Uh oh!
There was an error while loading. Please reload this page.
Optimization on LinearScan::buildPhysRegRecords - #83862
Conversation
ghost
commented
Mar 23, 2023
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch, @kunalspathak Issue Detailsintroducing separate macro on higher 16 zmm registers and lower 16 zmm registers, skipping non-AVX512 register if AVX512 not available.
|
Ruihan-Yin
commented
Mar 24, 2023
@dotnet-policy-service agree company="Intel Corporation" |
8ff92c7 to
f640316Compare…2 register if AVX512 not available.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Ruihan-Yin
commented
Mar 27, 2023
some tests failed, but it seems not to be caused by the changes in this PR. |
tannergooding
commented
Mar 28, 2023
Failure is unrelated and being handled by #84012 |
Uh oh!
There was an error while loading. Please reload this page.
fix the offset value when allocating upper registers, it should be the length of the lower register group.
Uh oh!
There was an error while loading. Please reload this page.
Ruihan-Yin
commented
Mar 31, 2023
@tannergooding Hi, is there anything further we need to do? Or this PR is ready to be merged. |
kunalspathak
left a comment
There was a problem hiding this comment.
LGTM
It seems the TP regression shows up because helix machines has EVEX encoding and this PR adds the extra check for it? @tannergooding - is that the right conclusion?
Ruihan-Yin
commented
Apr 3, 2023
Thanks all! |

introducing separate macro on upper 16 zmm registers and lower 16 zmm registers, skipping upper 16 registers unless on 64-bit system.