[llvm] b73476c - [SLP] Make sure MinVF is a power-of-2 by using PowerOf2Ceil.

[Florian Hahn via llvm-commits]

Author: Florian Hahn
Date: 2024-04-16T13:29:35+01:00
New Revision: b73476c7843f21966acb2fb5cab8515d9ec02905

URL: https://github.com/llvm/llvm-project/commit/b73476c7843f21966acb2fb5cab8515d9ec02905
DIFF: https://github.com/llvm/llvm-project/commit/b73476c7843f21966acb2fb5cab8515d9ec02905.diff

LOG: [SLP] Make sure MinVF is a power-of-2 by using PowerOf2Ceil.

This should ensure we explore the same VFs as before 6d66db3890a18e39.

Fixes https://github.com/llvm/llvm-project/issues/88640.

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    llvm/test/Transforms/SLPVectorizer/X86/trunc-store-value-ty-not-power-of-2.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index c63b500f546f3b..d0bcdceae392bd 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -15155,8 +15155,8 @@ bool SLPVectorizerPass::vectorizeStores(ArrayRef<StoreInst *> Stores,
       Type *ValueTy = StoreTy;
       if (auto *Trunc = dyn_cast<TruncInst>(Store->getValueOperand()))
         ValueTy = Trunc->getSrcTy();
-      unsigned MinVF = TTI->getStoreMinimumVF(
-          R.getMinVF(DL->getTypeSizeInBits(StoreTy)), StoreTy, ValueTy);
+      unsigned MinVF = PowerOf2Ceil(TTI->getStoreMinimumVF(
+          R.getMinVF(DL->getTypeStoreSizeInBits(StoreTy)), StoreTy, ValueTy));
 
       if (MaxVF < MinVF) {
         LLVM_DEBUG(dbgs() << "SLP: Vectorization infeasible as MaxVF (" << MaxVF

diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/trunc-store-value-ty-not-power-of-2.ll b/llvm/test/Transforms/SLPVectorizer/X86/trunc-store-value-ty-not-power-of-2.ll
index 81b4ee40e7fdf3..2f0fad70b593b5 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/trunc-store-value-ty-not-power-of-2.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/trunc-store-value-ty-not-power-of-2.ll
@@ -107,3 +107,36 @@ define void @test_4_trunc_i24_to_i16(i24 %x, ptr %A) {
   store i16 %t, ptr %gep.3, align 1
   ret void
 }
+
+%struct.d = type { [3 x i8], [3 x i8], [2 x i8] }
+
+; Test case for https://github.com/llvm/llvm-project/issues/88640.
+define void @test_access_i24_directly(ptr %src, ptr noalias %dst) "target-cpu"="btver2" {
+; CHECK-LABEL: define void @test_access_i24_directly(
+; CHECK-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load i64, ptr [[SRC]], align 8
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[TMP0]] to i24
+; CHECK-NEXT:    [[GEP_SRC:%.*]] = getelementptr inbounds [[STRUCT_D:%.*]], ptr [[SRC]], i64 0, i32 1
+; CHECK-NEXT:    [[BF_LOAD:%.*]] = load i24, ptr [[GEP_SRC]], align 1
+; CHECK-NEXT:    [[BF_VALUE:%.*]] = and i24 [[TMP1]], 8388607
+; CHECK-NEXT:    [[BF_CLEAR:%.*]] = and i24 [[BF_LOAD]], -8388608
+; CHECK-NEXT:    [[BF_SET:%.*]] = or disjoint i24 [[BF_CLEAR]], [[BF_VALUE]]
+; CHECK-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds [[STRUCT_D]], ptr [[DST]], i64 0, i32 1
+; CHECK-NEXT:    store i24 [[BF_SET]], ptr [[GEP_DST]], align 1
+; CHECK-NEXT:    store i24 0, ptr [[DST]], align 8
+; CHECK-NEXT:    ret void
+;
+entry:
+  %0 = load i64, ptr %src, align 8
+  %1 = trunc i64 %0 to i24
+  %gep.src = getelementptr inbounds %struct.d, ptr %src, i64 0, i32 1
+  %bf.load = load i24, ptr %gep.src, align 1
+  %bf.value = and i24 %1, 8388607
+  %bf.clear = and i24 %bf.load, -8388608
+  %bf.set = or disjoint i24 %bf.clear, %bf.value
+  %gep.dst = getelementptr inbounds %struct.d, ptr %dst, i64 0, i32 1
+  store i24 %bf.set, ptr %gep.dst, align 1
+  store i24 0, ptr %dst, align 8
+  ret void
+}