Thank. Yeah, I've seen BLT through it's use in RAJA. You're right, it'd
work, but I'm reluctant to stray away from "vanilla" cmake too much, since
other people will have to deal with this code, and cmake itself is already
quite a step up from the regular Makefiles that they're used to.
--Kai
On
On Tue, Jul 2, 2019 at 2:48 PM Robert Maynard
wrote:
> get_target_property(source_files SOURCES)
> set_source_files_properties(${source_files} PROPERTIES LANGUAGE CUDA)
>
Thanks, that'll work for the time being, I can put this into a
function/macro and it'll look alright.
--Kai
--
Powered by
Compiling C++ code with CUDA compiler practically means only to implicitly
include a bunch of CUDA-specific headers like cuda_runtime_api.h. Other
than that, it's the same as compiling with the host C++ compiler. Thus, you
can get the desired behavior by explicitly including those headers in case
o
In general I go with the source property approach, since you can pass
it a collection of files to be marked as CUDA.
If you are aware of when all sources have been added to a target you
can easily mark them all as cuda with:
get_target_property(source_files SOURCES)
set_source_files_properties(${
For background, a bunch of projects help writing portable C++ code that can
be compiled into CUDA device code as one option, e.g. hemi, kokkos, RAJA (
https://devblogs.nvidia.com/simple-portable-parallel-c-hemi-2/). As a
consequence, if available those source files need to be compiled with the
CUDA
