release/18.x: [SLP]Fix a crash if the argument of call was affected by minbitwidth analysis

[AtariDreams]

Need to support proper type conversion for function arguments to avoid compiler crash.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: AtariDreams (AtariDreams)

Changes

Need to support proper type conversion for function arguments to avoid compiler crash.

---

Full diff: https://github.com/llvm/llvm-project/pull/86731.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp (+20-1)
• (added) llvm/test/Transforms/SLPVectorizer/X86/call-arg-reduced-by-minbitwidth.ll (+82)

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 0a9e2c7f49f55f..1fbd69e38eaeec 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -11653,12 +11653,12 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
       if (UseIntrinsic && isVectorIntrinsicWithOverloadTypeAtArg(ID, -1))
         TysForDecl.push_back(
             FixedVectorType::get(CI->getType(), E->Scalars.size()));
+      auto *CEI = cast<CallInst>(VL0);
       for (unsigned I : seq<unsigned>(0, CI->arg_size())) {
         ValueList OpVL;
         // Some intrinsics have scalar arguments. This argument should not be
         // vectorized.
         if (UseIntrinsic && isVectorIntrinsicWithScalarOpAtArg(ID, I)) {
-          CallInst *CEI = cast<CallInst>(VL0);
           ScalarArg = CEI->getArgOperand(I);
           OpVecs.push_back(CEI->getArgOperand(I));
           if (isVectorIntrinsicWithOverloadTypeAtArg(ID, I))
@@ -11671,6 +11671,25 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
           LLVM_DEBUG(dbgs() << "SLP: Diamond merged for " << *VL0 << ".\n");
           return E->VectorizedValue;
         }
+        auto GetOperandSignedness = [&](unsigned Idx) {
+          const TreeEntry *OpE = getOperandEntry(E, Idx);
+          bool IsSigned = false;
+          auto It = MinBWs.find(OpE);
+          if (It != MinBWs.end())
+            IsSigned = It->second.second;
+          else
+            IsSigned = any_of(OpE->Scalars, [&](Value *R) {
+              return !isKnownNonNegative(R, SimplifyQuery(*DL));
+            });
+          return IsSigned;
+        };
+        ScalarArg = CEI->getArgOperand(I);
+        if (cast<VectorType>(OpVec->getType())->getElementType() !=
+            ScalarArg->getType()) {
+          auto *CastTy = FixedVectorType::get(ScalarArg->getType(),
+                                              VecTy->getNumElements());
+          OpVec = Builder.CreateIntCast(OpVec, CastTy, GetOperandSignedness(I));
+        }
         LLVM_DEBUG(dbgs() << "SLP: OpVec[" << I << "]: " << *OpVec << "\n");
         OpVecs.push_back(OpVec);
         if (UseIntrinsic && isVectorIntrinsicWithOverloadTypeAtArg(ID, I))
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/call-arg-reduced-by-minbitwidth.ll b/llvm/test/Transforms/SLPVectorizer/X86/call-arg-reduced-by-minbitwidth.ll
new file mode 100644
index 00000000000000..27c9655f94d3c5
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/call-arg-reduced-by-minbitwidth.ll
@@ -0,0 +1,82 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-pc-windows-msvc19.34.0 < %s | FileCheck %s
+
+define void @test(ptr %0, i8 %1, i1 %cmp12.i) {
+; CHECK-LABEL: define void @test(
+; CHECK-SAME: ptr [[TMP0:%.*]], i8 [[TMP1:%.*]], i1 [[CMP12_I:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <8 x i1> poison, i1 [[CMP12_I]], i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i1> [[TMP2]], <8 x i1> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <8 x i8> poison, i8 [[TMP1]], i32 0
