[DevTSAN] Treat each work item as a thread for GPU device

[zhaomaosu]

• Update tests to make it more robust

[Copilot]

Pull Request Overview

This PR updates the ThreadSanitizer tests and the runtime to treat each work item as a thread on the GPU and to improve robustness. Key changes include:

• Removing the check_usm.cpp test file.
• Updating several tests to use a larger ND-range (128, 8) and increasing buffer sizes.
• Modifying the epoch increment and GPU thread model in tsan_rtl.cpp to treat each work item as a thread.

Reviewed Changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
sycl/test-e2e/ThreadSanitizer/check_usm.cpp	Removed obsolete test file.
sycl/test-e2e/ThreadSanitizer/check_unaligned_access.cpp	Updated ND-range size from 32 to 128 for enhanced test robustness.
sycl/test-e2e/ThreadSanitizer/check_shared_usm.cpp	Added test with updated ND-range and USM allocation for shared memory.
sycl/test-e2e/ThreadSanitizer/check_host_usm.cpp	Added test with updated ND-range and USM allocation for host memory.
sycl/test-e2e/ThreadSanitizer/check_device_usm.cpp	Added test with updated ND-range and USM allocation for device memory.
sycl/test-e2e/ThreadSanitizer/check_device_global.cpp	Updated ND-range from 32 to 128 for global memory access race check.
sycl/test-e2e/ThreadSanitizer/check_buffer.cpp	Increased static buffer size from 16 to 128.
sycl/test-e2e/ThreadSanitizer/check_access16.cpp	Updated ND-range from 32 to 128 for improved test stability.
libdevice/sanitizer/tsan_rtl.cpp	Adjusted epoch increment from atomicAdd to a post-increment operator and updated comment to reflect that each work item is now treated as a thread.

Comments suppressed due to low confidence (1)

libdevice/sanitizer/tsan_rtl.cpp:79

• Replacing the call to atomicAdd with a post-increment operator may compromise atomicity and thread safety. Consider reverting to the atomicAdd function or ensuring that this update is safe in the GPU work item context.

return TsanLaunchInfo->Clock[sid].clk_[sid]++;

[zhaomaosu]

Hi @intel/llvm-gatekeepers, this PR is ready to be merged. Thanks.