ChipStar 1.1 Released For Compiling & Running HIP/CUDA On SPIR-V

ChipStar 1.1 was released this past week as one of the open-source projects to help in porting HIP and CUDA applications to support the industry-standard SPIR-V. ChipStar acts to get HIP/CUDA codes working on SPIR-V with OpenCL or Intel's oneAPI Level Zero.

ChipStar was born out of combining the HIPCL and HIPLZ projects and has already proven capable of moving large HPC applications targeting HIP/CUDA to working with SPIR-V intermediate representation.

The wonderful versatility and support around Khronos Group's SPIR-V IR.

With last week's chipStar 1.1 release there is now support for making use of LLVM/Clang 17, support for the Intel Unified Shared Memory Extension with OpenCL, optimized atomic operations with OpenCL 3.0, improved asynchronous execution, improved portability, and various other new features and fixes.

The chipStar 1.1 release can be nearly twice as fast as chipStar 1.0 for some codebases while the average performance improvement over the prior release is about 30%.

Downloads and more details for chipStar 1.1 for bringing HIP/CUDA codes to a SPIR-V world can be found via GitHub.