[llvm] ac93b61 - [SLP]Fix PR59098: check if the vector type is scalarized for

[Alexey Bataev via llvm-commits]

Author: Alexey Bataev
Date: 2022-11-21T10:26:01-08:00
New Revision: ac93b611659d94fd94b0394e78888b1402cd5199

URL: https://github.com/llvm/llvm-project/commit/ac93b611659d94fd94b0394e78888b1402cd5199
DIFF: https://github.com/llvm/llvm-project/commit/ac93b611659d94fd94b0394e78888b1402cd5199.diff

LOG: [SLP]Fix PR59098: check if the vector type is scalarized for
extractelements.

If the resulting type is going to be scalarized, no need to adjust the
cost of removed extractelement and insert/extract subvector costs.
Otherwise, the compiler can crash because of the wrong type sizes.

Added: 
    llvm/test/Transforms/SLPVectorizer/X86/scalarazied-result.ll

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index dc67c9e42e30..cbf57366126e 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -6275,6 +6275,10 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
   // FIXME: it tries to fix a problem with MSVC buildbots.
   TargetTransformInfo *TTI = this->TTI;
   auto AdjustExtractsCost = [=](InstructionCost &Cost) {
+    // If the resulting type is scalarized, do not adjust the cost.
+    unsigned VecNumParts = TTI->getNumberOfParts(VecTy);
+    if (VecNumParts == VecTy->getNumElements())
+      return;
     DenseMap<Value *, int> ExtractVectorsTys;
     SmallPtrSet<Value *, 4> CheckedExtracts;
     for (auto *V : VL) {
@@ -6296,8 +6300,7 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
       if (!EEIdx)
         continue;
       unsigned Idx = *EEIdx;
-      if (TTI->getNumberOfParts(VecTy) !=
-          TTI->getNumberOfParts(EE->getVectorOperandType())) {
+      if (VecNumParts != TTI->getNumberOfParts(EE->getVectorOperandType())) {
         auto It =
             ExtractVectorsTys.try_emplace(EE->getVectorOperand(), Idx).first;
         It->getSecond() = std::min<int>(It->second, Idx);
@@ -6328,7 +6331,7 @@ InstructionCost BoUpSLP::getEntryCost(const TreeEntry *E,
       unsigned NumElts = VecTy->getNumElements();
       if (Data.second % NumElts == 0)
         continue;
-      if (TTI->getNumberOfParts(EEVTy) > TTI->getNumberOfParts(VecTy)) {
+      if (TTI->getNumberOfParts(EEVTy) > VecNumParts) {
         unsigned Idx = (Data.second / NumElts) * NumElts;
         unsigned EENumElts = EEVTy->getNumElements();
         if (Idx + NumElts <= EENumElts) {

diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/scalarazied-result.ll b/llvm/test/Transforms/SLPVectorizer/X86/scalarazied-result.ll
new file mode 100644
index 000000000000..53f17083bd4b
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/scalarazied-result.ll
@@ -0,0 +1,19 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu -S < %s | FileCheck %s
+
+define void @test() {
+; CHECK-LABEL: @test(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = extractelement <8 x half> zeroinitializer, i64 1
+; CHECK-NEXT:    [[TOBOOL:%.*]] = fcmp une half [[TMP0]], 0xH0000
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <8 x half> zeroinitializer, i64 1
+; CHECK-NEXT:    [[TOBOOL3:%.*]] = fcmp une half [[TMP1]], 0xH0000
+; CHECK-NEXT:    ret void
+;
+entry:
+  %0 = extractelement <8 x half> zeroinitializer, i64 1
+  %tobool = fcmp une half %0, 0xH0000
+  %1 = extractelement <8 x half> zeroinitializer, i64 1
+  %tobool3 = fcmp une half %1, 0xH0000
+  ret void
+}