issue opencl - segfault about silx HOT 5 CLOSED

some possible hints from @pierrepaleo

from clinfo there in /etc/OpenCL/vendors/intel64.icd there is a /opt/intel_opencl_icd/icd/libintelocl.so and CPU is AMD (cat /proc/cpuinfo)

from silx.

and:

• silx 2.0 tested with pyopencl 2024.1 and pyopencl 2023.1.4 -> both are failling
• silx 1.1 tested with pyopencl 2023.1.4 only -> works

from silx.

indeed it even seems that a simple call to python -c "from silx.opencl.utils.ExternalRessources import ExternalRessources" make it segfault. But the file on itself seems to have nothing 'exotic'

from silx.

We had a quick look with @t20100 and there is two issues:

• as @pierrepaleo noticed the opencl setup seems to be wrong (it has a /etc/OpenCL/Vendors/intel.icd when has a AMD CPU and not any intel device). Furthermore we also have a seg fault with the following code:

(test_opencl) opid19@iccid191:/nobackup/iccid1911/tomotools$ python /tmp_14_days/payno/test.py 
Choose platform:
[0] <pyopencl.Platform 'NVIDIA CUDA' at 0x1eb47f0>
[1] <pyopencl.Platform 'Intel(R) OpenCL' at 0x1ed01c0>
Choice [0]:1
Set the environment variable PYOPENCL_CTX='1' to avoid being asked again.
Segmentation fault (core dumped)

• then there is most likely a design issue with the silx.opencl.common.py module and more specifically about the OpenCL class.
  During import will be executed. And this is what create a seg fault during import. Even if no-opencl is used when importing silx.opencl or other module linked.
  Can we get your point of view on it @kif ? Do you think it would make sense to redesign this part ?

from silx.

Replaced by more focused issue #4088

from silx.