Skip to content

Conversation

@fhahn
Copy link
Contributor

@fhahn fhahn commented Aug 11, 2025

Use VPIRMetadata for VPInterleaveRecipe to preserve noalias metadata added by versioning.

This still uses InterleaveGroup's logic to preserve existing metadata from IR. This can be migrated separately.

Fixes #153006.

@llvmbot
Copy link
Member

llvmbot commented Aug 11, 2025

@llvm/pr-subscribers-llvm-transforms

Author: Florian Hahn (fhahn)

Changes

Use VPIRMetadata for VPInterleaveRecipe to preserve noalias metadata added by versioning.

This still uses InterleaveGroup's logic to preserve existing metadata from IR. This can be migrated separately.

Fixes #153006.


Patch is 24.48 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/153084.diff

6 Files Affected:

  • (modified) llvm/lib/Transforms/Vectorize/VPlan.h (+7-6)
  • (modified) llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp (+15)
  • (modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+18-9)
  • (modified) llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll (+33-26)
  • (modified) llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll (+17-15)
  • (modified) llvm/unittests/Transforms/Vectorize/VPlanTest.cpp (+1-1)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 5c3edee13020f..95acf156b1648 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -938,6 +938,9 @@ class VPIRMetadata {
   void addMetadata(unsigned Kind, MDNode *Node) {
     Metadata.emplace_back(Kind, Node);
   }
+
+  /// Intersect the this VPIRMetada objet with \p MD, keeping only metadata nodes in both.
+  void intersect(const VPIRMetadata &MD);
 };
 
 /// This is a concrete Recipe that models a single VPlan-level instruction.
@@ -2389,7 +2392,7 @@ class LLVM_ABI_FOR_TEST VPBlendRecipe : public VPSingleDefRecipe {
 /// or stores into one wide load/store and shuffles. The first operand of a
 /// VPInterleave recipe is the address, followed by the stored values, followed
 /// by an optional mask.
-class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase {
+class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase, public VPIRMetadata  {
   const InterleaveGroup<Instruction> *IG;
 
   /// Indicates if the interleave group is in a conditional block and requires a
@@ -2403,10 +2406,8 @@ class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase {
 public:
   VPInterleaveRecipe(const InterleaveGroup<Instruction> *IG, VPValue *Addr,
                      ArrayRef<VPValue *> StoredValues, VPValue *Mask,
-                     bool NeedsMaskForGaps, DebugLoc DL)
-      : VPRecipeBase(VPDef::VPInterleaveSC, {Addr},
-                     DL),
-
+                     bool NeedsMaskForGaps, const VPIRMetadata &MD, DebugLoc DL)
+      : VPRecipeBase(VPDef::VPInterleaveSC, {Addr}, DL), VPIRMetadata(MD),
         IG(IG), NeedsMaskForGaps(NeedsMaskForGaps) {
     // TODO: extend the masked interleaved-group support to reversed access.
     assert((!Mask || !IG->isReverse()) &&
@@ -2429,7 +2430,7 @@ class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase {
 
   VPInterleaveRecipe *clone() override {
     return new VPInterleaveRecipe(IG, getAddr(), getStoredValues(), getMask(),
-                                  NeedsMaskForGaps, getDebugLoc());
+                                  NeedsMaskForGaps, *this,getDebugLoc());
   }
 
   VP_CLASSOF_IMPL(VPDef::VPInterleaveSC)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index e34cab117f321..910556d9c8589 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -1471,6 +1471,17 @@ void VPIRMetadata::applyMetadata(Instruction &I) const {
     I.setMetadata(Kind, Node);
 }
 
+void VPIRMetadata::intersect(const VPIRMetadata &Other) {
+  SmallVector<std::pair<unsigned, MDNode *>> MetadataUnion;
+  for (const auto &[KindA, MDA] : Metadata) {
+    for (const auto &[KindB, MDB] : Other.Metadata) {
+      if (KindA == KindB && MDA == MDB)
+        MetadataUnion.emplace_back(KindA, MDA);
+    }
+  }
+  Metadata = std::move(MetadataUnion);
+}
+
 void VPWidenCallRecipe::execute(VPTransformState &State) {
   assert(State.VF.isVector() && "not widening");
   assert(Variant != nullptr && "Can't create vector function.");
@@ -3512,6 +3523,8 @@ void VPInterleaveRecipe::execute(VPTransformState &State) {
     } else
       NewLoad = State.Builder.CreateAlignedLoad(VecTy, ResAddr,
                                                 Group->getAlign(), "wide.vec");
+    applyMetadata(*NewLoad);
+    // TODO: Also manage existing metadata using VPIRMetadata.
     Group->addMetadata(NewLoad);
 
     ArrayRef<VPValue *> VPDefs = definedValues();
@@ -3616,6 +3629,8 @@ void VPInterleaveRecipe::execute(VPTransformState &State) {
     NewStoreInstr =
         State.Builder.CreateAlignedStore(IVec, ResAddr, Group->getAlign());
 
+  applyMetadata(*NewStoreInstr);
+  // TODO: Also manage existing metadata using VPIRMetadata.
   Group->addMetadata(NewStoreInstr);
 }
 
