- Infos im HLRS Wiki sind nicht rechtsverbindlich und ohne Gewähr -
- Information contained in the HLRS Wiki is not legally binding and HLRS is not responsible for any damages that might result from its use -

Valgrind

From HLRS Platforms
Revision as of 12:50, 29 August 2023 by Hpcjohig (talk | contribs) (→‎Hawk special)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search
Valgrind is an award-winning instrumentation framework for building dynamic analysis tools. There are Valgrind tools that can automatically detect many memory management and threading bugs, and profile your programs in detail. You can also use Valgrind to build new tools.
Valgrind-logo.png
Developer: Valgrind Developers
Platforms: NEC Nehalem Cluster
Category: Debugger
License: GNU General Public License, version 2
Website: Valgrind homepage


Usage

In order to use Valgrind, you need to first load it:

module load tools/valgrind/3.6.1


To check your application with Valgrind, please use following command format (take ls command as an example):

valgrind ls


Please note, the default checking tool is Memcheck. If you want to use other tools in Valgrind, you have to specify it via --tool=<name> option.


Valgrind can also be used to check parallel application. For example, in order to check an MPI application, please follow the steps:

  • First reserve compute nodes:
qsub -I -lnodes=16:nehalem:ppn=8,walltime=1:0:0


  • Load all necessary tools:
module load tools/valgrind/3.6.1 mpi/openmpi/1.5.4-gnu-4.6.1


  • Run your MPI application with Valgrind:
mpirun -np 4 valgrind ./your_app


Note: For Open MPI there exists a suppression file which can be used with the following command
mpirun -np 4 valgrind --suppressions=$(dirname $(which mpirun))/../share/openmpi/openmpi-valgrind.supp  ./app


If you want to run Callgrind and later visualize the output in Kcachegrind, you have to do following steps:

  • Run your application with Callgrind:
valgrind --tool=callgrind ls


The above command will generate an output file with name callgrind.out.xxxx, where xxxx is the pid of the checked application.

  • Load Kcachegrind
module load tools/kcachegrind/0.7.0


  • Open Kcachegrind with the generated Callgrind output file. Or open Kcachegrind and load the callgrind file in the GUI.
kcachegrind callgrind.out.4076


You will see the Kcachegrind window similar to this:

Example of using Kcachegrind with Callgrind.

Hawk special

To work with valgrind on Hawk with mpt MPI you should set the following environment variables to disable xpmem and switch to TCP for communication

export MPI_XPMEM_ENABLED=false
export MPI_USE_TCP=true


In case valgrind cannot find a local debuginfo file it will try downloading it via HTTP. Since Hawk is not directly connected to the internet this download attempt can stall valgrind. This can be avoided by unsetting the DEBUGINFOD_URLS environment variable

export DEBUGINFOD_URLS=


See also

External links