> it seems it worked well, and the performance difference between CPU > and GPU is notable. Yes it is a test case for me and I will upload some real OpenCL test with pyfai, silx and pyvkfft thanks a lot for the rocm ci infra :)) > Also, since there is no direct dependency on ROCm packages, our CI will > never trigger for clpeak. I will need to adapt our scheduler to account > for opencl-icd. the dependencies are only in the test. This is not enough ? Fred