Re: OT : Approximate / fast math libraries ?

Hi,

On Friday 31 August 2007 8:48:49 pm Globe Trotter wrote:
Btw, I usually use -O3 rather than -O2. I agree with the other poster: I am
not sure getting rid of precision is a great idea.

I don't have overall control, so cannot just switch to -03 myself. I did try
it out though and it didn't make a lot of difference. Maybe I should
investigate that a little deeper though to make sure something wasn't wrong.

Also, I agree with you in general on the precision, just in one particular
case I know I need much less than I am currently getting, and using atan2
seems to be taking significant cpu (w.r.t. the rest of the method, i.e. 50%
or so).

Chris


Trotter

----- Original Message ----
From: Chris Jones <jonesc@xxxxxxxxxxxxxxxxx>
To: For users of Fedora <fedora-list@xxxxxxxxxx>
Sent: Friday, August 31, 2007 2:41:56 PM
Subject: Re: OT : Approximate / fast math libraries ?

Hi,

Thanks for your feedback.

What exactly is your need? Contact me off-list and maybe I can
help. Have you profiled your code? I have found that people
often do not actually know where their code is spending its
time. I once sped up an app which was universally acknowledged
to be slow "because it uses floating point." I sped it up 3x.

Yes, I have profiled the code, quite extensively, using the
valgrind/calltree application. From this I know this that I'm know I've
tidying up this to the point where its hard to find big improvements, the
cpu time is fairly well spread around, not isolated in a few places. So am
now looking a a few places where math calls are taking more time than I
would hope. I'm not going to get factors in speed in the overall
application, but I hope in a few places things can be improvemed a lot
locally.

Also, the project is not small, massive in fact, and I'm only writting one
small part. If you are interested you can find it here

http://lhcb-release-area.web.cern.ch/LHCb-release-area/DOC/brunel/releases/
latest/doxygen/index.html

It also has to be supported on a *lot* of hardware. Basically gcc 3.2.3
based Scientific Linux 3 machines, gcc 3.4.6 SL4 machines (32 and 64 bit)
and (not my decision), windows VC 7.1. I cannot rely on for instance SSE
math calls etc.

Taking an example from another thread, one place I'm trying to understand
is where I use atan2 see

http://www.hep.phy.cam.ac.uk/~jonesc/atan2.png

for the profiler output. atan2 is taking 50% of the time of this method.
Not here I don't need that much precision on the result - say +-
O(2*pi/100). Anything you can suggest here - The code is here

http://www.hep.phy.cam.ac.uk/~jonesc/RichPhotonRecoUsingCKEstiFromRadius.cp
p

( note though its full of internal classes etc...)

cheers Chris

I modified the parsing routines it used, not the floating point.

Mike
--
p="p=%c%s%c;main(){printf(p,34,p,34);}";main(){printf(p,34,p,34);}
Oppose globalization and One World Governments like the UN.
This message made from 100% recycled bits.
You have found the bank of Larn.
I can explain it for you, but I can't understand it for you.
I speak only for myself, and I am unanimous in that!

--
fedora-list mailing list
fedora-list@xxxxxxxxxx
To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list






___________________________________________________________________________
_________ Pinpoint customers who are looking for what you sell.
http://searchmarketing.yahoo.com/


--
fedora-list mailing list
fedora-list@xxxxxxxxxx
To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-list

Relevant Pages