Fix GCC OpenACC memory error in batch spline GPU code#222
Open
krystophny wants to merge 2 commits intomainfrom
Open
Fix GCC OpenACC memory error in batch spline GPU code#222krystophny wants to merge 2 commits intomainfrom
krystophny wants to merge 2 commits intomainfrom
Conversation
krystophny
temporarily deployed
to
github-pages
GitHub Actions
Inactive
krystophny
temporarily deployed
to
github-pages
GitHub Actions
Inactive
Remove SAVE attribute from work arrays in batch_interpolate_2d and batch_interpolate_3d. The persistent work arrays caused GPU memory errors (cuStreamSynchronize: illegal memory access) on the second iteration when using GCC with nvptx offload target. The fix allocates work arrays fresh on each call and properly manages their GPU memory lifetime with enter/exit data directives. This ensures clean GPU memory state between calls, avoiding stale mappings that caused the memory corruption. Also adds !$acc wait after exit data delete operations in destroy subroutines to ensure proper synchronization before host memory is freed. This fixes the memory error that occurred when running test_batch_splines with GCC OpenACC (GPU mode): Test 1 (Meiss) passed but Test 2 (Albert) failed with illegal memory access. With this fix, all tests pass.
Remove allocation reuse optimization - always allocate fresh to avoid potential stale GPU memory mappings. This is simpler and more robust for the GCC OpenACC nvptx backend. Note: GCC 16 nvptx still has memory issues on second iteration that require further investigation. Use ACC_DEVICE_TYPE=host as workaround.
krystophny
force-pushed
the
fix/openacc-wait-after-delete
branch
from
9ab4de3 to
28d69ed
Compare
Contributor
CI Feedback 🧐A test triggered by this PR failed. Here is an AI-generated analysis of the failure:
|
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
1 participant
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.
Summary
Problem
When using GCC with OpenACC nvptx offload target, the batch spline tests failed with:
The error occurred on the second iteration (Test 2: Albert field batching) while Test 1 (Meiss field batching) passed. The root cause was persistent (SAVE) work arrays that maintained stale GPU memory mappings between calls.
Solution
Changed work arrays from persistent (SAVE) to local allocation on each call. This ensures:
Testing
With this fix, all batch spline tests pass with GCC OpenACC GPU backend: