Problem about mpi_alltoall!

Hi:
   We have a PC cluster connected with Myrinet. Each PC is a SMP with four processors running linux 6.2. We've installed GM-1.4pre48 and mpich-1.2..4. 
   We run the example fpi in just one PC using four processes with one process per processor. If the source code of fpi.f is not modified,it runs ok. But if we modify fpi.f as follows:
   1.add variable definition:
      interger ngm
      parameter (ngm=scalenum)
      real gm1(ngm*4), gm2(ngm*4)
   2.add the following to just before "go to 10" in fpi.f:
      call mpi_alltoall(gm1(1),ngm,mpi_real,
                        gm2(1),ngm,mpi_real,
                        mpi_comm_world,ierr)
The results are as follows:
   1.scalenum=100000, it's ok.
   2.scalenum=10000000, the PC locked and no response for keyboard and mouse.
   
   We modify fpi.f as this just to show you our problems with our own program. We have run another program which has the same problem. So maybe these two problems are the same. Is there anything wrong with the implementation of mpi_alltoall or anything else?
   Please check it and help me!

Best regards.
yours Geoge.

   
   

_____________________________________________
35¿îÊÖ»úÍøÉÏ×îµÍ¼Û  http://shopping.263.net/hotsale/topmobile/
ÍøÉÏ»¯×±Æ·×¨Âô³¡   http://shopping.263.net/category04.htm