Mtrace - Revision history

Matthb2 at 22:28, 1 May 2014

2014-05-01T22:28:02Z

Matthb2 at 20:39, 15 July 2013

2013-07-15T20:39:09Z

Matthb2 at 20:38, 15 July 2013

2013-07-15T20:38:20Z

Matthb2 at 20:34, 15 July 2013

2013-07-15T20:34:21Z

Matthb2: Created page with "If you suspect that you have a memory leak, it is possible to get libc to log heap allocations to a text file using the "mtrace" call. == Serial == At it's simplest, to use mt..."

2013-07-15T20:33:31Z

Created page with "If you suspect that you have a memory leak, it is possible to get libc to log heap allocations to a text file using the "mtrace" call. == Serial == At it's simplest, to use mt..."

New page

If you suspect that you have a memory leak, it is possible to get libc to log heap allocations to a text file using the "mtrace" call.

== Serial ==

At it's simplest, to use mtrace you need to get your code to call the mtrace() function from the "mcheck.h" header as near to the beginning of execution as possible (or if you suspect you know where the leak is, right before the leak).

#include <mcheck.h>

int main(int x, char** y) { mcheck(); return(0); }

Then, prior to running your program, you must set the "MALLOC_TRACE" environment variable to the name of the file you'd like mtrace() to write it's information to.

export MALLOC_TRACE=memory_allocations.dat
./a.out

Once you get the output, you can use the "mtrace" script to post process the results and try to tell you where any leaked memory was allocated

mtrace ./a.out ./memory_allocations.date

== Parallel ==

Using mtrace() in parallel is similar to serial, but you need to ensure that each process gets a different value for MALLOC_TRACE
You may be able to do this by passing options to mpirun, but it's probably simpler to write some code to do this:

#include <mcheck.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>

int main(int argc, char** argv)
{
int rank;
char* tracepath
char* cwd = get_current_dir_name();
MPI_Init(&argc, &argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
asprintf(&tracepath, "%s/mtrace.dat.%d", cwd, rank);
mtrace()
/*....*/
MPI_Finalize();
}

With this code applied, you can then simply run your program normally and analyze the results as before:

mpirun -np 8 ./a.out
mtrace ./a.out mtrace.dat.0
mtrace ./a.out mtrace.dat.1
#...

← Older revision		Revision as of 20:34, 15 July 2013
Line 47:		Line 47:
	mtrace ./a.out mtrace.dat.1		mtrace ./a.out mtrace.dat.1
	#...		#...
		+
		+	In PHASTA, you should probably put this code in common/phasta.cc

@@ Line 36: / Line 36: @@
       MPI_Comm_rank(MPI_COMM_WORLD, &rank);
       asprintf(&tracepath, "%s/mtrace.dat.%d", cwd, rank);
       mtrace()
       /*....*/

@@ Line 16: / Line 16: @@
 Once you get the output, you can use the "mtrace" script to post process the results and try to tell you where any leaked memory was allocated
-   mtrace ./a.out ./memory_allocations.date
+   mtrace ./a.out ./memory_allocations.dat
 == Parallel ==

@@ Line 7: / Line 7: @@
    #include <mcheck.h>
-   int main(int x, char** y) { mcheck(); return(0); }
+   int main(int x, char** y) { mtrace(); return(0); }
 Then, prior to running your program, you must set the "MALLOC_TRACE" environment variable to the name of the file you'd like mtrace() to write it's information to.