[Omp] Good OpenMP Debuggers?

[Dieter an Mey]

Hello Michael,

I quickly covered OpenMP debugging recently.

A few slides are in the web:
http://www.rz.rwth-aachen.de/computing/events/2006/sunhpc_2006/index.php#Agenda

We also took some videos which are not yet completely post-processed.
So you will miss the TV life demo.

Using Solaris and Sun Studio you can use TotalView (prefered) or the 
SunIDE debugger (based on dbx) on Sparc and Opteron to debug OpenMP.

So Solaris on Opteron may be an intesting choice for your own boxes.

But as SunStudio is now free and available for Linux as well, it may be 
interesting fro your Opteron boxes as well. But I did not try debugging 
OpenMP recently with this combination.

best regards
Dieter



Meadows, Lawrence F schrieb:
> Michael:
> 
> A lot of it seems to depend on getting the right version match between
> debuggers and compilers. I've
> had pretty good luck with most of the debuggers you list at various
> times with the "right" version of the compilers on the "right"
> platform, and bad luck at other times.
> 
> Most compilers (PGI excepted) "outline" parallel regions; so usually
> debugging inside a parallel region is like debugging a subroutine.
> So if the compiler generates good symbolic information, then the
> debugger doesn't have a lot of extra work to do for local variables.
> Going up-scope to shared variables can be harder.
> 
> Giving the illusion that a parallel region is called from the lexical
> point where it occurs is harder; usually the code representing
> the region is called from some kind of "idle" loop in a slave thread.
> Totalview tries to do the right thing with that, and I'm sure that
> other debuggers do so as well; but it is still a fiction as far
> as what is really going on. This is probably why you are having
> trouble stepping into parallel regions.
> 
> Finally, different platforms have different levels of control
> per-thread. So that makes things harder. For example, single
> stepping a thread can be a painful experience. gdb seems to
> be particularly annoying here.
> 
> Good luck. Just remember that printf() is your friend, but hope
> it doesn't take a lock :)
> 
> Larry
> 
>> -----Original Message-----
>> From: omp-bounces at openmp.org [mailto:omp-bounces at openmp.org] 
>> On Behalf Of Michael Suess
>> Sent: Thursday, July 06, 2006 6:41 AM
>> To: Omp at openmp.org
>> Subject: [Omp] Good OpenMP Debuggers?
>>
>> Hi,
>>
>> I have been searching for a good OpenMP debugger recently, and 
>> was not really 
>> satisfied with the results so far. Some did not let me step through my 
>> parallel regions, some did not show my private variables and some just 
>> crashed constantly or did not execute my programs correctly at 
>> all. I have 
>> tried Totalview, DDT, the Sun Studio debugger and the Portland Group 
>> debugger. Oh and gdb and the Intel Debugger I have tried some 
>> time ago. Have 
>> I missed any good ones? Or am I just too stupid to use them 
>> correctly?? 
>>
>> Some of these debuggers do a really great job with 
>> MPI-programs, yet threaded 
>> debugging seems to be a totally different story. What do you 
>> use if you have 
>> to debug OpenMP-code?
>>
>> Thanks for your answers,
>> Michael Süß
>> _______________________________________________
>> Omp mailing list
>> Omp at openmp.org
>> http://openmp.org/mailman/listinfo/omp
>>
> _______________________________________________
> Omp mailing list
> Omp at openmp.org
> http://openmp.org/mailman/listinfo/omp
> 


-- 
--------------------------------------------------------------------
Dieter an Mey
High Performance Computing               Hochleistungsrechnen
RWTH Aachen University                   Rechen- und Kommunikations-
Center for Computing and Communication   zentrum der RWTH Aachen
phone: ++49-(0)241-80-24377              Seffenter Weg 23
fax:   ++49-(0)241-80-22134              52074 Aachen, Germany
email: anmey at rz.rwth-aachen.de
--------------------------------------------------------------------