Search papers, labs, and topics across Lattice.
RaFI is a CUDA and MPI-based framework enabling simplified development of data-parallel GPU applications where work items (rays) migrate between GPUs. It provides a CUDA kernel interface for forwarding work items, abstracting away CUDA and MPI management. The framework's potential is demonstrated through several example applications, suggesting its utility in complex simulation scenarios.
Building multi-GPU data-parallel applications just got easier: RaFI handles the CUDA/MPI plumbing for you, letting you focus on your kernels.
We present RaFI, a CUDA and MPI based software framework that simplifies the task of building GPU-enabled data-parallel software where rays or similar work items need to migrate between different GPUs. RaFI provides a simple interface for CUDA kernels to forward such work items to other GPUs, while under the hood managing all the CUDA and MPI related work required to make this happen. We describe RaFI's motivation and implementation, and show its potential in several example applications.
