PR #5655 Add GPU kernels for ROCm EP

Add GPU kernels for ROCm EP #5655

weixingzhang merged 6 commits into master from wezhan/rocm1

Adding back ROCm kernels by reverting 3a997e7a9 "move rocm kernels to…

e10ed50f

like lamb, multi-tensor-apply needs to be disabled for IsAllFiniteOp …

09b9ce42

weixingzhang requested a review 5 years ago

weixingzhang added training

weixingzhang assigned

jessebenson 5 years ago

weixingzhang assigned

suffiank 5 years ago

weixingzhang assigned

SherlockNoMad 5 years ago

weixingzhang assigned

mrry 5 years ago

jessebenson dismissed these changes on 2020-10-31

address comments from code review.

fdb46af5

weixingzhang dismissed their stale review via fdb46af5 5 years ago

Use hipMemsetAsync and add checks on HIP calls.

087afa0f

jessebenson dismissed these changes on 2020-11-03

move the generated files to build folder.

38af53aa

weixingzhang dismissed their stale review via 38af53aa 5 years ago

Merge branch 'master' into wezhan/rocm1

f029d60d

jessebenson approved these changes on 2020-11-06

weixingzhang merged fff85a6a into master 5 years ago

weixingzhang deleted the wezhan/rocm1 branch 5 years ago

Reviewers

jessebenson

Assignees

mrry

suffiank

jessebenson

SherlockNoMad

Labels

training

Milestone

No milestone