[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: profiler and/or debugger, loops vs vectorization

From: David Bateman
Subject: Re: profiler and/or debugger, loops vs vectorization
Date: Wed, 05 Jul 2006 17:59:33 +0200
User-agent: Mozilla Thunderbird 1.0.6-7.6.20060mdk (X11/20050322)

David Grohmann wrote:

> David Bateman wrote:
>>David Grohmann wrote:
>>>So I have some code that runs in matlab (on a P3 500 mhz 512 MB ram 
>>>windows box) about 133 times faster (33 minutes to complete something 
>>>instead of 15 seconds)  than it does in octave( on a 4 processor 3.6 ghz 
>>>xeon machine with an ungodly amount of ram linux box). my guess is that 
>>>it is the use of loops and not vectorization.
>>>Is such a large performance hit likely caused by loops vs vectorization 
>>>or might there be another explanation besides that?
>>>I've searched online and haven't found anything to profile m files with, 
>>>other than in matlab. I have profiled it in matlab and sure enough it 
>>>does run the lines in the loops many (hundreds of thousands of times) 
>>>times but only one of the loops actually takes up a lot of run time. 
>>>This would probably be a good candidate for vectorizing. I suspect.
>>>So is there anything out there for profiling or stepping through code in 
>>>an m file in octave?
>>>I know I haven't posted the offending code, that is because I currently 
>>>don't have the authorization to do that. So my general question is how 
>>>do people profile their code and what not. I'm not asking you to fix my 
>>>code for me.
>>>Thank you for any help you provide,
>>My guess is that you are running a cygwin binary version of octave.
>>There is a major bug in the setjmp/longjmp exception handler of g++ for
>>versions later than 3.2 that causes major pain to octave. As far as I
>>know this bug in cygwin as not yet been addressed. The result is
>>sssllloooowwww code..
> Octave was on a linux machine, matlab was on a windows machine.
> However it seems that vectorizing the code makes up the difference. I
> didn't know that such a performance boost would happen with vectorization.
The setjmp/longjmp issue is felt most in loops on cygwin..


David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary

reply via email to

[Prev in Thread] Current Thread [Next in Thread]