[CK_Tile] Support for various group sizes Preshuffle quant for 2d block scale gemm #3445
+334
−54
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Split cpp file to reduce building time - Support multiple GemmConfig
- Update Readme
- Add support for rowcol and tensor GEMM operations
- Update README
amd-khushbu
requested review from
ThomasNing,
afagaj,
andriy-ca,
aosewski,
asleepzzz,
bartekxk,
carlushuang,
cgmillette,
coderfeli,
geyyer,
illsilin,
poyenc,
qianfengz,
shumway,
tenpercent and
vidyasagar-amd
as code owners
7 tasks
ThomasNing
requested changes
Dec 19, 2025
Contributor
ThomasNing
left a comment
ThomasNing
left a comment
There was a problem hiding this comment.
Left a few comments. Please also add the tests and next step I believe to integrate that with the preshuffled.
| if constexpr(NPerQ <= WarpGemm::kN) | ||
| { | ||
| constexpr auto N1 = BlockGemmShape::kK / KPerQ; | ||
| constexpr auto N0 = WarpGemm::kN / NPerQ; |
Contributor
There was a problem hiding this comment.
I think the variable naming here with N1 = Kvalues and k0 = nwarps will cause confusion. Could we restructure the variable names and add the comments?
| constexpr auto K0 = KPerTile / K1; | ||
| constexpr auto KR = 1; | ||
|
|
||
| return make_static_tile_distribution( |
Contributor
There was a problem hiding this comment.
Could we also add the partition reasoning of the different condition of tile distribution?
| ck_tile::integer_least_multiple(warp_n * bqk_per_block, get_warp_size()); | ||
| constexpr auto tile_window_height = block_n / warp_n; | ||
| auto block_n_idx = i_n / block_n; | ||
| constexpr auto tile_window_height = |
Contributor
There was a problem hiding this comment.
Could I understand the reason of this condition?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Proposed changes
This PR introduces Preshuffle Quant for group size N = (1, 8, 16, 32, 64) for both prefill and decode shapes for 2d block scale Gemm.
Changes include:
Checklist
Please put an
xinto 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.clang-formaton all changed filesDiscussion
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