Add the nvidia-cuda-toolkit Debian package path to search path

[sylvestre.ledru]

Summary

Reported here:
http://bugs.debian.org/882505

Patch by Andreas Beckmann

[Hahnfeld]

The change looks good in general and is small enough and last in the candidates list to not break working setups.

What concerns me is that we have to do the same for every distribution (I know that Arch Linux for example has CUDA installed in /opt/cuda, see https://www.archlinux.org/packages/community/x86_64/cuda/). In addition many HPC systems use the modules system to install multiple versions in parallel. As such we should make the installation detection smarter:

1. We could have an environment variable that systems can set to the appropriate directory.
2. We could try to automatically deduce the install path: If we find ptxas in the PATH and it is in a directory bin, its parent is a good candidate to start searching.

I think 2. would work for Arch Linux which adds /opt/cuda/bin to PATH IIRC. I'm not sure about Debian though, @sylvestre.ledru? If that doesn't work, we have to add the candidate statically anyway...

[sylvestre.ledru]

Debian packages don't update the PATH and we are aiming at providing packages working out of the box.

In general, yeah, we have to do that for every distros... :/

[Hahnfeld]

In that case I don't see a way to be "clever", so we have to add the static candidate. I'll let @tra or @jlebar have a final look.

[tra]

I'm reluctant to add a distribution-specific search path unconditionally.

We do have Distro.IsDebian() https://github.com/llvm-mirror/clang/blob/9f9177d3ef72580ca29e8844327f63d7aa1908af/include/clang/Driver/Distro.h#L112

Adding distribution-specific path for particular distribution would be fine with me.

[tra]

lib/Driver/ToolChains/Cuda.cpp

81

No need for a named temporary. if (Distro(D.getVFS).IsDebian()) should do.

Also, please add a comment describing why Debian needs a special case here and include reference to the bug report which has the details.

[Hahnfeld]

On a related note please add context to your patch as described here: https://llvm.org/docs/Phabricator.html#requesting-a-review-via-the-web-interface

[jlebar]

I defer to tra on this.

lib/Driver/ToolChains/Cuda.cpp

if (Args.hasArg(clang::driver::options::OPT_cuda_path_EQ)) {
     for (const char *Ver : Versions)
       CudaPathCandidates.push_back(
           D.SysRoot + "/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v" +
           Ver);
   } else {
     CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda");
     for (const char *Ver : Versions)
       CudaPathCandidates.push_back(D.SysRoot + "/usr/local/cuda-" + Ver);
+
+    if (Distro(D.getVFS).IsDebian())
+      // Special case for Debian to have nvidia-cuda-toolkit work
+      // out of the box. More info on http://bugs.debian.org/882505
+      CudaPathCandidates.push_back(D.SysRoot + "/usr/lib/cuda");
   }
 
   for (const auto &CudaPath : CudaPathCandidates) {
     if (CudaPath.empty() || !D.getVFS().exists(CudaPath))
       continue;
 
     InstallPath = CudaPath;
     BinPath = CudaPath + "/bin";