Create utility functions for launching kernels similar to CUDA and HIP backends.

[kris-rowe]

Details

The CUDA and HIP backends use several utility functions for launching kernels, given the grid and block sizes. For consistency, it would be helpful to refactor the SYCL backends to match this usage pattern.

Utility functions for launching kernels could be defined using templates, e.g.

template<typename KernelName, int dim, typename... ArgTypes>
void launchKernel(sycl::queue& q, sycl::range<dim> kernel_range, ArgTypes... args) {
  q.submit([&](sycl::handler& cgh){
    KernelName launched_kernel(args...);
    cgh.parallel_for(kernel_range,launched_kernel);
  });
}

provide the kernels are defined as function objects.

Todo

• Turn existing lambdas into function objects and move to header files
• Define templates for launching kernels