ginkgo
ginkgo copied to clipboard
Bccoo gpu
Hi,
The balanced and compressed COO format (https://doi.org/10.1002/cpe.6515 , https://doi.org/10.1007/978-3-030-71593-9_7) work for all executors:
- reference and omp can use element and block compressions.
- cuda, hip and dpcpp only use block compression.
I would like to confirm if the format is properly implemented.
Thanks in advance!!
Best regards,
José I. Aliaga
label!
The label reg:testing has been added, assuming it means the test directories should be revised. The same for the label reg:benchmarking, but in this case only spmv benchmark has been modified. Thanks in advance!!
All tests have been successfully passed, except some of them using MPI. But MPI problems doesn't occur if the codes are launched properly using mpirun. Rebasing will probably fix these problems.
Which is the next step?
The first comments have been solved, and the new version is in the repository
I have included all of Thomas and Mike's comments. I will now push all the changes to the repository.