Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[VectorCombine][X86] foldShuffleOfCastops - fold shuffle(cast(x),cast(y)) -> cast(shuffle(x,y)) iff cost efficient #87510

Merged
merged 1 commit into from
Apr 4, 2024
Merged

[VectorCombine][X86] foldShuffleOfCastops - fold shuffle(cast(x),cast(y)) -> cast(shuffle(x,y)) iff cost efficient #87510

merged 1 commit into from
Apr 4, 2024

Conversation

RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented Apr 3, 2024

Based off the existing foldShuffleOfBinops fold

Fixes #67803

@llvmbot
Copy link
Member

llvmbot commented Apr 3, 2024

@llvm/pr-subscribers-llvm-transforms

Author: Simon Pilgrim (RKSimon)

Changes

Based off the existing foldShuffleOfBinops fold

Fixes #67803

Full diff: https://github.com/llvm/llvm-project/pull/87510.diff

3 Files Affected:

  • (modified) llvm/lib/Transforms/Vectorize/VectorCombine.cpp (+55)
  • (modified) llvm/test/Transforms/PhaseOrdering/X86/pr67803.ll (+1-5)
  • (modified) llvm/test/Transforms/VectorCombine/X86/shuffle-of-casts.ll (+26-32)
diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index af5e7c9bc385ca..f5d5048a64bcaf 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -112,6 +112,7 @@ class VectorCombine {
   bool foldSingleElementStore(Instruction &I);
   bool scalarizeLoadExtract(Instruction &I);
   bool foldShuffleOfBinops(Instruction &I);
+  bool foldShuffleOfCastops(Instruction &I);
   bool foldShuffleFromReductions(Instruction &I);
   bool foldTruncFromReductions(Instruction &I);
   bool foldSelectShuffle(Instruction &I, bool FromReduction = false);
@@ -1432,6 +1433,59 @@ bool VectorCombine::foldShuffleOfBinops(Instruction &I) {
   return true;
 }
 
