Re: Performance of Integer Multiplication on PIII (Results forgcc-2.95& Athlon)

[pete at ltoi dot iap dot physik dot tu-darmstadt dot de]

Hi Paolo,

On Mon, 5 Nov 2001, Paolo Carlini wrote:

> > {Hi Paolo (Carlini) listening: What was the quintessence of your flops
>
> > example?}
>
> Hi Peter!
> Yes I'm listening, but, honestly, after 12 hours of work in the lab and
> a tough libstdc++ patch in the work, I have trouble keeping up with this
> thread and your dozens of numbers ;-)
>
> But right now I have some spare time and I would be glad to test on my
> PII the patch that Jan prepared.
>
> Could you possibly forward it to me? Moreover, could you possibly point
> me quickly to the testcase and indicate the most relevant optimization
> switches?

you might consider using the source attached to

  http://gcc.gnu.org/ml/gcc/2001-11/msg00255.html

{scroll to the bottom} {crosscheck with the original, in case i screwed
 something}

So, far the "Blow away all Intel stuff" (;-) sorry) options was

Base:= -march=athlon -O2

Base & -fomit-frame-pointer -malign-loops=2 -malign-jumps=2
-mpreferred-stack-boundary=2

and the "read" used was read2.c {The Hormer-schemata}

But, checking here with the PII-400 still sucks:

hgcc -s -o imul imul.c read_empty.c read2.c -march=i686 -O2
-fomit-frame-pointer -malign-loops=2 malign-jumps=2
-mpreferred-stack-boundary=2

yields:

Loop: 1.18, Code: 2.57
Clocks: 19.15

Oh, hey, not so much this time with -mpreferred-stack-boundary=2
i.e.
  2.57*400/500= 2.056

  2.056/1.97  = 1.044

only 4% away from intel's icc ... this may be due to a pure PIII enhancement
over PII ...

Thus, Kevin, you might give that a try.

Peter