https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88711
--- Comment #8 from Jan Hubicka <hubicka at ucw dot cz> --- > "I committed a patch that causes a regression in the Fortran > testsuite. Clearly, the problem is with Fortran not my patch. > I don't care about Fortran or respect those that work on the > Fortran FE. Therefore, I just XFAIL the regressing test > case and call it Good(tm)." > > Thanks. > > How about reverting your patch? The patch fixes obvious bug. That bug made inliner heuristics lucky for this particular testcase (which was introduced by me because I also do care about Fortran perfomrance) but also it caused problems elsewhere (I noticed it while analyzing 10% slowdown of Firefox HTML parser). There is no way to make inline heuristics perfect for everyone (it would not be called heuristics then). Sadly I do not see way to improve this particular testcase without simply increasing inline limits (would increase code size and make others unhappy), adding some logic telling inliner that Fortran functions should be inlined more (which was discussed before but was not considered coolest direction ofthe attack) or teching inliner to recognize quite complex interaction between individual functions inlined that is hard to do. Bumping up --param inline-insns-auto limit helps this particular testcase. Honza
