Branch: refs/heads/main
Revision: 7fff9f5
Author: e-kayrakli
Link: Add a new GPU test for bulk data transfer performance by e-kayrakli · Pull Request #19569 · chapel-lang/chapel · GitHub
Log Message:
Merge pull request #19569 from e-kayrakli/gpu-data-pp
Add a new GPU test for bulk data transfer performance
This PR adds 3 GPU tests.
-
two are used for CPU/GPU worksharing and one of them closely resembles a code
I have in the release notes for 1.26. -
a performance test I used for understanding whether host-to-device bulk
transfer works (it does!), and to assess its impact on performance for cases
where the data is offloaded for a specific computation, and than copied back
to the host. This test currently runs in correctness mode as we don't have GPU
performance testing infrastructure set up yet. GPU performance testing is
requested in https://github.com/Cray/chapel-private/issues/2564.
The tests pass with the GPU locale model.
[Reviewed by @stonea]
Modified Files:
A test/gpu/native/dataPingPong.chpl
A test/gpu/native/dataPingPong.compopts
A test/gpu/native/dataPingPong.execopts
A test/gpu/native/dataPingPong.good
A test/gpu/native/dataPingPong.prediff
A test/gpu/native/multiGPU/EXECOPTS
A test/gpu/native/multiGPU/PREDIFF
A test/gpu/native/multiGPU/README
A test/gpu/native/multiGPU/multiGPU.numlaunches
A test/gpu/native/multiGPU/worksharing.chpl
A test/gpu/native/multiGPU/worksharing.numlaunches
A test/gpu/native/multiGPU/worksharingBasic.chpl
A test/gpu/native/multiGPU/worksharingBasic.numlaunches
R test/gpu/native/multiGPU/multiGPU.execopts
R test/gpu/native/multiGPU/multiGPU.prediff
Compare: https://github.com/chapel-lang/chapel/compare/182f1035d759...7fff9f529f38