New Intel error on jlse

[mrnorman]

Using current main, hash: d706e2f44fb8f5651cb2f6f3748e6fd3261a70ae

qsub -I -t 120 -n 1 -q arcticus

source source jlse_gpu_O3.sh
make -j
make test

[ac.normanmr@arcticus09:~/YAKL/unit/build/machines/jlse] >:O ./CArray/CArray 
terminate called after throwing an instance of 'sycl::_V1::runtime_error'
  what():  No device of requested type 'info::device_type::gpu' available. Please check https://software.intel.com/content/www/us/en/develop/articles/intel-oneapi-dpcpp-system-requirements.html -1 (PI_ERROR_DEVICE_NOT_FOUND)
Aborted

[mrnorman]

@abagusetty, @omarkahmed, any ideas what might cause this?

[abagusetty]

that is new one, will take a look. arcticus is deprecated hardware to be replaced with florentia (aka PVC). Looks like a new environment variable to be updated in the script.

[mrnorman]

I got a florentia_debug node, and things mostly pass there with that environment file. At least that bug isn't present.

[omarkahmed]

@mrnorman , this looks like an issue where the newer user mode driver (UMD) breaks on this deprecated GPU (ATS). Try executing the following after loading your modules: module switch -f intel_compute_runtime/release/agama-devel-551 intel_compute_runtime/release/agama-prerelease-475

[abagusetty]

Great, any ATS related issues wouldn't be relevant. Do you think we should drop the ATS files in the test-suite and move to florentia instead as that is more relevant ?

[mrnorman]

I'm fine with that. We do seem to have a problem with the streams unit test on florentia still: https://github.com/mrnorman/YAKL/issues/103

[mrnorman]

Moved to Kokkos or omitted by streams removal.