This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] for PR 18687


Hello,

> >>> no, this should have been purely compile time improvement.  I should
> >>> have time to check what's going on during this week.
> >>
> >> This is also a consequence of ivopts having *no* tests in the testsuite.
> >
> > we have 5 tests in the testsuite that test all major functionality
> > of ivopts.
> 
> 
> Yes, but I guess you fixed around 20 bugs (if not more) in Bugzilla related
> to performance problems with ivopts for which a testcase was never
> committed.

because in most cases it would be quite difficult, if not impossible,
and practically useless.  Many of the problems are only reproducible in testcases
in that there are many ivs and their uses, and are related to suboptimal
choice of new basic induction variables.  The tests would need to be
strictly platform specific, and even then they would break often due to changes in
unrelated parts of compiler much more often than detecting any real
problem.  Even if there is a change in the way ivs are chosen by the
ivopts heuristics, it does not necessarily mean that something is
broken; it's just a heuristics.

Zdenek


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]