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 9eeecb378a9fc..da33e7bd2e8e0 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2561,12 +2561,23 @@ void VPlanTransforms::createInterleaveGroups(
   VPDominatorTree VPDT;
   VPDT.recalculate(Plan);
   for (const auto *IG : InterleaveGroups) {
-    SmallVector<VPValue *, 4> StoredValues;
-    for (unsigned i = 0; i < IG->getFactor(); ++i)
-      if (auto *SI = dyn_cast_or_null<StoreInst>(IG->getMember(i))) {
-        auto *StoreR = cast<VPWidenStoreRecipe>(RecipeBuilder.getRecipe(SI));
+    // Get or create the start address for the interleave group.
+    auto *Start =
+        cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(IG->getMember(0)));
+
+   VPIRMetadata InterleaveMD(*Start);
+   SmallVector<VPValue *, 4> StoredValues;
+    for (unsigned i = 0; i < IG->getFactor(); ++i) {
+      Instruction *MemI = IG->getMember(i);
+      if (!MemI)
+        continue;
+      VPWidenMemoryRecipe *MemR = cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(MemI));
+      if (!MemR)
+        continue;
+      if (auto *StoreR = dyn_cast<VPWidenStoreRecipe>(MemR))
         StoredValues.push_back(StoreR->getStoredValue());
-      }
+      InterleaveMD.intersect(*MemR);
+    }
 
     bool NeedsMaskForGaps =
         (IG->requiresScalarEpilogue() && !ScalarEpilogueAllowed) ||
@@ -2581,9 +2592,6 @@ void VPlanTransforms::createInterleaveGroups(
             getLoadStorePointerOperand(IRInsertPos)->stripPointerCasts()))
       NW = Gep->getNoWrapFlags().withoutNoUnsignedWrap();
 
-    // Get or create the start address for the interleave group.
-    auto *Start =
-        cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(IG->getMember(0)));
     VPValue *Addr = Start->getAddr();
     VPRecipeBase *AddrDef = Addr->getDefiningRecipe();
     if (AddrDef && !VPDT.properlyDominates(AddrDef, InsertPos)) {
@@ -2618,8 +2626,9 @@ void VPlanTransforms::createInterleaveGroups(
       ReversePtr->insertBefore(InsertPos);
       Addr = ReversePtr;
     }
+
     auto *VPIG = new VPInterleaveRecipe(IG, Addr, StoredValues,
-                                        InsertPos->getMask(), NeedsMaskForGaps, InsertPos->getDebugLoc());
+                                        InsertPos->getMask(), NeedsMaskForGaps, InterleaveMD, InsertPos->getDebugLoc());
     VPIG->insertBefore(InsertPos);
 
