Crash in test_slice on Intel CL

[alexfikl]

Describe the bug

test_slice still seems to be crashing on Intel CL even after #638.

To Reproduce

https://github.com/inducer/pyopencl/actions/runs/3273678656/jobs/5386201682

Environment (please complete the following information):

• OS: Windows, Linux
• ICD Loader and version: ocl-icd 2.3.1
• ICD and version: intel-opencl-rt 2022.1.0
• CPU/GPU: Intel(R) Xeon(R) Platinum 8370C CPU
• Python version: 3.10.8
• PyOpenCL version: master

cc: @isuruf

[inducer]

Cannot reproduce locally with the following conda env: https://gist.github.com/inducer/6009379264cd7c23ee0bdb9306a18a3b

[isuruf]

Fixed in #641

@inducer, we need to run the test a few times to reproduce the failure.

[alexfikl]

I managed to hit this again in
https://github.com/inducer/pyopencl/actions/runs/3388095221/jobs/5629660314
Any sense if it's related or a new issue?

[inducer]

IDK. The Intel CL runtime is unreasonably buggy. I don't know how much more time I'd like to waste chasing their regressions. I've banned another version in #653, maybe that'll help?

[inducer]

FWIW, the test_slice thing was never a pyopencl issue AFAICT.

[alexfikl]

IDK. The Intel CL runtime is unreasonably buggy. I don't know how much more time I'd like to waste chasing their regressions. I've banned another version in #653, maybe that'll help?

Is this their CPU runtime? Either way, agreed it's not worth that much effort at this point.

[inducer]

Yeah. Their GPU runtime is of far higher quality in my experience.