PR for llvm/llvm-project#65044 #674
Conversation
(cherry picked from commit 2fce8f74b3c04bbf7e941cfed7604edae2b573b6)
… in ComplexDeinterleavingPass When replacing ComplexDeinterleavingPass::ReductionOperation, we can do it either from the Real or Imaginary part. The correct way is to take whichever is later in the BasicBlock, but before the patch, we just always took the Real part. Fixes llvm/llvm-project#65044 Differential Revision: https://reviews.llvm.org/D159209 (cherry picked from commit e2cb07c322e85604dc48f9caec52b3570db0e1d8)
|
@danilaml What do you think about merging this PR to the release branch? |
|
LGTM |
|
@tru not sure why this hasn't made it into rc4. Was it because one of the merge checks had a timeout failure (unrelated)? |
|
@danilaml no it seems like our automation never added it to my project view and I missed it. how safe is this to add in the -final release? |
|
It's a codegen issue for new enough AArch64 CPUs. @igogo-x86 might have a better idea, but it looks safe enough to me (but I also don't have a need for it to end up in a release since I can just cherry-pick it myself). |
|
@tru This patch is fixing quite a severe issue and is unlikely to introduce new bugs. So it is better to add it to the release. |
+1, in Chromium we've been tracking down a miscompile in pgo-instrumented builds for arm64 mac (crbug.com/1473604), and just yesterday concluded that this commit fixes it. |
resolves llvm/llvm-project#65044