POCL Is Maturing Well For Running OpenCL On The CPU

POCL, the Portable Computing Language, continues making headway for executing OpenCL kernels on CPUs. POCL now supports much of OpenCL 1.x and continues being refined for better support.

We've been covering POCL on Phoronix since its start nearly three years ago and with the POCL 0.9 release at the start of the year it implements much of OpenCL 1.2 while being MIT-licensed and running on CPUs through the use of LLVM. With my recent Intel Beignet OpenCL Linux testing I decided to give the latest POCL Git a go to see how it's working these days.

While POCL 0.9 is the latest stable release from January, POCL 0.10 is nearing completion and POCL 0.11 is already under development. POCL 0.10 includes support for LLVM/Clang 3.5, has an experimental CMake build system, improves thread safety, improvements were done to its OpenCL kernel compiler, various CL built-in functions were improved, and there was a number of OpenCL run-time / platform improvements. The latest change-log for POCL can be found via their Git code on GitHub ahead of the official 0.10 tagging.

In my testing of the Git code for POCL paired with LLVM/Clang 3.5 RCs, it's been running well for OpenCL on the CPU when testing the new AMD FX CPUs and the Intel Core i7 4790K.

In terms of when we could end up seeing POCL deployed out-of-the-box on Linux distributions for providing (fallback) OpenCL support on CPUs, Fedora has packaged POCL since last year while with Fedora 21 there should be out-of-the-box CL support that would include Intel's Beignet and Gallium3D Clover too.

Stay tuned for more OpenCL benchmarks on Phoronix in the near future. Learn more about the Portable Computing Language at PortableCL.org.