I want to start using SYCL, but so far I found that one needs to install ComputeCpp and it only supports Ubuntu, CentOS, and Windows. What are the alternatives for using SYCL on MacOS Catalina?
Multiple SYCL implementation support MacOS today, although they only support CPU devices.
GPU device support is not available because Apple does not support the back-end dependencies available on Linux or Windows.
triSYCL supports MacOS but only CPU execution. I use it frequently. You may need to switch C++ compilers since the native toolchain might not work. https://github.com/triSYCL/triSYCL
hipSYCL also supports CPU execution and should work on MacOS. Pull request #271 should have details.
We did some work to get Intel DPC++ to build on MacOS (see Issue #258) but it’s not there yet. Issue #982 is tracking MacOS support.
I routinely use all the SYCL compilers for Linux on my Mac laptop inside of virtual machines but that is still CPU-only execution.
The most promising route to GPU support of SYCL on MacOS requires translating to Metal Shader Language (MSL), because Apple OpenCL doesn’t support SPIR-V. This can be done with SPIRV-Cross but nobody has implemented it in any SYCL compiler as far as I know.
Note that translating kernels from SPIR-V to MSL is only part of the answer. The other requirement is runtime support for kernel loading, etc. Some of these issues are apparent in https://github.com/bobpepin/opencl-osx, although this is merely a problem statement, not a solution.
If Apple were to support CUDA (unlikely, based on https://gizmodo.com/apple-and-nvidia-are-over-1840015246), then the PTX back-ends of hipSYCL or DPC++ might work, but I am not aware of any attempts along these lines with older Macs that support(ed) CUDA.