POCL 1.5-RC1 Released As The Portable OpenCL Implementation For CPUs + Other Targets

POCL 1.5 is on the way for release in April as the first feature update to this Portable OpenCL implementation since the previous release last September.

POCL for those that don't know about it is a portable OpenCL implementation that can be run on CPUs of various architectures. Beyond that, this OpenCL 1.2~2.0 implementation has also gained support for running OpenCL on NVIDIA GPUs over CUDA, on AMD GPUs via HSA, and other accelerator targets thanks to building off LLVM's Clang.

POCL 1.5-RC1 brings support for the newly-released LLVM / Clang 10.0 compiler, refactoring of the convert_T() OpenCL functions, and other tracing/profiling improvements.

The convert_T work for POCL 1.5 better jives with LLVM's auto-vectorization criteria and can lead to better SIMD ISA use on CPUs like Arm where up to a ~5.5x improvement can be seen in tight loops.

Those wanting to try out POCL 1.5-RC1 on any of the supported targets can grab the latest code via GitHub. Those wishing to learn more about this open-source project can do so at PortableCL.org.