POCL Is Working On OpenCL-Over-CUDA With New Backend
POCL, the Portable Computing Language project, aiming to provide a portable OpenCL implementation for usage against different CPUs and back-ends, has been working on CUDA support.
There is now an experimental NVIDIA CPU back-end making use of the LLVM NVPTX back-end and CUDA driver API. So far this experimental back-end is good enough for running some OpenCL programs via POCL on CUDA.
This is a very interesting project especially with NVIDIA being slow to implement new versions of OpenCL. Recently we have seen their drivers pick up a bit more OpenCL 2.0 support, but this CUDA back-end could allow for running new OpenCL features not otherwise supported by their native OpenCL driver. This back-end can already handle SPIR/SYCL code too.
Those wanting to learn more about this new CUDA back-end for POCL can do so by visiting PortableCL.org.
There is now an experimental NVIDIA CPU back-end making use of the LLVM NVPTX back-end and CUDA driver API. So far this experimental back-end is good enough for running some OpenCL programs via POCL on CUDA.
This is a very interesting project especially with NVIDIA being slow to implement new versions of OpenCL. Recently we have seen their drivers pick up a bit more OpenCL 2.0 support, but this CUDA back-end could allow for running new OpenCL features not otherwise supported by their native OpenCL driver. This back-end can already handle SPIR/SYCL code too.
Those wanting to learn more about this new CUDA back-end for POCL can do so by visiting PortableCL.org.
7 Comments