llvm-project

07115651 - [AMDGPU] Fix flat D16 patterns handling of cpol. NFCI. (#150551)

[AMDGPU] Fix flat D16 patterns handling of cpol. NFCI. (#150551) GlobalSAddr and ScratchSVAddr produce 4 operands, not 3 as before. The last one is CPol which is curently ignored by D16 patterns. I was unable to catch it on any wrongdoing with these patterns, but in general it is dangerous to ignore it.