On Thu, 26 Jun 2003, Pak, Anne O wrote:
> i use malloc to allocate memory for variable that i use to store data
> that i eventually send out to other nodes using MPI_Send.
>
> if i don't call free at the end, my program works fine, but if i do,
> (even if i call it after i complete MPI_Send), i get
> MPI process rank 0 (n26, p22769) caught a SIGSEGV.
>
> for example:
>
> double *temp;
> temp = (double *)malloc(3*sizeof(double));
> for (i=0;i<3;i++)
> temp[i]=i;
> MPI_Send(temp,3,MPI_DOUBLE,1,tag,comm);
> free(temp);
>
> is there a trick to this? do i need to call MPI_Barrier or something?
That's definitely odd, and from the code snipit shown here, should not
happen. When MPI_Send returns, the temp buffer is totally back in your
control, so you should be able to free it.
This is a typical symptom of some other kind of memory badness elsewhere
in your code. Can you run it through a memory-checking debugger, such as
valgrind, Purify, or bcheck? If that's not an option, you might want to
see if someone is accidentally trying to use the temp buffer after you
free it -- try assigning it to NULL after the free statement. Or
assigning it to NULL *instead* of the free statement. You might want to
try to run this inside a debugger (or attach a debugger after the program
starts) and see exactly where the seg fault is occuring.
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