     unsigned J = 0;
diff --git a/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll b/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
index 6d562be03a9b0..41b7488cc7300 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
@@ -146,7 +146,7 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
 ; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 [[TMP9]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = shl i64 [[INDEX]], 4
 ; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 [[TMP11]]
-; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x float>, ptr [[TMP26]], align 4
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x float>, ptr [[TMP26]], align 4, !alias.scope [[META4:![0-9]+]]
 ; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 0, i32 8>
 ; CHECK-NEXT:    [[STRIDED_VEC14:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 1, i32 9>
 ; CHECK-NEXT:    [[STRIDED_VEC15:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 2, i32 10>
@@ -167,10 +167,10 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
 ; CHECK-NEXT:    [[TMP41:%.*]] = shufflevector <2 x float> [[TMP35]], <2 x float> [[TMP37]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:    [[TMP42:%.*]] = shufflevector <4 x float> [[TMP40]], <4 x float> [[TMP41]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <8 x float> [[TMP42]], <8 x float> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>
-; CHECK-NEXT:    store <8 x float> [[INTERLEAVED_VEC]], ptr [[TMP28]], align 4
+; CHECK-NEXT:    store <8 x float> [[INTERLEAVED_VEC]], ptr [[TMP28]], align 4, !alias.scope [[META7:![0-9]+]], !noalias [[META4]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP43]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT:    br i1 [[TMP43]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
 ; CHECK-NEXT:    br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -216,7 +216,7 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
 ; CHECK-NEXT:    store float [[MUL_4]], ptr [[GEP_11]], align 4
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], [[ARG1]]
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP10:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -397,12 +397,12 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
 ; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
 ; CHECK-NEXT:    [[TMP51:%.*]] = lshr exact i64 [[OFFSET_IDX]], 1
 ; CHECK-NEXT:    [[TMP52:%.*]] = getelementptr nusw i32, ptr [[B]], i64 [[TMP51]]
-; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP52]], align 4
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP52]], align 4, !alias.scope [[META11:![0-9]+]], !noalias [[META14:![0-9]+]]
 ; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
 ; CHECK-NEXT:    [[STRIDED_VEC34:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
 ; CHECK-NEXT:    [[TMP56:%.*]] = getelementptr i32, ptr [[A]], i64 [[OFFSET_IDX]]
 ; CHECK-NEXT:    [[TMP54:%.*]] = getelementptr i32, ptr [[B]], <4 x i64> [[VEC_IND]]
-; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> [[TMP54]], i32 4, <4 x i1> splat (i1 true), <4 x i32> poison), !alias.scope [[META6:![0-9]+]], !noalias [[META9:![0-9]+]]
+; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> [[TMP54]], i32 4, <4 x i1> splat (i1 true), <4 x i32> poison), !alias.scope [[META16:![0-9]+]], !noalias [[META14]]
 ; CHECK-NEXT:    [[TMP58:%.*]] = shufflevector <4 x i32> [[STRIDED_VEC]], <4 x i32> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:    [[TMP59:%.*]] = shufflevector <4 x i32> [[STRIDED_VEC34]], <4 x i32> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <4 x i32> [[WIDE_MASKED_GATHER]], <4 x i32> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -410,11 +410,11 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
 ; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <8 x i32> [[TMP60]], <8 x i32> zeroinitializer, <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:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP61]], <16 x i32> [[TMP62]], <32 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, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
 ; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <32 x i32> [[TMP63]], <32 x i32> poison, <32 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>
-; CHECK-NEXT:    store <32 x i32> [[INTERLEAVED_VEC]], ptr [[TMP56]], align 4
+; CHECK-NEXT:    store <32 x i32> [[INTERLEAVED_VEC]], ptr [[TMP56]], align 4, !alias.scope [[META14]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 32)
 ; CHECK-NEXT:    [[TMP64:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP64]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
+; CHECK-NEXT:    br i1 [[TMP64]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[SCALAR_PH]]
 ; CHECK:       [[SCALAR_PH]]:
@@ -455,7 +455,7 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
 ; CHECK-NEXT:    store i32 0, ptr [[GEP_A_7]], align 4
 ; CHECK-NEXT:    [[IV_NEXT_7]] = add nuw nsw i64 [[IV]], 8
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], [[N]]
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP12:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP19:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -521,7 +521,7 @@ define void @interleave_store_double_i64(ptr %dst) {
 ; CHECK-NEXT:    store <4 x double> [[INTERLEAVED_VEC]], ptr [[TMP1]], align 8
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
-; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
+; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[EXIT:.*]]
 ; CHECK:       [[SCALAR_PH]]:
@@ -535,7 +535,7 @@ define void @interleave_store_double_i64(ptr %dst) {
 ; CHECK-NEXT:    store double 0.000000e+00, ptr [[GEP_0]], align 8
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], 1
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP14:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP21:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -642,7 +642,7 @@ define void @interleave_store_i64_double_2(ptr %dst) {
 ; CHECK-NEXT:    store <4 x double> [[INTERLEAVED_VEC]], ptr [[TMP1]], align 8
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
-; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
+; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[EXIT:.*]]
 ; CHECK:       [[SCALAR_PH]]:
@@ -656,7 +656,7 @@ define void @interleave_store_i64_double_2(ptr %dst) {
 ; CHECK-NEXT:    store double 0.000000e+00, ptr [[GEP_1]], align 8
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], 1
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP23:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -687,17 +687,24 @@ attributes #1 = { "min-legal-vector-width"="0" "target-cpu"="cascadelake" }
 ; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
 ; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
 ; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
-; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
-; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]]}
-; CHECK: [[META6]] = !{[[META7:![0-9]+]]}
-; CHECK: [[META7]] = distinct !{[[META7]], [[META8:![0-9]+]]}
-; CHECK: [[META8]] = distinct !{[[META8]], !"LVerDomain"}
-; CHECK: [[META9]] = !{[[META10:![0-9]+]]}
-; CHECK: [[META10]] = distinct !{[[META10]], [[META8]]}
-; CHECK: [[LOOP11]] = distinct !{[[LOOP11]], [[META1]], [[META2]]}
-; CHECK: [[LOOP12]] = distinct !{[[LOOP12]], [[META1]]}
-; CHECK: [[LOOP13]] = distinct !{[[LOOP13]], [[META1]], [[META2]]}
-; CHECK: [[LOOP14]] = distinct !{[[LOOP14]], [[META2]], [[META1]]}
-; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]], [[META2]]}
-; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META2]], [[META1]]}
+; CHECK: [[META4]] = !{[[META5:![0-9]+]]}
+; CHECK: [[META5]] = distinct !{[[META5]], [[META6:![0-9]+]]}
+; CHECK: [[META6]] = distinct !{[[META6]], !"LVerDomain"}
+; CHECK: [[META7]] = !{[[META8:![0-9]+]]}
+; CHECK: [[META8]] = distinct !{[[META8]], [[META6]]}
+; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META1]], [[META2]]}
+; CHECK: [[LOOP10]] = distinct !{[[LOOP10]], [[META1]]}
+; CHECK: [[META11]] = !{[[META12:![0-9]+]]}
+; CHECK: [[META12]] = distinct !{[[META12]], [[META13:![0-9]+]]}
+; CHECK: [[META13]] = distinct !{[[META13]], !"LVerDomain"}
+; CHECK: [[META14]] = !{[[META15:![0-9]+]]}
+; CHECK: [[META15]] = distinct !{[[META15]], [[META13]]}
+; CHECK: [[META16]] = !{[[META17:![0-9]+]]}
+; CHECK: [[META17]] = distinct !{[[META17]], [[META13]]}
+; CHECK: [[LOOP18]] = distinct !{[[LOOP18]], [[META1]], [[META2]]}
+; CHECK: [[LOOP19]] = distinct !{[[LOOP19]], [[META1]]}
+; CHECK: [[LOOP20]] = distinct !{[[LOOP20]], [[META1]], [[META2]]}
+; CHECK: [[LOOP21]] = distinct !{[[LOOP21]], [[META2]], [[META1]]}
+; CHECK: [[LOOP22]] = distinct !{[[LOOP22]], [[META1]], [[META2]]}
+; CHECK: [[LOOP23]] = distinct !{[[LOOP23]], [[META2]], [[META1]]}
 ;.
diff --git a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll
index 65148b0babcd7..9c102ae0dd9b7 100644
--- a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll
+++ b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll
@@ -130,17 +130,17 @@ define void @ir_tbaa_different(ptr %base, ptr %end, ptr %src) {
 ; CHECK-NEXT:    [[TMP11:%.*]] = load float, ptr [[SRC]], align 4, !alias.scope [[META14:![0-9]+]]
 ; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[TMP11]], i64 0
 ; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
-; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <4 x float>, ptr [[NEXT_GEP]], align 4
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <4 x float>, ptr [[NEXT_GEP]], align 4, !alias.scope [[META17:![0-9]+]], !noalias [[META14]]
 ; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <4 x float> [[WIDE_VEC]], <4 x float> poison, <2 x i32> <i32 0, i32 2>
 ; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <4 x float> [[WIDE_VEC]], <4 x float> poison, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:    [[TMP6:%.*]] = fmul <2 x float> [[STRIDED_VEC]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = fmul <2 x float> [[STRIDED_VEC3]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> [[TMP7]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x float> [[TMP8]], <4 x float> poison, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
-; CHECK-NEXT:    store <4 x float> [[INTERLEAVED_VEC]], ptr [[NEXT_GEP]], align 4
+; CHECK-NEXT:    store <4 x float> [[INTERLEAVED_VEC]], ptr [[NEXT_GEP]], align 4, !alias.scope [[META17]], !noalias [[META14]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]
+; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:   ...
[truncated]

