Fix precompilation #52
Conversation
|
Test this please |
|
@williamfgc Can you rerun? |
|
Test this please |
|
Ok, that bot doesn't listen to me :-). |
|
Test this please |
|
The MacOS one seems to be a CI issue. |
|
Test this please |
|
@Geekdude do you know if cousteau is up? I couldn't run CI. Thanks! @michel2323 yes, the macos error is related to brew update. |
|
Test this please |
|
@michel2323 @williamfgc Looks like I got the tests running again on Cousteau, but the test is failing. |
|
@williamfgc Glad there's a CI. I had a typo that I only fixed in CUDA. Aaand a rerun please :-) . |
|
Test this please |
|
@Geekdude I think this is the CI failing. Something with ROCm 6.0. |
|
Test this please |
|
@michel2323 can you rebase main and push force this branch? I am trying a generic rocm module. |
michel2323
force-pushed
the
fix_precompilation
branch
from
6c4f711 to
7238b49
Compare
|
@williamfgc Rebased and squashed into one commit. |
|
Test this please |
|
@michel2323, Steve updated the ROCM version on Cousteau to rocm/6.0.2. Since only one working version will be deployed (different versions of ROCM don't play nice with each other), I will update the CI to load the latest ROCM instead of a specific version. |
|
Test this please |
|
Looks like there is still an error, but it is no longer erroring on the module load. |
michel2323
force-pushed
the
fix_precompilation
branch
from
7238b49 to
6d12657
Compare
|
This should work now. Ran the tests on an AMD GPU. |
|
Test this please |
|
Thanks @Geekdude and @michel2323 for taking care of CI issues. I will merge, the macos error is way upstream. |
This fixes the precompilation when using an extension for CUDA, ROCm, or oneAPI, by specializing the methods and not overwriting them.