Charm++ parallel programming system

[Forrest Iandola <fiandola AT gmail.com>]

Hi Eduardo,

I've gotten useful profiling results just by compiling the program with gprof (-pg) and debug (-g).  Also, be sure to compile charm with the -g -pg options.  From your charm build directory (eg. netlinux-x86_64), this is what recompiling charm with gprof would look like:
cd ~/charm/tmp
make clean
make LIBS OPTS="-O0 -g -pg" 

After you run your program, you will get a gmon.out file for each core.  So, if you have 2 cores, you'll get gmon.out.0 and gmon.out.1.  You can convert a gmon.out file into a readable file with this command line argument: gprof pgm gmon.out.1 > readableProfilingData.1.txt

My gprof pingpong makefile and some recent pingpong gprof data are here on TMS: https://charm.cs.uiuc.edu/private/tms/listlog.php?param=106&sortBy=postedDate&flag=0

If gprof isn't giving you the sort of output you want, you might want to try some of these other profiling tools: https://wiki.alcf.anl.gov/index.php/Quick_Reference_Guide#Performance_Tools

Forrest

On Mon, Mar 8, 2010 at 11:26 AM, Phil Miller <mille121 AT illinois.edu> wrote:

On Mon, Mar 8, 2010 at 10:05, Eduardo Rocha Rodrigues
<errodrigues AT inf.ufrgs.br> wrote:
> Does anyone know how to make a subroutine profile (like gprof) of a
> AMPI program?

I believe Forrest was simply using gprof for some of his tests.

Phil

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

-- 
Forrest Iandola
College of Engineering - Computer Science
University of Illinois at Urbana-Champaign
http://www.forrestiandola.com