[all-commits] [llvm/llvm-project] b5281a: [X86] Avoid returning the same shuffle operation f...

[Phoebe Wang via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: b5281afe42246680ede4d563227df3657d501028
      https://github.com/llvm/llvm-project/commit/b5281afe42246680ede4d563227df3657d501028
  Author: Phoebe Wang <phoebe.wang at intel.com>
  Date:   2023-10-29 (Sun, 29 Oct 2023)

  Changed paths:
    M llvm/lib/Target/X86/X86ISelLowering.cpp
    A llvm/test/CodeGen/X86/shuffle-combine-crash-5.ll

  Log Message:
  -----------
  [X86] Avoid returning the same shuffle operation for broadcast (#70592)

This is to fix a crash since aab8b2eb080d, which generates a new pattern
```
      t35: v8i32 = xor t11, t14
    t36: v8i32 = vector_shuffle<0,1,0,1,0,1,0,1> t35, undef:v8i32
```

The pattern exposed a bug introduced since f885c08034, which breaks
element widen but doesn't handle the broadcast case.

The patch just solved the crash issue. I observed performance regression
cased by above patches in the test, which may need further
investigation.