[SLP] Fix crash on attempt to access on invalid iterator state.

[vdmitrie]

Summary

The patch fixes corner case when no of scalar instructions
required scheduling for vectorized node.

[ABataev]

Could you try this one instead:

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 82563091f7ce..84c30da74d3d 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -9074,8 +9074,8 @@ Instruction &BoUpSLP::getLastInstructionInBundle(const TreeEntry *E) {

   // Set the insert point to the beginning of the basic block if the entry
   // should not be scheduled.
-  if (E->State != TreeEntry::NeedToGather &&
-      (doesNotNeedToSchedule(E->Scalars) ||
+  if (doesNotNeedToSchedule(E->Scalars) ||
+      (E->State != TreeEntry::NeedToGather &&
        all_of(E->Scalars, isVectorLikeInstWithConstOps))) {
     Instruction *InsertInst;
     if ((E->getOpcode() == Instruction::GetElementPtr &&

[vdmitrie]

alternative approach to fix the issue

[ABataev]

LG

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

for (Value *V : E->Scalars) {
       if (NodeA->getDFSNumIn() > NodeB->getDFSNumIn())
         FirstInst = I;
     }
     return FirstInst;
   };
 
   // Set the insert point to the beginning of the basic block if the entry
   // should not be scheduled.
-  if (E->State != TreeEntry::NeedToGather &&
-      (doesNotNeedToSchedule(E->Scalars) ||
+  if (doesNotNeedToSchedule(E->Scalars) ||
+      (E->State != TreeEntry::NeedToGather &&
        all_of(E->Scalars, isVectorLikeInstWithConstOps))) {
     Instruction *InsertInst;
     if ((E->getOpcode() == Instruction::GetElementPtr &&
          any_of(E->Scalars,
                 [](Value *V) {
                   return !isa<GetElementPtrInst>(V) && isa<Instruction>(V);
                 })) ||
         all_of(E->Scalars, [](Value *V) {

llvm/test/Transforms/SLPVectorizer/X86/no-scheduled-instructions.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt -S -passes=slp-vectorizer -mattr=+avx -mtriple=x86_64 < %s | FileCheck %s
+
+define void @test() {
+; CHECK-LABEL: define void @test
+; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    br i1 undef, label [[BB1:%.*]], label [[BB2:%.*]]
+; CHECK:       bb1:
+; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.vector.reduce.mul.v8i32(<8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 undef, i32 undef, i32 undef, i32 undef>)
+; CHECK-NEXT:    call void @f(i32 noundef [[TMP1]])
+; CHECK-NEXT:    br label [[BB2]]
+; CHECK:       bb2:
+; CHECK-NEXT:    ret void
+;
+  %i27 = extractelement <4 x i32> poison, i64 0
+  %i28 = extractelement <4 x i32> poison, i64 1
+  %i29 = extractelement <4 x i32> poison, i64 2
+  %i30 = extractelement <4 x i32> poison, i64 3
+  %i31 = extractelement <4 x i32> zeroinitializer, i64 0
+  %i32 = extractelement <4 x i32> zeroinitializer, i64 1
+  %i33 = extractelement <4 x i32> zeroinitializer, i64 2
+  %i34 = extractelement <4 x i32> zeroinitializer, i64 3
+  br i1 undef, label %bb1, label %bb2
+
+bb1:
+  %i11 = mul nsw i32 %i28, %i27
+  %i12 = mul nsw i32 %i11, %i29
+  %i13 = mul nsw i32 %i12, %i30
+  %i14 = mul nsw i32 %i13, %i31
+  %i15 = mul nsw i32 %i14, %i32
+  %i16 = mul nsw i32 %i15, %i33
+  %i17 = mul nsw i32 %i16, %i34
+  call void @f(i32 noundef %i17)
+  br label %bb2
+
+bb2:
+  ret void
+}
+
+declare void @f(i32)