Skip to content

Conversation

@samremes
Copy link
Contributor

@samremes samremes commented Dec 4, 2025

Proposed changes

Enables more layouts for BQuant GEMM. Row/Col for A is tested, and support for Row/Col for B and BQ is added.

Checklist

Please put an x into the boxes that apply. You can also fill these out after creating the PR. If you're not sure, please don't hesitate to ask.

  • I have added tests relevant to the introduced functionality, and the unit tests are passing locally
  • I have added the test to REGRESSION_TESTS list defined at the top of CMakeLists.txt in tests/CMakeLists.txt, IF the test takes more than 30 seconds to run.
  • I have added inline documentation which enables the maintainers with understanding the motivation
  • I have removed the stale documentation which is no longer relevant after this pull request
  • (If this change is user-facing) I have added release notes which provide the end users with a brief summary of the improvement from this pull request
  • I have run clang-format on all changed files
  • Any dependent changes have been merged

Discussion

If this is a relatively large or complex change, feel free to start a discussion by explaining why you chose the solution you did and what alternatives you considered

@samremes samremes added the WIP label Dec 4, 2025
Copy link
Contributor

@ThomasNing ThomasNing left a comment

Choose a reason for hiding this comment

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

Thank you for the contributions. LGTM except the above comments.

static_assert(std::is_same_v<BQLayout, tensor_layout::gemm::ColumnMajor>);
// PreshuffleQuant currently assumes ColumnMajor layout
// For RowMajor, the preshuffle logic would need adjustment
static_assert(std::is_same_v<BQLayout, tensor_layout::gemm::ColumnMajor>,
Copy link
Contributor

Choose a reason for hiding this comment

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

@amd-khushbu Please not this part as our discussion yesterday. With the preshuffle quant our tensor layout should actually only be RowMajor. We could modify that in your PR.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I'll revert the comment here to avoid creating any confusion.

decltype(make_static_distributed_tensor<ADataType>(ABlockTileDistr{}));
using BBlockTile =
decltype(make_static_distributed_tensor<BDataType>(BBlockTileDistr{}));
decltype(make_static_distributed_tensor<ADataType>(BBlockTileDistr{}));
Copy link
Contributor

Choose a reason for hiding this comment

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

Here should be BDataType?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I think this should be the non-pkint4 datatype, which should be the same as ADataType, but maybe it should be made more clear somehow. I'll create an alias for the type name, so it avoids confusion.

b_dram_block_window_tmp,
[](const BDataType& b) { return b; },
// Note: BDataType gets converted to ADataType during loading
[](const ADataType& b) { return b; },
Copy link
Contributor

Choose a reason for hiding this comment

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

That condition should only happen when the BDataType is pk_int4 right?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants