Re: [Gomp-discuss] Profiling.

[Lars Segerlund]

 It might be cpu dependent, as there are counters for the number of 
executed fops on some cpu's and not on other. This is obviously the best 
solution.

 If this is lacking, it is possible to generate breakpoints, ( stubs ), 
and count the instructions in the asm ( well not realy but you know what 
i mean ), and use this timings to get an accurate count. Or to implement 
a fops counter which is increased at strategical points.


 / Lars Segerlund.

Biagio Lucini wrote:
> Hi Lars - Thank you for reviving this mailing list :-)
>
> On Mon, 28 Apr 2003, Lars Segerlund wrote:
>
>
> >  I found this on slash today, does it look like what we need, I gave it
> > a glance , and it seems to be a good fit for our requirements.
> >
> >  Any comments ?
>
>
> It looks like a nice tool. There are several around, but many of them are
> commercial. By the way, totally unrelated (almost): does anybody know how
> to measure the performance in (M)Flops for a given code?
>
> Thanks
> Biagio
>
>
> _______________________________________________
> Gomp-discuss mailing list
> address@hidden
> http://mail.nongnu.org/mailman/listinfo/gomp-discuss