+/// Try to convert "shuffle (castop), (castop)" with a shared castop operand into
+/// "castop (shuffle)".
+bool VectorCombine::foldShuffleOfCastops(Instruction &I) {
+  Value *V0, *V1;
+  ArrayRef<int> Mask;
+  if (!match(&I, m_Shuffle(m_OneUse(m_Value(V0)), m_OneUse(m_Value(V1)),
+                           m_Mask(Mask))))
+    return false;
+
+  auto *C0 = dyn_cast<CastInst>(V0);
+  auto *C1 = dyn_cast<CastInst>(V1);
+  if (!C0 || !C1)
+    return false;
+
+  Instruction::CastOps Opcode = C0->getOpcode();
+  if (Opcode == Instruction::BitCast || Opcode != C1->getOpcode() ||
+      C0->getSrcTy() != C1->getSrcTy())
+    return false;
+
+  auto *ShuffleDstTy = dyn_cast<FixedVectorType>(I.getType());
+  auto *CastDstTy = dyn_cast<FixedVectorType>(C0->getDestTy());
+  auto *CastSrcTy = dyn_cast<FixedVectorType>(C0->getSrcTy());
+  if (!ShuffleDstTy || !CastDstTy || !CastSrcTy ||
+      CastDstTy->getElementCount() != CastSrcTy->getElementCount())
+    return false;
+
+  auto *NewCastSrcTy =
+      FixedVectorType::get(CastSrcTy->getScalarType(), Mask.size());
+
+  // Try to replace a castop with a shuffle if the shuffle is not costly.
+  TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
+  InstructionCost OldCost =
+      TTI.getCastInstrCost(Opcode, CastDstTy, CastSrcTy,
+                           TTI::CastContextHint::None, CostKind) +
+      TTI.getCastInstrCost(Opcode, CastDstTy, CastSrcTy,
+                           TTI::CastContextHint::None, CostKind);
+  OldCost += TTI.getShuffleCost(TargetTransformInfo::SK_PermuteTwoSrc,
+                                CastDstTy, Mask, CostKind);
+  InstructionCost NewCost = TTI.getCastInstrCost(
+      Opcode, ShuffleDstTy, NewCastSrcTy, TTI::CastContextHint::None, CostKind);
+  NewCost += TTI.getShuffleCost(TargetTransformInfo::SK_PermuteTwoSrc,
+                                CastSrcTy, Mask, CostKind);
+  if (NewCost > OldCost)
+    return false;
+
+  Value *Shuf =
+      Builder.CreateShuffleVector(C0->getOperand(0), C1->getOperand(0), Mask);
+  Value *Cast = Builder.CreateCast(Opcode, Shuf, ShuffleDstTy);
+  // TODO Copy IR flags?
+  replaceValue(I, *Cast);
+  return true;
+}
+
 /// Given a commutative reduction, the order of the input lanes does not alter
 /// the results. We can use this to remove certain shuffles feeding the
 /// reduction, removing the need to shuffle at all.
@@ -1986,6 +2040,7 @@ bool VectorCombine::run() {
         break;
       case Instruction::ShuffleVector:
         MadeChange |= foldShuffleOfBinops(I);
+        MadeChange |= foldShuffleOfCastops(I);
         MadeChange |= foldSelectShuffle(I);
         break;
       case Instruction::BitCast:
diff --git a/llvm/test/Transforms/PhaseOrdering/X86/pr67803.ll b/llvm/test/Transforms/PhaseOrdering/X86/pr67803.ll
index 495ec0a633999a..45e411d733169a 100644
--- a/llvm/test/Transforms/PhaseOrdering/X86/pr67803.ll
+++ b/llvm/test/Transforms/PhaseOrdering/X86/pr67803.ll
@@ -9,11 +9,7 @@ define <4 x i64> @PR67803(<4 x i64> %x, <4 x i64> %y, <4 x i64> %a, <4 x i64> %b
 ; CHECK-NEXT:    [[TMP0:%.*]] = bitcast <4 x i64> [[X:%.*]] to <8 x i32>
 ; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i64> [[Y:%.*]] to <8 x i32>
 ; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt <8 x i32> [[TMP0]], [[TMP1]]
-; CHECK-NEXT:    [[CMP_I21:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:    [[SEXT_I22:%.*]] = sext <4 x i1> [[CMP_I21]] to <4 x i32>
-; CHECK-NEXT:    [[CMP_I:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT:    [[SEXT_I:%.*]] = sext <4 x i1> [[CMP_I]] to <4 x i32>
-; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x i32> [[SEXT_I22]], <4 x i32> [[SEXT_I]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP3:%.*]] = sext <8 x i1> [[TMP2]] to <8 x i32>
 ; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>
 ; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <32 x i8> [[TMP5]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:    [[TMP7:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>
diff --git a/llvm/test/Transforms/VectorCombine/X86/shuffle-of-casts.ll b/llvm/test/Transforms/VectorCombine/X86/shuffle-of-casts.ll
index 3a7c3318d2d978..669f282becc535 100644
--- a/llvm/test/Transforms/VectorCombine/X86/shuffle-of-casts.ll
+++ b/llvm/test/Transforms/VectorCombine/X86/shuffle-of-casts.ll
@@ -6,9 +6,8 @@
 
 define <16 x i32> @concat_zext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
 ; CHECK-LABEL: @concat_zext_v8i16_v16i32(
-; CHECK-NEXT:    [[X0:%.*]] = zext <8 x i16> [[A0:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[X1:%.*]] = zext <8 x i16> [[A1:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <8 x i32> [[X0]], <8 x i32> [[X1]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i16> [[A0:%.*]], <8 x i16> [[A1:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[R:%.*]] = zext <16 x i16> [[TMP1]] to <16 x i32>
 ; CHECK-NEXT:    ret <16 x i32> [[R]]
 ;
   %x0 = zext <8 x i16> %a0 to <8 x i32>
@@ -19,9 +18,8 @@ define <16 x i32> @concat_zext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
 
 define <16 x i32> @concat_sext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
 ; CHECK-LABEL: @concat_sext_v8i16_v16i32(
-; CHECK-NEXT:    [[X0:%.*]] = sext <8 x i16> [[A0:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[X1:%.*]] = sext <8 x i16> [[A1:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <8 x i32> [[X0]], <8 x i32> [[X1]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i16> [[A0:%.*]], <8 x i16> [[A1:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[R:%.*]] = sext <16 x i16> [[TMP1]] to <16 x i32>
 ; CHECK-NEXT:    ret <16 x i32> [[R]]
 ;
   %x0 = sext <8 x i16> %a0 to <8 x i32>
@@ -32,9 +30,8 @@ define <16 x i32> @concat_sext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
 
