CUDA: Add support for Clang compiler

When crosscompiling we pass the sysroot.

We need to try various architecture flags. Clang doesn't automatically
select one that works.  First try the ones that are more likely to work
for modern installations:

* <=sm_50 is deprecated since CUDA 10.2, try sm_52 first for
          future compatibility.
* <=sm_20 is removed since CUDA 9.0, try sm_30.

Otherwise fallback to Clang's current default. Currently that's `sm_20`,
the lowest it supports.

Separable compilation isn't supported yet.

Fixes: #16586

1 files changed, 14 insertions, 0 deletions

diff --git a/Source/cmGeneratorTarget.cxx b/Source/cmGeneratorTarget.cxx
index a44126c..75790c8 100644
--- a/Source/cmGeneratorTarget.cxx
+++ b/Source/cmGeneratorTarget.cxx
@@ -3179,6 +3179,20 @@ void cmGeneratorTarget::AddCUDAArchitectureFlags(std::string& flags) const
 
       flags += "]";
     }
+  } else if (compiler == "Clang") {
+    for (CudaArchitecture& architecture : architectures) {
+      flags += " --cuda-gpu-arch=sm_" + architecture.name;
+
+      if (!architecture.real) {
+        Makefile->IssueMessage(
+          MessageType::WARNING,
+          "Clang doesn't support disabling CUDA real code generation.");
+      }
+
+      if (!architecture.virtual_) {
+        flags += " --no-cuda-include-ptx=sm_" + architecture.name;
+      }
+    }
   }
 }