The HipSYCL Quest for Universal SYCL* Binaries
In current SYCL* implementations (such as Data Parallel C++ [DPC++] or hipSYCL) constructing universal binaries can be an arduous task. The multipass compilation model requires parsing the code separately for each targeted back end, and parsing once per target device in the case of AMD* hardware. This can inflate compile times to impractical levels.
This talk discusses the new hipSYCL generic single-pass compiler, which can generate a universal binary while parsing the source code only a single time. With runtime performance levels comparable to current compilers, the new compiler can substantially outperform other SYCL compilers' compile time, and provides instant binary portability across NVIDIA*, AMD*, and Intel® GPUs, including Intel® Data Center GPU Max Series. The talk also presents the first hipSYCL performance numbers on Intel Data Center GPU Max Series.
Speaker
Aksel Alpay is a researcher and software engineer from Heidelberg University, where he works on high performance computing topics. In particular, he is the creator and lead developer of the hipSYCL SYCL implementation, and also engages within the SYCL working group at the Khronos Group* to advance the language.
Product and Performance Information
Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.