This will allow targets more flexibility to replace the register allocator core passes. In a future commit, AMDGPU will run the core register assignment passes twice, and will also want to disallow using the standard -regalloc option. llvm-svn: 356506
46 KiB
46 KiB