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <8 x i8> [[TMP4]], <8 x i8> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    br label [[PRE:%.*]]
+; CHECK:       pre:
+; CHECK-NEXT:    [[TMP6:%.*]] = zext <8 x i8> [[TMP5]] to <8 x i32>
+; CHECK-NEXT:    [[TMP7:%.*]] = call <8 x i32> @llvm.umax.v8i32(<8 x i32> [[TMP6]], <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>)
+; CHECK-NEXT:    [[TMP8:%.*]] = trunc <8 x i32> [[TMP7]] to <8 x i8>
+; CHECK-NEXT:    [[TMP9:%.*]] = add <8 x i8> [[TMP8]], <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
+; CHECK-NEXT:    [[TMP10:%.*]] = select <8 x i1> [[TMP3]], <8 x i8> [[TMP9]], <8 x i8> [[TMP5]]
+; CHECK-NEXT:    store <8 x i8> [[TMP10]], ptr [[TMP0]], align 1
+; CHECK-NEXT:    br label [[PRE]]
+;
+entry:
+  %idx11 = getelementptr i8, ptr %0, i64 1
+  %idx22 = getelementptr i8, ptr %0, i64 2
+  %idx33 = getelementptr i8, ptr %0, i64 3
+  %idx44 = getelementptr i8, ptr %0, i64 4
+  %idx55 = getelementptr i8, ptr %0, i64 5
+  %idx66 = getelementptr i8, ptr %0, i64 6
+  %idx77 = getelementptr i8, ptr %0, i64 7
+  br label %pre
+
+pre:
+  %conv.i = zext i8 %1 to i32
+  %2 = tail call i32 @llvm.umax.i32(i32 %conv.i, i32 1)
+  %.sroa.speculated.i = add i32 %2, 1
+  %intensity.0.i = select i1 %cmp12.i, i32 %.sroa.speculated.i, i32 %conv.i
+  %conv14.i = trunc i32 %intensity.0.i to i8
+  store i8 %conv14.i, ptr %0, align 1
+  %conv.i.1 = zext i8 %1 to i32
+  %3 = tail call i32 @llvm.umax.i32(i32 %conv.i.1, i32 1)
+  %ss1 = add i32 %3, 1
+  %ii1 = select i1 %cmp12.i, i32 %ss1, i32 %conv.i.1
+  %conv14.i.1 = trunc i32 %ii1 to i8
+  store i8 %conv14.i.1, ptr %idx11, align 1
+  %conv.i.2 = zext i8 %1 to i32
+  %4 = tail call i32 @llvm.umax.i32(i32 %conv.i.2, i32 1)
+  %ss2 = add i32 %4, 1
+  %ii2 = select i1 %cmp12.i, i32 %ss2, i32 %conv.i.2
+  %conv14.i.2 = trunc i32 %ii2 to i8
+  store i8 %conv14.i.2, ptr %idx22, align 1
+  %conv.i.3 = zext i8 %1 to i32
+  %5 = tail call i32 @llvm.umax.i32(i32 %conv.i.3, i32 1)
+  %ss3 = add i32 %5, 1
+  %ii3 = select i1 %cmp12.i, i32 %ss3, i32 %conv.i.3
+  %conv14.i.3 = trunc i32 %ii3 to i8
+  store i8 %conv14.i.3, ptr %idx33, align 1
+  %conv.i.4 = zext i8 %1 to i32
+  %6 = tail call i32 @llvm.umax.i32(i32 %conv.i.4, i32 1)
+  %ss4 = add i32 %6, 1
+  %ii4 = select i1 %cmp12.i, i32 %ss4, i32 %conv.i.4
+  %conv14.i.4 = trunc i32 %ii4 to i8
+  store i8 %conv14.i.4, ptr %idx44, align 1
+  %conv.i.5 = zext i8 %1 to i32
+  %7 = tail call i32 @llvm.umax.i32(i32 %conv.i.5, i32 1)
+  %ss5 = add i32 %7, 1
+  %ii5 = select i1 %cmp12.i, i32 %ss5, i32 %conv.i.5
+  %conv14.i.5 = trunc i32 %ii5 to i8
+  store i8 %conv14.i.5, ptr %idx55, align 1
+  %conv.i.6 = zext i8 %1 to i32
+  %8 = tail call i32 @llvm.umax.i32(i32 %conv.i.6, i32 1)
+  %ss6 = add i32 %8, 1
+  %ii6 = select i1 %cmp12.i, i32 %ss6, i32 %conv.i.6
+  %conv14.i.6 = trunc i32 %ii6 to i8
+  store i8 %conv14.i.6, ptr %idx66, align 1
+  %conv.i.7 = zext i8 %1 to i32
+  %9 = tail call i32 @llvm.umax.i32(i32 %conv.i.7, i32 1)
+  %ss7 = add i32 %9, 1
+  %ii7 = select i1 %cmp12.i, i32 %ss7, i32 %conv.i.7
+  %conv14.i.7 = trunc i32 %ii7 to i8
+  store i8 %conv14.i.7, ptr %idx77, align 1
+  br label %pre
+}

[alexey-bataev]

LG

[tstellar]

Hi @AtariDreams (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix.