[Fix][TOPI] Fuse GPU sort grid dimensions to avoid CUDA launch failures on large inputs#19938
Open
bujjibabukatta wants to merge 1 commit into
Open
[Fix][TOPI] Fuse GPU sort grid dimensions to avoid CUDA launch failures on large inputs#19938bujjibabukatta wants to merge 1 commit into
bujjibabukatta wants to merge 1 commit into
Conversation
…es on large inputs
Contributor
|
Warning Gemini encountered an error creating the review. You can try again by commenting |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The GPU sort/topk kernels currently map nthread_bx to blockIdx.x and nthread_by to blockIdx.y. For large inputs (e.g. topk on a tensor of shape (60, 1000, 1000)), the product nthread_by * nbz can easily exceed 65535 which is the hardware limit for gridDim.y on CUDA, causing CUDA_ERROR_INVALID_VALUE at kernel launch.
This patch fuses the two block dimensions into a single blockIdx.x axis (limit 2^31-1) and reconstructs virtual bx/by indices inside the kernel with modulo and integer division. All five call sites of _get_threads are updated consistently.
Fixes #19549