Wu Yingcong

[UR] Test for unified runtime PR

1

UR PR: https://github.com/oneapi-src/unified-runtime/pull/1385

UR PR: https://github.com/oneapi-src/unified-runtime/pull/1637

[DeviceSanitizer] refactor options handling and fix use-after-free related problems

1

This patch: - refactor options handling. - for use-after-free, do not try to get allocated/released info when quarantine is not enabled(no such info anyway). - for findAllocInfoByAddress(), add an assertion...

Implement urKernelGetSuggestedLocalWorkSize

29

This PR try to implement the API `urKernelGetSuggestedLocalWorkSize`, discussed in https://github.com/oneapi-src/unified-runtime/issues/1270. SYCLOS PR: https://github.com/intel/llvm/pull/12902 Also fix: - For Level-Zero: when `LocalWorkSize` is provided, `urEnqueueKernelLaunch()` will read `LocalWorkSize` without respecting `workDim`.

[DeviceSanitizer] Handle the case of urMemGetNativeHandle getting a nullptr Device

1

We follow the implementation of L0 adapter and use the first device in the context instead.

[DeviceSanitizer] Exlcude shadow memory from coredump file for CPU device.

3

Currently, the coredump file includes shadow memory, which makes the coredump file crazily big and extremely slow to write to disk. This patch would not include the shadow memory in...

[github process] Why doesn't UR repo use "squash and merge" instead of merge all the PR's commits.

2

In intel/llvm repo, "squash and merge" is used over simply merging the PR branch into main branch. This will help keep a simple and clear main branch. I wonder why...

[UR][CUDA] enqueue//enqueue-test/urEnqueueKernelLaunchIncrementMultiDeviceTest fails

1

### Describe the bug CI job `Unified Runtime Pre Commit / Adapters (CUDA, UR_CUDA, -u 1001 --privileged --cap-add SYS_ADM` failed and still fails for multiple rerun. - https://github.com/intel/llvm/actions/runs/15703193260/job/44305438571?pr=19009 Failures: ```...