 define <8 x i32> @concat_sext_v4i1_v8i32(<4 x i1> %a0, <4 x i1> %a1) {
 ; CHECK-LABEL: @concat_sext_v4i1_v8i32(
-; CHECK-NEXT:    [[X0:%.*]] = sext <4 x i1> [[A0:%.*]] to <4 x i32>
-; CHECK-NEXT:    [[X1:%.*]] = sext <4 x i1> [[A1:%.*]] to <4 x i32>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i32> [[X0]], <4 x i32> [[X1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i1> [[A0:%.*]], <4 x i1> [[A1:%.*]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[R:%.*]] = sext <8 x i1> [[TMP1]] to <8 x i32>
 ; CHECK-NEXT:    ret <8 x i32> [[R]]
 ;
   %x0 = sext <4 x i1> %a0 to <4 x i32>
@@ -45,9 +42,8 @@ define <8 x i32> @concat_sext_v4i1_v8i32(<4 x i1> %a0, <4 x i1> %a1) {
 
 define <8 x i16> @concat_trunc_v4i32_v8i16(<4 x i32> %a0, <4 x i32> %a1) {
 ; CHECK-LABEL: @concat_trunc_v4i32_v8i16(
-; CHECK-NEXT:    [[X0:%.*]] = trunc <4 x i32> [[A0:%.*]] to <4 x i16>
-; CHECK-NEXT:    [[X1:%.*]] = trunc <4 x i32> [[A1:%.*]] to <4 x i16>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i16> [[X0]], <4 x i16> [[X1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[A0:%.*]], <4 x i32> [[A1:%.*]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[R:%.*]] = trunc <8 x i32> [[TMP1]] to <8 x i16>
 ; CHECK-NEXT:    ret <8 x i16> [[R]]
 ;
   %x0 = trunc <4 x i32> %a0 to <4 x i16>
@@ -58,9 +54,8 @@ define <8 x i16> @concat_trunc_v4i32_v8i16(<4 x i32> %a0, <4 x i32> %a1) {
 
 define <8 x ptr> @concat_inttoptr_v4i32_v8iptr(<4 x i32> %a0, <4 x i32> %a1) {
 ; CHECK-LABEL: @concat_inttoptr_v4i32_v8iptr(
-; CHECK-NEXT:    [[X0:%.*]] = inttoptr <4 x i32> [[A0:%.*]] to <4 x ptr>
-; CHECK-NEXT:    [[X1:%.*]] = inttoptr <4 x i32> [[A1:%.*]] to <4 x ptr>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x ptr> [[X0]], <4 x ptr> [[X1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i32> [[A0:%.*]], <4 x i32> [[A1:%.*]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[R:%.*]] = inttoptr <8 x i32> [[TMP1]] to <8 x ptr>
 ; CHECK-NEXT:    ret <8 x ptr> [[R]]
 ;
   %x0 = inttoptr <4 x i32> %a0 to <4 x ptr>
@@ -71,9 +66,8 @@ define <8 x ptr> @concat_inttoptr_v4i32_v8iptr(<4 x i32> %a0, <4 x i32> %a1) {
 
 define <16 x i64> @concat_ptrtoint_v8i16_v16i32(<8 x ptr> %a0, <8 x ptr> %a1) {
 ; CHECK-LABEL: @concat_ptrtoint_v8i16_v16i32(
-; CHECK-NEXT:    [[X0:%.*]] = ptrtoint <8 x ptr> [[A0:%.*]] to <8 x i64>
-; CHECK-NEXT:    [[X1:%.*]] = ptrtoint <8 x ptr> [[A1:%.*]] to <8 x i64>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <8 x i64> [[X0]], <8 x i64> [[X1]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x ptr> [[A0:%.*]], <8 x ptr> [[A1:%.*]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[R:%.*]] = ptrtoint <16 x ptr> [[TMP1]] to <16 x i64>
 ; CHECK-NEXT:    ret <16 x i64> [[R]]
 ;
   %x0 = ptrtoint <8 x ptr> %a0 to <8 x i64>
@@ -83,11 +77,16 @@ define <16 x i64> @concat_ptrtoint_v8i16_v16i32(<8 x ptr> %a0, <8 x ptr> %a1) {
 }
 
 define <8 x double> @concat_fpext_v4f32_v8f64(<4 x float> %a0, <4 x float> %a1) {
-; CHECK-LABEL: @concat_fpext_v4f32_v8f64(
-; CHECK-NEXT:    [[X0:%.*]] = fpext <4 x float> [[A0:%.*]] to <4 x double>
-; CHECK-NEXT:    [[X1:%.*]] = fpext <4 x float> [[A1:%.*]] to <4 x double>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x double> [[X0]], <4 x double> [[X1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT:    ret <8 x double> [[R]]
+; SSE-LABEL: @concat_fpext_v4f32_v8f64(
+; SSE-NEXT:    [[TMP1:%.*]] = shufflevector <4 x float> [[A0:%.*]], <4 x float> [[A1:%.*]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; SSE-NEXT:    [[R:%.*]] = fpext <8 x float> [[TMP1]] to <8 x double>
+; SSE-NEXT:    ret <8 x double> [[R]]
+;
+; AVX-LABEL: @concat_fpext_v4f32_v8f64(
+; AVX-NEXT:    [[X0:%.*]] = fpext <4 x float> [[A0:%.*]] to <4 x double>
+; AVX-NEXT:    [[X1:%.*]] = fpext <4 x float> [[A1:%.*]] to <4 x double>
+; AVX-NEXT:    [[R:%.*]] = shufflevector <4 x double> [[X0]], <4 x double> [[X1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; AVX-NEXT:    ret <8 x double> [[R]]
 ;
   %x0 = fpext <4 x float> %a0 to <4 x double>
   %x1 = fpext <4 x float> %a1 to <4 x double>
@@ -112,9 +111,8 @@ define <16 x float> @concat_fptrunc_v8f64_v16f32(<8 x double> %a0, <8 x double>
 
 define <16 x i32> @rconcat_sext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
 ; CHECK-LABEL: @rconcat_sext_v8i16_v16i32(
-; CHECK-NEXT:    [[X0:%.*]] = sext <8 x i16> [[A0:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[X1:%.*]] = sext <8 x i16> [[A1:%.*]] to <8 x i32>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <8 x i32> [[X0]], <8 x i32> [[X1]], <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i16> [[A0:%.*]], <8 x i16> [[A1:%.*]], <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[R:%.*]] = sext <16 x i16> [[TMP1]] to <16 x i32>
 ; CHECK-NEXT:    ret <16 x i32> [[R]]
 ;
   %x0 = sext <8 x i16> %a0 to <8 x i32>
@@ -127,9 +125,8 @@ define <16 x i32> @rconcat_sext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
 
 define <8 x double> @interleave_fpext_v4f32_v8f64(<4 x float> %a0, <4 x float> %a1) {
 ; CHECK-LABEL: @interleave_fpext_v4f32_v8f64(
-; CHECK-NEXT:    [[X0:%.*]] = fpext <4 x float> [[A0:%.*]] to <4 x double>
-; CHECK-NEXT:    [[X1:%.*]] = fpext <4 x float> [[A1:%.*]] to <4 x double>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x double> [[X0]], <4 x double> [[X1]], <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x float> [[A0:%.*]], <4 x float> [[A1:%.*]], <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
+; CHECK-NEXT:    [[R:%.*]] = fpext <8 x float> [[TMP1]] to <8 x double>
 ; CHECK-NEXT:    ret <8 x double> [[R]]
 ;
   %x0 = fpext <4 x float> %a0 to <4 x double>
@@ -184,6 +181,3 @@ define <16 x i32> @concat_sext_zext_v8i16_v16i32(<8 x i16> %a0, <8 x i16> %a1) {
   %r = shufflevector <8 x i32> %x0, <8 x i32> %x1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
   ret <16 x i32> %r
 }
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; AVX: {{.*}}
-; SSE: {{.*}}

@github-actions GitHub Actions
Copy link

github-actions bot commented Apr 3, 2024 •
edited
Loading

✅ With the latest revision this PR passed the C/C++ code formatter.

goldsteinn
goldsteinn reviewed Apr 3, 2024
View reviewed changes
llvm/lib/Transforms/Vectorize/VectorCombine.cpp Outdated
return false;

Instruction::CastOps Opcode = C0->getOpcode();
if (Opcode == Instruction::BitCast || Opcode != C1->getOpcode() ||
Copy link
Contributor

@goldsteinn goldsteinn Apr 3, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

elementwise bitcast should be okay? (m_ElementWiseBitCast)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Eventually yes - except that we have other folds that fold bitcast(shuffle(x,y)) -> shuffle(bitcast(x),bitcast(y)) - I think what we're really after is whatever direction reduces the number of bitcasts, but we're not there yet :(

@RKSimon RKSimon force-pushed the shuffle_of_casts branch from a12447a to 160c237 Compare April 3, 2024 18:20
RKSimon added a commit that referenced this pull request Apr 3, 2024
@RKSimon
[VectorCombine][X86] Add additional tests for #87510
b15d27e 
Add zext nneg tests and check we don't fold casts with different src types
RKSimon added a commit that referenced this pull request Apr 3, 2024
@RKSimon
[VectorCombine][X86] shuffle-of-casts.ll - adjust zext nneg tests to …
d53b829 
…improve costs for testing

Improves SSE vs AVX test results for #87510
@RKSimon RKSimon force-pushed the shuffle_of_casts branch from 160c237 to cc8b45f Compare April 3, 2024 21:31
dtcxzyw added a commit to dtcxzyw/llvm-opt-benchmark that referenced this pull request Apr 4, 2024
@dtcxzyw
pre-commit: test PR87510
1ca7ddc 
PR Link: llvm/llvm-project#87510
@dtcxzyw dtcxzyw mentioned this pull request Apr 4, 2024
Closed
@RKSimon
[VectorCombine][X86] foldShuffleOfCastops - fold shuffle(cast(x),cast…
364e915 
…(y)) -> cast(shuffle(x,y)) iff cost efficient

Based off the existing foldShuffleOfBinops fold

Fixes llvm#67803
@RKSimon RKSimon force-pushed the shuffle_of_casts branch from cc8b45f to 364e915 Compare April 4, 2024 09:08
dtcxzyw
dtcxzyw approved these changes Apr 4, 2024
View reviewed changes
Copy link
Member

@dtcxzyw dtcxzyw left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp Show resolved Hide resolved
@RKSimon RKSimon merged commit 212b2bb into llvm:main Apr 4, 2024
3 of 4 checks passed
@RKSimon RKSimon deleted the shuffle_of_casts branch April 4, 2024 10:22
davemgreen added a commit to davemgreen/llvm-project that referenced this pull request Apr 7, 2024
@davemgreen
[AArch64] Add costs for ST3 and ST4 instructions, modelled as store(s…
2faddc9 
…huffle).

This tries to add some costs for the shuffle in a ST3/ST4 instruction, which
are represented in LLVM IR as store(interleaving shuffle). In order to detect
the store, it needs to add a CxtI context instruction to check the users of the
shuffle. LD3 and LD4 are added, LD2 should be a zip1 shuffle, which will be
added in another patch.

It should help fix some of the regressions from llvm#87510.
@davemgreen davemgreen mentioned this pull request Apr 7, 2024
Merged
davemgreen added a commit to davemgreen/llvm-project that referenced this pull request Apr 8, 2024
@davemgreen
[AArch64] Add costs for ST3 and ST4 instructions, modelled as store(s…
fcb91ed 
…huffle).

This tries to add some costs for the shuffle in a ST3/ST4 instruction, which
are represented in LLVM IR as store(interleaving shuffle). In order to detect
the store, it needs to add a CxtI context instruction to check the users of the
shuffle. LD3 and LD4 are added, LD2 should be a zip1 shuffle, which will be
added in another patch.

It should help fix some of the regressions from llvm#87510.
davemgreen added a commit that referenced this pull request Apr 9, 2024
@davemgreen
[AArch64] Add costs for ST3 and ST4 instructions, modelled as store(s…
4ac2721 
…huffle). (#87934)

This tries to add some costs for the shuffle in a ST3/ST4 instruction,
which are represented in LLVM IR as store(interleaving shuffle). In
order to detect the store, it needs to add a CxtI context instruction to
check the users of the shuffle. LD3 and LD4 are added, LD2 should be a
zip1 shuffle, which will be added in another patch.

It should help fix some of the regressions from #87510.
RKSimon added a commit that referenced this pull request Apr 11, 2024
@RKSimon
[VectorCombine][X86] Extend shuffle(bitcast(x),bitcast(y)) test coverage
32b95a3 
As discussed on #87510 the intention is to fold shuffle(bitcast(x),bitcast(y)) -> bitcast(shuffle(x,y)), but it must not interfere with existing bitcast(shuffle(bitcast(x),bitcast(y))) folds.
RKSimon added a commit that referenced this pull request Apr 11, 2024
@RKSimon
[VectorCombine][X86] Extend bitcast(shuffle(x,y)) test coverage
b60974d 
As discussed on #87510 the intention is only to fold bitcast(shuffle(x,y)) -> shuffle(bitcast(x),bitcast(y)) if we won't actually increase the number of bitcasts (i.e. x or y is already bitcasted from the correct type).
@davemgreen davemgreen mentioned this pull request Apr 15, 2024
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@goldsteinn goldsteinn goldsteinn left review comments

@dtcxzyw dtcxzyw dtcxzyw approved these changes

@nikic nikic Awaiting requested review from nikic

@phoebewang phoebewang Awaiting requested review from phoebewang

Assignees
No one assigned
Labels
llvm:transforms vectorizers
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

[VectorCombine][X86] Poor handling of compare-select patterns with AVX2 spoofing on AVX1 targets
4 participants
@RKSimon @llvmbot @dtcxzyw @goldsteinn