Thiemo Seufer 写道: > Maybe. The 'mips3' maps to -march=r4000, it would assume more memory > latency and a slower integer divider then -march=r4600. > Just like Ralf has said, -march might not lead to significant performance difference, but -mtune= may. Two years ago, I found that adding a machine.def to descripe pipeline and resources for loongson2 and -mtune=loongson boost many programs by 2-5%。 > > Thiemo > > >
