--- MITgcm/eesupp/src/exch_z_rx_cube.template 2001/03/28 19:48:51 1.1 +++ MITgcm/eesupp/src/exch_z_rx_cube.template 2001/05/29 14:01:36 1.2 @@ -0,0 +1,170 @@ +C $Header: /home/ubuntu/mnt/e9_copy/MITgcm/eesupp/src/Attic/exch_z_rx_cube.template,v 1.2 2001/05/29 14:01:36 adcroft Exp $ +C $Name: $ + +#include "CPP_EEOPTIONS.h" + + SUBROUTINE EXCH_Z_RX_CUBE( + U array, + I myOLw, myOLe, myOLn, myOLs, myNz, + I exchWidthX, exchWidthY, + I simulationMode, cornerMode, myThid ) +C /==========================================================\ +C | SUBROUTINE EXCH_Z_RX_CUBE | +C | o Control edge exchanges for RX array. | +C |==========================================================| +C | | +C | Controlling routine for exchange of XY edges of an array | +C | distributed in X and Y. The routine interfaces to | +C | communication routines that can use messages passing | +C | exchanges, put type exchanges or get type exchanges. | +C | This allows anything from MPI to raw memory channel to | +C | memmap segments to be used as a inter-process and/or | +C | inter-thread communiation and synchronisation | +C | mechanism. | +C | Notes -- | +C | 1. Some low-level mechanisms such as raw memory-channel | +C | or SGI/CRAY shmem put do not have direct Fortran bindings| +C | and are invoked through C stub routines. | +C | 2. Although this routine is fairly general but it does | +C | require nSx and nSy are the same for all innvocations. | +C | There are many common data structures ( myByLo, | +C | westCommunicationMode, mpiIdW etc... ) tied in with | +C | (nSx,nSy). To support arbitray nSx and nSy would require | +C | general forms of these. | +C | | +C \==========================================================/ + IMPLICIT NONE + +C == Global data == +#include "SIZE.h" +#include "EEPARAMS.h" +#include "EESUPPORT.h" +#include "EXCH.h" + +C == Routine arguments == +C array - Array with edges to exchange. +C myOLw - West, East, North and South overlap region sizes. +C myOLe +C myOLn +C myOLs +C exchWidthX - Width of data region exchanged in X. +C exchWidthY - Width of data region exchanged in Y. +C Note -- +C 1. In theory one could have a send width and +C a receive width for each face of each tile. The only +C restriction woul be that the send width of one +C face should equal the receive width of the sent to +C tile face. Dont know if this would be useful. I +C have left it out for now as it requires additional +C bookeeping. +C simulationMode - Forward or reverse mode exchange ( provides +C support for adjoint integration of code. ) +C cornerMode - Flag indicating whether corner updates are +C needed. +C myThid - Thread number of this instance of S/R EXCH... + INTEGER myOLw + INTEGER myOLe + INTEGER myOLs + INTEGER myOLn + INTEGER myNz + INTEGER exchWidthX + INTEGER exchWidthY + INTEGER simulationMode + INTEGER cornerMode + INTEGER myThid + _RX array(1-myOLw:sNx+myOLe, + & 1-myOLs:sNy+myOLn, + & myNZ, nSx, nSy) + +C == Local variables == +C theSimulationMode - Holds working copy of simulation mode +C theCornerMode - Holds working copy of corner mode + INTEGER theSimulationMode + INTEGER theCornerMode + INTEGER I,J,K,repeat + INTEGER bl,bt,bn,bs,be,bw + +C == Statement function == + INTEGER tilemod + tilemod(I)=1+mod(I-1+6,6) + + theSimulationMode = simulationMode + theCornerMode = cornerMode + +C For now tile<->tile exchanges are sequentialised through +C thread 1. This is a temporary feature for preliminary testing until +C general tile decomposistion is in place (CNH April 11, 2001) + CALL BAR2( myThid ) + IF ( myThid .EQ. 1 ) THEN + + DO repeat=1,2 + + DO bl = 1, 5, 2 + + bt = bl + bn=tilemod(bt+2) + bs=tilemod(bt-1) + be=tilemod(bt+1) + bw=tilemod(bt-2) + + DO K = 1, myNz + DO J = 1, sNy+1 + DO I = 0, exchWidthX-1 + +C Tile Odd:Odd+2 [get] [North<-West] + array(J,sNy+I+1,K,bt,1) = array(I+1,sNy+2-J,K,bn,1) +C Tile Odd:Odd+1 [get] [East<-West] + array(sNx+I+1,J,K,bt,1) = array(I+1,J,K,be,1) + +cs- these above loop should really have the same range the lower one + ENDDO + DO I = 1, exchWidthX-0 +cs- but this replaces the missing I/O routines for now + +C Tile Odd:Odd-1 [get] [South<-North] + array(J,1-I,K,bt,1) = array(J,sNy+1-I,K,bs,1) +C Tile Odd:Odd-2 [get] [West<-North] + array(1-I,J,K,bt,1) = array(sNx+2-J,sNy+1-I,K,bw,1) + + ENDDO + ENDDO + ENDDO + + bt = bl+1 + bn=tilemod(bt+1) + bs=tilemod(bt-2) + be=tilemod(bt+2) + bw=tilemod(bt-1) + + DO K = 1, myNz + DO J = 1, sNy+1 + DO I = 0, exchWidthX-1 + +C Tile Even:Even+1 [get] [North<-South] + array(J,sNy+I+1,K,bt,1) = array(J,I+1,K,bn,1) +C Tile Even:Even+2 [get] [East<-South] + array(sNx+I+1,J,K,bt,1) = array(sNx+2-J,I+1,K,be,1) + +cs- these above loop should really have the same range the lower one + ENDDO + DO I = 1, exchWidthX-0 +cs- but this replaces the missing I/O routines for now + +C Tile Even:Even-2 [get] [South<-East] + array(J,1-I,K,bt,1) = array(sNx+1-I,sNy+2-J,K,bs,1) +C Tile Even:Even-1 [get] [West<-East] + array(1-I,J,K,bt,1) = array(sNx+1-I,J,K,bw,1) + + ENDDO + ENDDO + ENDDO + + ENDDO + + ENDDO + + ENDIF + CALL BAR2(myThid) + + RETURN + END