[PATCH] D19894: AMDGPU: Fix not counting shader input registers
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Thu May 5 15:17:27 PDT 2016
arsenm added a comment.
In http://reviews.llvm.org/D19894#422566, @nhaehnle wrote:
> I'm not sure I understand the design of addArgUserReg. Is the CurReg parameter supposed to be correct or not? If yes, why is there a separate return value? If not, why is it there in the first place? It also seems like the alignment of SGPRs could just be done by a numerical rounding up instead of a loop.
It's supposed to be ensuring the number of used registers is consistent with the reported register by the generated calling convention (there was supposed to be an assert in here). I was trying to avoid assumptions based on the value of the register enums. It's kind of gross, but I'm not sure of a better way to express the connection between the two separate ways the calling convention registers are tracked
http://reviews.llvm.org/D19894
More information about the llvm-commits
mailing list