Re: Proposed testcase for PR 7344

[Jan Hubicka <jh at suse dot cz>]

> On Thu, Oct 17, 2002 at 05:04:48PM +0200, Michael Matz wrote:
> > Hi,
> > 
> > On Thu, 17 Oct 2002, Nathanael Nerode wrote:
> > 
> > > This is intended to go in the gcc.dg directory.  Comments welcome, since
> > > I'm pretty new to testsuite work.  I titled it switch-2.c since I don't
> > > like the by-date naming convention, and it's a pretty generic test.
> > 
> > How many minutes are we talking about?  Would it also be enough to reduce
> > the number of cases to 5000 (or whatever), so that the effect is still
> > visible, but not _that_ annoying?  Already 20001226-1 takes a huge part of
> > time of the overall runtime of the testsuite, having more such tests is
> > not necessarily productive.
> When it *passes*, it takes 28 seconds.  I don't think that's a problem.
> 
> When it *fails*, it takes an unknown length of time.

I don't think this is resonable way to handle it.  28seconds on one
machine can be 5 minutes on others and we run testsuite for instance on
m68k that has no machines with enought computation power available
(if I want to build gcc on my m68k box, it takes about a week).

Also testsuite takes long time.  If you makes it 10 times as slow, it is
regression, but still you won't notice that in compilation time of whole
testsuite unless you are really curefull.

Perhaps we can add such testcases into kind of automated tracking that
will produce graphs like we do for SPEC?  That seems to be more
productive to deal with performance issues.
> 
> --Nathanael
> 
> > 
> > 
> > Ciao,
> > Michael.
> >