Implement vpblendvb optimization in the JIT#69509
Conversation
|
Tagging subscribers to this area: @dotnet/area-system-runtime-intrinsics Issue DetailsFixes #67039 This optimizes calls to ConditonalSelect that use a "per-element mask" intrinsic as the condition vector. Thank you @tannergooding for collaborating with me on this one. Reviewers, please double check that the new
|
dakersnar
changed the title
src/tests/JIT/Regression/JitBlue/Runtime_67039/Runtime_67039.cs
Outdated
Show resolved
Hide resolved
|
Changes look correct to me. Left a note where a couple of intrinsics are missing the flag and where we could improve the opportunistic check. |
|
This needs an additional review from someone on @dotnet/jit-contrib |
|
@dotnet/jit-contrib would love to get this in before the preview5 snap if anyone is available for a review. |
AndyAyersMS
left a comment
AndyAyersMS
left a comment
There was a problem hiding this comment.
I didn't carefully check the instruction table changes, but everything else looks ok.
ghost
locked as resolved and limited conversation to collaborators
4 participants
Fixes #67039
This optimizes calls to ConditonalSelect that use a "per-element mask" intrinsic as the condition vector. Thank you @tannergooding for collaborating with me on this one. Reviewers, please double check that the new
HW_Flag_ReturnsPerElementMaskwas correctly added to all relevant intrinsics (insrc/coreclr/jit/hwintrinsiclistxarch.h). Style nitpicks also appreciated.