fail on CUDA and there are comments saying this is due to name mangling for free function kernels not working with PTX.
The error logs as seen in the CUDA CI run of this PR(https://github.com/intel/llvm/pull/16081) say that an exception is thrown with a kernel not found error so this may or may not be a name mangling issue. More investigation is needed.
To reproduce
Simply run llvm-lit with the tests above on a CUDA platform. As access to a CUDA platform may be hard to get, usually I will simply submit a draft PR where I remove the UNSUPPORTED directive from one of these tests and then look at the CI results for the CUDA runner.
Environment
OS: Any
Target device and vendor: CUDA
DPC++ version: dc181bbb888044dab158a7cbe9644ea53a42d926 (latest commit as of writing this)
Dependencies version: [e.g. the output of sycl-ls --verbose]
@lbushi25, please rename this issue in order to prevent further confusion. As discussed elsewhere, the problem doesn't appear to be related to name mangling.
Describe the bug
The E2E tests
fail on CUDA and there are comments saying this is due to name mangling for free function kernels not working with PTX. The error logs as seen in the CUDA CI run of this PR(https://github.com/intel/llvm/pull/16081) say that an exception is thrown with a kernel not found error so this may or may not be a name mangling issue. More investigation is needed.
To reproduce
Simply run
llvm-lit
with the tests above on a CUDA platform. As access to a CUDA platform may be hard to get, usually I will simply submit a draft PR where I remove the UNSUPPORTED directive from one of these tests and then look at the CI results for the CUDA runner.Environment
sycl-ls --verbose
]Additional context
No response