On Fri, Oct 26, 2012 at 4:51 AM, Jakub Jelinek <ja...@redhat.com> wrote: > On Fri, Oct 26, 2012 at 04:33:07AM -0700, H.J. Lu wrote: >> You are right. reflect test took a long time to compile. >> It has nothing to with LRA. > > About the nothing to do with LRA I'm not sure, because on Tuesday my > make -j48 -k check (slowish 16way box, yes,rtl checking) took just 30 > minutes and on Thursday already 1h20m. It seems to be mostly > var-tracking clobber_overlapping_mems, but before LRA it clearly generated > different code that didn't have as many MEM accesses or MEM based tracked > values as it does now, otherwise it wouldn't take that long. > > Jakub
I opened: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55092 It is a serious LRA regression. -- H.J.
