[TRITON] Standardize get_config for GEMM kernels and have a unified format for their configs - #1587
Merged
Merged
Conversation
Boss2002n
commented
Dec 9, 2025
Contributor
There was a problem hiding this comment.
Pull request overview
This PR refactors GEMM configuration management by introducing a centralized get_gemm_config utility function to replace manual configuration loading across multiple kernel files. The configuration keys in JSON files are standardized from descriptive names like "small", "medium", and "large" to systematic M-dimension-based keys like "M_LEQ_8", "M_LEQ_32", and "any". This consolidation removes redundant caching logic and file handling code, improving maintainability and consistency across the codebase.
Reviewed changes
Copilot reviewed 106 out of 135 changed files in this pull request and generated 6 comments.
| File | Description |
|---|---|
| aiter/ops/triton/utils/gemm_config_utils.py | New utility module providing centralized GEMM configuration loading with get_gemm_config and compute_splitk_params functions |
| aiter/ops/triton/configs/gemm/*.json | Standardized configuration keys from descriptive names to M-dimension-based format (e.g., "small" → "M_LEQ_16", "xlarge" → "any") |
| aiter/ops/triton/_triton_kernels/*.py | Replaced manual configuration loading with calls to centralized get_gemm_config utility across 20+ kernel files |
Comments suppressed due to low confidence (1)
aiter/ops/triton/configs/gemm/gfx950-BATCHED_GEMM_PREQUANT-AFP4WFP4-N=128-K=512.json:1
- An entire configuration block for "small_M16" was removed. While consolidation is good, ensure that cases where M is between 16 and 32 are still properly handled by the M_LEQ_32 configuration, as the removed config had different parameter values.
{
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Boss2002n
force-pushed
the
amd/satya/standardise_GEMM_configs
branch
from
December 10, 2025 07:06
657e056 to
90323e1
Compare
Boss2002n
requested review from
azaidy,
brunomazzottiamd,
cagrikymk and
lucas-santos-amd
December 10, 2025 13:54
Contributor
Author
|
Sorry about the massive PR :( |
Contributor
Author
|
Note: The new files changed view for GitHub does not render the file diffs, we have to switch back to the old view for the diff to be generated due to the large number of files |
Boss2002n
force-pushed
the
amd/satya/standardise_GEMM_configs
branch
from
December 12, 2025 13:41
90323e1 to
72bd2a3
Compare
brunomazzottiamd
requested changes
Dec 12, 2025
brunomazzottiamd
left a comment
Contributor
There was a problem hiding this comment.
I have no strong objections to this PR, it's OK! However I have some suggestions to harden get_gemm_config. It's a very important function, called by all GEMM kernels. I think it's worth investing in it to avoid future bugs.
cagrikymk
reviewed
Dec 12, 2025
Boss2002n
force-pushed
the
amd/satya/standardise_GEMM_configs
branch
3 times, most recently
from
December 14, 2025 21:31
2fe2bf1 to
24ea90d
Compare
…nfig` utility across multiple GEMM kernel files Updated configuration keys in JSON files to improve clarity and consistency, changing "large" and "small" to more descriptive keys like "M_GEQ_4096" and "any" Removed redundant caching logic and file handling 1st set of files
brunomazzottiamd
previously approved these changes
Dec 15, 2025
brunomazzottiamd
previously approved these changes
Dec 15, 2025
cagrikymk
previously approved these changes
Dec 15, 2025
Boss2002n
dismissed stale reviews from cagrikymk, azaidy, and brunomazzottiamd
via
December 15, 2025 23:00
1c0d6cb
cagrikymk
approved these changes
Dec 15, 2025
zhuyuhua-v
pushed a commit
that referenced
this pull request
Dec 17, 2025
…ormat for their configs (#1587) * Replaced manual configuration loading with a centralized `get_gemm_config` utility across multiple GEMM kernel files Updated configuration keys in JSON files to improve clarity and consistency, changing "large" and "small" to more descriptive keys like "M_GEQ_4096" and "any" Removed redundant caching logic and file handling * enable LRU caching * fix typecasting from list to tuple
ZhangLirong-amd
pushed a commit
that referenced
this pull request
Dec 29, 2025
…ormat for their configs (#1587) * Replaced manual configuration loading with a centralized `get_gemm_config` utility across multiple GEMM kernel files Updated configuration keys in JSON files to improve clarity and consistency, changing "large" and "small" to more descriptive keys like "M_GEQ_4096" and "any" Removed redundant caching logic and file handling * enable LRU caching * fix typecasting from list to tuple
valarLip
pushed a commit
that referenced
this pull request
Mar 18, 2026
…ormat for their configs (#1587) * Replaced manual configuration loading with a centralized `get_gemm_config` utility across multiple GEMM kernel files Updated configuration keys in JSON files to improve clarity and consistency, changing "large" and "small" to more descriptive keys like "M_GEQ_4096" and "any" Removed redundant caching logic and file handling * enable LRU caching * fix typecasting from list to tuple
1 task
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.
Replaced manual configuration loading with a centralized
get_gemm_configutility for GEMM kernel filesUpdated configuration keys in JSON files to be consistent, changing "large" and "small" to more descriptive keys like "M_GEQ_4096" and "any"
Removed redundant caching logic and file handling