------- Comment #4 from pluto at agmk dot net 2006-02-16 15:32 ------- (In reply to comment #3) > I don't know if this is really valid code as long double on x86_64 uses 387 so > turning off 387 explicitly says no 387 instructions. In fact I think the > regression was caused by actually -mno-387 working correctly in turning off > the > 387 instructions from being emitting. >
testcase from #c2 works with -m32 -mno-80387 -mno-fp-ret-in-387 -msoft-float but fails with -m64. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26223