Fix sddmm2 when nnz=0 #300
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
Also add cudaGetLastError which was missing
blefaudeux
reviewed
May 10, 2022
| @@ -158,14 +158,14 @@ def test_sddmm_sputnik(device): | |||
|
|
|||
|
|
|||
| @cuda_only | |||
| @pytest.mark.parametrize("prob", [0.5, 1]) | |||
There was a problem hiding this comment.
thanks for adding this, nice catch
blefaudeux
reviewed
May 10, 2022
| @@ -539,6 +546,7 @@ torch::Tensor sddmm_cuda_csr( | |||
| D2.data_ptr<float>(), | |||
| out.data_ptr<float>()); | |||
| } | |||
| AT_CUDA_CHECK(cudaGetLastError()); | |||
blefaudeux
approved these changes
May 10, 2022
There was a problem hiding this comment.
LGTM, thanks a lot @fmassa for diving in and the very quick fix
|
the mypy error is fixed on main, will be fine on landing |
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Outdated suggestions cannot be applied.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.
One of the internal implementations of sampled dense dense matrix multiplication (sddmm) that we have had two issues:
cudaGetLastErrorcall in the end of the functions. So errors in this kernel would only be reported at the next function invocation, misleading the true location of the issue.The PR fixes this by returning early when
nnz=0(which is fine as there is no data in the tensor anyway), and also add cudaGetLastError which was missing.Should fix the errors present in #263