[VectorCombine] Add foldShuffleToIdentity #88693

davemgreen · 2024-04-15T07:56:33Z

This patch adds a basic version of a combine that attempts to remove shuffles that when combined simplify away to an identity shuffle. For example:
%ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
%at = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
%abt = fneg <4 x half> %at
%abb = fneg <4 x half> %ab
%r = shufflevector <4 x half> %abt, <4 x half> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
By looking through the shuffles and fneg, it can be simplified to:
%r = fneg <8 x half> %a

The code tracks each lane starting from the original shuffle, keeping a track of a vector of {src, idx}. As we propagate up through the instructions we will either look through intermediate instructions (binops and unops) or see a collections of lanes that all have the same src and incrementing idx (an identity). We can also see a single value with identical lanes, which we can treat like a splat.

Only the basic version is added here, handling identities, splats, binops and unops. In follow-up patches other instructions can be added such as constants, intrinsics, cmp/sel and zext/sext/trunc.

llvmbot · 2024-04-15T07:57:04Z

@llvm/pr-subscribers-llvm-transforms

Author: David Green (davemgreen)

Changes

This patch adds a basic version of a combine that attempts to remove shuffles that when combined simplify away to an identity shuffle. For example:
%ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
%at = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
%abt = fneg <4 x half> %at
%abb = fneg <4 x half> %ab
%r = shufflevector <4 x half> %abt, <4 x half> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
By looking through the shuffles and fneg, it can be simplified to:
%r = fneg <8 x half> %a

The code tracks each lane starting from the original shuffle, keeping a track of a vector of {src, idx}. As we propagate up through the instructions we will either look through intermediate instructions (binops and unops) or see a collections of lanes that all have the same src and incrementing idx (an identity). We can also see a single value with identical lanes, which we can treat like a splat.

Only the basic version is added here, handling identities, splats, binops and unops. In follow-up patches other instructions can be added such as constants, intrinsics, cmp/sel and zext/sext/trunc.

Patch is 27.66 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/88693.diff

4 Files Affected:

(modified) llvm/lib/Transforms/Vectorize/VectorCombine.cpp (+141)
(modified) llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll (+22-112)
(modified) llvm/test/Transforms/VectorCombine/X86/pr67803.ll (+1-4)
(modified) llvm/test/Transforms/VectorCombine/X86/select-shuffle.ll (+1-2)

diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index e0e2f50c89adad..12e37f67d8b264 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -113,6 +113,7 @@ class VectorCombine {
   bool scalarizeLoadExtract(Instruction &I);
   bool foldShuffleOfBinops(Instruction &I);
   bool foldShuffleOfCastops(Instruction &I);
+  bool foldShuffleToIdentity(Instruction &I);
   bool foldShuffleFromReductions(Instruction &I);
   bool foldTruncFromReductions(Instruction &I);
   bool foldSelectShuffle(Instruction &I, bool FromReduction = false);
@@ -1547,6 +1548,145 @@ bool VectorCombine::foldShuffleOfCastops(Instruction &I) {
   return true;
 }
 
