Expose various floating-point intrinsics for Avx512F and Avx512DQ #85716
Conversation
ghost
added
area-CodeGen-coreclr
|
Note regarding the This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change. |
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsThis exposes some instructions unique to the AVX512 family of instructions making progress towards completing:
There will be a separate PR to utilize some of these in our scalar math APIs. For example, Likewise
|
| case NI_SSE41_CeilingScalar: | ||
| case NI_AVX_Ceiling: | ||
| { | ||
| FALLTHROUGH; |
There was a problem hiding this comment.
I had it split "explicitly" here to help visualize the groupings between Ceiling vs RoundToPositiveInfinity (and likewise Floor vs RoundToNegativeInfinity), particularly since Ceiling/Floor don't have AVX512 equivalents.
|
This PR is part of a regression I just filed: #85930 |
|
I wouldn't expect it, this was a zero spmi diffs change since it just added new APIs, it didn't update any existing paths to use them. |
ghost
locked as resolved and limited conversation to collaborators
3 participants
This exposes some instructions unique to the AVX512 family of instructions making progress towards completing:
There will be a separate PR to utilize some of these in our scalar math APIs. For example,
vrangecan be used to implement a faster/correctMax/MaxMagnitude/MaxNumber/MaxNumberMagnitudeandMin/MinMagnitude/MinNumber/MinNumberMagnitude(where-as currently we can only accelerate on x86/x64 if one input is constant).Likewise
vfixupcan be used to handle many complex branching conditions where various edge cases are being handled.