Repository navigation
Compatibility with libsycl.so.6 #1022
Description
Activity
Maybe
dpnpandnumba_dpexin the same channel will be affected too ?In the meantime the dpcpp runtime can be pinned in conda, e.g with
export CONDA_DPEX_ENV_NAME=my-dpex-env conda create --yes --name $CONDA_DPEX_ENV_NAME \ --channel dppy/label/dev \ --channel conda-forge \ --channel intel \ `# TODO: remove <2023.0.0 pin after dppy/label/dev is updated` \ numba-dpex "intel::dpcpp_linux-64<2023.0.0"dpctl 0.14.1dev1 built with DPC++ 2023.0.0 has been uploaded to dppy/label/dev. It would be compatible with
libsycl.so.6Thank you for the bumps.
I'm getting this new error though:
AttributeError: 'dpctl._sycl_device.SyclDevice' object has no attribute 'sub_group_sizes'(using the custom build from this Dockerfile)
When on CPU (
SYCL_DEVICE_FILTER=cpu) I'm also havingdpctl._sycl_queue.SyclKernelSubmitError: Kernel submission to Sycl queue failed.errors. How can I debug such errors ? (reported in #1023)
It seems the conda package on channel
dppy/label/devis compatible withlibsycl.so.5but notlibsycl.so.6. Butinteljust released2023.0.0versions of all the dependencies oncondaandpypiand it shipslibsycl.so.6thus breaking thedpctlconda distribution:Would it be possible to update the build to work with
libsycl.so.6?