Seems the relative paths were wrong basically all around such that only
compiling files in the top-level directory would work. I've modified
CudaOnly.SeparateCompilation to cover this.
Fixes#22482.
Architecture 30 was removed with CUDA 11, so most of the CUDA tests fail with
it.
Remove setting the architecture and bump the minimum version to 3.18, so
CMP0104 takes effect and we can rely on the default architecture, which is
guaranteed to be compilable.
Use of __ldg() in ProperLinkFlags was removed as it only affects performance
and is available only on sm_35 and above.
Testing the functionality of CUDA_ARCHITECTURES is already covered by
CudaOnly.Architecture and CudaOnly.CompileFlags.
The change in commit v3.9.0-rc4~3^2 (VS: Improve workaround for CUDA
-Xcompiler placement bug, 2017-06-21) accidentally appended to the
`AdditionalOptions` as if it were a `;`-separated list, but it is
actually a command-line string. Append with a space instead.
While at it, fix the same problem for the `AdditionalOptions` added to
`CudaLink` by commit v3.9.0-rc3~1^2 (CUDA: When linking device code
suppress CUDA 8.0+ deprecation warnings, 2017-06-09).
Fixes: #17008
In commit v3.9.0-rc1~431^2~6 (VS: Place CUDA host compiler options in
proper project file fields, 2017-03-07) we worked around a bug in the
CUDA VS integration by dropping `AdditionalCompilerOptions`. However,
this silently drops `-Xcompiler=` options given by the user that don't
map to one of CudaCompile's dedicated settings. Improve the workaround
to instead put the remaining `AdditionalCompilerOptions` into the
`AdditionalOptions` field behind `-Xcompiler=` ourselves.
