For C compilers like gcc and clang it's comparing against -O0. Anyway, there are two issues with that long-term trend in compiler design from my perspective. Th

For C compilers like gcc and clang it's comparing against -O0. Anyway, there are two issues with that long-term trend in compiler design from my perspective. The first is that the optimizing path is too slow and they don't offer an intermediate point on the Pareto frontier that reflects my trade-off preferences because too much of the slowness is architectural in origin. But the second and much worse problem is that the non-optimizing path is barely any faster. You can't make a fast non-optimizing or somewhat-optimizing compiler by taking the architecture of a heavyweight optimizing compiler and setting some internal flags and skipping a few passes. You need the fast compilation path to completely dictate the architecture from end to end, with the constraint of still supporting optimizing backends, and you need a development culture that treats the fast path's performance as inviolate and sacrosanct.