I find a case that Machine-CSE could but does not perform the optimization
%24 = MOVi32imm 1
%23 = COPY %24
%38 = MOVi32imm 1
%37 = COPY %38
%7 = PHI %23, %bb.1, %37, %bb.3, …
The second MOVi32imm instruction could be eliminated by the CSE, but two COPYs in bb.1 and bb.3 prevent the problem. CSE thinks extending %24 to %37 increases the register pressure. Machine-CSE already supports TrivialCopyPropagation. It seems running this propogation before machine-cse can simply fix this problem. Is there any other advice?