Value Range Propagation patch (was: Re: Optimize C / R)

Toon Moene toon@moene.indiv.nluug.nl
Thu Jul 18 08:12:00 GMT 2002


Steven Bosscher wrote:

> Op wo 17-07-2002, om 13:01 schreef Toon Moene:

> > Wow, thanks for digging into this - I had already given up, thinking "this
> > might be something for John Wehle's Value Range Propagation pass" (about
> > which I haven't heard since a year, BTW.
> >
> > It would be nice if this could be solved without that sledgehammer.
> 
> If you want to swing the sledgehammer anyway, I have a VRP patch that
> should apply without problems to the mainline.

> Why wasn't it included at the time? Too slow?

My impression was that John felt that it was "not ready yet".  It had a
tendency to spend too much time on too little code.  I remember seeing it
driving up compile time for one routine of our NWP code to 5 minutes, while
on execution the same routine would be wizzed by in milliseconds,
regardless of optimization level.

So it needs some heuristics, I'd say.

-- 
Toon Moene - mailto:toon@moene.indiv.nluug.nl - phoneto: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
Maintainer, GNU Fortran 77: http://gcc.gnu.org/onlinedocs/g77_news.html
Join GNU Fortran 95: http://g95.sourceforge.net/ (under construction)



More information about the Gcc-patches mailing list