Skip to content

Disable the SkyLakeX DGEMMIxCOPY kernels as well#2111

Merged
martin-frbg merged 2 commits intoOpenMathLib:developfrom
martin-frbg:issue1955
May 5, 2019
Merged

Disable the SkyLakeX DGEMMIxCOPY kernels as well#2111
martin-frbg merged 2 commits intoOpenMathLib:developfrom
martin-frbg:issue1955

Conversation

@martin-frbg
Copy link
Collaborator

as a stopgap measure for numpy/numpy#13401 as mentioned in #1955


DGEMMINCOPY = dgemm_ncopy_8_skylakex.c
DGEMMITCOPY = dgemm_tcopy_8_skylakex.c
#DGEMMITCOPY = dgemm_tcopy_8_skylakex.c
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not familiar with the code base. Can you explain what gemm_tcopy_8 does and how it's different from gemm_tcopy_4?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

that one takes four elements of input at a time and other eight. 4x8 would mean taking that big rectangular block from input.

Copy link
Contributor

@isuruf isuruf May 5, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Doesn't that mean DGEMMINCOPY and DGEMMITCOPY both need to be gemm_*copy_4 variants instead of gemm_*copy_8 variants?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nope.

@martin-frbg martin-frbg changed the title Disable the SkyLakeX DGEMMITCOPY kernel as well Disable the SkyLakeX DGEMMIxCOPY kernels as well May 5, 2019
@martin-frbg martin-frbg merged commit 6a8b426 into OpenMathLib:develop May 5, 2019
@martin-frbg martin-frbg added this to the 0.3.7 milestone May 5, 2019
@tylerjereddy
Copy link
Contributor

Cool--note that an earlier version of 0.3.7.dev actually caused an additional 14 new linear algebra test failures in numpy/numpy#13466

Maybe some kind of weird behavior with partial disable of the problematic kernels?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants