1 |
C $Header: /u/gcmpack/MITgcm/eesupp/src/cumulsum_z_tile.F,v 1.1 2011/07/01 18:18:31 jmc Exp $ |
2 |
C $Name: $ |
3 |
|
4 |
#include "PACKAGES_CONFIG.h" |
5 |
#include "CPP_EEOPTIONS.h" |
6 |
|
7 |
C-- File cumulsum_z_tile.F: Routines that perform cumulated sum |
8 |
C on a tiled array, corner grid-cell location |
9 |
C Contents |
10 |
C o CUMULSUM_Z_TILE_RL |
11 |
C o CUMULSUM_Z_TILE_RS <- not yet coded |
12 |
|
13 |
C---+----1----+----2----+----3----+----4----+----5----+----6----+----7-|--+----| |
14 |
CBOP |
15 |
C !ROUTINE: CUMULSUM_Z_TILE_RL |
16 |
|
17 |
C !INTERFACE: |
18 |
SUBROUTINE CUMULSUM_Z_TILE_RL( |
19 |
O psiZ, psiLoc, |
20 |
I dPsiX, dPsiY, myThid ) |
21 |
|
22 |
C !DESCRIPTION: |
23 |
C *==========================================================* |
24 |
C | SUBROUTINE CUMULSUM\_Z\_TILE\_RL |
25 |
C | o Handle cumulated sum for _RL tile data. |
26 |
C *==========================================================* |
27 |
C | Cumulate sum on tiled array, corner grid-cell location: |
28 |
C | Starts from 1rst tile and, going through all tiles & all |
29 |
C | the processes, add increment in both directions |
30 |
C *==========================================================* |
31 |
|
32 |
C !USES: |
33 |
IMPLICIT NONE |
34 |
|
35 |
C == Global data == |
36 |
#include "SIZE.h" |
37 |
#include "EEPARAMS.h" |
38 |
#include "EESUPPORT.h" |
39 |
#include "CUMULSUM.h" |
40 |
|
41 |
C !INPUT/OUTPUT PARAMETERS: |
42 |
C == Routine arguments == |
43 |
C psiZ :: results of cumulated sum, corresponds to tile South-East corner |
44 |
C psiLoc :: cumulated sum at special locations |
45 |
C dPsiX :: tile increment in X direction |
46 |
C dPsiY :: tile increment in Y direction |
47 |
C myThid :: my Thread Id. number |
48 |
_RL psiZ (nSx,nSy) |
49 |
_RL psiLoc(2) |
50 |
_RL dPsiX (nSx,nSy) |
51 |
_RL dPsiY (nSx,nSy) |
52 |
INTEGER myThid |
53 |
|
54 |
C !LOCAL VARIABLES: |
55 |
#ifndef ALLOW_EXCH2 |
56 |
C == Local variables == |
57 |
C bi,bj :: tile indices |
58 |
C- type declaration of: loc[1,2]Buf and shareBufCS[1,2]_R8 : |
59 |
C all 4 needs to have the same length as MPI_DOUBLE_PRECISION |
60 |
INTEGER bi,bj |
61 |
INTEGER nf |
62 |
#ifdef ALLOW_USE_MPI |
63 |
INTEGER biG, bjG, npe, np1 |
64 |
INTEGER lbuf1, lbuf2, idest, itag, ready_to_receive |
65 |
INTEGER istatus(MPI_STATUS_SIZE), ierr |
66 |
Real*8 loc1Buf (nSx,nSy) |
67 |
Real*8 loc2Buf(2,nSx,nSy) |
68 |
Real*8 globalBuf(3,nSx*nPx,nSy*nPy) |
69 |
#endif /* ALLOW_USE_MPI */ |
70 |
#endif /* ALLOW_EXCH2 */ |
71 |
CEOP |
72 |
|
73 |
#ifdef ALLOW_EXCH2 |
74 |
CALL W2_CUMULSUM_Z_TILE_RL( |
75 |
O psiZ, psiLoc, |
76 |
I dPsiX, dPsiY, myThid ) |
77 |
RETURN |
78 |
#else /* ALLOW_EXCH2 */ |
79 |
|
80 |
C-- write input into shared-buffer array |
81 |
DO bj = myByLo(myThid), myByHi(myThid) |
82 |
DO bi = myBxLo(myThid), myBxHi(myThid) |
83 |
shareBufCS2_R8(1,bi,bj) = dPsiX(bi,bj) |
84 |
shareBufCS2_R8(2,bi,bj) = dPsiY(bi,bj) |
85 |
ENDDO |
86 |
ENDDO |
87 |
psiLoc(1) = 0. |
88 |
psiLoc(2) = 0. |
89 |
|
90 |
C-- Master thread cannot start until everyone is ready: |
91 |
CALL BAR2( myThid ) |
92 |
_BEGIN_MASTER( myThid ) |
93 |
|
94 |
#ifdef ALLOW_USE_MPI |
95 |
#ifdef ALWAYS_USE_MPI |
96 |
IF ( .TRUE. ) THEN |
97 |
#else |
98 |
IF ( usingMPI ) THEN |
99 |
#endif |
100 |
|
101 |
lbuf1 = nSx*nSy |
102 |
lbuf2 = 2*lbuf1 |
103 |
idest = 0 |
104 |
itag = 0 |
105 |
ready_to_receive = 0 |
106 |
|
107 |
IF ( mpiMyId.NE.0 ) THEN |
108 |
|
109 |
C-- All proceses except 0 wait to be polled then send local array |
110 |
#ifndef DISABLE_MPI_READY_TO_RECEIVE |
111 |
CALL MPI_RECV (ready_to_receive, 1, MPI_INTEGER, |
112 |
& idest, itag, MPI_COMM_MODEL, istatus, ierr) |
113 |
#endif |
114 |
CALL MPI_SEND (shareBufCS2_R8, lbuf2, MPI_DOUBLE_PRECISION, |
115 |
& idest, itag, MPI_COMM_MODEL, ierr) |
116 |
|
117 |
C-- All proceses except 0 receive result from process 0 |
118 |
CALL MPI_RECV (shareBufCS1_R8, lbuf1, MPI_DOUBLE_PRECISION, |
119 |
& idest, itag, MPI_COMM_MODEL, istatus, ierr) |
120 |
|
121 |
ELSE |
122 |
|
123 |
C-- Process 0 fills-in its local data |
124 |
np1 = 1 |
125 |
DO bj=1,nSy |
126 |
DO bi=1,nSx |
127 |
biG = (mpi_myXGlobalLo(np1)-1)/sNx+bi |
128 |
bjG = (mpi_myYGlobalLo(np1)-1)/sNy+bj |
129 |
globalBuf(1,biG,bjG) = shareBufCS2_R8(1,bi,bj) |
130 |
globalBuf(2,biG,bjG) = shareBufCS2_R8(2,bi,bj) |
131 |
ENDDO |
132 |
ENDDO |
133 |
|
134 |
C-- Process 0 polls and receives data from each process in turn |
135 |
DO npe = 1, numberOfProcs-1 |
136 |
#ifndef DISABLE_MPI_READY_TO_RECEIVE |
137 |
CALL MPI_SEND (ready_to_receive, 1, MPI_INTEGER, |
138 |
& npe, itag, MPI_COMM_MODEL, ierr) |
139 |
#endif |
140 |
CALL MPI_RECV (loc2Buf, lbuf2, MPI_DOUBLE_PRECISION, |
141 |
& npe, itag, MPI_COMM_MODEL, istatus, ierr) |
142 |
|
143 |
C-- Process 0 gathers the local arrays into a global array. |
144 |
np1 = npe + 1 |
145 |
DO bj=1,nSy |
146 |
DO bi=1,nSx |
147 |
biG = (mpi_myXGlobalLo(np1)-1)/sNx+bi |
148 |
bjG = (mpi_myYGlobalLo(np1)-1)/sNy+bj |
149 |
globalBuf(1,biG,bjG) = loc2Buf(1,bi,bj) |
150 |
globalBuf(2,biG,bjG) = loc2Buf(2,bi,bj) |
151 |
ENDDO |
152 |
ENDDO |
153 |
ENDDO |
154 |
|
155 |
C-- Cumulate Sum over all tiles: |
156 |
globalBuf(3,1,1) = 0. |
157 |
bj = 1 |
158 |
DO bi = 1,nSx*nPx-1 |
159 |
