Home Workaround for nvcc pthread issue

Workaround for nvcc pthread issue

I recently upgraded various pieces of software on my work PC. Afterwards, I found that our C++/CUDA projects wouldn’t build. The following error was reported:

nvcc fatal : Unknown option 'pthread'

I was attempting to build it in CLion 2018.2.5 using the bundled CMake 3.12.2. My OS is Ubuntu 18.04 and I’m using CUDA 9.1.

So far, I haven’t been able to find a proper solution. However, I have found a simple workaround. If you know of a better solution then please let me know in the comments.

The problem

This issue seems to occur when using the FindThreads module in the CMake file. It adds a “-pthread” configuration flag which the CUDA compiler (nvcc) doesn’t understand.

I haven’t had time to dig into the underlying cause so I don’t know why this has started occurring relatively recently. However, it is a known problem and there have been some attempts to resolve it.

The workaround

For now, downgrading CMake seems to avoid the problem. I downgraded to CMake 3.10.2, although more recent versions may also work.

If you’re using Ubuntu 18.04, then you can install CMake using apt. This currently installs version 3.10.2:

sudo apt install cmake

Other distributions/repositories may have other versions available. If it doubt, consult the CMake website.

If you build your projects from within CLion then you will need to instruct it to use the correct CMake instance. You can do this by going to “File → Settings → Build, Execution, Deployment → Toolchains”. For CMake, select “Custom CMake executable”, and then give it the path to your manually installed version of CMake. On Linux, this is likely to be /usr/bin/cmake. You may want to do this in a new toolchain so that you don’t affect other projects.

See also

This post is licensed under CC BY 4.0 by the author.