Various SYCLcompat test failures

[joeatodd]

Describe the bug

Recently we have started seeing failures in the following tests:

• syclcompat/util/util_shift_sub_group_right.cpp
• syclcompat/memory/usm_allocations.cpp
• syclcompat/launch/launch_policy_lmem.cpp
• syclcompat/math/math_ops.cpp

These are occurring across various PRs unrelated to these tests, so I believe the breakage is already in sycl.

https://github.com/intel/llvm/actions/runs/11363119797/job/31606773901?pr=15719
https://github.com/intel/llvm/actions/runs/11364896963/job/31612247815?pr=15572
https://github.com/intel/llvm/actions/runs/11370543686/job/31630960112?pr=15728
https://github.com/intel/llvm/actions/runs/11367014317/job/31619261890?pr=15721

I am working on this with high priority. I have made this issue to help unblock unrelated PRs.

To reproduce

No response

Environment

• OS: [e.g Windows/Linux]
• Target device and vendor: [e.g. Intel GPU]
• DPC++ version: [e.g. commit hash or output of clang++ --version]
• Dependencies version: [e.g. the output of sycl-ls --verbose]

Additional context

No response

[pbalcer]

This might be related to the change that was already reverted here: #15720

Are you still seeing this issue after that revert?

[joeatodd]

@pbalcer thanks, I think you're right! I will wait to confirm before closing.

[joeatodd]

Seems to be fixed. I pulled tip into this PR and it's passing.

[pbalcer]

@joeatodd was there any other patch that went into that could have addressed the syclcompat issues in your opinion? We are hoping to reintroduce the reverted patch (see @winstonzhang-intel PR here: oneapi-src/unified-runtime#2234), and we'd like to avoid reintroducing the same problem.

[joeatodd]

@pbalcer I don't know to be honest; the only symptom was that 3 quite different tests started failing all at once. I would suggest that if the PR passes CI it could be reintroduced and then if these tests start failing again we'll have fairly high confidence of the root cause.

[pbalcer]

@joeatodd No worries, thanks!