[SYCL] Fix race in kernel_impl::getName

[uditagarwal97]

No description provided.

[sarnex]

I think this is throwing the below then hanging in check-sycl or check-sycl-unittests

/__w/llvm/llvm/build/bin/clang --driver-mode=g++  -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=gfx90a -fsycl /__w/llvm/llvm/src/sycl/test/regression/queue_submit.cpp -o /__w/llvm/llvm/build/tools/sycl/test/regression/Output/queue_submit.cpp.tmp.out
# executed command: /__w/llvm/llvm/build/bin/clang --driver-mode=g++ -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=gfx90a -fsycl /__w/llvm/llvm/src/sycl/test/regression/queue_submit.cpp -o /__w/llvm/llvm/build/tools/sycl/test/regression/Output/queue_submit.cpp.tmp.out
# .---command stderr------------
# | clang: warning: argument unused during compilation: '-Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=gfx90a' [-Wunused-command-line-argument]
# `-----------------------------
# RUN: at line 4
/__w/llvm/llvm/build/tools/sycl/test/regression/Output/queue_submit.cpp.tmp.out
# executed command: /__w/llvm/llvm/build/tools/sycl/test/regression/Output/queue_submit.cpp.tmp.out
# .---command stderr------------
# | terminate called after throwing an instance of 'sycl::_V1::exception'
# |   what():  The kernel bundle does not contain the kernel identified by kernelId.
# `-----------------------------
# error: command failed with exit status: -6
--

[Copilot]

Pull request overview

This PR fixes a race condition in kernel_impl::getName() that could occur when multiple threads concurrently access the same kernel object. The fix ensures thread-safe lazy initialization of the kernel name using std::call_once.

Key changes:

• Added thread-safe lazy initialization to kernel_impl::getName() using std::call_once
• Added regression test to verify concurrent kernel submission to the same queue works correctly

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
sycl/source/detail/kernel_impl.cpp	Implements thread-safe lazy initialization of kernel name using std::call_once and std::once_flag
sycl/test-e2e/Regression/queue_submit.cpp	Adds regression test that submits the same kernel from multiple threads to verify the race condition fix

[sergey-semenov]

LGTM overall

[uditagarwal97]

Failure in SpecConstants/2020/nested-non-packed-struct.cpp on CUDA is unrelated: #17650