globalBuf(3,1+bi,bj) = globalBuf(3,bi,bj) |
160 |
& + globalBuf(1,bi,bj) |
161 |
ENDDO |
162 |
DO bj = 1,nSy*nPy-1 |
163 |
DO bi = 1,nSx*nPx |
164 |
globalBuf(3,bi,1+bj) = globalBuf(3,bi,bj) |
165 |
& + globalBuf(2,bi,bj) |
166 |
ENDDO |
167 |
ENDDO |
168 |
|
169 |
C-- Process 0 fills-in its local data |
170 |
np1 = 1 |
171 |
DO bj=1,nSy |
172 |
DO bi=1,nSx |
173 |
biG = (mpi_myXGlobalLo(np1)-1)/sNx+bi |
174 |
bjG = (mpi_myYGlobalLo(np1)-1)/sNy+bj |
175 |
shareBufCS1_R8(bi,bj) = globalBuf(3,biG,bjG) |
176 |
ENDDO |
177 |
ENDDO |
178 |
|
179 |
C-- Process 0 sends result to all other processes |
180 |
DO npe = 1, numberOfProcs-1 |
181 |
C- fill local array with relevant portion of global array |
182 |
np1 = npe + 1 |
183 |
DO bj=1,nSy |
184 |
DO bi=1,nSx |
185 |
biG = (mpi_myXGlobalLo(np1)-1)/sNx+bi |
186 |
bjG = (mpi_myYGlobalLo(np1)-1)/sNy+bj |
187 |
loc1Buf(bi,bj) = globalBuf(3,biG,bjG) |
188 |
ENDDO |
189 |
ENDDO |
190 |
CALL MPI_SEND (loc1Buf, lbuf1, MPI_DOUBLE_PRECISION, |
191 |
& npe, itag, MPI_COMM_MODEL, ierr) |
192 |
|
193 |
ENDDO |
194 |
|
195 |
ENDIF |
196 |
|
197 |
ELSEIF (useCubedSphereExchange) THEN |
198 |
#else /* not USE_MPI */ |
199 |
IF (useCubedSphereExchange) THEN |
200 |
#endif /* ALLOW_USE_MPI */ |
201 |
|
202 |
C-- assume 1 tile / face, from bi=1 to 6, no MPI |
203 |
shareBufCS1_R8(1,1) = 0. |
204 |
bj = 1 |
205 |
DO bi = 1,nSx-1 |
206 |
nf = 1 + MOD(1+bi,2) |
207 |
shareBufCS1_R8(1+bi,bj) = shareBufCS1_R8(bi,bj) |
208 |
& + shareBufCS2_R8(nf,bi,bj) |
209 |
ENDDO |
210 |
C- fill in missing corner: 1 = North-West corner of face 1 |
211 |
C- 2 = South-East corner of face 2 |
212 |
bi = 1 |
213 |
psiLoc(1) = shareBufCS1_R8(bi,bj) + shareBufCS2_R8(2,bi,bj) |
214 |
bi = MIN(2,nSx) |
215 |
psiLoc(2) = shareBufCS1_R8(bi,bj) + shareBufCS2_R8(1,bi,bj) |
216 |
|
217 |
ELSE |
218 |
|
219 |
C-- Cumulate Sum over all tiles: |
220 |
shareBufCS1_R8(1,1) = 0. |
221 |
bj = 1 |
222 |
DO bi = 1,nSx-1 |
223 |
shareBufCS1_R8(1+bi,bj) = shareBufCS1_R8(bi,bj) |
224 |
& + shareBufCS2_R8(1,bi,bj) |
225 |
ENDDO |
226 |
DO bj = 1,nSy-1 |
227 |
DO bi = 1,nSx |
228 |
shareBufCS1_R8(bi,1+bj) = shareBufCS1_R8(bi,bj) |
229 |
& + shareBufCS2_R8(2,bi,bj) |
230 |
ENDDO |
231 |
ENDDO |
232 |
|
233 |
ENDIF |
234 |
|
235 |
_END_MASTER( myThid ) |
236 |
C-- Everyone wait for Master thread to be ready |
237 |
CALL BAR2( myThid ) |
238 |
|
239 |
C-- set result for every threads |
240 |
DO bj = myByLo(myThid), myByHi(myThid) |
241 |
DO bi = myBxLo(myThid), myBxHi(myThid) |
242 |
psiZ(bi,bj) = shareBufCS1_R8(bi,bj) |
243 |
ENDDO |
244 |
ENDDO |
245 |
|
246 |
RETURN |
247 |
#endif /* ALLOW_EXCH2 */ |
248 |
END |