[bgl-discuss] Odd MPI-IO behavior.

[Robert Jacob <jacob@xxxxxxxxxxx>]

Hello,

We are trying to read some data files for the Parallel Ocean Program using 
MPI-IO on BG/L and seeing an odd bug.   The bug seems to be caused by using 
a monotonic decreasing displacement array in the 
MPI_type_create_indexed_block call.   The code works fine on an IBM SP4 
with GPFS but doesn't work on the Argonne and NCAR BG/L's.   The bug 
manifests as a "hole" in the input data field.

A Fortran code fragment illustrating the calls are below.   Some possible 
questions:
1.  Is a monotonic decreasing array legal in the displacement array for 
MPI_type_create_indexed_block?
2.  If so, is this possible a bug in BG/L's MPI?

Thanks,
Rob Jacob
John Dennis


!----------------------------------------------------------------------- 
--------------
!  This displacement generates a bug on NCAR's and Argonne's BGL
!   but works on NCAR's IBM Power4.
displace = [48 40 32 24 16 8 0];

!  This displacement works on NCAR's and Argonne's BGL
! displace = [0 8 16 24 32 40 48];


lenBLOCKS = 24
baseTYPE = MPI_REAL8
numBLOCKS = SIZE(displace)

!---------------------------------
! Create the elementary structure
!---------------------------------
call MPI_Type_contiguous(lenBLOCKS,baseTYPE,elemTYPE,ierr)
call MPI_Type_commit(elemTYPE,ierr)

!---------------------------
! Create the file structure
!---------------------------
call
MPI_Type_create_indexed_block(numBLOCKS,1,displace,elemTYPE,fileTYPE,ier r)
call MPI_Type_commit(fileTYPE,ierr)

!-----------------------
!  Set the file view
!-----------------------
call MPI_File_set_view(File%fh,offset, elemTYPE, fileTYPE,
datarep,File%info,ierr)

allocate(IOBUF(numBLOCKS*lenBLOCKS))
IOBUF = specialVALUE
!-----------------------------
! Read out distributed array
!-----------------------------
call MPI_file_read_all(File%fh,IOBUF,numBLOCKS,elemTYPE,fstatus,ierr)
call MPI_get_count(fstatus,elemTYPE,cnt,ierr)

!------------------------------------------------------
! FAILURE:  Occurs when:
!	COUNT(specialVALUE) != 0
!		and
!	  numBLOCKS == cnt
!------------------------------------------------------
print *,'COUNT(specialVALUE) :=', COUNT(IOBUF .eq. specialVALUE)
print *,'numBLOCKS, cnt: ',numBLOCKS,cnt

!----------------------------------------------------------------------- 
--------------

- --------------------------------------------------------------------
To add or remove yourself from this mailing list, use the 'notifyme'
command on any BGL machine. To remove: notifyme -n, to add: notifyme -y.