When a CUDA sdk doesn't have nvcc, defer to the existence of
a version.txt file. When we do this fall back we also reconstruct
the CUDA version via version.txt
Fixes#20643
Clang isn't very good at finding the installed CUDA toolkit.
The upstream recommendation is that we should pass the toolkit explicitly.
Additionally:
* Avoids Clang having to search for the toolkit on every invocation.
* Allows the user to use a toolkit from a non-standard location by simply
setting CUDAToolkit_ROOT. The same way as with FindCUDAToolkit.
Clang wants the directory containing the device library and version.txt as the
toolkit path.
We thus pass the newly introduced CUDAToolkit_LIBRARY_ROOT as the toolkit path.
We save CUDAToolkit_ROOT_DIR and CUDAToolkit_LIBRARY_ROOT on Clang to have them
available in try_compile() and avoid unnecessary re-searching or a possibly
different installation being found in FindCUDAToolkit.
This however means that the selected toolkit can't be changed after the initial
language enablement.
We now determine CUDA compiler ID before doing actual detection, as we don't
want to spend time finding the CUDA toolkit for NVIDIA.
Implements #20754.
We can avoid searching for this since CUDAToolkit_TARGET_DIR always contains
the include/ directory. But add a warning just in case.
Also apply this in CMakeDetermineCUDACompiler for Clang code.
On scattered installations version.txt and nvvm are located at this location.
This may be useful to users and will allow us in the future to parse
version.txt instead of invoking nvcc to figure out the CUDA toolkit version.
We also add it to CMakeDetermineCUDACompiler in preparation for future use by
Clang code.
A portion of FindCUDAToolkit was previously split in commit dc2eae1f
(FindCUDAToolkit: Factor out discovery code into a separate file, 2020-04-22)
out into Internal/CUDAToolkit to allow re-use of the code in
CMakeDetermineCUDACompiler for Clang support.
This has turned out to be a bad solution due to Clang requiring quite a bit of
special handling and special handling for NVCC leaking out from
Internal/CUDAToolkit into the Clang code using it.
Thus it seems better to re-unify this code and duplicate the parts of the code
necessary for Clang where it's required. This will help us get logic correct
for both NVCC and CUDA handling. We can still unify the common parts in the
future once the code has matured.
This ended up after the searching after commit dc2eae1 (FindCUDAToolkit: Factor
out discovery code into a separate file, 2020-04-22).
Move it back to where it was and should be.
This allows for re-use in other parts of the code, that require the CUDA
toolkit location, but can't or may not want to use the full
`FindCUDAToolkit`.
Update `_CUDAToolkit_find_and_add_import_lib` to create and add
dependencies to a target in one step that shared a single guard against
repeated definitions. Otherwise we were adding dependencies again on
every call.
Fixes: #20282
It is not a requirement to have shared|static consistent across your
CUDA libraries (e.g curand, nppc ) and your CUDA runtime library.
It is entirely allowable to use a static nppc and a shared runtime.
