[DeviceASAN] Fix ASAN with kernel assert

[AllanZyne]

LLVM: intel/llvm#16256

Even when the kernel is not instrumented, we still need to execute the same routine just like instrumented kernel.
That's because some of arguments have been intercepted by the sanitizer layer and are no longer compatible with the native level zero's value. For example, membuffer.

[zhaomaosu]

lgtm

[AllanZyne]

Hi @oneapi-src/unified-runtime-maintain, please review and add this to 2025.1.
Thanks!

[pbalcer]

just a few nits, but I do have a general comment - this looks like a pretty big change in logic, with the not instrumented kernels now executing more code that was previously short-circuited with if (not instrumented) return KernelLaunch(...). Which might be fine, but there's no explanation in the commit description.

Do you have SYCL tests that exercise this path with non-instrumented kernels?

[AllanZyne]

just a few nits, but I do have a general comment - this looks like a pretty big change in logic, with the not instrumented kernels now executing more code that was previously short-circuited with if (not instrumented) return KernelLaunch(...). Which might be fine, but there's no explanation in the commit description.

Ok, I will add explanation.

Do you have SYCL tests that exercise this path with non-instrumented kernels?

Thanks for reminder, I enhanced the kernel-filter test.

[AllanZyne]

Hi @pbalcer, the commit description was updated.