Conversation
| "-//xla/backends/gpu/runtime:all_reduce_test_amdgpu_any" | ||
| "-//xla/tests:collective_ops_e2e_test_amdgpu_any" | ||
| "-//xla/tests:collective_ops_test_amdgpu_any" | ||
| "-//xla/tools/multihost_hlo_runner:functional_hlo_runner_test_amdgpu_any" |
There was a problem hiding this comment.
@zoranjovanovic-ns IIUC, we skip these tests due to openxla#39373 ?
There was a problem hiding this comment.
Initial reason for creating openxla#39373 was failing //xla/service/gpu:float_support_test. Test itself start passing after hipblaslt was disabled on upstream, thus new test case was added to xla/backends/gpu/transforms/gemm_rewriter.cc in PR.
There was a problem hiding this comment.
but upstream has enabled "hipblaslt" 2 days ago openxla@ba4f5f2 you mean because of this?
There was a problem hiding this comment.
but upstream has enabled "hipblaslt" 2 days ago openxla@ba4f5f2 you mean because of this?
hipblaslt is re-enabled, thus openxla#39373 might be necessary for //xla/service/gpu:float_support_test test to pass on MI200.
I did not saw relation of that PR to remaining tests from the list.
There was a problem hiding this comment.
This is not a way we want to skip the tests. Check our current script. It uses --test_filter.
Lets use that mechanism instead
There was a problem hiding this comment.
There was a problem hiding this comment.
Also upstream we have just 1 failing test but here we have to exclude quite some of them. Some configs must be missing.
4 participants
No description provided.