https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70432
Richard Biener <rguenth at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |NEW Last reconfirmed| |2016-03-29 Ever confirmed|0 |1 --- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> --- Confirmed. GIMPLE level "scheduling" could fix this (basically reduce # of live regs anywhere, esp. across calls). Looks like -fschedule-insns doesn't do the trick here (not with -fsched-pressure either) possibly because there's no heuristic to make a different schedule profitable (or it never considers to schedule sth before calls). Leaving as RTL scheduling issue.