#121472)

These can generally be emitted using an ext instruction or mov from the
high half. The half half extracts can be free depending on the users,
but that is not handled here, just the basic costs. It originally
included all subvector extracts, but that was toned-down to just
half-vector extracts to try and help the mid end not breakup high/low
extracts without having the SLP vectorizer create a mess using other
shuffles.

Author: davemgreen

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

@@ -4738,10 +4738,21 @@ InstructionCost AArch64TTIImpl::getShuffleCost(
   }
 
   Kind = improveShuffleKindFromMask(Kind, Mask, Tp, Index, SubTp);
-  // Treat extractsubvector as single op permutation.
   bool IsExtractSubvector = Kind == TTI::SK_ExtractSubvector;
-  if (IsExtractSubvector && LT.second.isFixedLengthVector())
+  // A sebvector extract can be implemented with a ext (or trivial extract, if
+  // from lane 0). This currently only handles low or high extracts to prevent
+  // SLP vectorizer regressions.
+  if (IsExtractSubvector && LT.second.isFixedLengthVector()) {
+    if (LT.second.is128BitVector() &&
+        cast<FixedVectorType>(SubTp)->getNumElements() ==
+            LT.second.getVectorNumElements() / 2) {
+      if (Index == 0)
+        return 0;
+      if (Index == LT.second.getVectorNumElements() / 2)
+        return 1;
+    }
     Kind = TTI::SK_PermuteSingleSrc;
+  }
 
   // Check for broadcast loads, which are supported by the LD1R instruction.
   // In terms of code-size, the shuffle vector is free when a load + dup get

llvm/test/Analysis/CostModel/AArch64/shuffle-extract.ll

@@ -15,23 +15,23 @@ define void @extract_half() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i8_hi = shufflevector <8 x i8> poison, <8 x i8> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i8_lo = shufflevector <16 x i8> poison, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_mi = shufflevector <16 x i8> poison, <16 x i8> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_hi = shufflevector <16 x i8> poison, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_hi = shufflevector <16 x i8> poison, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i16_lo = shufflevector <2 x i16> poison, <2 x i16> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_hi = shufflevector <2 x i16> poison, <2 x i16> poison, <1 x i32> <i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i16_lo = shufflevector <4 x i16> poison, <4 x i16> poison, <2 x i32> <i32 0, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i16_mi = shufflevector <4 x i16> poison, <4 x i16> poison, <2 x i32> <i32 1, i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i16_hi = shufflevector <4 x i16> poison, <4 x i16> poison, <2 x i32> <i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i16_lo = shufflevector <8 x i16> poison, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_mi = shufflevector <8 x i16> poison, <8 x i16> poison, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_hi = shufflevector <8 x i16> poison, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_hi = shufflevector <8 x i16> poison, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i16_lo = shufflevector <16 x i16> poison, <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16i16_mi = shufflevector <16 x i16> poison, <16 x i16> poison, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16i16_hi = shufflevector <16 x i16> poison, <16 x i16> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_lo = shufflevector <2 x i32> poison, <2 x i32> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_hi = shufflevector <2 x i32> poison, <2 x i32> poison, <1 x i32> <i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_lo = shufflevector <4 x i32> poison, <4 x i32> poison, <2 x i32> <i32 0, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_mi = shufflevector <4 x i32> poison, <4 x i32> poison, <2 x i32> <i32 1, i32 2>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_hi = shufflevector <4 x i32> poison, <4 x i32> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_hi = shufflevector <4 x i32> poison, <4 x i32> poison, <2 x i32> <i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i32_lo = shufflevector <8 x i32> poison, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8i32_mi = shufflevector <8 x i32> poison, <8 x i32> poison, <4 x i32> <i32 2, i32 3, i32 4, i32 5>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8i32_hi = shufflevector <8 x i32> poison, <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
@@ -114,19 +114,19 @@ define void @extract_qtr() {
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_mi = shufflevector <8 x i16> poison, <8 x i16> poison, <2 x i32> <i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_hi = shufflevector <8 x i16> poison, <8 x i16> poison, <2 x i32> <i32 4, i32 5>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i16_lo = shufflevector <16 x i16> poison, <16 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16i16_mi = shufflevector <16 x i16> poison, <16 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_mi = shufflevector <16 x i16> poison, <16 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16i16_hi = shufflevector <16 x i16> poison, <16 x i16> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_lo = shufflevector <4 x i32> poison, <4 x i32> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_mi = shufflevector <4 x i32> poison, <4 x i32> poison, <1 x i32> <i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_hi = shufflevector <4 x i32> poison, <4 x i32> poison, <1 x i32> <i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i32_lo = shufflevector <8 x i32> poison, <8 x i32> poison, <2 x i32> <i32 0, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8i32_mi = shufflevector <8 x i32> poison, <8 x i32> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_mi = shufflevector <8 x i32> poison, <8 x i32> poison, <2 x i32> <i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8i32_hi = shufflevector <8 x i32> poison, <8 x i32> poison, <2 x i32> <i32 4, i32 5>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i32_lo = shufflevector <16 x i32> poison, <16 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16i32_mi = shufflevector <16 x i32> poison, <16 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16i32_hi = shufflevector <16 x i32> poison, <16 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i64_lo = shufflevector <4 x i64> poison, <4 x i64> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_mi = shufflevector <4 x i64> poison, <4 x i64> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_mi = shufflevector <4 x i64> poison, <4 x i64> poison, <1 x i32> <i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_hi = shufflevector <4 x i64> poison, <4 x i64> poison, <1 x i32> <i32 2>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i64_lo = shufflevector <8 x i64> poison, <8 x i64> poison, <2 x i32> <i32 0, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i64_mi = shufflevector <8 x i64> poison, <8 x i64> poison, <2 x i32> <i32 2, i32 3>

