[OMPI devel] *neighbor_alltoall* are broken

2014-05-23 Thread Gilles Gouaillardet
Folks,

i noticed that *neighbor_alltoall* are now broken.

the bug is in the way parameters are checked (i revamped this and did the
wrong thing for neighbor communications :-()

this bug is only visible when the number of tasks become large
(this can explain why i did not detect this on my VM ...)

i am sorry for the mess and now working on a fix

Gilles


[OMPI devel] Still problems with del_procs in trunkj

2014-05-23 Thread Rolf vandeVaart
I am still seeing problems with del_procs with openib.  Do we believe 
everything should be working?  This is with the latest trunk (updated 1 hour 
ago).

[rvandevaart@drossetti-ivy0 examples]$ mpirun --mca btl_openib_if_include 
mlx5_0:1 -np 2 -host drossetti-ivy0,drossetti-ivy1 connectivity_cConnectivity 
test on 2 processes PASSED.
connectivity_c: ../../../../../ompi/mca/btl/openib/btl_openib.c:1151: 
mca_btl_openib_del_procs: Assertion 
`((opal_object_t*)endpoint)->obj_reference_count == 1' failed.
connectivity_c: ../../../../../ompi/mca/btl/openib/btl_openib.c:1151: 
mca_btl_openib_del_procs: Assertion 
`((opal_object_t*)endpoint)->obj_reference_count == 1' failed.
--
mpirun noticed that process rank 1 with PID 28443 on node drossetti-ivy1 exited 
on signal 11 (Segmentation fault).
--
[rvandevaart@drossetti-ivy0 examples]$ 
---
This email message is for the sole use of the intended recipient(s) and may 
contain
confidential information.  Any unauthorized review, use, disclosure or 
distribution
is prohibited.  If you are not the intended recipient, please contact the 
sender by
reply email and destroy all copies of the original message.
---


[OMPI devel] Threshold for pinning down user-buffers

2014-05-23 Thread Siddhartha Jana
Hi,
Is my understanding correct?
While servicing MPI_Send  using eager protocol over openib, there is a
data-size threshold below which it is preferred to copy the contents of the
user-buffer to a pre-pinned buffer instead of directly pinning down the
user-buffer. If this is correct, how does one determine this threshold?

As always, thanks for the help
Sid