The ABINIT Users Mailing List ( CLOSED )

[Eric Roman <ESRoman@berkeley.edu>]

1.
I've seen the exact same problem with abinit 4.3 on our IBM SP.  I posted a
note a few months ago about this, received some sympathy from other SP users,
but never got a resolution from the abinit team.
 
I don't think 4.3 (parallel) was ever successfully tested on an SP.

The problem is that abinit is trying to receive a message into a receive
buffer, but tells MPI that the length of the receive buffer is zero.

Though this is not a problem for a zero-byte send, the matching message in
my case is between 100-200 bytes.  (seen by wrapping the mpi_recv).

I think the parallel version running on one processor works in both cases.
I'll have to check my notes for sure.

2.
abinit 4.2 (parallel) works fine on the IBM SP.  Same makefile_macros.
No sign of this problem.

(Didn't try 4.2 w/ LAM).

3.
I can reproduce it on a PC cluster using the LAM MPI libraries.  (I see it in
MPI_Recv).

Here's the LAM error corresponding to the MPI_Recv case:

%tail alp.log
================================================================================

 ----iterations are completed or convergence reached----

 outwf  : write wavefunction to file alpo_DS11_WFK
-P-0000  leave_test : synchronization done...
MPI_Recv: message truncated (rank 0, MPI_COMM_WORLD)
Rank (0, MPI_COMM_WORLD): Call stack within LAM:
Rank (0, MPI_COMM_WORLD):  - MPI_Recv()
Rank (0, MPI_COMM_WORLD):  - main()

Saw this in outwf.f.  It was near the send at line 362 in abinit v4.3.2.
I _think_ I got this message any time I had prtwfk set to 1.

4.
My suggestion is to try version 4.2.  (I cannot use it, since I need a
feature in 4.3.)

Good luck,
Eric

-- 
Eric Roman                       Department of Physics
510-642-7302                     UC Berkeley