bader wrote:

> > If I get right, I can't reproduce these issues because I don't have NVIDIA 
> > SDK installed on my system and all these tests are just skipped as 
> > unsupported. I'll try to figure out how to configure my environment to 
> > reproduce these failures.
> 
> It looks like [Joseph addressed your 
> question](https://github.com/llvm/llvm-project/pull/202457#pullrequestreview-4454098782).
>  If I can provide any more info about what I'm seeing, let me know.

Unfortunately, it doesn't help. 

I tried a docker container with installed CUDA SDK with following commands:

cmake -G Ninja -S llvm -B offload-build -C offload/cmake/caches/Offload.cmake 
-DLLVM_INCLUDE_TESTS=ON -DOFFLOAD_INCLUDE_TESTS=ON 
-DLIBOMPTARGET_FORCE_NVIDIA_TESTS=ON
ninja -C offload-build check-openmp

Still see no failures. Considering that `check-libomptarget` target doesn't 
exist, I suspect there still some differences in envonment.

I think the best course of actions is to avoid sharing the functionality 
between `clagn-sycl-linker` and `clang-nvlink-wrapper`.

https://github.com/llvm/llvm-project/pull/201253
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to