[MITgcm-support] using gather/scatter to work with global fields

Jean-Michel Campin jmc at ocean.mit.edu
Mon Aug 27 10:19:55 EDT 2012


Hi Katy,

few comments:
1) the BARRIER calls (either coded with the macro "_BARRIER" or
  explicit call to BAR2) are itended to syncronise threads.
  Unless you you are running multi-threads (nTx > 1 or nTy > 1 in eedata),
  these barrier calls will do nothing.

2) S/R GATHER_2D_R4 is used to write output when the default readBinaryPrec 
  is used (i.e., =32) with single CPU IO (useSingleCpuIO=.TRUE.).
  It has been used & tested by many people in many different set-up,
  so a bug is not very likely; but anyway, it would be useful to check that, 
  in your set-up, using these 2 options (readBinaryPrec=32 and 
  useSingleCpuIO=.TRUE.) give you the right output files.
  Assuming this test passes, it means we can focus on (3):

3) How to (re-) use  S/R GATHER_2D_R4 in a customised piece of code
  within MITgcm.
 I would suggest to pass by MIT one day, so that we can get a more precised
 idea of the purpose of this piece of code and how to make it to work.
 Since few persons from AER are coming regularly to MIT, I imagine it 
 should be possible to make this visit.

Cheers,
Jean-Michel

On Thu, Aug 23, 2012 at 01:40:55PM -0400, Katherine Quinn wrote:
> Hi all,
> Within the MITgcm code (not matlab) I need to gather tiled variables
> into a global array, specifically I'm trying to gather XC and YC.
> I'm using the subroutine gather_2d_r4, which suggests "barrier
> calls" before and after but it seems no matter what various barrier
> calls I use, when I look at the global array there's chunks of tiles
> missing.  Other relevant info - the MITgcm setup is for ecco version
> 4 runs, with all_exch2 turned on.
> Here's the relevant portions of my code, please suggest what I'm
> doing wrong:
> 
> C     == Global variables ===
> #include "SIZE.h"
> #include "EEPARAMS.h"
> #include "EESUPPORT.h"
> #include "PARAMS.h"
> #ifdef ALLOW_EXCH2
> # include "W2_EXCH2_SIZE.h"
> # include "W2_EXCH2_TOPOLOGY.h"
> # include "W2_EXCH2_PARAMS.h"
> #endif /* ALLOW_EXCH2 */
> #include "EEBUFF_SCPU.h"
> #include "GRID.h"
> 
> C     INPUT/OUTPUT PARAMETERS:
> C     == Routine arguments ==
> C     myThid - Thread number for this instance of the routine.
>       INTEGER  myThid
> 
> C     LOCAL VARIABLES:
>       LOGICAL zeroBuff, useExch2ioLayOut
>       INTEGER xSize, ySize
> #ifdef ALLOW_EXCH2
>       _RL  XC_global(exch2_global_Nx,exch2_global_Ny)
>       _RL  YC_global(exch2_global_Nx,exch2_global_Ny)
> #else  /* ALLOW_EXCH2 */
>       _RS  XC_global(Nx,Ny)
>       _RS  YC_global(Nx,Ny)
> #endif /* ALLOW_EXCH2 */
> 
> C Set dimensions and flags (e.g. mdsio_write_field.F)
>       xSize = Nx
>       ySize = Ny
>       useExch2ioLayOut = .FALSE.
>       zeroBuff = .TRUE.
> #ifdef ALLOW_EXCH2
>       IF ( W2_useE2ioLayOut ) THEN
>          xSize = exch2_global_Nx
>          ySize = exch2_global_Ny
>          useExch2ioLayOut = .TRUE.
>       ENDIF
> #endif /* ALLOW_EXCH2 */
> 
> C gather XC and YC into global model grid
> 
>       CALL BAR2( myThid )
>       _BEGIN_MASTER( myThid )
>       CALL GATHER_2D_R4(
>      O                  XC_global,
>      I                  XC,
>      I                  xSize, ySize,
>      I                  useExch2ioLayOut,
>      I                  zeroBuff,
>      I                  myThid )
>       _END_MASTER( myThid )
>       CALL BAR2( myThid )
> 
>       _BEGIN_MASTER( myThid )
>       CALL GATHER_2D_R4(
>      O                  YC_global,
>      I                  YC,
>      I                  xSize, ySize,
>      I                  useExch2ioLayOut,
>      I                  zeroBuff,
>      I                  myThid )
>       _END_MASTER( myThid )
>       CALL BAR2( myThid )
> 
> 
> 
> -- 
> Katherine J. Quinn
> Atmospheric and Environmental Research         voice: 781-761-2234
> 131 Hartwell Avenue                              fax: 781-761-2299
> Lexington, MA  02421-3126                     e-mail: kquinn at aer.com
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support



More information about the MITgcm-support mailing list