New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[ROCm] Add ROCm support for CSR Sparse Matrix Ops #34800
[ROCm] Add ROCm support for CSR Sparse Matrix Ops #34800
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good thanks! I just a some minor nits.
tensorflow/python/kernel_tests/linalg/sparse/csr_sparse_matrix_grad_test.py
Outdated
Show resolved
Hide resolved
tensorflow/python/kernel_tests/linalg/sparse/csr_sparse_matrix_ops_test.py
Show resolved
Hide resolved
@chsigg, please re-review. thanks as always :) |
@rthadur gentle ping |
@deven-amd Can you please resolve conflicts? Thanks! |
The failures are because either * the subtests require support for complex type (which is not yet supported by ROCm) * or they require a GPU kernel implementation for the SparseMatrixAdd op (which is also not supported by ROCm, because the underlying hipSPARSE API routine - csrgeam - does not exist). There are also a couple of subtests commented out because hipSPARSE API errors out with an unknown error for them. Those will be looked into and fixed soon
bac4f54
to
5d1ccc1
Compare
@gbaned , rebased the PR to remove the merge-conflict. please merge. thanks |
bump. |
@gbaned, gentle ping |
@deven-amd here are some internal internal, can you please check once.
|
@rthadur , can you post the contents of then generated file thanks |
also please post the bazel build command that leads to this error |
@chsigg can you please assist ? |
…ocm_csr_sparse_matrix_support PiperOrigin-RevId: 289617600 Change-Id: Ic1aa3714126d7b867295ae386b6be643c1dc83e4
This PR adds ROCm support for CSR Sparse Matrix Ops.
The PR has 6 commits which organize the changes as per functionality being changed. Please review the commits indidvidually.
The file
cuda_sparse.h
should ideally be renamed togpu_sparse.h
once this PR is taken. I have not made that change a part of this PR. I can add a commit to make that change, assuming the reviewers are okay with it...please let me know if that is needed./cc @whchung @chsigg