+// Starting from a shuffle, look up through operands tracking the shuffled index
+// of each lane. If we can simplify away the shuffles to identities then
+// do so.
+bool VectorCombine::foldShuffleToIdentity(Instruction &I) {
+  FixedVectorType *Ty = dyn_cast<FixedVectorType>(I.getType());
+  if (!Ty || !isa<Instruction>(I.getOperand(0)) ||
+      !isa<Instruction>(I.getOperand(1)))
+    return false;
+
+  using InstLane = std::pair<Value *, int>;
+
+  auto LookThroughShuffles = [](Value *V, int Lane) -> InstLane {
+    while (auto *SV = dyn_cast<ShuffleVectorInst>(V)) {
+      unsigned NumElts =
+          cast<FixedVectorType>(SV->getOperand(0)->getType())->getNumElements();
+      int M = SV->getMaskValue(Lane);
+      if (M < 0)
+        return {nullptr, -1};
+      else if (M < (int)NumElts) {
+        V = SV->getOperand(0);
+        Lane = M;
+      } else {
+        V = SV->getOperand(1);
+        Lane = M - NumElts;
+      }
+    }
+    return InstLane{V, Lane};
+  };
+
+  auto GenerateInstLaneVectorFromOperand =
+      [&LookThroughShuffles](const SmallVector<InstLane> &Item, int Op) {
+        SmallVector<InstLane> NItem;
+        for (InstLane V : Item) {
+          NItem.emplace_back(
+              !V.first
+                  ? InstLane{nullptr, -1}
+                  : LookThroughShuffles(
+                        cast<Instruction>(V.first)->getOperand(Op), V.second));
+        }
+        return NItem;
+      };
+
+  SmallVector<InstLane> Start;
+  for (unsigned M = 0; M < Ty->getNumElements(); ++M)
+    Start.push_back(LookThroughShuffles(&I, M));
+
+  SmallVector<SmallVector<InstLane>> Worklist;
+  Worklist.push_back(Start);
+  SmallPtrSet<Value *, 4> IdentityLeafs, SplatLeafs;
+
+  while (!Worklist.empty()) {
+    SmallVector<InstLane> Item = Worklist.pop_back_val();
+
+    // If we found an undef first lane then bail out to keep things simple.
+    if (!Item[0].first)
+      return false;
+
+    // Look for an identity value.
+    if (Item[0].second == 0 && Item[0].first->getType() == Ty &&
+        all_of(drop_begin(enumerate(Item)), [&](const auto &E) {
+          return !E.value().first || (E.value().first == Item[0].first &&
+                                      E.value().second == (int)E.index());
+        })) {
+      IdentityLeafs.insert(Item[0].first);
+      continue;
+    }
+    // Look for a splat value.
+    if (all_of(drop_begin(Item), [&](InstLane &IL) {
+          return !IL.first ||
+                 (IL.first == Item[0].first && IL.second == Item[0].second);
+        })) {
+      SplatLeafs.insert(Item[0].first);
+      continue;
+    }
+
+    // We need each element to be the same type of value, and check that each
+    // element has a single use.
+    if (!all_of(drop_begin(Item), [&](InstLane IL) {
+          if (!IL.first)
+            return true;
+          if (isa<Instruction>(IL.first) &&
+              !cast<Instruction>(IL.first)->hasOneUse())
+            return false;
+          return IL.first->getValueID() == Item[0].first->getValueID() &&
+                 (!isa<IntrinsicInst>(IL.first) ||
+                  cast<IntrinsicInst>(IL.first)->getIntrinsicID() ==
+                      cast<IntrinsicInst>(Item[0].first)->getIntrinsicID());
+        }))
+      return false;
+
+    // Check the operator is one that we support.
+    if (isa<BinaryOperator>(Item[0].first)) {
+      Worklist.push_back(GenerateInstLaneVectorFromOperand(Item, 0));
+      Worklist.push_back(GenerateInstLaneVectorFromOperand(Item, 1));
+    } else if (isa<UnaryOperator>(Item[0].first)) {
+      Worklist.push_back(GenerateInstLaneVectorFromOperand(Item, 0));
+    } else {
+      return false;
+    }
+  }
+
+  // If we got this far, we know the shuffles are superfluous and can be
+  // removed. Scan through again and generate the new tree of instructions.
+  std::function<Value *(const SmallVector<InstLane> &)> generate =
+      [&](const SmallVector<InstLane> &Item) -> Value * {
+    if (IdentityLeafs.contains(Item[0].first) &&
+        all_of(drop_begin(enumerate(Item)), [&](const auto &E) {
+          return !E.value().first || (E.value().first == Item[0].first &&
+                                      E.value().second == (int)E.index());
+        })) {
+      return Item[0].first;
+    } else if (SplatLeafs.contains(Item[0].first)) {
+      if (auto ILI = dyn_cast<Instruction>(Item[0].first))
+        Builder.SetInsertPoint(*ILI->getInsertionPointAfterDef());
+      else if (isa<Argument>(Item[0].first))
+        Builder.SetInsertPointPastAllocas(I.getParent()->getParent());
+      SmallVector<int, 16> Mask(Ty->getNumElements(), Item[0].second);
+      return Builder.CreateShuffleVector(Item[0].first, Mask);
+    }
+
+    auto *I = cast<Instruction>(Item[0].first);
+    SmallVector<Value *> Ops;
+    unsigned E = I->getNumOperands();
+    for (unsigned Idx = 0; Idx < E; Idx++)
+      Ops.push_back(generate(GenerateInstLaneVectorFromOperand(Item, Idx)));
+    Builder.SetInsertPoint(I);
+    if (auto BI = dyn_cast<BinaryOperator>(I))
+      return Builder.CreateBinOp((Instruction::BinaryOps)BI->getOpcode(),
+                                 Ops[0], Ops[1]);
+    if (auto UI = dyn_cast<UnaryOperator>(I))
+      return Builder.CreateUnOp((Instruction::UnaryOps)UI->getOpcode(), Ops[0]);
+    llvm_unreachable("Unhandled instruction in generate");
+  };
+
+  Value *V = generate(Start);
+  replaceValue(I, *V);
+  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.
@@ -2103,6 +2243,7 @@ bool VectorCombine::run() {
         MadeChange |= foldShuffleOfBinops(I);
         MadeChange |= foldShuffleOfCastops(I);
         MadeChange |= foldSelectShuffle(I);
+        MadeChange |= foldShuffleToIdentity(I);
         break;
       case Instruction::BitCast:
         MadeChange |= foldBitcastShuffle(I);
diff --git a/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll b/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
index d96dfec849167d..47f52a341df24f 100644
--- a/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
+++ b/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
@@ -5,10 +5,7 @@ target triple = "aarch64"
 
 define <8 x i8> @trivial(<8 x i8> %a) {
 ; CHECK-LABEL: @trivial(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i8> [[AT]], <4 x i8> [[AB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    ret <8 x i8> [[R]]
+; CHECK-NEXT:    ret <8 x i8> [[R:%.*]]
 ;
   %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
   %at = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
@@ -18,13 +15,7 @@ define <8 x i8> @trivial(<8 x i8> %a) {
 
 define <8 x i8> @add(<8 x i8> %a, <8 x i8> %b) {
 ; CHECK-LABEL: @add(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BB:%.*]] = shufflevector <8 x i8> [[B:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[BT:%.*]] = shufflevector <8 x i8> [[B]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[ABT:%.*]] = add <4 x i8> [[AT]], [[BT]]
-; CHECK-NEXT:    [[ABB:%.*]] = add <4 x i8> [[AB]], [[BB]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i8> [[ABT]], <4 x i8> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[R:%.*]] = add <8 x i8> [[A:%.*]], [[B:%.*]]
 ; CHECK-NEXT:    ret <8 x i8> [[R]]
 ;
   %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -83,13 +74,7 @@ define <8 x i8> @wrong_lanes(<8 x i8> %a, <8 x i8> %b) {
 
 define <8 x half> @fadd(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @fadd(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x half> [[A]], <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BB:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[BT:%.*]] = shufflevector <8 x half> [[B]], <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[ABT:%.*]] = fadd <4 x half> [[AT]], [[BT]]
-; CHECK-NEXT:    [[ABB:%.*]] = fadd <4 x half> [[AB]], [[BB]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[ABT]], <4 x half> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[A:%.*]], [[B:%.*]]
 ; CHECK-NEXT:    ret <8 x half> [[R]]
 ;
   %ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -104,11 +89,7 @@ define <8 x half> @fadd(<8 x half> %a, <8 x half> %b) {
 
 define <8 x half> @fneg(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @fneg(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x half> [[A]], <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[ABT:%.*]] = fneg <4 x half> [[AT]]
-; CHECK-NEXT:    [[ABB:%.*]] = fneg <4 x half> [[AB]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[ABT]], <4 x half> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[R:%.*]] = fneg <8 x half> [[A:%.*]]
 ; CHECK-NEXT:    ret <8 x half> [[R]]
 ;
   %ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -138,12 +119,8 @@ define <8 x i8> @abs(<8 x i8> %a) {
 
 define <8 x half> @splat0(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @splat0(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x half> [[A]], <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BS:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[ABT:%.*]] = fadd <4 x half> [[AT]], [[BS]]
-; CHECK-NEXT:    [[ABB:%.*]] = fadd <4 x half> [[AB]], [[BS]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[ABT]], <4 x half> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret <8 x half> [[R]]
 ;
   %ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -157,12 +134,8 @@ define <8 x half> @splat0(<8 x half> %a, <8 x half> %b) {
 
 define <8 x half> @splat2(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @splat2(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x half> [[A]], <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BS:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <4 x i32> <i32 2, i32 2, i32 2, i32 2>
-; CHECK-NEXT:    [[ABT:%.*]] = fadd <4 x half> [[AT]], [[BS]]
-; CHECK-NEXT:    [[ABB:%.*]] = fadd <4 x half> [[AB]], [[BS]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[ABT]], <4 x half> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <8 x i32> <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret <8 x half> [[R]]
 ;
   %ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -176,12 +149,8 @@ define <8 x half> @splat2(<8 x half> %a, <8 x half> %b) {
 
 define <8 x half> @splatandidentity(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @splatandidentity(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x half> [[A]], <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BS:%.*]] = shufflevector <8 x half> [[A]], <8 x half> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[ABT:%.*]] = fadd <4 x half> [[AT]], [[BS]]
-; CHECK-NEXT:    [[ABB:%.*]] = fadd <4 x half> [[AB]], [[BS]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[ABT]], <4 x half> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[A]], [[TMP1]]
 ; CHECK-NEXT:    ret <8 x half> [[R]]
 ;
   %ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -195,11 +164,9 @@ define <8 x half> @splatandidentity(<8 x half> %a, <8 x half> %b) {
 
