[MigraphX EP] [ROCm EP] Upstream ROCm changes for bugfixes and features#23249

Open

TedThemistokleous wants to merge 20 commits intomicrosoft:mainfrom ROCm:upstream_rocm_commits

+13,421-72

Commits on Jan 2, 2025

[VitisAI] change all support tensor type from ir 9 to ir 10 (microsoft#23204 )

authored and
Ted Themistokleous
committed
add HIP language compile option -parallel-jobs=2

jeffdaily
authored and
Ted Themistokleous
committed
update rocm-ci-pipeline-env.Dockerfile to work with internal builds

jeffdaily
authored and
Ted Themistokleous
committed
skip all MHA tests on ROCm EP

jeffdaily
authored and
Ted Themistokleous
committed
use custom linker script hip_fatbin_insert

jeffdaily
authored and
Ted Themistokleous
committed
work-around upstream pytorch changing fromDLPack to take non-const pointer

jeffdaily
authored and
Ted Themistokleous
committed
Bundle MIGraphX with ROCm when built together (#47 )

apwojcik
authored and
Ted Themistokleous
committed
Add support for MIGraphX Exhaustive tune flag in MIGraphX EP (#46 ) (#50 )

TedThemistokleous
and
Ted Themistokleous
committed
Rocm6.3 turn off ort ops for migx (#55 )

TedThemistokleous
authored and
Ted Themistokleous
committed
Ensure we support all inputs for MatMulInteger and ConvInteger. Limit… (#53 )

TedThemistokleous
and
Ted Themistokleous
committed
fixup! Ensure we support all inputs for MatMulInteger and ConvInteger. Limit… (#53 ) (#56 )

TedThemistokleous
authored and
Ted Themistokleous
committed
Remove default noopt for migraphx. use -o no_opt instead on select scripts (#58 ) (#59 )

TedThemistokleous
and
Ted Themistokleous
committed
Update ROCM version detection code (#67 )

xinyazhang
authored and
Ted Themistokleous
committed
Force hipify to use copied version from rocm-6.3.0-14776 build (#69 )

TedThemistokleous
authored and
Ted Themistokleous
committed
Fix SWDEV-491378

xinyazhang
authored and
Ted Themistokleous
committed
Fix CudaKernelTest.*SoftmaxGrad* part of SWDEV-477109 (#63 )

xinyazhang
authored and
Ted Themistokleous
committed
Add Einsum to supported op list

ikalinic
authored and
Ted Themistokleous
committed
Fixes SWDEV-483388 (#74 )

xinyazhang
authored and
Ted Themistokleous
committed

Commits on Jan 3, 2025

Add safety check to that TransposeBigMLFloat16 test passes (#77 )

sstamenk
authored and
Ted Themistokleous
committed
Move BFLoat16 transpose check to HipBlas instead of rocblas
Ted Themistokleous
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MigraphX EP] [ROCm EP] Upstream ROCm changes for bugfixes and features#23249

Commits on Jan 2, 2025

[VitisAI] change all support tensor type from ir 9 to ir 10 (microsoft#23204 )

add HIP language compile option -parallel-jobs=2

update rocm-ci-pipeline-env.Dockerfile to work with internal builds

skip all MHA tests on ROCm EP

use custom linker script hip_fatbin_insert

work-around upstream pytorch changing fromDLPack to take non-const pointer

Bundle MIGraphX with ROCm when built together (#47 )

Add support for MIGraphX Exhaustive tune flag in MIGraphX EP (#46 ) (#50 )

Rocm6.3 turn off ort ops for migx (#55 )

Ensure we support all inputs for MatMulInteger and ConvInteger. Limit… (#53 )

fixup! Ensure we support all inputs for MatMulInteger and ConvInteger. Limit… (#53 ) (#56 )

Remove default noopt for migraphx. use -o no_opt instead on select scripts (#58 ) (#59 )

Update ROCM version detection code (#67 )

Force hipify to use copied version from rocm-6.3.0-14776 build (#69 )

Fix SWDEV-491378

Fix CudaKernelTest.SoftmaxGrad part of SWDEV-477109 (#63 )

Add Einsum to supported op list

Fixes SWDEV-483388 (#74 )

Commits on Jan 3, 2025

Add safety check to that TransposeBigMLFloat16 test passes (#77 )

Move BFLoat16 transpose check to HipBlas instead of rocblas