Skip jagged_softmax large-grid tests on ROCm (exceed HIP 2^32 launch limit) - #6022
Open
gchalump wants to merge 1 commit into
Open
Skip jagged_softmax large-grid tests on ROCm (exceed HIP 2^32 launch limit)#6022gchalump wants to merge 1 commit into
gchalump wants to merge 1 commit into
Conversation
…limit)
Summary:
test_jagged_softmax_{forward,backward}_large_grid intentionally build an
oversized launch grid (D=20000, B=2047 => grid.x * grid.y * block =
20000 * 2047 * 128 = 5.24B threads) to exercise the large-grid path. On ROCm this
exceeds HIP's hard 2^32 threads-per-launch limit and the launch is rejected
(CUDA silently wraps, so it passes there). This surfaced on gfx942 / ROCm 7.1 in
run 29418021816.
Rather than special-case the kernel launch for this synthetic oversized grid,
skip the two tests on ROCm via skipIfRocm. The large-grid path stays exercised
on CUDA; ROCm coverage of jagged_softmax at normal grid sizes is unaffected.
Differential Revision: D112352791
Contributor
|
@gchalump has exported this pull request. If you are a Meta employee, you can view the originating Diff in D112352791. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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:
test_jagged_softmax_{forward,backward}_large_grid intentionally build an
oversized launch grid (D=20000, B=2047 => grid.x * grid.y * block =
20000 * 2047 * 128 = 5.24B threads) to exercise the large-grid path. On ROCm this
exceeds HIP's hard 2^32 threads-per-launch limit and the launch is rejected
(CUDA silently wraps, so it passes there). This surfaced on gfx942 / ROCm 7.1 in
run 29418021816.
Rather than special-case the kernel launch for this synthetic oversized grid,
skip the two tests on ROCm via skipIfRocm. The large-grid path stays exercised
on CUDA; ROCm coverage of jagged_softmax at normal grid sizes is unaffected.
Differential Revision: D112352791