| View previous topic :: View next topic |
| Author |
Message |
JNpederzani
Joined: 09 Feb 2006 Posts: 1
|
Posted: Mon Jun 04, 2007 2:07 pm Post subject: Profiling: subroutine cost from different parents |
|
|
Hello,
I've been trying for a while now, but don't seem to find a way. Let's say I have a subroutine A that is called by subroutine B and C. At the moment using pgprof I'm able to show the total cost for A but not divided in the two categories cost while called from B and cost while called from C. I've tried compiling with almost all flags, and I've read to PGI_tools guide, and the is no direct reference to this feature, so I'm thinking that maybe it's not possible.
Thanks for you help
Jean-Noel |
|
| Back to top |
|
 |
mkcolg
Joined: 30 Jun 2004 Posts: 5001 Location: The Portland Group Inc.
|
Posted: Mon Jun 04, 2007 2:41 pm Post subject: |
|
|
Hi Jean-Noel,
While PGPROF doesn't support call graph profiling, gprof (Linux) and CodeAnylst (Windows) do.
For gprof, compile your code using the PGI compilers and the "-pg" flag to instrument the binary with gprof-style profiling. After you run your binary, a "gmon.out" file will be created and can be used with the gprof utility.
CodeAnalyst can be found at http://developer.amd.com/cawin.jsp.
Hope this helps,
Mat |
|
| Back to top |
|
 |
PeteBradley
Joined: 10 Apr 2006 Posts: 11
|
Posted: Wed Jul 15, 2009 2:02 pm Post subject: |
|
|
+1 for wanting this feature directly in pgprof. It can be really important for big codes.
Pete |
|
| Back to top |
|
 |
mkcolg
Joined: 30 Jun 2004 Posts: 5001 Location: The Portland Group Inc.
|
Posted: Thu Jul 16, 2009 9:17 am Post subject: |
|
|
Hi Pete,
I sent a note to our lead tools engineer about this request. He said that we do have "callpath" profiling on list of future enhancements but so far there has been only a few requests for it. He'll see if he can give it a higher priority but it still may be a ways away.
- Mat |
|
| Back to top |
|
 |
El Depurador
Joined: 26 Mar 2008 Posts: 1
|
Posted: Tue Aug 04, 2009 7:49 pm Post subject: |
|
|
I would like to second PeteBradley's request.
As a user of a big code, I could sure use this feature!
Cheers,
Jon |
|
| Back to top |
|
 |
|