CUDA: Do not device link if target has no CUDA usage

When CUDA is enabled, and a pure non-CUDA target has
CMAKE_CUDA_SEPARABLE_COMPILATION enabled, don't actually perform
the device linking step, as it will fail. A target that has
CMAKE_CUDA_SEPARABLE_COMPILATION enabled must also have CUDA
usage (either itself, or something it links to).

Fixes: #20182

1 files changed, 5 insertions, 0 deletions

diff --git a/Tests/Cuda/SeparableCompCXXOnly/main.cpp b/Tests/Cuda/SeparableCompCXXOnly/main.cpp
new file mode 100644
index 0000000..8135246
--- /dev/null
+++ b/Tests/Cuda/SeparableCompCXXOnly/main.cpp
@@ -0,0 +1,5 @@
+
+int main(int, char const* [])
+{
+  return 0;
+}