llvm/test/Transforms/SLPVectorizer/AArch64/reduce-fadd.ll

@@ -80,16 +80,11 @@ define half @reduce_fast_half8(<8 x half> %vec8) {
 ; NOFP16-LABEL: define half @reduce_fast_half8(
 ; NOFP16-SAME: <8 x half> [[VEC8:%.*]]) #[[ATTR0]] {
 ; NOFP16-NEXT:  [[ENTRY:.*:]]
-; NOFP16-NEXT:    [[ELT4:%.*]] = extractelement <8 x half> [[VEC8]], i64 4
-; NOFP16-NEXT:    [[ELT5:%.*]] = extractelement <8 x half> [[VEC8]], i64 5
-; NOFP16-NEXT:    [[ELT6:%.*]] = extractelement <8 x half> [[VEC8]], i64 6
-; NOFP16-NEXT:    [[ELT7:%.*]] = extractelement <8 x half> [[VEC8]], i64 7
 ; NOFP16-NEXT:    [[TMP0:%.*]] = shufflevector <8 x half> [[VEC8]], <8 x half> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; NOFP16-NEXT:    [[TMP1:%.*]] = call fast half @llvm.vector.reduce.fadd.v4f16(half 0xH0000, <4 x half> [[TMP0]])
-; NOFP16-NEXT:    [[OP_RDX:%.*]] = fadd fast half [[TMP1]], [[ELT4]]
-; NOFP16-NEXT:    [[OP_RDX1:%.*]] = fadd fast half [[ELT5]], [[ELT6]]
-; NOFP16-NEXT:    [[OP_RDX2:%.*]] = fadd fast half [[OP_RDX]], [[OP_RDX1]]
-; NOFP16-NEXT:    [[OP_RDX3:%.*]] = fadd fast half [[OP_RDX2]], [[ELT7]]
+; NOFP16-NEXT:    [[TMP2:%.*]] = shufflevector <8 x half> [[VEC8]], <8 x half> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; NOFP16-NEXT:    [[TMP3:%.*]] = call fast half @llvm.vector.reduce.fadd.v4f16(half 0xH0000, <4 x half> [[TMP2]])
+; NOFP16-NEXT:    [[OP_RDX3:%.*]] = fadd fast half [[TMP1]], [[TMP3]]
 ; NOFP16-NEXT:    ret half [[OP_RDX3]]
 ;
 ; FULLFP16-LABEL: define half @reduce_fast_half8(