Skip to content

[hipblaslt] Fix tensilelite-tests build after MX API signature changes#7277

Open
amd-songpiao wants to merge 1 commit into
ROCm:developfrom
amd-songpiao:bugfix/tensilelite-mx-tests-build
Open

[hipblaslt] Fix tensilelite-tests build after MX API signature changes#7277
amd-songpiao wants to merge 1 commit into
ROCm:developfrom
amd-songpiao:bugfix/tensilelite-mx-tests-build

Conversation

@amd-songpiao
Copy link
Copy Markdown
Contributor

Two MX test files added in PR #7017 broke the next day when API renames landed without updating their call sites. CI missed it because tensilelite-tests is gated behind TENSILELITE_BUILD_TESTING=ON (default OFF).

  • MXDataGen_test.cpp (5 sites): generateMXInput gained a hipDataType scaleType 2nd param; pass HIP_R_8F_UE8M0 (UE8M0), matching the canonical mapping in DataInitialization.cpp.
  • MXScalePadding_test.cpp (2 sites): setMXScaleA/B gained a rocisa::DataType mxType 1st param; pass rocisa::DataType::E8, matching tensile_host.cpp's Block_*_UE8M0 cases.

Two MX test files added in PR ROCm#7017 broke the next day when API renames
landed without updating their call sites. CI missed it because
tensilelite-tests is gated behind TENSILELITE_BUILD_TESTING=ON (default OFF).

- MXDataGen_test.cpp (5 sites): generateMXInput gained a `hipDataType
  scaleType` 2nd param; pass HIP_R_8F_UE8M0 (UE8M0), matching the
  canonical mapping in DataInitialization.cpp.
- MXScalePadding_test.cpp (2 sites): setMXScaleA/B gained a
  `rocisa::DataType mxType` 1st param; pass rocisa::DataType::E8,
  matching tensile_host.cpp's Block_*_UE8M0 cases.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant