Grouped conv bwd weight with grouped gemm #2304
Merged
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
bartekxk
requested review from
ThomasNing,
afagaj,
andriy-ca,
aosewski,
asleepzzz,
carlushuang,
coderfeli,
geyyer,
illsilin,
poyenc,
qianfengz and
tenpercent
as code owners
aosewski
reviewed
Jun 10, 2025
...tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
Outdated
Show resolved
Hide resolved
| group_id = index_t((left + right) / 2); | ||
| } | ||
|
|
||
| if(gemm_kernel_args[group_id].HasMainKBlockLoop_) |
Collaborator
There was a problem hiding this comment.
This will likely result in scratch... It would be better to keep HasMainKBlockLoop tparam. Can we make such assumption that all groups have the same HasMainKBlockLoop value?
Contributor
Author
There was a problem hiding this comment.
HasMainKBlockLoop can be different for each gemm so we cant assume that. I dont see any scratch with such implementation
...tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
Outdated
Show resolved
Hide resolved
...tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
Outdated
Show resolved
Hide resolved
...tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
Outdated
Show resolved
Hide resolved
...tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
Show resolved
Hide resolved
...tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
Outdated
Show resolved
Hide resolved
Comment on lines
99
to
106
| this->conv_params.push_back( | ||
| {2, 2, 2, 16, 16, {3, 3}, {28, 28}, {2, 2}, {1, 1}, {1, 1}, {1, 1}}); | ||
| this->conv_params.push_back( | ||
| {2, 2, 2, 16, 16, {3, 3}, {28, 28}, {6, 6}, {1, 1}, {1, 1}, {1, 1}}); | ||
| this->conv_params.push_back( | ||
| {2, 2, 2, 16, 16, {6, 6}, {28, 28}, {6, 6}, {1, 1}, {1, 1}, {1, 1}}); | ||
| this->conv_params.push_back( | ||
| {2, 2, 2, 16, 16, {4, 8}, {28, 28}, {4, 8}, {1, 1}, {1, 1}, {1, 1}}); |
Collaborator
There was a problem hiding this comment.
How many groups per each case would you test here? Do you cover as well case with just one group? And a case where you exceed the MaxKernelArgsNum ?
Contributor
Author
There was a problem hiding this comment.
It covers 4, 9 (but with ZTilde *YTilde * XTilde = 36), 36, 32. So each case. I will add a description
andriy-ca
approved these changes
Jun 11, 2025
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
Change multi kernel launch to grouped gemm to avoid launching a lot of kernels with one stream. It could cause the situation that a lot of workgroups are idle.
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