Repository navigation
fix(rocSHMEM): select correct wave size, fix descrepancies in functional tests, #AIROCSHMEM-560 - #14971
Merged
Conversation
✅ All Policy Checks Passed
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
Contributor
There was a problem hiding this comment.
🟢 Approval recommended
There are no active review issues for the wave-count test configuration changes.
No open issues 🎉
What changed in this PR
Updates rocSHMEM’s wave-granularity functional tests to preserve the intended wave count on both 32- and 64-wide GPU architectures.
Changes:
- Replaces fixed thread counts with
NUM_WFin CTest definitions for wave RMA, signal, synchronization, fence, and heatmap tests. - Mirrors those changes in the shell driver using
WAVE_SIZE, while preserving fixed-thread baseline cases.
| File | Description |
|---|---|
projects/rocshmem/tests/functional_tests/CTestFunctionalTests.cmake |
Uses runtime-derived wave counts for affected test configurations. |
projects/rocshmem/scripts/functional_tests/driver.sh |
Calculates corresponding thread counts from the existing WAVE_SIZE variable. |
Review effort: Balanced
Resource usage report (4gfx1201) |
Resource usage report (8gfx942) |
amirsojoodi
changed the title
Resource usage report (4gfx1201) |
Resource usage report (8gfx942) |
edgargabriel
approved these changes
Nov 5, 2026
Contributor
TheRock Submodule Bump ActivityNewest first
|
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Several wave-granularity functional tests hardcoded thread counts (64/128/256/1024) to mean "N wavefronts," assuming a 64-wide wave. On gfx1250 and other 32-wide architectures this silently doubled the intended wave count per workgroup.
CTestFunctionalTests.cmake: migratedwaveput,waveputnbi,waveget,wavegetnbi,teamwavesync,teamwavebarrier,fence_putwavesignal,fence_putwavenbichunksfrom literalTHREADStoNUM_WFwhere the test genuinely creates one ROCSHMEM context per wave (MAX_NUM_CONTEXTS = WORKGROUPS * NUM_WF), which derives the thread count from the HIP-runtime-detectedwarpSize(--num-wf); the same mechanismbroadcast_wave/fcollect_wave/tile_*_wave*already use. Also migrated theTHREADS 64data point ofwaveputsignal,waveputsignalnbi, andwavesignalfetchtoNUM_WF 1so it means "1 wave" on every architecture.NUM_WFmigration for entries that are actually workgroup-scope (one context per workgroup, not per wave):teamwavesync/teamwavebarrier'sWORKGROUPS 32/39rows,fence_putwavesignal'sWORKGROUPS 8/32rows, and alltile_*_wg_*rows. For these,NUM_WFwould have scaledMAX_NUM_CONTEXTSby wave count and changed test semantics, not just the thread count; they keep literalTHREADSvalues (256/1024) representing a fixed max workgroup size.driver.sh: mirrored the survivingNUM_WFmigrations using the existingWAVE_SIZEshell variable (already detected viarocminfofor gfx1100/gfx1201/gfx1250), and mirrored the reverted entries with the same literalTHREADSvalues as the cmake file, so the two stay in lockstep.tile_put_wave_3d/tile_get_wave_3d/tile_put_wave_4d/tile_get_wave_4dare declaredNUM_WF 4in cmake but driver.sh was still passing$WAVE_SIZE(1 wave) instead of$WAVE_SIZE * 4.The baseline
THREADS 32entries of the signal tests are untouched; that literal is shared with the non-waveputsignal/wgputsignalvariants, not wave-size-derived.Issue Tracking
JIRA ID: AIROCSHMEM-560