Hi,
I just recompiled PovRAy 3.6 using not default GCC 3.2 and -march=i686,
but gcc-4.0 and -march=k8 (but still in 32b mode).

Benchmark was a bit faster - about 34 minutes instead of 38 (Amd64 2800, 1
GB ram), but the renderer images differed a bit - it looked liked photos
where more blurred in gcc4/k8. Is gcc4 misscompiling povray?

I will repeat tests more precisely soon.

-- 

Post a reply to this message

Raf256 schrieb:
> Hi,
> I just recompiled PovRAy 3.6 using not default GCC 3.2 and -march=i686,
> but gcc-4.0 and -march=k8 (but still in 32b mode).
> 
> Benchmark was a bit faster - about 34 minutes instead of 38 (Amd64 2800, 1
> GB ram), but the renderer images differed a bit - it looked liked photos
> where more blurred in gcc4/k8. Is gcc4 misscompiling povray?

I think -march=k8 uses the SSE unit to do floating point calculations. 
GCC 3.2 uses the x87 FPU I think.

-- 

mat### [at] matweide
http://www.matwei.de

Post a reply to this message

=?ISO-8859-2?Q?Matthias_Wei=DFer?= <mat### [at] matweide> wrote:
> Raf256 schrieb:
> > Hi,
> > I just recompiled PovRAy 3.6 using not default GCC 3.2 and -march=i686,
> > but gcc-4.0 and -march=k8 (but still in 32b mode).
> >
> > Benchmark was a bit faster - about 34 minutes instead of 38 (Amd64 2800, 1
> > GB ram), but the renderer images differed a bit - it looked liked photos
> > where more blurred in gcc4/k8. Is gcc4 misscompiling povray?
>
> I think -march=k8 uses the SSE unit to do floating point calculations.
> GCC 3.2 uses the x87 FPU I think.
>
> --

> mat### [at] matweide
> http://www.matwei.de

To truly get a comparison, you would need to do the following:
-march=k8 -msse2 -mfpmath=sse2
On both compilers.  Otherwise, you're getting completely different code
produced.

....Chambers

Post a reply to this message