[llvm] ef1d19b - [SLP]Fix PR89438: check for all tree entries for the resized value.
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 22 06:55:47 PDT 2024
Author: Alexey Bataev
Date: 2024-04-22T06:38:38-07:00
New Revision: ef1d19b0a58f92241963a1403905e8b0a8484b5f
URL: https://github.com/llvm/llvm-project/commit/ef1d19b0a58f92241963a1403905e8b0a8484b5f
DIFF: https://github.com/llvm/llvm-project/commit/ef1d19b0a58f92241963a1403905e8b0a8484b5f.diff
LOG: [SLP]Fix PR89438: check for all tree entries for the resized value.
Need to check all possible entries, before trying looking for the
minbitwidth in the user node. Otherwise we may incorrectly get
signedness info.
Added:
Modified:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
llvm/test/Transforms/SLPVectorizer/X86/gather-node-same-reduced.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 1b56bb7b600c26..e70612ae49f7de 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -13139,31 +13139,55 @@ Value *BoUpSLP::vectorizeTree(
assert(Vec->getType()->isIntOrIntVectorTy() &&
PrevVec->getType()->isIntOrIntVectorTy() &&
"Expected integer vector types only.");
- std::optional<std::pair<unsigned long, bool>> Res;
- if (const TreeEntry *BaseTE = getTreeEntry(TE->Scalars.front())) {
- SmallVector<const TreeEntry *> BaseTEs;
- if (BaseTE->isSame(TE->Scalars))
- BaseTEs.push_back(BaseTE);
- auto It = MultiNodeScalars.find(TE->Scalars.front());
- if (It != MultiNodeScalars.end()) {
- for (const TreeEntry *MNTE : It->getSecond())
- if (MNTE->isSame(TE->Scalars))
- BaseTEs.push_back(MNTE);
+ std::optional<bool> IsSigned;
+ for (Value *V : TE->Scalars) {
+ if (const TreeEntry *BaseTE = getTreeEntry(V)) {
+ auto It = MinBWs.find(BaseTE);
+ if (It != MinBWs.end()) {
+ IsSigned = IsSigned.value_or(false) || It->second.second;
+ if (*IsSigned)
+ break;
+ }
+ for (const TreeEntry *MNTE : MultiNodeScalars.lookup(V)) {
+ auto It = MinBWs.find(MNTE);
+ if (It != MinBWs.end()) {
+ IsSigned = IsSigned.value_or(false) || It->second.second;
+ if (*IsSigned)
+ break;
+ }
+ }
+ if (IsSigned.value_or(false))
+ break;
+ // Scan through gather nodes.
+ for (const TreeEntry *BVE : ValueToGatherNodes.lookup(V)) {
+ auto It = MinBWs.find(BVE);
+ if (It != MinBWs.end()) {
+ IsSigned = IsSigned.value_or(false) || It->second.second;
+ if (*IsSigned)
+ break;
+ }
+ }
+ if (IsSigned.value_or(false))
+ break;
+ if (auto *EE = dyn_cast<ExtractElementInst>(V)) {
+ IsSigned =
+ IsSigned.value_or(false) ||
+ !isKnownNonNegative(EE->getVectorOperand(), SimplifyQuery(*DL));
+ continue;
+ }
+ if (IsSigned.value_or(false))
+ break;
}
- const auto *BaseIt = find_if(BaseTEs, [&](const TreeEntry *BaseTE) {
- return MinBWs.contains(BaseTE);
- });
- if (BaseIt != BaseTEs.end())
- Res = MinBWs.lookup(*BaseIt);
}
- if (!Res) {
- assert(MinBWs.contains(TE->UserTreeIndices.front().UserTE) &&
- "Expected user in MinBWs.");
- Res = MinBWs.lookup(TE->UserTreeIndices.front().UserTE);
+ if (IsSigned.value_or(false)) {
+ // Final attempt - check user node.
