Panda is a c++ template library that provides a framework to build powerful data processing applications that rely on multiple accelerators and threads. Its main feature is that it allows you to specify, either via a configuration file or programtically, exaclty which compute resources you wish to apply to any particular compute task.
PANDA is available through git:
git clone git@gitlab.com:SKA-TDT/panda.git
Contact Chris Williams at Oxford University AstroPhysics for any info.