clang + cuda version issue

Project Infrastructure LLVM Dev List Archives
1

I’m trying to get up and running with llvm/clang + cuda. llvm 11.0.1 successfully installed.

When I try to compile the simple example shown in llvm examples I get

clang-11: warning: Unknown CUDA version. cuda.h: CUDA_VERSION=11020. Assuming the latest supported version 10.1 [-Wunknown-cuda-version]
axpy.cu:23:7: error: use of undeclared identifier cudaConfigureCall
axpy<<<1, kDataLen>>>(a, device_x, device_y);

I have cuda 11.2 installed however clang apparently needs cuda 10.1 and I’m running fedora 33 which is too new for cuda 10.1. cuda 10.1 is supported on fedora 29. I can’t downgrade to fedora 29.

Is there a combo of llvm clang + cuda that works with fedora 33?

Is there any plan for llvm to use the latest cuda? on what time frame?

Thanks
Burlen

2

Hi Burlen,

it might be better to repost the question on cfe-dev@lists.llvm.org, which is the clang frontend
developer list.

~ Johannes