|
From: Ashley P. <as...@qu...> - 2006-03-01 20:05:46
|
I can't help you here other than to say it should just work (tm). This could be a problem with valgrind loading debug information from intel compiled binaries, in which case it would be helpful if you gave a version number but you'll have to wait for one of the developers to look at it (I'm a fellow MPI user, not a valgrind hacker). I've seen something similar with icc 9.0 compiling c code but haven't managed to narrow down a reproducer yet :( Oh and it's data corruption you are looking for, not a memory leak. Ashley, On 1 Mar 2006, at 19:13, Yeliang Zhang wrote: > Hi, > > Thank you very much for your email. I have compiled my code with "-g > -O0" switches but valgrind doesn't point me a line number for the > memory leak. Do you know how to detect the memory leak here? The > output is not helpful at all. > > ==32067== Invalid write of size 8 > ==32067== at 0x804FCFF: MAIN__ (in > /home/ylzhang/hydrology/parallel_3/mmox) > ==32067== by 0x804DE00: main (in > /home/ylzhang/hydrology/parallel_3/mmox) > ==32067== Address 0x4370E60 is 8 bytes before a block of size 8 > alloc'd > ==32067== at 0x400446D: malloc (vg_replace_malloc.c:149) > ==32067== by 0x40896CC: for_allocate (in > /opt/intel/fc/9.0/lib/libifcore.so.5) > ==32067== by 0x4089638: for_alloc_allocatable (in > /opt/intel/fc/9.0/lib/libifcore.so.5) > ==32067== by 0x804E149: MAIN__ (in > /home/ylzhang/hydrology/parallel_3/mmox) > ==32067== by 0x804DE00: main (in > /home/ylzhang/hydrology/parallel_3/mmox) > ==32067== > > > Thank you. > > Yeliang > -----Original Message----- > From: Ashley Pittman [mailto:as...@qu...] > Sent: Wednesday, March 01, 2006 11:53 AM > To: Yeliang Zhang > Cc: val...@li... > Subject: [POSSIBLE SPAM] Re: [Valgrind-users] mpi program with valgrind > Importance: Low > > > On Wed, 2006-03-01 at 11:00 -0700, Yeliang Zhang wrote: >> Hi, >> >> May anybody tell me how to valgrind an MPI program? >> >> I am using "mpirun -np 1 valgrind ./a.out" now, am I right? > > That's very close to being correct if not correct. It depends a lot on > which resource manager you are using but as you are only using one > process then there isn't a problem in this case. There is potential > for > only the first process being run under valgrind if you were using a > different process count however. > >> The result of the execution is shown below. It gives a lot of error >> on intel fortran compiler library but very few information regarding >> my program. Is this normal? > > What makes you think they are errors in the intel fortran library? The > stack trace points at MAIN__ which is presumably your code? > > Ashley, |