I've been having problems with my 3D matrix transpose program. I'm using
MPI_Type_indexed in order to allign specific blocks that I want to send and
receive across one or multiple nodes of a cluster. Up to few days ago I was
able to run my program without any errors. However several test cases on
t
Bogdan,
As far as I can tell your code is correct, and the problem is coming from
Open MPI. More specifically, I used alloca in the optimization stage in
MPI_Type_commit, and as your arrays of length were too large, alloca failed
and lead to a segfault. I fixed in the trunk (3c489ea), and this wil
Actually, you are not the first to encounter the problem with
'MPI_Type_indexed' for very large datatypes. I also run with a 1.6
release, and solved the problem by switching to
'MPI_Type_Create_Hindexed' for the datatype. The critical difference is
that the displacements for 'MPI_type_indexed' ar
Hello George,
So is it safe for me to assume that my code is good and that you will
remove this bug from next OpenMPI version?
Also I would like to know which future OpenMPI version will incorporate
this fix (so I can try my code in fixed version)?
Thank you,
Bogdan Sataric
email: bogdan
Hello Tom,
Actually I have tried using: MPI_Type_Create_Hindexed but the same problem
persisted for the same matrix dimensions.
Displacements array values are not a problem. Matrix of a size 800x640x480
creates type that is a bit less then 4GB large in case of complex datatype.
It definitely fits
On Thu, Mar 5, 2015 at 6:22 PM, Bogdan Sataric
wrote:
> Hello George,
>
> So is it safe for me to assume that my code is good and that you will
> remove this bug from next OpenMPI version?
>
Yes I think it is safe to assume your code is correct (or at least it
follows the specifications you desc
