[Fix][TIR] LowerThreadAllreduce warp reduction mask #17307
Merged
Conversation
This file contains 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
The warp reduction implemented by "shuffle down" primitive takes a mask denoting the active threads within the warp that participate in this shuffle. Previously we compute the mask, while in practice we find that it results in "CUDA illegal instruction" error on NVIDIA H100 GPU when the mask is set, and the issue is gone if we do not update the mask. Therefore, this PR updates the allreduce lowering to remove the mask update. Confirmed the correctness on the following devices: * NVIDIA H100, * NVIDIA RTX 4090, * AMD Radeon 7900 XTX, * Apple M2 Ultra.
MasterJH5574
force-pushed
the
tvm-dev/2024-08-27-lower-allreduce-h100
branch
from
5345b05 to
b04fcf1
Compare
vinx13
approved these changes
Aug 28, 2024
|
awesome, I just came across this bug today.. |
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Outdated suggestions cannot be applied.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.
The warp reduction implemented by "shuffle down" primitive takes a mask denoting the active threads within the warp that participate in this shuffle.
Previously we compute the mask, while in practice we find that it results in "CUDA illegal instruction" error on NVIDIA H100 GPU when the mask is set, and the issue is gone if we do not update the mask. Therefore, this PR updates the allreduce lowering to remove the mask update.
Confirmed the correctness on the following devices: