On Mar 10, 1:37 pm, Bill Hart <goodwillh...@googlemail.com> wrote:
> Both of those proposed changes actually sound really useful to me. +1 from me.
>
> Bill.
>
> 2009/3/10 Cactus <rieman...@googlemail.com>:
>
>
>
> > Hi All,
>
> > I have been using speed to track down Windows MPIR performance issues
> > and I have made local changes to speed that might be of more general
> > interest. I hence wnat top get your views on what I am doing before I
> > add the capabilities to speed in MPIR.
>
> > Firstly it is painful to do a run of lengths from say 1 to 10 in steps
> > of 1 followed by 10 to 1000 in steps of 10 because the step size
> > option -t is global to each sequence specified. But the first
> > sequence needs a step size of 1 and the second needs 10.
>
> > So, in addition to the 'start-end' notation, I have added 'start(step)
> > end' so that a group such as 100(100)1000 will go from 100 to 1000 in
> > steps of 100.
>
> > Secondly I am in the process of adding an option that changes the
> > printout from a list of timings to the coefficients of the equation of
> > a least squares polynomial curve fit to the results so that we get a
> > [0], a[1] ... in:
>
> > cycles = a[0] + a[1] * n + a[2] * n^2 + a[3] * n^3 .....
>
> > where n is the number of limbs. I have done this because, once the
> > lists of timings are known to be smooth, comparing these coefficients
> > is much easier than comparing long lists of values. I could add a
> > general curve fitting capability but that would get quite hard so I am
> > thinking only of doing up to a quadratic or cubic polynomial.
>
> > Does anyone see benefit in, or have any objection to, either or both
> > of these capabilities being added to MPIR's speed?
>
> > Brian
I have now modified speed.c in SVN trunk to allow limb lengths (-
s ...) to include items in the form 'start(step)end'
It turned out to be difficult to add the curve fitting option to speed
itself so I have produced a Python program to do this.
I have upload this program - run-speed.py - but anyone trying it
should be aware that it is very new and is likely to contain errors.
It requires the new version of speed with my recent 'start(step)end'
modification.
At the moment run-speed.py expects Python 2.6 or later. It expects to
run on Linux in the directory in which speed.exe is located. In the
Windows VC++ build it runs in the build.vc9 directory.
I will be interested in any problems and in any results it produces
(please report machine and its speed).
Brian
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"mpir-devel" group.
To post to this group, send email to mpir-devel@googlegroups.com
To unsubscribe from this group, send email to
mpir-devel+unsubscr...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/mpir-devel?hl=en
-~----------~----~----~----~------~----~------~--~---
