[SYCL][UR][L0] Don't overwrite context's devices

[jandres742]

They are only set at contextCreate time.

[smaslov-intel]

LGTM

[jandres742]

@intel/llvm-gatekeepers : please merge when possible.

[psalz]

If I may, what is the intent behind this change? Should it have any user-facing consequences?

I'm asking because it appears that it somehow broke our nightly CI run for Celerity. It seems that kernels on the second GPU no longer execute.

What is also notable is that the device names listed are different.

Broken run:

[2023-08-15 10:11:04.562] [1] [info] Using platform 'Intel(R) Level-Zero', device 'Intel(R) Arc(TM) A770 Graphics' (automatically selected platform 0, device 1)

Working run:

[2023-08-14 22:13:43.544] [1] [info] Using platform 'Intel(R) Level-Zero', device 'Intel(R) Graphics [0x56a0]' (automatically selected platform 0, device 1)

[steffenlarsen]

Tag @jandres742 about the above comment.

[jandres742]

If I may, what is the intent behind this change? Should it have any user-facing consequences?

I'm asking because it appears that it somehow broke our nightly CI run for Celerity. It seems that kernels on the second GPU no longer execute.

What is also notable is that the device names listed are different.

Broken run:

[2023-08-15 10:11:04.562] [1] [info] Using platform 'Intel(R) Level-Zero', device 'Intel(R) Arc(TM) A770 Graphics' (automatically selected platform 0, device 1)

Working run:

[2023-08-14 22:13:43.544] [1] [info] Using platform 'Intel(R) Level-Zero', device 'Intel(R) Graphics [0x56a0]' (automatically selected platform 0, device 1)

if those tests were passing before with this fix, it means this change was just allowing them to pass incorrectly. could you please file an internal bug tracker, to take a closer look at those tests?

[psalz]

This does not appear to be an issue with Celerity. I have built a small reproducer and opened an issue.

[jandres742]

I have found some tests failing after this patch. Possible fix in #11122