The PoCL 1.3 project release (Portable Computing Language OpenCL) is now available, advancing the implementation of the OpenCL standard, independent of graphics accelerator manufacturers, and allowing various backends to execute OpenCL kernels on different types of graphics and central processors. The project's code is distributed under the MIT license. Support is provided for platforms including X86_64, MIPS32, ARM v7, AMD HSA APU, and various specialized TTA (Transport Triggered Architecture) processors with VLIW architecture.
The implementation of the OpenCL kernel compiler is based on LLVM, and Clang is used as the frontend for OpenCL C. To ensure proper portability and performance, the OpenCL kernel compiler can generate combined functions that can utilize various hardware resources for parallel execution of code, such as VLIW, superscalar execution, SIMD, SIMT, multicore, and multithreading. ICD drivers are supported.
(Installable Client Driver). Backends are available for operation through CPU, ASIP (TCE/TTA), HSA architecture-based GPUs, and NVIDIA GPUs (CUDA).
The new version adds support for LLVM/Clang 8.0. ICD (Installable Client Driver) support has been implemented for macOS. The ability to build pocl without backend drivers for CPU has been realized. An initial support for compiling ISA over HSA runtime has been introduced for HSA (Heterogeneous System Architecture). The codebase has been cleaned up, including the removal of the Vecmathlib library and discontinuation of support for earlier LLVM versions (below 6.0).
Source: opennet.ru
