hipSYCL Becomes Open SYCL For Targeting All Major CPUs & GPUs
Developer Aksel Alpay announced on Thursday that hipSYCL has become Open SYCL while retaining the same code-base and focus. Open SYCL is a modern implementation of Khronos' SYCL that seeks to support CPU and GPU targets from all major vendors and suited for a variety of use-cases. Open SYCL compiles kernels to a unified code representation that is then lowered at run-time to the target device such as for NVIDIA CUDA via PTX, AMD ROCm via AMDGCN, and Intel graphics via SPIR-V. Open SYCL supports CPU execution via any processor with OpenMP compiler support in its library-only compilation mode or when augmented by LLVM can run with any LLVM-supported processor or other LLVM back-ends.
With Intel backing SYCL for their oneAPI effort, efforts like Open SYCL are all the more important and of relevance for opening up ecosystem support across hardware vendors and implementations. There are also other similar projects in this space like triSYCL and ComputeCpp.
The project formerly known as hipSYCL is now hosted at OpenSYCL/OpenSYCL on GitHub.