[PATCH] D118376: [x86] try harder to scalarize a vector load with extracted integer op uses

Sat Feb 12 07:59:50 PST 2022

spatel added a comment.

In D118376#3316151 <https://reviews.llvm.org/D118376#3316151>, @haowei wrote:

> The latest patch no longer causes clang-refactor test failures on stage2 mac build on Fuchsia builders.

Great - thank you for checking!

================
Comment at: llvm/test/CodeGen/X86/extractelement-load.ll:341-345
+; This test is reduced from a C source example that showed a miscompile:
+; https://github.com/llvm/llvm-project/issues/53695
+; The scalarized loads from 'zero' in the AVX asm must occur before
+; the vector store to 'zero' overwrites the values.
+; If compiled to a binary, this test should return 0 if correct.
----------------
This is the new test based on @dyung 's reduction. In the old version of this patch, a `movl` load from `@zero` was placed after the `vmovaps %ymm0, zero(%rip)`

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D118376/new/

https://reviews.llvm.org/D118376