[PATCH] AMDGPU: Fix VGPR spilling

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Tue Oct 13 15:34:07 PDT 2015


On 10/13/2015 01:39 PM, Tom Stellard via llvm-commits wrote:
> This will also reserve the wrong register for non-compute shaders.  We should
> be choosing the register to reserve by calling  CCInfo.getFirstUnallocated()
> at the end of SITargetLowering::LowerFormalArguments().
I tested it as working with piglit on graphics shaders, it just required 
the incrementing of NumUserSGPRs which isn't great


More information about the llvm-commits mailing list