[SLPVectorizer] Miscompile at O3

[dtcxzyw]

Reproducer: https://gist.github.com/dtcxzyw/117dc735db97658ea852d30fd896242a

> bin/clang -O0 test.c -w -I/usr/include/csmith && ./a.out
checksum = BDCB7A78
> bin/clang -O3 test.c -w -mllvm -inline-threshold=100000 -I/usr/include/csmith && ./a.out
checksum = 9198FC41

llvm version: 1193f7d

[dtcxzyw]

Reproducer: https://alive2.llvm.org/ce/z/GCXoe9

; bin/opt -passes=slp-vectorizer test.ll -S

target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define i1 @main(i32 %g.promoted.i.i, i16 %d.promoted.i.i) {
entry:
  %0 = and i16 %d.promoted.i.i, 1
  %xor.i.i = xor i32 %g.promoted.i.i, 1
  %conv1.i.i = trunc i32 %xor.i.i to i8
  %notsub.i = add i8 %conv1.i.i, -1
  %cmp.i.i = icmp sgt i8 %notsub.i, -3
  %conv3.i.i = zext i1 %cmp.i.i to i32
  %cmp4.i.i = icmp sgt i32 %xor.i.i, %conv3.i.i
  %conv1.1.i.i = trunc i32 %g.promoted.i.i to i8
  %notsub25.i = add i8 %conv1.1.i.i, -1
  %cmp.1.i.i = icmp sgt i8 %notsub25.i, -3
  %conv3.1.i.i = zext i1 %cmp.1.i.i to i32
  %cmp4.1.i.i = icmp sgt i32 %g.promoted.i.i, %conv3.1.i.i
  %notsub26.i = add i8 %conv1.1.i.i, -9
  %cmp.i17.i = icmp sgt i8 %notsub26.i, -3
  %conv3.i18.i = zext i1 %cmp.i17.i to i32
  %cmp4.i19.i = icmp sgt i32 %g.promoted.i.i, %conv3.i18.i
  %notsub27.i = add i8 %conv1.i.i, -9
  %cmp.1.i22.i = icmp sgt i8 %notsub27.i, -3
  %conv3.1.i23.i = zext i1 %cmp.1.i22.i to i32
  %cmp4.1.i24.i = icmp sgt i32 %xor.i.i, %conv3.1.i23.i
  %1 = and i1 %cmp4.i19.i, %cmp4.1.i24.i
  %2 = and i1 %cmp4.i.i, %1
  %3 = and i1 %cmp4.1.i.i, %2
  ret i1 %3
}

Output:

define i1 @main(i32 %g.promoted.i.i, i16 %d.promoted.i.i) {
entry:
  %0 = and i16 %d.promoted.i.i, 1
  %xor.i.i = xor i32 %g.promoted.i.i, 1
  %1 = trunc i32 %g.promoted.i.i to i8
  %2 = insertelement <2 x i8> poison, i8 %1, i32 0
  %3 = trunc i32 %xor.i.i to i8
  %4 = insertelement <2 x i8> %2, i8 %3, i32 1
  %5 = shufflevector <2 x i8> %4, <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
  %6 = add <4 x i8> %5, <i8 -9, i8 -9, i8 -1, i8 -1>
  %7 = icmp sgt <4 x i8> %6, <i8 -3, i8 -3, i8 -3, i8 -3>
  %8 = zext <4 x i1> %7 to <4 x i8>
  %9 = sext <2 x i8> %4 to <2 x i32>
  %10 = shufflevector <2 x i32> %9, <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
  %11 = zext <4 x i8> %8 to <4 x i32>
  %12 = icmp sgt <4 x i32> %10, %11
  %13 = call i1 @llvm.vector.reduce.and.v4i1(<4 x i1> %12)
  ret i1 %13
}

cc @alexey-bataev