@llvmbot
Copy link
Member

llvmbot commented Aug 11, 2025

@llvm/pr-subscribers-vectorizers

Author: Florian Hahn (fhahn)

Changes

Use VPIRMetadata for VPInterleaveRecipe to preserve noalias metadata added by versioning.

This still uses InterleaveGroup's logic to preserve existing metadata from IR. This can be migrated separately.

Fixes #153006.


Patch is 24.48 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/153084.diff

6 Files Affected:

  • (modified) llvm/lib/Transforms/Vectorize/VPlan.h (+7-6)
  • (modified) llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp (+15)
  • (modified) llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp (+18-9)
  • (modified) llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll (+33-26)
  • (modified) llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll (+17-15)
  • (modified) llvm/unittests/Transforms/Vectorize/VPlanTest.cpp (+1-1)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 5c3edee13020f..95acf156b1648 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -938,6 +938,9 @@ class VPIRMetadata {
   void addMetadata(unsigned Kind, MDNode *Node) {
     Metadata.emplace_back(Kind, Node);
   }
+
+  /// Intersect the this VPIRMetada objet with \p MD, keeping only metadata nodes in both.
+  void intersect(const VPIRMetadata &MD);
 };
 
 /// This is a concrete Recipe that models a single VPlan-level instruction.
@@ -2389,7 +2392,7 @@ class LLVM_ABI_FOR_TEST VPBlendRecipe : public VPSingleDefRecipe {
 /// or stores into one wide load/store and shuffles. The first operand of a
 /// VPInterleave recipe is the address, followed by the stored values, followed
 /// by an optional mask.
-class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase {
+class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase, public VPIRMetadata  {
   const InterleaveGroup<Instruction> *IG;
 
   /// Indicates if the interleave group is in a conditional block and requires a
@@ -2403,10 +2406,8 @@ class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase {
 public:
   VPInterleaveRecipe(const InterleaveGroup<Instruction> *IG, VPValue *Addr,
                      ArrayRef<VPValue *> StoredValues, VPValue *Mask,
-                     bool NeedsMaskForGaps, DebugLoc DL)
-      : VPRecipeBase(VPDef::VPInterleaveSC, {Addr},
-                     DL),
-
+                     bool NeedsMaskForGaps, const VPIRMetadata &MD, DebugLoc DL)
+      : VPRecipeBase(VPDef::VPInterleaveSC, {Addr}, DL), VPIRMetadata(MD),
         IG(IG), NeedsMaskForGaps(NeedsMaskForGaps) {
     // TODO: extend the masked interleaved-group support to reversed access.
     assert((!Mask || !IG->isReverse()) &&
@@ -2429,7 +2430,7 @@ class LLVM_ABI_FOR_TEST VPInterleaveRecipe : public VPRecipeBase {
 
   VPInterleaveRecipe *clone() override {
     return new VPInterleaveRecipe(IG, getAddr(), getStoredValues(), getMask(),
-                                  NeedsMaskForGaps, getDebugLoc());
+                                  NeedsMaskForGaps, *this,getDebugLoc());
   }
 
   VP_CLASSOF_IMPL(VPDef::VPInterleaveSC)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index e34cab117f321..910556d9c8589 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -1471,6 +1471,17 @@ void VPIRMetadata::applyMetadata(Instruction &I) const {
     I.setMetadata(Kind, Node);
 }
 
+void VPIRMetadata::intersect(const VPIRMetadata &Other) {
+  SmallVector<std::pair<unsigned, MDNode *>> MetadataUnion;
+  for (const auto &[KindA, MDA] : Metadata) {
+    for (const auto &[KindB, MDB] : Other.Metadata) {
+      if (KindA == KindB && MDA == MDB)
+        MetadataUnion.emplace_back(KindA, MDA);
+    }
+  }
+  Metadata = std::move(MetadataUnion);
+}
+
 void VPWidenCallRecipe::execute(VPTransformState &State) {
   assert(State.VF.isVector() && "not widening");
   assert(Variant != nullptr && "Can't create vector function.");
@@ -3512,6 +3523,8 @@ void VPInterleaveRecipe::execute(VPTransformState &State) {
     } else
       NewLoad = State.Builder.CreateAlignedLoad(VecTy, ResAddr,
                                                 Group->getAlign(), "wide.vec");
+    applyMetadata(*NewLoad);
+    // TODO: Also manage existing metadata using VPIRMetadata.
     Group->addMetadata(NewLoad);
 
     ArrayRef<VPValue *> VPDefs = definedValues();
@@ -3616,6 +3629,8 @@ void VPInterleaveRecipe::execute(VPTransformState &State) {
     NewStoreInstr =
         State.Builder.CreateAlignedStore(IVec, ResAddr, Group->getAlign());
 
+  applyMetadata(*NewStoreInstr);
+  // TODO: Also manage existing metadata using VPIRMetadata.
   Group->addMetadata(NewStoreInstr);
 }
 
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 9eeecb378a9fc..da33e7bd2e8e0 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2561,12 +2561,23 @@ void VPlanTransforms::createInterleaveGroups(
   VPDominatorTree VPDT;
   VPDT.recalculate(Plan);
   for (const auto *IG : InterleaveGroups) {
-    SmallVector<VPValue *, 4> StoredValues;
-    for (unsigned i = 0; i < IG->getFactor(); ++i)
-      if (auto *SI = dyn_cast_or_null<StoreInst>(IG->getMember(i))) {
-        auto *StoreR = cast<VPWidenStoreRecipe>(RecipeBuilder.getRecipe(SI));
+    // Get or create the start address for the interleave group.
+    auto *Start =
+        cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(IG->getMember(0)));
+
+   VPIRMetadata InterleaveMD(*Start);
+   SmallVector<VPValue *, 4> StoredValues;
+    for (unsigned i = 0; i < IG->getFactor(); ++i) {
+      Instruction *MemI = IG->getMember(i);
+      if (!MemI)
+        continue;
+      VPWidenMemoryRecipe *MemR = cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(MemI));
+      if (!MemR)
+        continue;
+      if (auto *StoreR = dyn_cast<VPWidenStoreRecipe>(MemR))
         StoredValues.push_back(StoreR->getStoredValue());
-      }
+      InterleaveMD.intersect(*MemR);
+    }
 
     bool NeedsMaskForGaps =
         (IG->requiresScalarEpilogue() && !ScalarEpilogueAllowed) ||
@@ -2581,9 +2592,6 @@ void VPlanTransforms::createInterleaveGroups(
             getLoadStorePointerOperand(IRInsertPos)->stripPointerCasts()))
       NW = Gep->getNoWrapFlags().withoutNoUnsignedWrap();
 
-    // Get or create the start address for the interleave group.
-    auto *Start =
-        cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(IG->getMember(0)));
     VPValue *Addr = Start->getAddr();
     VPRecipeBase *AddrDef = Addr->getDefiningRecipe();
     if (AddrDef && !VPDT.properlyDominates(AddrDef, InsertPos)) {
@@ -2618,8 +2626,9 @@ void VPlanTransforms::createInterleaveGroups(
       ReversePtr->insertBefore(InsertPos);
       Addr = ReversePtr;
     }
+
     auto *VPIG = new VPInterleaveRecipe(IG, Addr, StoredValues,
-                                        InsertPos->getMask(), NeedsMaskForGaps, InsertPos->getDebugLoc());
+                                        InsertPos->getMask(), NeedsMaskForGaps, InterleaveMD, InsertPos->getDebugLoc());
     VPIG->insertBefore(InsertPos);
 
