You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Unfortunately, it's not as simple as modifying the CMakeLists file.
There are several issues with using CUDA 12.5 in Chrono:
the latest version of Thrust (2.4.0) included with the CUDA 12.5 toolkit does not allow thrust/ headers in host-only TUs when THRUST_DEVICE_SYSTEM=CUDA and only allows thrust/ headers in host-only TUs when THRUST_DEVICE_SYSTEM={CPP, OMP, TBB}.
This will require a refactoring of the internal multicore collision detection algorithms and of the Chrono::Multicore module, both of which rely on Thrust with the OpenMP backend.
the API for cusparse and cublas has changed, with some functionality (e.g. incomplete decomposition preconditioners) completely removed.
This will require a refactoring of the Chrono::FSI module (and possibly the Chrono::GPU module).
All of the above will have to happen at some point, but I do not have the bandwidth to tackle any of it for quite a while.
I saw during configuration that chrono does not support cuda 12.5 yet. Is that a hard constraint, or should I modify the cmakefile and try it?
The text was updated successfully, but these errors were encountered: