> I would hazard a guess that the authors simply didn't consider the > multi-hard reg case. Essentially if the original set reached an > extension, then obviously the original set got there unharmed and the > extended destination should reach as well -- except that doesn't apply > to multi-word hard regs.
This pass started as a Zero-Extension Elimination pass for x86-64 and was only considering implicit SI->DI extensions initially. The algorithm was tailored to this specific pattern and I agree that the pass should be reimplemented. -- Eric Botcazou