+ auto It = MinBWs.find(TE->UserTreeIndices.front().UserTE);
+ if (It != MinBWs.end())
+ IsSigned = It->second.second;
}
- assert(Res && "Expected user node or perfect diamond match in MinBWs.");
- bool IsSigned = Res->second;
- Vec = Builder.CreateIntCast(Vec, PrevVec->getType(), IsSigned);
+ assert(IsSigned &&
+ "Expected user node or perfect diamond match in MinBWs.");
+ Vec = Builder.CreateIntCast(Vec, PrevVec->getType(), *IsSigned);
}
PrevVec->replaceAllUsesWith(Vec);
PostponedValues.try_emplace(Vec).first->second.push_back(TE);
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/gather-node-same-reduced.ll b/llvm/test/Transforms/SLPVectorizer/X86/gather-node-same-reduced.ll
index b03eb9e67254be..42ed26d82e036d 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/gather-node-same-reduced.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/gather-node-same-reduced.ll
@@ -82,3 +82,95 @@ define i64 @test(ptr %p) {
store i8 %55, ptr %3, align 1
ret i64 0
}
+
+define i64 @test1(ptr %p) {
+; CHECK-LABEL: define i64 @test1(
+; CHECK-SAME: ptr [[P:%.*]]) {
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[P]], i64 12
+; CHECK-NEXT: [[TMP2:%.*]] = xor <4 x i32> zeroinitializer, zeroinitializer
+; CHECK-NEXT: [[TMP3:%.*]] = xor <4 x i32> [[TMP2]], zeroinitializer
+; CHECK-NEXT: [[TMP4:%.*]] = xor <4 x i32> [[TMP3]], zeroinitializer
+; CHECK-NEXT: [[TMP5:%.*]] = xor <4 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT: [[TMP6:%.*]] = xor <4 x i32> [[TMP5]], zeroinitializer
+; CHECK-NEXT: [[TMP7:%.*]] = xor <4 x i32> [[TMP6]], zeroinitializer
+; CHECK-NEXT: [[TMP8:%.*]] = xor <4 x i32> [[TMP7]], zeroinitializer
+; CHECK-NEXT: [[TMP9:%.*]] = xor <4 x i32> [[TMP8]], zeroinitializer
+; CHECK-NEXT: [[TMP10:%.*]] = xor <4 x i32> [[TMP9]], zeroinitializer
+; CHECK-NEXT: [[TMP11:%.*]] = xor <4 x i32> [[TMP10]], zeroinitializer
+; CHECK-NEXT: [[TMP12:%.*]] = trunc <4 x i32> [[TMP11]] to <4 x i8>
+; CHECK-NEXT: store <4 x i8> [[TMP12]], ptr [[TMP1]], align 1
+; CHECK-NEXT: ret i64 0
+;
+ %1 = getelementptr i8, ptr %p, i64 13
+ %2 = getelementptr i8, ptr %p, i64 14
+ %3 = getelementptr i8, ptr %p, i64 15
+ %4 = getelementptr i8, ptr %p, i64 12
+ %5 = zext i8 0 to i32
+ %6 = and i32 %5, 0
+ %.not866 = icmp eq i32 %6, 0
+ %7 = select i1 %.not866, i32 0, i32 0
+ %8 = xor i32 0, %7
+ %9 = zext i8 0 to i32
+ %10 = and i32 %9, 0
+ %.not869 = icmp eq i32 %10, 0
+ %11 = select i1 %.not869, i32 0, i32 0
+ %12 = xor i32 0, %11
+ %13 = zext i8 0 to i32
+ %14 = and i32 %13, 0
+ %.not871 = icmp eq i32 %14, 0
+ %15 = select i1 %.not871, i32 0, i32 0
+ %16 = xor i32 0, %15
+ %17 = zext i8 0 to i32
+ %18 = and i32 %17, 0
+ %.not874 = icmp eq i32 %18, 0
+ %19 = select i1 %.not874, i32 0, i32 0
+ %20 = xor i32 0, %19
+ %21 = xor i32 %13, 0
+ %22 = xor i32 %21, 0
+ %23 = xor i32 %22, 0
+ %24 = xor i32 %23, 0
+ %25 = xor i32 %24, 0
+ %26 = xor i32 %25, 0
+ %27 = xor i32 %26, %8
+ %28 = xor i32 %27, 0
+ %29 = xor i32 %28, 0
+ %30 = xor i32 %29, 0
+ %31 = trunc i32 %30 to i8
+ store i8 %31, ptr %4, align 1
+ %32 = xor i32 %13, 0
+ %33 = xor i32 %32, 0
+ %34 = xor i32 %33, 0
+ %35 = xor i32 %34, 0
+ %36 = xor i32 %35, 0
+ %37 = xor i32 %36, 0
+ %38 = xor i32 %37, %20
+ %39 = xor i32 %38, 0
+ %40 = xor i32 %39, 0
+ %41 = xor i32 %40, 0
+ %42 = trunc i32 %41 to i8
+ store i8 %42, ptr %1, align 1
+ %43 = xor i32 %9, 0
+ %44 = xor i32 %43, 0
+ %45 = xor i32 %44, 0
+ %46 = xor i32 %45, 0
+ %47 = xor i32 %46, 0
+ %48 = xor i32 %47, 0
+ %49 = xor i32 %48, %16
+ %50 = xor i32 %49, 0
+ %51 = xor i32 %50, 0
+ %52 = xor i32 %51, 0
+ %53 = trunc i32 %52 to i8
+ store i8 %53, ptr %2, align 1
+ %54 = xor i32 %43, 0
+ %55 = xor i32 %54, 0
+ %56 = xor i32 %55, 0
+ %57 = xor i32 %56, 0
+ %58 = xor i32 %57, 0
+ %59 = xor i32 %58, %12
+ %60 = xor i32 %59, 0
+ %61 = xor i32 %60, 0
+ %62 = xor i32 %61, 0
+ %63 = trunc i32 %62 to i8
+ store i8 %63, ptr %3, align 1
+ ret i64 0
+}
More information about the llvm-commits
mailing list