Extending SYCL support to multi-vendor GPU and Accelerator

This report mainly introduces how they extend the cross architecture programming language DPC++ to AMD GPU platform. The report first briefly analyzes the basic architecture and constituent modules of AMD GPU computing platforms ROCM and DPC++. Based on the above analysis, the report describes how they integrate AMD device compilation tool chain and DPC + + SYCL compilation tool chain, and realize the mapping from SYCL runtime to AMD device runtime, so as to complete the support of DPC++ for AMD GPU. At the end of the report, he will provide some sample programs to demonstrate their work.