ValeevGroup / tiledarray

A massively-parallel, block-sparse tensor framework written in C++
GNU General Public License v3.0
247 stars 51 forks source link

Fix CUDA compilation with Cray Wrappers #416

Closed wavefunction91 closed 10 months ago

wavefunction91 commented 11 months ago

Cray wrappers don't populate MPI_<LANG>_INCLUDE_DIRECTORIES because they're implicit, and to avoid incompatibilities, it makes sense to delegate to the compiler. The problem is that this information is not propagated to the CUDA compiler. Found a simpler fix than the one suggested in https://github.com/ValeevGroup/tiledarray/issues/413: simply append the implicit CMAKE_CXX_IMPLICIT_INCLUDE_DIRECTORIES as source-level compilation features for the CUDA files.

Closes https://github.com/ValeevGroup/tiledarray/issues/413