Re: Let's use gcc-4.2, not 4.1 -- OpenMP

On Sunday 17 December 2006 15:44, Ulrich Spoerlein wrote:
> David O'Brien wrote:
> > On Fri, Dec 15, 2006 at 07:14:53PM +0100, Stefan Ehmann wrote:
> > > > CPU: AMD Athlon(TM) XP 2700+ (2166.44-MHz 686-class CPU)
> >
> > ..
> >
> > > Settings/Compiler           | gcc-3.4 | gcc-4.1 | gcc-4.2
> > > ----------------------------+---------+---------+---------
> > > -O2                         |   6.46s |   6.67s |   6.38s
> > > -O2 -funroll-loops          |   4.44s |   4.16s |   4.02s
> > > -O2 -march=athlon-xp -fun.. |   4.39s |   4.38s |   4.26s
> > > -O3                         |   6.14s |   5.23s |   5.16s
> > > -O3 -funroll-loops          |   4.24s |   4.87s |   4.95s
> > > -O3 -march=athlon-xp -fun.. |   4.19s |   4.90s |   5.07s
> >
> > A fine example that -O3 isn't always better than -O2.
> > I wonder if you're blowing the L2 cache.  IIRC, all Athlon XP 2700+
> > are the Thoughbread core, which has only 256KB L2.
>
> I'd be very much interested in -Os numbers. It should help with the
> cache ...

While -Os -funroll-loops seems a weird combination:

Settings/Compiler           | gcc-3.4 | gcc-4.1 | gcc-4.2
----------------------------+---------+---------+---------
-Os                         |   6.96s |   6.48s |   6.69s
-Os -funroll-loops          |   5.01s |   4.63s |   4.58s
-Os -march=athlon-xp -fun   |   4.93s |   4.69s |   4.64s

We probably should stop exploiting my simple test or perform it properly if 
there's really any interest (e.g. larger number of programs, different CPUs, 
something better than time(1); also my computer was up to 0.05s slower than 
on Friday :-))

Also, for most "normal" programs, there won't be that much difference between 
compilers and/or settings.