RAJA generic device policies

[artv3]

Since there is a discrepancy between RAJA/HIP and SYCL ordering it may be convenient to define raja device policy alias to simplify transitions between backends. It could be implemented as alias accordingly to the backend enabled. This would reduce IF/DEFS in user code.

RAJA Execution Policies	CUDA/HIP Execution Policies	SYCL Execution Policies
device_exec<BLOCK_SIZE>	cuda/hip_exec<BLOCK_SIZE>	sycl_exec<WORK_GROUP_SIZE>
device_launch_t	cuda/hip_launch_t	sycl_launch_t
device_global_size_x_direct<nx_threads>	cuda/hip_global_size_x_direct<nx_threads>	sycl_global_2<WORK_GROUP_SIZE>
device_global_size_y_direct<ny_threads>	cuda/hip_global_size_y_direct<ny_threads>	sycl_global_1<WORK_GROUP_SIZE>
device_global_size_z_direct<nz_threads>	cuda/hip_global_size_z_direct<nz_threads>	sycl_global_0<WORK_GROUP_SIZE>
device_global_thread_x	cuda_global_thread_x	sycl_global_item_2
device_global_thread_y	cuda_global_thread_y	sycl_global_item_1
device_global_thread_z	cuda_global_thread_z	sycl_global_item_0
device_thread_x_direct	cuda/hip_thread_x_direct	sycl_local_2_direct
device_thread_y_direct	cuda/hip_thread_y_direct	sycl_local_1_direct
device_thread_z_direct	cuda/hip_thread_z_direct	sycl_local_0_direct
device_thread_x_loop	cuda/hip_thread_x_loop	sycl_local_2_loop
device_thread_y_loop	cuda/hip_thread_y_loop	sycl_local_1_loop
device_thread_z_loop	cuda/hip_thread_z_loop	sycl_local_0_loop
device_block_x_direct	cuda/hip_block_x_direct	sycl_group_2_direct
device_block_y_direct	cuda/hip_block_y_direct	sycl_group_1_direct
device_block_z_direct	cuda/hip_block_z_direct	sycl_group_0_direct
device_block_x_loop	cuda/hip_block_x_loop	sycl_group_2_loop
device_block_y_loop	cuda/hip_block_y_loop	sycl_group_1_loop
device_block_z_loop	cuda/hip_block_z_loop	sycl_group_0_loop

Teams and threads are ordered differently between CUDA/Sycl. One option is to consider a policy or alias to enable more explicit naming convention. CUDA-like, or SYCL-like or layout_left{right}

[artv3]

@rhornung67 @MrBurmark , this is what I had in mind.

[rhornung67]

I think the indices for sycl_global_* need to be reversed; i.e., device_global_size_x_direct should correspond to sycl_global_2, not zero, etc....

[artv3]

I think the indices for sycl_global_* need to be reversed; i.e., device_global_size_x_direct should correspond to sycl_global_2, not zero, etc....

Ah thats right! Thanks --fixed!