 define <8 x half> @splattwice(<8 x half> %a, <8 x half> %b) {
 ; CHECK-LABEL: @splattwice(
-; CHECK-NEXT:    [[AS:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[BS:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[AB1:%.*]] = fadd <4 x half> [[AS]], [[BS]]
-; CHECK-NEXT:    [[AB2:%.*]] = fadd <4 x half> [[AS]], [[BS]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[AB1]], <4 x half> [[AB2]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[TMP2]], [[TMP1]]
 ; CHECK-NEXT:    ret <8 x half> [[R]]
 ;
   %as = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> zeroinitializer
@@ -212,13 +179,7 @@ define <8 x half> @splattwice(<8 x half> %a, <8 x half> %b) {
 
 define <8 x i8> @undeflane(<8 x i8> %a, <8 x i8> %b) {
 ; CHECK-LABEL: @undeflane(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BB:%.*]] = shufflevector <8 x i8> [[B:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[BT:%.*]] = shufflevector <8 x i8> [[B]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[ABT:%.*]] = add <4 x i8> [[AT]], [[BT]]
-; CHECK-NEXT:    [[ABB:%.*]] = add <4 x i8> [[AB]], [[BB]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i8> [[ABT]], <4 x i8> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 poison, i32 1, i32 0>
+; CHECK-NEXT:    [[R:%.*]] = add <8 x i8> [[A:%.*]], [[B:%.*]]
 ; CHECK-NEXT:    ret <8 x i8> [[R]]
 ;
   %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -284,18 +245,9 @@ define <8 x i8> @constantdiff2(<8 x i8> %a) {
 
 define <8 x i8> @inner_shuffle(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c) {
 ; CHECK-LABEL: @inner_shuffle(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:    [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT:    [[BB:%.*]] = shufflevector <8 x i8> [[B:%.*]], <8 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT:    [[BT:%.*]] = shufflevector <8 x i8> [[B]], <8 x i8> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT:    [[CS:%.*]] = shufflevector <8 x i8> [[C:%.*]], <8 x i8> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT:    [[ABT:%.*]] = mul <4 x i8> [[AT]], [[BT]]
-; CHECK-NEXT:    [[ABB:%.*]] = mul <4 x i8> [[AB]], [[BB]]
-; CHECK-NEXT:    [[ABT2:%.*]] = shufflevector <4 x i8> [[ABT]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[ABB2:%.*]] = shufflevector <4 x i8> [[ABB]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[ABT3:%.*]] = add <4 x i8> [[ABT2]], [[CS]]
-; CHECK-NEXT:    [[ABB3:%.*]] = add <4 x i8> [[ABB2]], [[CS]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i8> [[ABT3]], <4 x i8> [[ABB3]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[C:%.*]], <8 x i8> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = mul <8 x i8> [[A:%.*]], [[B:%.*]]
+; CHECK-NEXT:    [[R:%.*]] = add <8 x i8> [[TMP2]], [[TMP1]]
 ; CHECK-NEXT:    ret <8 x i8> [[R]]
 ;
   %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
@@ -338,14 +290,9 @@ define <8 x i8> @extrause_add(<8 x i8> %a, <8 x i8> %b) {
 
 define <8 x i8> @extrause_shuffle(<8 x i8> %a, <8 x i8> %b) {
 ; CHECK-LABEL: @extrause_shuffle(
-; CHECK-NEXT:    [[AB:%.*]] = shufflevector <8 x i8> [[B:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[BT:%.*]] = shufflevector <8 x i8> [[B]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
-; CHECK-NEXT:    [[BB:%.*]] = shufflevector <8 x i8> [[B1:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[BT1:%.*]] = shufflevector <8 x i8> [[B1]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
+; CHECK-NEXT:    [[BT1:%.*]] = shufflevector <8 x i8> [[B1:%.*]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
 ; CHECK-NEXT:    call void @use(<4 x i8> [[BT1]])
-; CHECK-NEXT:    [[ABT:%.*]] = add <4 x i8> [[BT]], [[BT1]]
-; CHECK-NEXT:    [[ABB:%.*]] = add <4 x i8> [[AB]], [[BB]]
-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x i8> [[ABT]], <4 x i8> [[ABB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[R:%.*]] = add <8 x i8> [[A:%.*]], [[B1]]
 ; CHECK-NEXT:    ret <8 x i8> [[R]]
 ;
   %ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -543,52 +490,15 @@ define void @v8f64interleave(i64 %0, ptr %1, ptr %x, double %z) {
 ; CHECK-LABEL: @v8f64interleave(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[Z:%.*]], i64 0
-; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x double> [[BROADCAST_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x double> [[BROADCAST_SPLATINSERT]], <2 x double> poison, <16 x i32> zeroinitializer
 ; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x double>, ptr [[TMP1:%.*]], align 8
-; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x double> [[WIDE_VEC]], <16 x double> poison, <2 x i32> <i32 0, i32 8>
-; CHECK-NEXT:    [[STRIDED_VEC27:%.*]] = shufflevector <16 x double> [[WIDE_VEC]], <16 x double> poison, <2 x i32> <i32 1, i32 9>
-; CHECK-NEXT:    [[STRIDE...
[truncated]

