https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61515
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> --- 4.8 and 4.9 at -O0 take ~32s and ~2.4GB of ram (and IRA taking 28% of the time) 4.8 at -O1 takes ~15min and ~5.5GB of ram (RTL loop invariant motion taking 80% of the time - ugh) 4.9 at -O1 takes ~5min and ~2GB of ram (30% compile time in SSA incremental, 27% in DF) Now -O[s2] will enable passes that are not really "tamed" and may show quadratic behavior (PRE and VRP for example, off the top of my head). You're supposed to drop to -O1 whenever you hit this kind of bug ... Building release checking trunk now.