Re: [PATCH] Split mantissa calculation loop and add branchprediction to mp multiplication

[Siddhesh Poyarekar <siddhesh at redhat dot com>]

On Thu, Jan 03, 2013 at 11:09:25AM -0600, Steven Munroe wrote:
> > 
> I was referring to this code from your note:

OK, thanks for clarifying.  That code snippet is from another patch
and is not related to this current patch.  I may have started that
confusion by introducing that snippet here, so I apologize.

> > (2) u does not exist since it is replaced by a much simpler operation,
> >     which results in that snippet looking like this:
> > 
> >     int64_t tmp = Z[k];
> >     for (i=i1,j=i2-1; i<i2; i++,j--)
> >       tmp += (int64_t) X[i]*Y[j];
> > 
> >     Z[k]  = (int) (tmp % (1 << 24));
> >     Z[--k] = (int) (tmp / (1 << 24));
> 
> I do performance analysis and tuning for a living and this is an obvious
> problem. oprofile will show this a hot spot.
> 

Let's have this discussion when I post the complete patch around that
snippet again - the snippet by itself is incomplete and says nothing.
I'll cc you on it and will also have powerpc numbers handy.  I'll also
set up a branch with all of the work I have done so far on this - I'll
need some time to get the patches in at least a moderately coherent
manner.

Thanks for your comments so far!

Siddhesh