RKSimon · 2024-04-15T10:08:08Z

Funnily enough, I have started work on a VectorCombine::foldShuffleOfShuffles fold that will sort of do this (and catch many of these test cases):

shuffle(shuffle(x,u,m1),shuffle(y,u,m2)) -> shuffle(x,y,m3)

But it doesn't recurse, so every fold has to be cost beneficial - are you seeing cases where we need the recursion capability?

davemgreen · 2024-04-16T13:11:50Z

Funnily enough, I have started work on a VectorCombine::foldShuffleOfShuffles fold that will sort of do this (and catch many of these test cases):

shuffle(shuffle(x,u,m1),shuffle(y,u,m2)) -> shuffle(x,y,m3)

But it doesn't recurse, so every fold has to be cost beneficial - are you seeing cases where we need the recursion capability?

Hi - I get the feeling that X86 has generally better shuffle support than other architectures. Even for AArch64, which generally does OK, there can be a big difference between a low-cost shuffle and a bad one. For something like MVE which is more constrained, the differences can be a lot larger often falling back to scalarization where it can't do much better.

It means we can't really take local steps and get to the optimal solution, because each step on it's own is worse for performance. It's only taken together that we end up with something better. It is not the motivating case but one that can come up a fair amount - consider the LD4/ST4 vectorization that we do for interleaving loads/stores. The load is various shuffle(loads) and the store a store(shuffle(shuffle, shuffle)). If the intervening instructions are all the same then the whole thing can be simplified away to continuous loads/stores, but a single shuffle moved only breaks the pattern for LD4/ST4. I think the v8f64interleave test shows something similar, but it involves splat'd vectors and wider interleaving groups.

I wanted to add some recursion limit to this code, to make sure it doesn't go too wrong. I will try and look into why #88743 causes regressions too.

RKSimon · 2024-04-16T14:22:02Z

Yes - we've never come up with a good way to accumulate candidate folds and only commit them if/when it becomes cost effective.

Please don't think I want to just take #88743 + #88899 instead of this, I think they're both needed, I just want to make sure we don't end up with a heavy fold constantly doing too much of the combining.

davemgreen · 2024-04-17T09:19:06Z

The trivial cases were just added for testing - many of them are already handled by instcombine. Some of those combines in instcombine are the ones causing the problem, as they already move shuffles through binops without considering the costs and breaking the canonical patterns for interleaving loads.

We were previously trying to fix that in the InterleavedAccess (through replaceBinOpShuffles), but that starts to become impossible as more and more folds happen in the mid-end.

That is separate from this patch though, which shouldn't create the same problems of breaking interleaving patterns without replacing them with identity shuffles, which should always be an improvement.

alexey-bataev · 2024-04-17T11:59:47Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+// of each lane. If we can simplify away the shuffles to identities then
+// do so.
+bool VectorCombine::foldShuffleToIdentity(Instruction &I) {
+  FixedVectorType *Ty = dyn_cast<FixedVectorType>(I.getType());


Suggested change

FixedVectorType *Ty = dyn_cast<FixedVectorType>(I.getType());

auto *Ty = dyn_cast<FixedVectorType>(I.getType());

alexey-bataev · 2024-04-17T12:00:31Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+          cast<FixedVectorType>(SV->getOperand(0)->getType())->getNumElements();
+      int M = SV->getMaskValue(Lane);
+      if (M < 0)
+        return {nullptr, -1};


Suggested change

return {nullptr, -1};

return {nullptr, PoisonMaskElem};

alexey-bataev · 2024-04-17T12:01:13Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+  };
+
+  auto GenerateInstLaneVectorFromOperand =
+      [&LookThroughShuffles](const SmallVector<InstLane> &Item, int Op) {


Suggested change

[&LookThroughShuffles](const SmallVector<InstLane> &Item, int Op) {

[&LookThroughShuffles](ArrayRef<InstLane> Item, int Op) {

alexey-bataev · 2024-04-17T12:01:38Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+        for (InstLane V : Item) {
+          NItem.emplace_back(
+              !V.first
+                  ? InstLane{nullptr, -1}


