Repository navigation
Conversation
Signed-off-by: tywuAMD <tianyuan.wu@amd.com>
Contributor
|
test is missing in test_CustomSchedule.py |
talumbau
self-requested a review
February 3, 2026 18:58
talumbau
requested changes
Feb 3, 2026
talumbau
left a comment
talumbau
left a comment
Collaborator
There was a problem hiding this comment.
The code overall looks good. A couple of changes to th test YAML are needed.
For trace/log directory this still needs:
- hipblaslt_bench run with CMS kernels added in and
--algo_method allrun so we can see the kernel against all others for a representative GEMM. - Rocprof att trace for CMS and non-CMS cases.
| - Range: [[128], [192], [1], [64, 64, 256]] | ||
| - Range: [[128], [192], [1], [1,1,64]] | ||
| - Range: [[128], [192], [1], [32, 64, 256]] | ||
| - Range: [[4096], [6144], [1], [64, 64, 256]] |
Collaborator
There was a problem hiding this comment.
Please don't add large GEMMs to the test file, especially the large M and N test values. These do not help check correctness.
| - Range: [[128], [192], [1], [1,1,64]] | ||
| - Range: [[128], [192], [1], [32, 64, 256]] | ||
| - Range: [[4096], [6144], [1], [64, 64, 256]] | ||
| - Exact: [2048, 3072, 1, 8192] |
Collaborator
There was a problem hiding this comment.
same issue here with this Exact GEMM test.
dileepr1
added a commit
that referenced
this pull request
May 29, 2026
Mirrors the legacy TEST_TO_IGNORE map in TheRock/build_tools/github_actions/test_executable_scripts/test_miopenprovider.py which excluded the miopen_plugin_integration_tests ctest entry on Windows+gfx110X only (TODO #3709). - shared/ctest/parse_test_categories.py: teach the parser an optional `_(windows|linux)` suffix on exclude_gpu_* keys. Keys without a suffix (e.g. exclude_gpu_gfx950) stay OS-agnostic and behave exactly as before, so every existing YAML continues to parse identically. Keys with a suffix are skipped at YAML-parse time on the wrong OS, so the entry never reaches the install file for that platform. The runtime label stays `ex_gpu_<arch>` so test_runner.py's GPU-arch matching needs no changes. Same change is independently being landed via users/dravindr/tf_rocprim (commit ef8eee2); whichever PR merges first, the other will be a clean no-op. - dnn-providers/miopen-provider/test_categories.yaml: add exclude_gpu_gfx110X_windows with test_patterns: ["*Integration*"]. All TEST_P classes in integration_tests/ are named IntegrationGpu* and none in tests/ contain "Integration", so the filter is a no-op on miopen_plugin_tests and a full exclusion on miopen_plugin_integration_tests, matching the legacy script's ctest --exclude-regex "miopen_plugin_integration_tests" exactly. Co-authored-by: Cursor <cursoragent@cursor.com>
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.
Motivation
Add CMS scheduling for TF32 192x256x32 NT.
Relevant PR:
TF32 192x256x32 TN: #3551
TF32 192x256x32 NN: #3677
Technical Details
Test Plan
UseCustomMainLoopSchedule: [0, 1]to compare CMS on and off.Test Result
Tensile
~15% uplift was observed when running Tensile
hipblaslt-bench
No gain versus the default kernel
Submission Checklist