Fix compile issue for Marin qqq on sm<8.0 #1651
Open
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1651
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit bb389e5 with merge base 6ffe236 ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
cc @psinger Can you also check if this PR fixes your issue? Thank you. |
alexsamardzic
approved these changes
Feb 3, 2025
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Outdated suggestions cannot be applied.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.
Closes #1648
Test on Google Colab T4 (sm75): https://colab.research.google.com/drive/107-fKXymnK-QNCvnfXggTK8SSqy9oz8o?usp=sharing
The main issue is that
TORCH_CHECK_NOT_IMPLEMENTED()should not be called inside__global__function. However, I also believe the aggressive use of__CUDA_ARCH__guard will lead to undefined behavior, as outlined in here (https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#cuda-arch). Note that__CUDA_ARCH__is not defined in host code.Following the practice from @alexsamardzic, I simply let the kernel empty when
__CUDA_ARCH__ < 800, and add a runtime check in the torch function.at::cuda::getCurrentDeviceProperties()call.Run
python benchmarks/benchmark_marlin_qqq.pyon 4070Ti SUPERmain
this PR