Suggested change

? InstLane{nullptr, -1}

? InstLane{nullptr, PoisonMaskElem}

alexey-bataev · 2024-04-17T12:03:13Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+    if (!all_of(drop_begin(Item), [&](InstLane IL) {
+          if (!IL.first)
+            return true;
+          if (isa<Instruction>(IL.first) &&


Use dyn_cast

alexey-bataev · 2024-04-17T12:04:43Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+
+  // If we got this far, we know the shuffles are superfluous and can be
+  // removed. Scan through again and generate the new tree of instructions.
+  std::function<Value *(const SmallVector<InstLane> &)> generate =


Suggested change

std::function<Value *(const SmallVector<InstLane> &)> generate =

auto Generate =

alexey-bataev · 2024-04-17T12:05:02Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+  // If we got this far, we know the shuffles are superfluous and can be
+  // removed. Scan through again and generate the new tree of instructions.
+  std::function<Value *(const SmallVector<InstLane> &)> generate =
+      [&](const SmallVector<InstLane> &Item) -> Value * {


Suggested change

[&](const SmallVector<InstLane> &Item) -> Value * {

[&](ArrayRef<InstLane> Item) -> Value * {

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

alexey-bataev · 2024-04-22T11:07:21Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+          if (auto *I = dyn_cast<Instruction>(IL.first); I && !I->hasOneUse())
+            return false;
+          return IL.first->getValueID() == Item[0].first->getValueID() &&
+                 (!isa<IntrinsicInst>(IL.first) ||
+                  cast<IntrinsicInst>(IL.first)->getIntrinsicID() ==
+                      cast<IntrinsicInst>(Item[0].first)->getIntrinsicID());


Suggested change

if (auto *I = dyn_cast<Instruction>(IL.first); I && !I->hasOneUse())

return false;

return IL.first->getValueID() == Item[0].first->getValueID() &&

(!isa<IntrinsicInst>(IL.first) ||

cast<IntrinsicInst>(IL.first)->getIntrinsicID() ==

cast<IntrinsicInst>(Item[0].first)->getIntrinsicID());

auto *I = dyn_cast<Instruction>(IL.first);

if (I && !I->hasOneUse())

return false;

return IL.first->getValueID() == Item[0].first->getValueID() &&

(!I || I->getIntrinsicID() ==

cast<IntrinsicInst>(Item[0].first)->getIntrinsicID());

I've changed it to something like this. The second one is an IntrinsicInst, to make sure the types of the intrinsics are the same.

alexey-bataev · 2024-04-22T11:07:53Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+  std::function<Value *(ArrayRef<InstLane>)> generate =
+      [&](ArrayRef<InstLane> Item) -> Value * {


Suggested change

std::function<Value *(ArrayRef<InstLane>)> generate =

[&](ArrayRef<InstLane> Item) -> Value * {

auto Generate =

[&](ArrayRef<InstLane> Item) -> Value * {

This is called inside the lambda, so needs a type up-front.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

alexey-bataev · 2024-04-22T11:13:22Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+    if (auto UI = dyn_cast<UnaryOperator>(I))
+      return Builder.CreateUnOp((Instruction::UnaryOps)UI->getOpcode(), Ops[0]);
+    llvm_unreachable("Unhandled instruction in generate");


Suggested change

if (auto UI = dyn_cast<UnaryOperator>(I))

return Builder.CreateUnOp((Instruction::UnaryOps)UI->getOpcode(), Ops[0]);

llvm_unreachable("Unhandled instruction in generate");

assert(isa<UnaryOperator>(I) && "Expected either Unary or Binary Op.");

return Builder.CreateUnOp((Instruction::UnaryOps)I->getOpcode(), Ops[0]);

I was planning to add a number of other types added here in follow patches. This one just contains the basics to make sure the general ideal is working OK. I can make it so that UnaryInstructions are left as the else (even if I personally prefer the structure of keeping them all consistent :) I find it more readable. )

… shuffles (#88899) Refactor to be closer to foldShuffleOfCastops - sibling patch to #88743 that can be used to address some of the issues identified in #88693

RKSimon

Do you intend to support intrinsics at some point? I notice you have a abs test

RKSimon · 2024-04-24T13:17:52Z

llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll

-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[AB1]], <4 x half> [[AB2]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[TMP2]], [[TMP1]]


This looks like we might need a foldBinopOfShuffles as well......

