This patch updates the hipSYCL compilation in portBLAS including :
Necessary naming updates hipSYCL -> AdaptiveCpp)
Cmake changes to account for AdaptiveCpp compilation approaches
Relevant compatibility considerations of SYCL 2020 features and other extensions still not supported by AdaptiveCpp.
Updated documentation/readme.
Compilation flows / Device targets compiled successfully at this stage :
intel CPU : omp default flow (host), and generic SPIRV/OpenCL flow.
intel GPU : generic OpenCL/SPIRV flow, and generic SPIRV/LevelZero flow.
Nvidia GPU : generic PTX/cuda flow.
AMD GPU : generic HIP/amdgcn flow. (Tested with rocm 5.6.1)
Tests and benchmarks are conditionally disabled for some operators depending on the target device and AdaptiveCpp flow. Remaining tests (in each case) are passing for all tested devices above.
This patch updates the hipSYCL compilation in portBLAS including :
Compilation flows / Device targets compiled successfully at this stage :
Tests and benchmarks are conditionally disabled for some operators depending on the target device and AdaptiveCpp flow. Remaining tests (in each case) are passing for all tested devices above.