Repository navigation
fix(rocfft): fix GPU faults of fused Bluestein with unit batch distance - #13020
Merged
regan-amd merged 3 commits intoOct 12, 2026
Conversation
❌ PR Check — Action Required
📖 Need help? See the Policy FAQ for details on every check and how to fix failures. |
Pre-commit check failed⛔ pre-commit failed Please run locally:
This repo uses |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## users/regan-amd/fix_fused_bluestein_buffer_assignment #13020 +/- ##
=========================================================================================
+ Coverage 71.88% 72.01% +0.13%
=========================================================================================
Files 2876 2876
Lines 475561 475562 +1
Branches 70287 70288 +1
=========================================================================================
+ Hits 341856 342473 +617
+ Misses 109940 109319 -621
- Partials 23765 23770 +5
*This pull request uses carry forward flags. Click here to find out more.
🚀 New features to boost your workflow:
|
malcolmroberts
approved these changes
Oct 8, 2026
regan-amd
force-pushed
the
users/regan-amd/fix_fused_bluestein_unit_distance
branch
from
October 8, 2026 21:09
ef7e865 to
47bab4d
Compare
evetsso
approved these changes
Oct 8, 2026
regan-amd
force-pushed
the
users/regan-amd/fix_fused_bluestein_unit_distance
branch
from
October 9, 2026 05:35
47bab4d to
7b21aae
Compare
eng-flavio-teixeira
approved these changes
Oct 9, 2026
Fused multi-kernel Bluestein plans with interleaved batches (idist = odist = 1, istride = ostride = batch) fail in rocfft_plan_create or fault on the GPU. Add a suite covering such layouts (forward/inverse, single/double, in-place/out-of-place), plus general non-unit stride/dist cases.
The L1D_CC "dist 1" special case reshapes its children's kernel launch geometry, swapping the batch and sub-FFT dimensions. Fused Bluestein stages have already set their Bluestein strides and dists against the original geometry, so after the reshape the kernels index past the Bluestein buffer and fault on the GPU. Skip the reshape for fused Bluestein nodes.
regan-amd
force-pushed
the
users/regan-amd/fix_fused_bluestein_unit_distance
branch
from
October 9, 2026 18:31
7b21aae to
ebd3410
Compare
regan-amd
deleted the
users/regan-amd/fix_fused_bluestein_unit_distance
branch
October 12, 2026 15:57
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
JIRA ID: AIFFT-13, AIFFT-551 (issue found while working on that)
1D complex transforms of lengths involving large prime factors that use the multi-kernel fused Bluestein algorithm fail for inner-batched layouts (unit batch distance, i.e. idist = odist = 1 and istride = ostride = batch). On develop, such plans either fail in
rocfft_plan_createor fault on the GPU at execution. With the previous PR in this stack, plan creation succeeds, but execution still faults.Technical Details
The L1D_CC "dist 1" special case (
CC1DNode::AssignParams_internal) reshapes its children's kernel launch geometry, swapping the batch and sub-FFT dimensions so that the batch dimension is the adjacent one. Fused Bluestein stages have already set their Bluestein strides and dists against the original geometry, so after the reshape their kernels index past the Bluestein buffer (GPU memory access fault in thesbcc_*_chirp_mulkernels).The reshape is now skipped for fused Bluestein nodes (
BT_MULTI_KERNEL_FUSED), keeping their launch geometry consistent with their Bluestein strides. This keeps the fused path (no fallback to non-fused Bluestein).Test Plan
fused_bluestein_non_unit_strideaccuracy suite (9 tokens):prime_*,adhoc*,fused_bluestein*and all*_idist_1_*suites in full, plus a 5% sample of all other single-device suites.Test Result
rocfft_plan_create;complex_inverse_len_8191_single_op_batch_8_..._idist_1_odist_1faults on the GPU (sbcc_inv_chirp_mul).Submission Checklist