All these tests were just added as examples for testing this combine, many of the smaller ones like this are already optimized by instcombine. They were only added to make sure we were testing the different cases that can come up when folding the shuffles to identities.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

davemgreen

Do you intend to support intrinsics at some point? I notice you have a abs test

Yep, those and a few other operations. I wanted to make sure the basics were working before adding the others but they are hopefully relatively straight forward.

davemgreen · 2024-04-22T13:33:56Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+          if (auto *I = dyn_cast<Instruction>(IL.first); I && !I->hasOneUse())
+            return false;
+          return IL.first->getValueID() == Item[0].first->getValueID() &&
+                 (!isa<IntrinsicInst>(IL.first) ||
+                  cast<IntrinsicInst>(IL.first)->getIntrinsicID() ==
+                      cast<IntrinsicInst>(Item[0].first)->getIntrinsicID());


I've changed it to something like this. The second one is an IntrinsicInst, to make sure the types of the intrinsics are the same.

davemgreen · 2024-04-22T13:34:11Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+  std::function<Value *(ArrayRef<InstLane>)> generate =
+      [&](ArrayRef<InstLane> Item) -> Value * {


This is called inside the lambda, so needs a type up-front.

davemgreen · 2024-04-22T13:34:53Z

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

+    if (auto UI = dyn_cast<UnaryOperator>(I))
+      return Builder.CreateUnOp((Instruction::UnaryOps)UI->getOpcode(), Ops[0]);
+    llvm_unreachable("Unhandled instruction in generate");


I was planning to add a number of other types added here in follow patches. This one just contains the basics to make sure the general ideal is working OK. I can make it so that UnaryInstructions are left as the else (even if I personally prefer the structure of keeping them all consistent :) I find it more readable. )

davemgreen · 2024-04-24T13:46:48Z

llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll

-; CHECK-NEXT:    [[R:%.*]] = shufflevector <4 x half> [[AB1]], <4 x half> [[AB2]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x half> [[B:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x half> [[A:%.*]], <8 x half> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R:%.*]] = fadd <8 x half> [[TMP2]], [[TMP1]]


All these tests were just added as examples for testing this combine, many of the smaller ones like this are already optimized by instcombine. They were only added to make sure we were testing the different cases that can come up when folding the shuffles to identities.

llvm/lib/Transforms/Vectorize/VectorCombine.cpp

davemgreen · 2024-05-01T12:38:05Z

Rebased and ping. Thanks

RKSimon

Final question - do we need handling for DivRem instructions to make sure we don't introduce poison mask elements? see 4cc9c6d

This patch adds a basic version of a combine that attempts to fold away shuffles that when combines simplify away to an identity shuffle. For example: %ab = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> %at = shufflevector <8 x half> %a, <8 x half> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4> %abt = fneg <4 x half> %at %abb = fneg <4 x half> %ab %r = shufflevector <4 x half> %abt, <4 x half> %abb, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0> By looking through the shuffles, it can be simplified to: %r = fneg <8 x half> %a The code tracks each lane starting from the original shuffle, keeping a track of a vector of {src, idx}. As we propagate up through the instructions we will either look through intermediate instructions (binops and unops) or see a collections of lanes that all have the same src and incrementing idx (an identity). We can also see a single value with identical lanes, which we can treat like a splat. Only the basic version is added here, handling identites, splats, binops and unops. In follow-up patches other instructions can be added such as constants, intrinsics, cmp/sel and zext/sext/trunc.

davemgreen · 2024-05-02T14:28:07Z

Final question - do we need handling for DivRem instructions to make sure we don't introduce poison mask elements? see 4cc9c6d

Thanks - I hadn't considered div/rem as they don't as often come up for vectors. I can exclude them for now if that sounds OK, but as far as I understand they might be fine as we end up removing the shuffles so do not have poison lanes left over.
https://alive2.llvm.org/ce/z/NAyQty

RKSimon

LGTM - cheers

dyung · 2024-05-04T18:40:31Z

Hi @davemgreen, we are seeing an assertion failure in a few internal tests which I bisected back to your change. I have filed #91078 for the issue, can you take a look?

davemgreen · 2024-05-05T05:28:56Z

Thanks for the heads up - I'll take a look.

Zentrik · 2024-10-08T20:11:02Z

Hi @davemgreen, I'm seeing an assertion failure in foldShuffleToIdentity at #110382. I suspect this is due to the auto *Ty = dyn_cast<FixedVectorType>(I.getType()); at the beginning of the function but not sure. Could you take a look please?

davemgreen · 2024-10-09T05:22:16Z

Hi - yeah I'll take a look. Thanks for letting me know.

davemgreen requested review from RKSimon, rotateright, topperc and sjoerdmeijer April 15, 2024 07:56

llvmbot added vectorizers llvm:transforms labels Apr 15, 2024

RKSimon mentioned this pull request Apr 15, 2024

[VectorCombine] foldShuffleOfShuffles - fold "shuffle (shuffle x, undef), (shuffle y, undef)" -> "shuffle x, y" #88743

