[CUDA] Assume the latest known CUDA version if we've found an unknown one.

[tra]

Summary

This makes clang somewhat forward-compatible with new CUDA releases without
having to patch it for every minor release without adding any new functionality.

If an unknown version is found, clang issues a warning (can be disabled with -Wno-cuda-unknown-version)
and assumes that it has detected the latest known version. CUDA releases are usually supersets
of older ones feature-wise, so it should be sufficient to keep released clang versions
working with minor CUDA updates without having to upgrade clang, too.

[yaxunl]

LGTM. Thanks.

[tra]

Use std::string instead of Twine which can't be stored.

[yaxunl]

LGTM

[tra]

@hans : that's another candidate for 10.x cherry-pick, if you're OK with it.

[hans]

In D73231#1845096, @tra wrote:

@hans : that's another candidate for 10.x cherry-pick, if you're OK with it.

Sounds good to me. Cherry-picked in 5777899f146aaa53f598e6978702a7a2725200d7

[yaxunl]

clang/lib/Driver/ToolChains/Cuda.cpp

51

We got some issue with this warning.

Basically CudaInstallationDetector is a member of GNU and other host toolchain and this check is done in the ctor.

If a user has latest CUDA SDK installed, clang will always emit a warning. If the user has -Werror set, which is quite common, compilation will always fail.

Is it possible only do this check for CUDA compilation?

Thanks.

[tra]

clang/lib/Driver/ToolChains/Cuda.cpp

51

Agreed. Clang should not warn about this unless it's a CUDA compilation. I'll get it fixed.

[tra]

This patch should fix it: https://reviews.llvm.org/D76030