Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix remove rocblas tests #3100

Merged
merged 6 commits into from
May 22, 2024
Merged

Fix remove rocblas tests #3100

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
1880898
Fixed rocblas failed tests
tvukovic-amd May 17, 2024
b8ba443
Merge remote-tracking branch 'origin/develop' into fix_remove_rocblas…
tvukovic-amd May 17, 2024
1b19621
Fix format
tvukovic-amd May 17, 2024
eb15631
Add changes according to comments
tvukovic-amd May 20, 2024
dfb7450
Fix licencing and format
tvukovic-amd May 20, 2024
c97572a
Merge branch 'develop' into fix_remove_rocblas_tests
kahmed10 May 21, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions src/targets/gpu/target.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,6 +103,10 @@ std::vector<pass> target::get_passes(migraphx::context& gctx, const compile_opti
unsupported_fp8_ops.insert("dot");
unsupported_fp8_ops.insert("quant_dot");
}
#else
// mlir doesn't support fp8 dot
unsupported_fp8_ops.insert("dot");
unsupported_fp8_ops.insert("quant_dot");
Copy link
Member

@umangyadav umangyadav May 17, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

This shouldn't be necessary.

AMDMIGraphX/src/targets/gpu/fuse_mlir.cpp

Line 234 in 7011bda

return false;

MIGraphX doesn't pick rocMLIR for the FP8 dot/quant_dots. It would use rocBLAS for fp8 dots,quant_dots.

Do you have a case where rocBLAS is disabled in MIGraphX ?

Copy link
Collaborator

Choose a reason for hiding this comment

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

Yes, that is required when rocBLAS is not used (see MIGRAPHX_USE_ROCBLAS=OFF).

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

Yes, this is added for case where rocBLAS is disabled in MIGraphX. There is variable MIGRAPHX_USE_ROCBLAS in develop branch, so when it is set to OFF we need to add unsupported fp8 ops for dot and quant dot.

Copy link
Collaborator

Choose a reason for hiding this comment

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

This just duplicating code for both branches. It would be better to have rocblas_fp8_available function that just always returns false when rocblas is disabled.

#endif
// MIOpen doesn't have support for fp8 pooling yet.
unsupported_fp8_ops.insert("pooling");
Expand Down
2 changes: 1 addition & 1 deletion test/gpu/gemm_tune.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,7 +182,6 @@ TEST_CASE(gemm_tune_strided_lowered)
EXPECT(0 == solution_idx.to<std::size_t>());
#endif
}
#endif

TEST_CASE(gemm_tune_invalid_sol_index)
{
Expand Down Expand Up @@ -223,5 +222,6 @@ TEST_CASE(gemm_tune_invalid_sol_index)
EXPECT(0 != solution_idx.to<std::size_t>());
#endif
}
#endif

int main(int argc, const char* argv[]) { test::run(argc, argv); }
Loading