clover/nir precusor patches for amd nir compute support
I've gotten the basic nir compute backend on AMD into shape again, however it needs some changes across nir/clover lowering.
This contains those three patches.
- Due to llvm/amd backend global_group_size shouldn't be lowered in the frontend and there is a backend intrinsic
- Due to llvm/amd backend API all the shader args needs to be passed in as llvm IR function args, this is a lot easier if you have access to the derefs instead of lowered offsets so add an option to avoid lowering in frontend.
- work_dim should be lowered in the frontend for this backend as clover can add it to the args list.