-
fcfbef80 · Merge branch 'CUDA' into 'master'
- ... and 1 more commit. Compare ae05d616...fcfbef80
-
879ee872 · removed the unused `cuda_block_size` for the `gpu.generate_benchmark`
joined project
pycodegen / pystencils-benchmark
-
d38a9324 · using pytest skip if there is no nvcc or gpu available
Good point, I think skipping is the the most reasonable in that case
This is dangerous, as it makes the test pass, although it was not tested completely. I'd use pytest.fail
, pytest.skip
, or pytest.exit
, depending on...
-
1e542f17 · Skipping compiling and running cuda kernels if cuda or gpu is not
-
82ce1d7d · Fix the missing constants for the gpu main file and added a kernel ...
-
6e88d389 · Using cuda as a base for the docker container to also test the gpu
-
4b1f3f53 · fixed the _add_launch_bounds and also added some small tests
-
4ee400e9 · added the new packages to the setup.cfg and the new templates to the
-
9140da63 · Added a parameter to insert a launch bounds to the kernel
Alright, fine for me
I think it would be a bit cleaner to do that in another MR
Would it make sense to also add ROCm support? Mostly this would be just a renaming for example: #include <cuda_runtime.h>
--> #include <hip_runtime.h>
...
It means this:...
Thank you and thank you for your review