Reorganize target plugins
Tasks
-
Review OpenMP target API and interface with the Clang compiler -
Review CUDA plugin -
Review AMDGPU plugin (WIP) -
Review Johannes cleanup patch (WIP) -
Fix locking issues - Some critical regions are unnecessarily large
-
std::mutex
is used instead of a faster spinlock on small critical operations
-
Fix or mitigate dynamic memory issues - Kernel launches define multiple
std::vectors
- Kernel launches define multiple
This re-organization focuses on the CUDA and AMDGPU plugins, but it should into account the VE and GENERIC plugins too.