[Triton/Gluon] Add config-aware repr to the GEMM and conv1d kernels - #5095
Open
Boss2002n wants to merge 1 commit into
Open
[Triton/Gluon] Add config-aware repr to the GEMM and conv1d kernels#5095Boss2002n wants to merge 1 commit into
Boss2002n wants to merge 1 commit into
Conversation
Every launchable Triton kernel is required to set a config-aware repr via make_kernel_repr so a trace row can be matched back to the exact tuned config. Twelve launchable kernels in the GEMM feed-forward/fused and conv subtrees had none, so they showed up in traces under their bare kernel name: _ff_a16w16_fused_gated, _ff_a16w16_fused_ungated _fused_gemm_a8w8_blockscale_a16w16_kernel (+ its reduce) _fused_gemm_a8w8_blockscale_split_cat, ..._preshuffle_split_cat (+ its reduce) _fused_gemm_afp4wfp4_split_cat_reduce _causal_conv1d_fwd_kernel, _causal_conv1d_update_kernel _causal_conv1d_update_single_token_kernel _reshape_causal_conv1d_update_single_token_kernel Key lists follow the neighbouring kernels in each subtree: tuned block/tile and split-K constexprs plus the meaningful compile-time flags, with grid extents and shape-derived constexprs (GRID_MN*, GROUP_K/GROUP_N, BLOCK_SIZE_S3) excluded -- the launchers compute those from shapes, and the sibling reprs omit them too. A key that is not a real parameter renders as NONE in every trace name, so each list was cross-checked against the kernel's own signature; all 12 pass. Note _fused_gemm_a8w8_blockscale_a16w16_kernel does NOT carry the num_warps / num_stages / waves_per_eu / matrix_instr_nonkdim constexprs its afp4wfp4 twin has, so its list is correspondingly shorter.
Contributor
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
PR title tags: |
github-actions
Bot
changed the title
Contributor
There was a problem hiding this comment.
Pull request overview
This PR improves kernel traceability and debugging for Triton GEMM and conv1d kernels by attaching config-aware repr functions via make_kernel_repr, so compiled kernel names reflect key compile-time/meta parameters.
Changes:
- Added
make_kernel_repr(...)definitions for several GEMM fused/feed-forward Triton kernels and wired them into@triton.jit(repr=...). - Added config-aware
reprfor causal conv1d forward/update kernels (including single-token update paths). - Standardized kernel naming keys used in
reprto match existing kernel meta-parameter patterns (e.g.,BLOCK_SIZE_*,NUM_KSPLIT, feature flags).
Reviewed changes
Copilot reviewed 7 out of 7 changed files in this pull request and generated no comments.
Show a summary per file
| File | Description |
|---|---|
| aiter/ops/triton/_triton_kernels/gemm/fused/fused_gemm_afp4wfp4_split_cat.py | Adds config-aware repr to the split-cat reduce kernel (and uses make_kernel_repr consistently with other GEMM kernels). |
| aiter/ops/triton/_triton_kernels/gemm/fused/fused_gemm_a8w8_blockscale_split_cat.py | Adds make_kernel_repr import plus config-aware repr for main, preshuffle, and reduce split-cat kernels. |
| aiter/ops/triton/_triton_kernels/gemm/fused/fused_gemm_a8w8_blockscale_a16w16.py | Adds config-aware repr for the main FP8/BF16 fused kernel and its reduce kernel. |
| aiter/ops/triton/_triton_kernels/gemm/feed_forward/ff_a16w16_fused_ungated.py | Adds config-aware repr for the fused ungated FFN kernel. |
| aiter/ops/triton/_triton_kernels/gemm/feed_forward/ff_a16w16_fused_gated.py | Adds config-aware repr for the fused gated FFN kernel. |
| aiter/ops/triton/_triton_kernels/conv/causal_conv1d.py | Adds config-aware repr for causal conv1d forward and update kernels. |
| aiter/ops/triton/_triton_kernels/conv/causal_conv1d_update_single_token.py | Adds config-aware repr for single-token update and reshape+update kernels (helpers remain as device functions). |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
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
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.
No description provided.