Merged

RKSimon mentioned this pull request Apr 16, 2024

[VectorCombine] foldShuffleOfBinops - add support for length changing shuffles #88899

Merged

davemgreen force-pushed the gh-shuffleToIdentity branch from 37dda86 to 3aacf81 Compare April 17, 2024 11:22

alexey-bataev reviewed Apr 17, 2024

View reviewed changes

davemgreen force-pushed the gh-shuffleToIdentity branch from 93685b2 to 7babf9e Compare April 22, 2024 07:42

alexey-bataev reviewed Apr 22, 2024

View reviewed changes

davemgreen force-pushed the gh-shuffleToIdentity branch from 7babf9e to 839e2c1 Compare April 22, 2024 13:37

davemgreen force-pushed the gh-shuffleToIdentity branch from 839e2c1 to 7417cd3 Compare April 24, 2024 12:33

RKSimon reviewed Apr 24, 2024

View reviewed changes

davemgreen commented Apr 24, 2024

View reviewed changes

davemgreen force-pushed the gh-shuffleToIdentity branch from 7417cd3 to 26000de Compare May 1, 2024 12:37

RKSimon approved these changes May 1, 2024

View reviewed changes

davemgreen added 5 commits May 2, 2024 11:33

Address comments and fix failing test

3843999

Rebase and update phase-ordering tests

3aff7df

Address more comments

3306543

Rebase over other changes and fixup tests.

ee1e4df

Exclude div/rem

01af3d6

davemgreen force-pushed the gh-shuffleToIdentity branch from 26000de to 01af3d6 Compare May 2, 2024 14:28

RKSimon approved these changes May 2, 2024

View reviewed changes

davemgreen merged commit a4d1026 into llvm:main May 3, 2024
4 checks passed

davemgreen deleted the gh-shuffleToIdentity branch May 3, 2024 18:14

davemgreen mentioned this pull request May 18, 2024

InstSimplify: lookthru casts, binops in folding shuffles #92668

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[VectorCombine] Add foldShuffleToIdentity #88693

[VectorCombine] Add foldShuffleToIdentity #88693

davemgreen commented Apr 15, 2024

llvmbot commented Apr 15, 2024

RKSimon commented Apr 15, 2024

davemgreen commented Apr 16, 2024

RKSimon commented Apr 16, 2024

davemgreen commented Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 17, 2024

alexey-bataev Apr 22, 2024

davemgreen Apr 22, 2024

alexey-bataev Apr 22, 2024

davemgreen Apr 22, 2024

alexey-bataev Apr 22, 2024

davemgreen Apr 22, 2024

RKSimon left a comment

RKSimon Apr 24, 2024

davemgreen Apr 24, 2024

davemgreen left a comment

davemgreen Apr 22, 2024

davemgreen Apr 22, 2024

davemgreen Apr 22, 2024

davemgreen Apr 24, 2024

davemgreen commented May 1, 2024

RKSimon left a comment •

edited

Loading

davemgreen commented May 2, 2024

RKSimon left a comment

dyung commented May 4, 2024

davemgreen commented May 5, 2024

Zentrik commented Oct 8, 2024

davemgreen commented Oct 9, 2024

	FixedVectorType *Ty = dyn_cast<FixedVectorType>(I.getType());
	auto *Ty = dyn_cast<FixedVectorType>(I.getType());

	[&LookThroughShuffles](const SmallVector<InstLane> &Item, int Op) {
	[&LookThroughShuffles](ArrayRef<InstLane> Item, int Op) {

	std::function<Value *(const SmallVector<InstLane> &)> generate =
	auto Generate =

	[&](const SmallVector<InstLane> &Item) -> Value * {
	[&](ArrayRef<InstLane> Item) -> Value * {

		std::function<Value *(ArrayRef<InstLane>)> generate =
		[&](ArrayRef<InstLane> Item) -> Value * {

[VectorCombine] Add foldShuffleToIdentity #88693

[VectorCombine] Add foldShuffleToIdentity #88693

Conversation

davemgreen commented Apr 15, 2024

llvmbot commented Apr 15, 2024

RKSimon commented Apr 15, 2024

davemgreen commented Apr 16, 2024

RKSimon commented Apr 16, 2024

davemgreen commented Apr 17, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

RKSimon left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davemgreen left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davemgreen commented May 1, 2024

RKSimon left a comment • edited Loading

Choose a reason for hiding this comment

davemgreen commented May 2, 2024

RKSimon left a comment

Choose a reason for hiding this comment

dyung commented May 4, 2024

davemgreen commented May 5, 2024

Zentrik commented Oct 8, 2024

davemgreen commented Oct 9, 2024

RKSimon left a comment •

edited

Loading