|
[Sponsors] |
Taking advantage of GCCbs profilearcs optimization |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
June 28, 2007, 18:34 |
I was wondering if anybody has
|
#1 |
Guest
Posts: n/a
|
I was wondering if anybody has played around with using GCC's profile arcs optimization. I figured I would mention it since making OpenFOAM just a little faster would have big payoffs for serious computations.
how you use it. 1) compile your app as normal but add the flag -fprofile-generate 2) run your app on a _good_ sample case. when the app exits it will put a <source>.gcda file in the directory where the source is located for your app. 3) recompile your app. This time remove the -fprofile-generate flag and add the -fprofile-use flag. Gcc will now use the profile data in the gcda file to optimize your code. 4) Use your app as you normally would and enjoy the fact it runs faster now I haven't tried it on OpenFoam yet, but on other apps I have found a speed increase of 2 to 13%. |
|
June 28, 2007, 18:44 |
It would be interesting to com
|
#2 |
Senior Member
Alberto Passalacqua
Join Date: Mar 2009
Location: Ames, Iowa, United States
Posts: 1,912
Rep Power: 36 |
It would be interesting to compare the optimized version to the one compiled with Intel C++ Compiler.
Regards, A.
__________________
Alberto Passalacqua GeekoCFD - A free distribution based on openSUSE 64 bit with CFD tools, including OpenFOAM. Available as in both physical and virtual formats (current status: http://albertopassalacqua.com/?p=1541) OpenQBMM - An open-source implementation of quadrature-based moment methods. To obtain more accurate answers, please specify the version of OpenFOAM you are using. |
|
June 29, 2007, 14:11 |
You might see how GCC stacks u
|
#3 |
Guest
Posts: n/a
|
You might see how GCC stacks up against IntelC++ compiling with -ftree-vectorize flag. It tends to help when dealing with programs that use alot of math while chewing through chunks of memory. It also helps the performance of Mesa on non graphics card accelerated functions.
Since you're on an x86, you may also want to try the -mfpmath=sse,387 flag. It tells GCC to try and use both math units. It is a bit of a toss up on whether it will help or hurt performance |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Taking Course / Training | Huygen | Main CFD Forum | 1 | November 15, 2008 07:51 |
taking screen shots | c-dot | Siemens | 8 | December 1, 2007 19:03 |
What happened to Advantage CFD? | gocarts | Main CFD Forum | 0 | February 6, 2007 13:01 |
taking more information about a macro???? | Asghari | FLUENT | 0 | July 24, 2006 13:14 |
Taking average in CFX Post | TING | CFX | 0 | August 24, 2004 10:00 |