[CK] Add new fwd conv fp16/bf16 instances optimized for unit group size.#3670
Closed
vpietila-amd wants to merge 9 commits intodevelopfrom
Closed
[CK] Add new fwd conv fp16/bf16 instances optimized for unit group size.#3670vpietila-amd wants to merge 9 commits intodevelopfrom
vpietila-amd wants to merge 9 commits intodevelopfrom
Conversation
vpietila-amd
requested review from
a team,
Snektron,
ThomasNing,
afagaj,
andriy-ca,
aosewski,
asleepzzz,
bartekxk,
carlushuang,
cgmillette,
coderfeli,
geyyer,
illsilin,
poyenc,
qianfengz,
shumway and
vidyasagar-amd
as code owners
bartekxk
previously approved these changes
Jan 28, 2026
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds new FP16 and BF16 convolution instances optimized for unit group size (G=1) to improve performance. The new instances leverage the compute-optimized block GEMM pipeline and demonstrate significant performance improvements (up to 14% TFLOP increase).
Changes:
- Added two new BF16 instances optimized for G=1 with different block configurations (256x256x256 and 512x128x32)
- Added two new FP16 instances optimized for G=1 with matching block configurations
- Modified line endings to add commas for existing instances to accommodate new entries
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4> | ||
| // clang-format on | ||
| DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, F16, F16, F32, F16, DsDataTypes, F16, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 256, 128, 128, 64, 8, 8, 32, 32, 2, 2, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 0, 1, 1, S<1, 32, 1, 8>, 8, BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion::v4>, | ||
|
|
There was a problem hiding this comment.
Empty line contains trailing whitespace. Remove the trailing whitespace to maintain code cleanliness.
Suggested change
vpietila-amd
commented
Jan 29, 2026
...tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp
Outdated
Show resolved
Hide resolved
added 2 commits
January 29, 2026 10:19
Contributor
|
Imported to ROCm/rocm-libraries |
auto-merge was automatically disabled
February 3, 2026 22:01
Pull request was closed
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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
Added new FP16/BF16 instances that are optimized for group size = 1. The new instance use the compute optimized block GEMM pipeline.