Re: [OctDev] problem with ones and openmpi

2010-05-26 Thread Andrea Bressan
On Wed, 26 May 2010 10:03:29 + (GMT) Riccardo Corradini wrote: > It is because you are sending a "range" object and there is little > bug in both procedure. I will fix it as soon as I may. > Just browse the source code and you will see on your own. > .. > octave:3> typeinfo message > ans = sq

[OctDev] problem with ones and openmpi

2010-05-26 Thread Andrea Bressan
Hi all, I 'm facing an odd behavior using MPI_Send with messages created with ones. In particular if the message is created with message=[1 1]; it is successfully delivered, but if it is created with message=ones(1,2); the communication fails. Both MPI_Send and MPI_Recv return MPI_SUCCESS, bu

Re: [OctDev] problem with ones and openmpi

2010-05-26 Thread Riccardo Corradini
Hi Andrea, I posted the new versions of MPI_Send.cc and MPI_Rec.cc on svn on octave-forge. I am about to prepare a new release with the bug fix. Bests Riccardo --- Mer 26/5/10, Andrea Bressan ha scritto: Da: Andrea Bressan Oggetto: Re: problem with ones and openmpi A: "Riccardo Corradini" Cc:

Re: [OctDev] problem with ones and openmpi

2010-05-26 Thread Riccardo Corradini
No. the fix should use the vector of 3 elements and use the proper constructor of range class. Now I am able to receive but I am trying to understand how the constructor works ... int send_range(int t_id, MPI_Comm comm, Range range,ColumnVector rankrec, int mytag){    /* put base,limit,inc