Hi, Any news regarding this issue? See below an example of MPI_Status that causes the abort.
Thanks
status.MPI_SOURCE = -2
status.MPI_TAG = -32766
status.MPI_ERROR = 0
status.st_length = -1191808578
MPI_Get_count: invalid argument (rank 0, MPI_COMM_WORLD)
Rank (0, MPI_COMM_WORLD): Call stack within LAM:
Rank (0, MPI_COMM_WORLD): - MPI_Get_count()
Rank (0, MPI_COMM_WORLD): - main()
It seems that sometimes meaningless non-zero values are returned in MPI_Status.st_length that causes an abort in MPI_Get_count when the tested (MPI_Test) operation is a send (MPI_Isend). I know that I shouldn't call MPI_Get_count for sends but I do this for convenience. -2 is consistently returned in MPI_Status.MPI_SOURCE and -32766 in MPI_TAG so I guess I could use one of these to check if the request was a send. Is there any other way? Does -2 correspond to a named constant (-32766 is MPI_UNDEFINED)? Also, is there a standard on what to return for these values in this case (e.g. MPI_ANY_SOURCE and MPI_ANY_TAG are returned if the request is MPI_REQUEST_NULL).
Thanks,
Andras Laszloffy
MPI_Get_count: invalid argument (rank 1, MPI_COMM_WORLD)
Rank (1, MPI_COMM_WORLD): Call stack within LAM:
Rank (1, MPI_COMM_WORLD): - MPI_Get_count()
Rank (1, MPI_COMM_WORLD): - main()
---------------------------------
Do you Yahoo!?
Free online calendar with sync to Outlook(TM).
|