     unsigned J = 0;
diff --git a/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll b/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
index 6d562be03a9b0..41b7488cc7300 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/interleave-cost.ll
@@ -146,7 +146,7 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
 ; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[ARG]], i64 [[TMP9]]
 ; CHECK-NEXT:    [[TMP11:%.*]] = shl i64 [[INDEX]], 4
 ; CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[ARG2]], i64 [[TMP11]]
-; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x float>, ptr [[TMP26]], align 4
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x float>, ptr [[TMP26]], align 4, !alias.scope [[META4:![0-9]+]]
 ; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 0, i32 8>
 ; CHECK-NEXT:    [[STRIDED_VEC14:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 1, i32 9>
 ; CHECK-NEXT:    [[STRIDED_VEC15:%.*]] = shufflevector <16 x float> [[WIDE_VEC]], <16 x float> poison, <2 x i32> <i32 2, i32 10>
@@ -167,10 +167,10 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
 ; CHECK-NEXT:    [[TMP41:%.*]] = shufflevector <2 x float> [[TMP35]], <2 x float> [[TMP37]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:    [[TMP42:%.*]] = shufflevector <4 x float> [[TMP40]], <4 x float> [[TMP41]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <8 x float> [[TMP42]], <8 x float> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 1, i32 3, i32 5, i32 7>
-; CHECK-NEXT:    store <8 x float> [[INTERLEAVED_VEC]], ptr [[TMP28]], align 4
+; CHECK-NEXT:    store <8 x float> [[INTERLEAVED_VEC]], ptr [[TMP28]], align 4, !alias.scope [[META7:![0-9]+]], !noalias [[META4]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[TMP43:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP43]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT:    br i1 [[TMP43]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
 ; CHECK-NEXT:    br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -216,7 +216,7 @@ define void @geps_feeding_interleave_groups_with_reuse(ptr %arg, i64 %arg1, ptr
 ; CHECK-NEXT:    store float [[MUL_4]], ptr [[GEP_11]], align 4
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], [[ARG1]]
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP10:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -397,12 +397,12 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
 ; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
 ; CHECK-NEXT:    [[TMP51:%.*]] = lshr exact i64 [[OFFSET_IDX]], 1
 ; CHECK-NEXT:    [[TMP52:%.*]] = getelementptr nusw i32, ptr [[B]], i64 [[TMP51]]
-; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP52]], align 4
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP52]], align 4, !alias.scope [[META11:![0-9]+]], !noalias [[META14:![0-9]+]]
 ; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
 ; CHECK-NEXT:    [[STRIDED_VEC34:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
 ; CHECK-NEXT:    [[TMP56:%.*]] = getelementptr i32, ptr [[A]], i64 [[OFFSET_IDX]]
 ; CHECK-NEXT:    [[TMP54:%.*]] = getelementptr i32, ptr [[B]], <4 x i64> [[VEC_IND]]
-; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> [[TMP54]], i32 4, <4 x i1> splat (i1 true), <4 x i32> poison), !alias.scope [[META6:![0-9]+]], !noalias [[META9:![0-9]+]]
+; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> [[TMP54]], i32 4, <4 x i1> splat (i1 true), <4 x i32> poison), !alias.scope [[META16:![0-9]+]], !noalias [[META14]]
 ; CHECK-NEXT:    [[TMP58:%.*]] = shufflevector <4 x i32> [[STRIDED_VEC]], <4 x i32> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:    [[TMP59:%.*]] = shufflevector <4 x i32> [[STRIDED_VEC34]], <4 x i32> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <4 x i32> [[WIDE_MASKED_GATHER]], <4 x i32> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
@@ -410,11 +410,11 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
 ; CHECK-NEXT:    [[TMP62:%.*]] = shufflevector <8 x i32> [[TMP60]], <8 x i32> zeroinitializer, <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:    [[TMP63:%.*]] = shufflevector <16 x i32> [[TMP61]], <16 x i32> [[TMP62]], <32 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, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
 ; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <32 x i32> [[TMP63]], <32 x i32> poison, <32 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 1, i32 5, i32 9, i32 13, i32 17, i32 21, i32 25, i32 29, i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30, i32 3, i32 7, i32 11, i32 15, i32 19, i32 23, i32 27, i32 31>
-; CHECK-NEXT:    store <32 x i32> [[INTERLEAVED_VEC]], ptr [[TMP56]], align 4
+; CHECK-NEXT:    store <32 x i32> [[INTERLEAVED_VEC]], ptr [[TMP56]], align 4, !alias.scope [[META14]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 32)
 ; CHECK-NEXT:    [[TMP64:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP64]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]]
+; CHECK-NEXT:    br i1 [[TMP64]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[SCALAR_PH]]
 ; CHECK:       [[SCALAR_PH]]:
@@ -455,7 +455,7 @@ define void @geps_feeding_interleave_groups_with_reuse2(ptr %A, ptr %B, i64 %N)
 ; CHECK-NEXT:    store i32 0, ptr [[GEP_A_7]], align 4
 ; CHECK-NEXT:    [[IV_NEXT_7]] = add nuw nsw i64 [[IV]], 8
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], [[N]]
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP12:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP19:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -521,7 +521,7 @@ define void @interleave_store_double_i64(ptr %dst) {
 ; CHECK-NEXT:    store <4 x double> [[INTERLEAVED_VEC]], ptr [[TMP1]], align 8
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
-; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
+; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[EXIT:.*]]
 ; CHECK:       [[SCALAR_PH]]:
@@ -535,7 +535,7 @@ define void @interleave_store_double_i64(ptr %dst) {
 ; CHECK-NEXT:    store double 0.000000e+00, ptr [[GEP_0]], align 8
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], 1
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP14:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP21:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -642,7 +642,7 @@ define void @interleave_store_i64_double_2(ptr %dst) {
 ; CHECK-NEXT:    store <4 x double> [[INTERLEAVED_VEC]], ptr [[TMP1]], align 8
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
-; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
+; CHECK-NEXT:    br i1 true, label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:    br label %[[EXIT:.*]]
 ; CHECK:       [[SCALAR_PH]]:
@@ -656,7 +656,7 @@ define void @interleave_store_i64_double_2(ptr %dst) {
 ; CHECK-NEXT:    store double 0.000000e+00, ptr [[GEP_1]], align 8
 ; CHECK-NEXT:    [[IV_NEXT]] = add i64 [[IV]], 1
 ; CHECK-NEXT:    [[EC:%.*]] = icmp eq i64 [[IV]], 1
-; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-NEXT:    br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP23:![0-9]+]]
 ; CHECK:       [[EXIT]]:
 ; CHECK-NEXT:    ret void
 ;
@@ -687,17 +687,24 @@ attributes #1 = { "min-legal-vector-width"="0" "target-cpu"="cascadelake" }
 ; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
 ; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
 ; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
-; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
-; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]]}
-; CHECK: [[META6]] = !{[[META7:![0-9]+]]}
-; CHECK: [[META7]] = distinct !{[[META7]], [[META8:![0-9]+]]}
-; CHECK: [[META8]] = distinct !{[[META8]], !"LVerDomain"}
-; CHECK: [[META9]] = !{[[META10:![0-9]+]]}
-; CHECK: [[META10]] = distinct !{[[META10]], [[META8]]}
-; CHECK: [[LOOP11]] = distinct !{[[LOOP11]], [[META1]], [[META2]]}
-; CHECK: [[LOOP12]] = distinct !{[[LOOP12]], [[META1]]}
-; CHECK: [[LOOP13]] = distinct !{[[LOOP13]], [[META1]], [[META2]]}
-; CHECK: [[LOOP14]] = distinct !{[[LOOP14]], [[META2]], [[META1]]}
-; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]], [[META2]]}
-; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META2]], [[META1]]}
+; CHECK: [[META4]] = !{[[META5:![0-9]+]]}
+; CHECK: [[META5]] = distinct !{[[META5]], [[META6:![0-9]+]]}
+; CHECK: [[META6]] = distinct !{[[META6]], !"LVerDomain"}
+; CHECK: [[META7]] = !{[[META8:![0-9]+]]}
+; CHECK: [[META8]] = distinct !{[[META8]], [[META6]]}
+; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META1]], [[META2]]}
+; CHECK: [[LOOP10]] = distinct !{[[LOOP10]], [[META1]]}
+; CHECK: [[META11]] = !{[[META12:![0-9]+]]}
+; CHECK: [[META12]] = distinct !{[[META12]], [[META13:![0-9]+]]}
+; CHECK: [[META13]] = distinct !{[[META13]], !"LVerDomain"}
+; CHECK: [[META14]] = !{[[META15:![0-9]+]]}
+; CHECK: [[META15]] = distinct !{[[META15]], [[META13]]}
+; CHECK: [[META16]] = !{[[META17:![0-9]+]]}
+; CHECK: [[META17]] = distinct !{[[META17]], [[META13]]}
+; CHECK: [[LOOP18]] = distinct !{[[LOOP18]], [[META1]], [[META2]]}
+; CHECK: [[LOOP19]] = distinct !{[[LOOP19]], [[META1]]}
+; CHECK: [[LOOP20]] = distinct !{[[LOOP20]], [[META1]], [[META2]]}
+; CHECK: [[LOOP21]] = distinct !{[[LOOP21]], [[META2]], [[META1]]}
+; CHECK: [[LOOP22]] = distinct !{[[LOOP22]], [[META1]], [[META2]]}
+; CHECK: [[LOOP23]] = distinct !{[[LOOP23]], [[META2]], [[META1]]}
 ;.
diff --git a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll
index 65148b0babcd7..9c102ae0dd9b7 100644
--- a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll
+++ b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-metadata.ll
@@ -130,17 +130,17 @@ define void @ir_tbaa_different(ptr %base, ptr %end, ptr %src) {
 ; CHECK-NEXT:    [[TMP11:%.*]] = load float, ptr [[SRC]], align 4, !alias.scope [[META14:![0-9]+]]
 ; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[TMP11]], i64 0
 ; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer
-; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <4 x float>, ptr [[NEXT_GEP]], align 4
+; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <4 x float>, ptr [[NEXT_GEP]], align 4, !alias.scope [[META17:![0-9]+]], !noalias [[META14]]
 ; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <4 x float> [[WIDE_VEC]], <4 x float> poison, <2 x i32> <i32 0, i32 2>
 ; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <4 x float> [[WIDE_VEC]], <4 x float> poison, <2 x i32> <i32 1, i32 3>
 ; CHECK-NEXT:    [[TMP6:%.*]] = fmul <2 x float> [[STRIDED_VEC]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP7:%.*]] = fmul <2 x float> [[STRIDED_VEC3]], [[BROADCAST_SPLAT]]
 ; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> [[TMP7]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
 ; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x float> [[TMP8]], <4 x float> poison, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
-; CHECK-NEXT:    store <4 x float> [[INTERLEAVED_VEC]], ptr [[NEXT_GEP]], align 4
+; CHECK-NEXT:    store <4 x float> [[INTERLEAVED_VEC]], ptr [[NEXT_GEP]], align 4, !alias.scope [[META17]], !noalias [[META14]]
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]
+; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
 ; CHECK-NEXT:   ...
[truncated]

