Repository navigation
fix(rocfft): fall back to gather-scatter when opt multi-device build throws - #11996
Merged
Merged
Conversation
✅ 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❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #11996 +/- ##
===========================================
+ Coverage 70.23% 70.25% +0.03%
===========================================
Files 2810 2810
Lines 462535 462543 +8
Branches 68105 68104 -1
===========================================
+ Hits 324828 324951 +123
+ Misses 114187 114091 -96
+ Partials 23520 23501 -19
*This pull request uses carry forward flags. Click here to find out more.
🚀 New features to boost your workflow:
|
regan-amd
marked this pull request as ready for review
September 12, 2026 00:48
Base automatically changed from
users/regan-amd/fft_bump_post_10_1
to
develop
September 12, 2026 05:25
…throws BuildOptMultiDevicePlan could let an exception escape to rocfft_plan_create_internal (e.g. get_most_balanced_proc_pair failing to factor an unbalanceable pencil grid), which returned rocfft_status_failure instead of falling through to the single-device plan path. Give it the same function-try-block exception handling that BuildMultiDevicePlan already had, so it logs the exception, discards any partial plan state, and returns false to let plan creation fall back as intended. Factor the shared handler into rocfft_plan_t::discard_failed_multi_device_plan and reuse it from both functions to align the two implementations and remove duplication. Add a multi-GPU adhoc regression token that exercises the throw, plus a changelog entry.
…throw invariant Remove the duplicated doc comment on discard_failed_multi_device_plan (kept on the header declaration) and document that BuildOptMultiDevicePlan must throw rather than return false past the first plan-state mutation, since BuildMultiDevicePlan no longer clears partial state on entry.
regan-amd
force-pushed
the
users/regan-amd/duct_tape_mgpu
branch
from
September 12, 2026 05:25
236148c to
844940d
Compare
evetsso
approved these changes
Sep 14, 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
BuildOptMultiDevicePlancould let an exception escape torocfft_plan_create_internal(e.g.get_most_balanced_proc_pairfailing to factor an unbalanceable pencil grid), which returnedrocfft_status_failureinstead of falling through to the single-device plan path.JIRA ID: AIFFT-539
Technical Details
BuildMultiDevicePlanalready had, so it logs the exception, discards any partial plan state, and returnsfalseto let plan creation fall back as intended.rocfft_plan_t::discard_failed_multi_device_planand reuse it from both functions to align the two implementations and avoid duplication.Test Plan
Added a multi-GPU ad-hoc regression token that exercises the throw.
Test Result
Tests pass for the added ad-hoc token on a system with 8
gfx90a(plan creation failed prior).Submission Checklist