Repository navigation
test(rocsparse): add ELL triangular solve and SELL SpMV unit tests (ROCM-33114) - #13879
Merged
kliegeois merged 12 commits intoNov 5, 2026
Merged
Conversation
These paths are only covered by YAML, which misses the slice-size boundaries and the partial last slice.
✅ All Checks Passed — Ready for Review
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
|
🎉 All checks passed! This PR is ready for review. |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## develop #13879 +/- ##
===========================================
+ Coverage 76.89% 77.02% +0.14%
===========================================
Files 2886 2883 -3
Lines 481037 479235 -1802
Branches 72377 72023 -354
===========================================
- Hits 369847 369118 -729
+ Misses 86832 85868 -964
+ Partials 24358 24249 -109
*This pull request uses carry forward flags. Click here to find out more. 🚀 New features to boost your workflow:
|
jsandham
reviewed
Oct 20, 2026
…3114) Drop the file-header note and say in the test comment what the transpose partial-last-slice case checks and what it does not.
ntrost57
reviewed
Oct 23, 2026
ntrost57
reviewed
Oct 23, 2026
ntrost57
reviewed
Oct 23, 2026
ntrost57
reviewed
Oct 23, 2026
ntrost57
reviewed
Oct 23, 2026
ntrost57
reviewed
Oct 23, 2026
ntrost57
reviewed
Oct 23, 2026
Every input is a small integer or dyadic fraction and every partial sum stays below 2^24, so the old 1e-4 absolute tolerance was already an exact comparison for most values. Say so and use EXPECT_EQ. Drop the unused <cmath> include.
…sition (ROCM-33114) Preset the zero-pivot output to -2 so a get_output that never writes it is caught, and require rocsparse_status_success exactly when the position is -1 and rocsparse_status_zero_pivot otherwise, as testing_sptrsv_ell does.
…s (ROCM-33114) run_ellsv and run_v2_spmv_sell are void helpers, so a failing ASSERT inside them only returned from the helper and the test went on to check stale output. Wrap every call in ASSERT_NO_FATAL_FAILURE.
…-33114) The m = slice = 256 cases ran one full slice with iter == 1, so neither half of the large-slice row_in_range guard could be false. Replace them with slice 128 and 129 at m = 600 and slice 300 at m = 700, in both directions: the last small-slice size, the first large-slice size with threads outside the slice, and two passes per slice, each with a partial last slice; m = 600 also leaves a surplus block in the launch grid. y now carries a sentinel tail past its length, so a non-transpose write past row m fails the test. Add a small-slice case per direction with 10 entries per row, so the strided loop takes a second step and every reduction level combines real data. build_bidiagonal_sell becomes build_banded_sell with a depth.
…M-33114) Every transpose case used beta = 0 on a zero y with m == n, so dropping the scale_array call before the transpose kernels, or giving it m instead of n, passed. Seed y with 7 in every beta = 0 case, and add a transpose case with beta = 3 and n > m, where y[m..n-1] only get the beta scaling.
ntrost57
reviewed
Oct 29, 2026
…CM-33114) ROCSPARSE_WITH_ELL_TRSV was retired with the 5.1 feature flags, so the EllSv tests were compiled out and only the skip stub ran. Drop the guards and the stub.
ntrost57
approved these changes
Oct 30, 2026
jsandham
reviewed
Nov 3, 2026
jsandham
approved these changes
Nov 3, 2026
…M-33114) run_ellsv created its sptrsv descriptor up front and destroyed it only on the last line, so any failed ASSERT_* in between returned early and leaked it. run_v2_spmv_sell and the EllSv test bodies had the same pattern for their spmat, dnvec and spmv descriptors. Hold them in the level2 RAII guards from unit_test_device_level2_common.hpp (sptrsv_descr_guard, spmv_descr_guard, spmat_guard, dnvec_guard), which the other level-2 unit tests already use.
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.
JIRA ID : ROCM-33114.
Summary
sellmv_device.h. It is based on develop after testintel: fix severe issue in sellmvt_device (created by kliegeoi (@kliegeois)) #13339, which guards the padding-thread read ofxin the small-slice transpose kernel.Test plan
rocsparse-unit-test-devicefor gfx1201;EllSv.*:SellMv.*and the full binary pass.row < mfrom the non-transpose large-slice guard makesSellMv.none_slice_size_boundariesfail.