@github-actions
Copy link

github-actions bot commented Aug 11, 2025

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

@fhahn fhahn force-pushed the vplan-interleave-group-metadata branch from 7af5853 to c8bfe91 Compare August 12, 2025 09:29
Copy link
Contributor

Choose a reason for hiding this comment

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

It looks like we're computing the metadata for member 0 twice. Could we peel out the first iteration?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Hmm, yep we are copying the metadata vector here and intersecting with itself below. Peeling the iteration won't work, as we need all iterations to also handle populating StoredValues.

I added a copy assignment operator, and assign the metadata for the first iteration, and intersect for the others.

Copy link
Contributor

Choose a reason for hiding this comment

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

Could this work?

    // Get or create the start address for the interleave group.
    auto *Start =
        cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(IG->getMember(0)));

    VPIRMetadata InterleaveMD(*Start);  
    SmallVector<VPValue *, 4> StoredValues;
    if (auto *StoreR = dyn_cast<VPWidenStoreRecipe>(Start))
       StoredValues.push_back(StoreR->getStoredValue());
       
    for (unsigned i = 1; i < IG->getFactor(); ++i) {
      Instruction *MemI = IG->getMember(i);
      if (!MemI)
        continue;
      VPWidenMemoryRecipe *MemR = 
          cast<VPWidenMemoryRecipe>(RecipeBuilder.getRecipe(MemI));
      if (!StoredValues.empty()) 
        StoredValues.push_back(cast<VPWidenStoreRecipe>(MemR)->getStoredValue());
      InterleaveMD.intersect(*MemR);
    }

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Sure, updated, thanks!

@fhahn fhahn force-pushed the vplan-interleave-group-metadata branch 2 times, most recently from 613a56f to 8367317 Compare August 14, 2025 18:23
Copy link
Collaborator

Choose a reason for hiding this comment

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

If we're passing this (even if to pass its inherited metadata), suffice to pass only it, as in copy constructor?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

We could, but that would mean adding another constructor that is only used by clone?

Copy link
Collaborator

Choose a reason for hiding this comment

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

Ok, then perhaps a trivial getMetadata() would help this look better than passing this to essentially a copy constructor along with a bunch of other redundant parameters.

Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
SmallVector<std::pair<unsigned, MDNode *>> MetadataUnion;
SmallVector<std::pair<unsigned, MDNode *>> MetadataIntersection;

?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Fixed, thanks

Comment on lines 1498 to 1499
Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
if (KindA == KindB && MDA == MDB)
MetadataUnion.emplace_back(KindA, MDA);
if (KindA == KindB && MDA == MDB) {
MetadataUnion.emplace_back(KindA, MDA);
break;
}

early continue to the next KindA, MDA?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done, thanks

Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
/// nodes in both.
/// nodes that are common to both.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Updated, thanks

Comment on lines 2595 to 2597
Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
if (!StoredValues.empty())
StoredValues.push_back(
cast<VPWidenStoreRecipe>(MemR)->getStoredValue());
if (auto *StoreR = dyn_cast<VPWidenStoreRecipe>(MemR))
StoredValues.push_back(StoreR->getStoredValue());

consistent with how Start is handled above?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done, thanks

Comment on lines +1496 to +1566
Copy link
Collaborator

Choose a reason for hiding this comment

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

Thoughts about reducing the O(|Metadata|*|Other.Metadata|) complexity here? Perhaps working with a temporary copy of Other.Metadata and erasing from it every element that is matched, which would work better if both Metadata and Other.Metadata are scanned in reverse, for the common case(?) where they are equal?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

One issue is that the entries are not neccessarily sorted. We could sort them first, or use a set, but in most cases the number of entries will be small. SmallSet also defaults to linear scan for low number of elements. I;ve left it simple for now

Copy link
Collaborator

Choose a reason for hiding this comment

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

Ok, deserves a note.
Sure the entries are unsorted, and expected to be few. Thought of retaining SmallVectors but erasing from a copy of Other every matched entry, iterating over both in reverse, aiming to be linear in (the frequent?) case the two metadata are ordered similarly and their intersection is large.

Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done thanks

Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
Instruction *MemI = IG->getMember(I);
Instruction *MemberI = IG->getMember(I);

(Mem also stands for Memory)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

done thanks

Copy link
Collaborator

Choose a reason for hiding this comment

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

Suggested change
VPWidenMemoryRecipe *MemR =
VPWidenMemoryRecipe *MemoryR =

Copy link
Contributor Author

Choose a reason for hiding this comment

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

done thanks

Copy link
Contributor

@Mel-Chen Mel-Chen left a comment

Choose a reason for hiding this comment

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

LGTM

Copy link
Contributor

Choose a reason for hiding this comment

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

Unrelated blank

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Removed, thanks

Copy link
Contributor Author

@fhahn fhahn left a comment

Choose a reason for hiding this comment

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

Just realized I didn't send earlier responses, done now

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Updated, thanks

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Fixed, thanks

Comment on lines 1498 to 1499
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done, thanks

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done thanks

Copy link
Contributor Author

Choose a reason for hiding this comment

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

done thanks

Copy link
Contributor Author

Choose a reason for hiding this comment

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

done thanks

Comment on lines 2595 to 2597
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Done, thanks

Comment on lines +1496 to +1566
Copy link
Contributor Author

Choose a reason for hiding this comment

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

One issue is that the entries are not neccessarily sorted. We could sort them first, or use a set, but in most cases the number of entries will be small. SmallSet also defaults to linear scan for low number of elements. I;ve left it simple for now

Copy link
Contributor Author

Choose a reason for hiding this comment

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

We could, but that would mean adding another constructor that is only used by clone?

@fhahn fhahn force-pushed the vplan-interleave-group-metadata branch from e5ebe09 to af1bdda Compare August 20, 2025 21:33
fhahn added 6 commits August 21, 2025 12:25
Use VPIRMetadata for VPInterleaveRecipe to preserve noalias metadata
added by versioning.

This still uses InterleaveGroup's logic to preserve existing metadata
from IR. This can be migrated separately.

Fixes llvm#153006.
@fhahn fhahn force-pushed the vplan-interleave-group-metadata branch from af1bdda to 7e6564f Compare August 21, 2025 11:28
@fhahn fhahn merged commit e41aaf5 into llvm:main Aug 21, 2025
9 checks passed
@fhahn fhahn deleted the vplan-interleave-group-metadata branch August 21, 2025 17:58
llvm-sync bot pushed a commit to arm/arm-toolchain that referenced this pull request Aug 21, 2025
Use VPIRMetadata for VPInterleaveRecipe to preserve noalias metadata
added by versioning.

This still uses InterleaveGroup's logic to preserve existing metadata
from IR. This can be migrated separately.

Fixes llvm/llvm-project#153006.

PR: llvm/llvm-project#153084
Copy link
Collaborator

@ayalz ayalz left a comment

Choose a reason for hiding this comment

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

Couple of post commit nits.

Comment on lines +1496 to +1566
Copy link
Collaborator

Choose a reason for hiding this comment

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

Ok, deserves a note.
Sure the entries are unsorted, and expected to be few. Thought of retaining SmallVectors but erasing from a copy of Other every matched entry, iterating over both in reverse, aiming to be linear in (the frequent?) case the two metadata are ordered similarly and their intersection is large.

Copy link
Collaborator

Choose a reason for hiding this comment

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

Ok, then perhaps a trivial getMetadata() would help this look better than passing this to essentially a copy constructor along with a bunch of other redundant parameters.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[LoopVectorize] Missing noalias metadata on store

4 participants