1 |
C $Header: /u/gcmpack/MITgcm/eesupp/src/global_sum.F,v 1.12 2004/03/27 03:51:51 edhill Exp $ |
2 |
C $Name: $ |
3 |
|
4 |
C-- File global_sum.F: Routines that perform global sum on an array |
5 |
C of thread values. |
6 |
C Contents |
7 |
C o GLOBAL_SUM_R4 |
8 |
C o GLOBAL_SUM_R8 |
9 |
C o GLOBAL_SUM_INT |
10 |
|
11 |
#include "CPP_EEOPTIONS.h" |
12 |
|
13 |
CBOP |
14 |
C !ROUTINE: GLOBAL_SUM_R4 |
15 |
|
16 |
C !INTERFACE: |
17 |
SUBROUTINE GLOBAL_SUM_R4( |
18 |
U sumPhi, |
19 |
I myThid ) |
20 |
|
21 |
C !DESCRIPTION: |
22 |
C *==========================================================* |
23 |
C | SUBROUTINE GLOBAL\_SUM\_R4 |
24 |
C | o Handle sum for real*4 data. |
25 |
C *==========================================================* |
26 |
C | Perform sum an array of one value per thread and then |
27 |
C | sum result of all the processes. |
28 |
C | Notes: |
29 |
C | Within a process only one thread does the sum. |
30 |
C | The same thread also does the inter-process sum for |
31 |
C | example with MPI and then writes the result into a shared |
32 |
C | location. All threads wait until the sum is avaiailable. |
33 |
C *==========================================================* |
34 |
|
35 |
C !USES: |
36 |
IMPLICIT NONE |
37 |
|
38 |
C == Global data == |
39 |
#include "SIZE.h" |
40 |
#include "EEPARAMS.h" |
41 |
#include "EESUPPORT.h" |
42 |
#include "GLOBAL_SUM.h" |
43 |
|
44 |
C !INPUT/OUTPUT PARAMETERS: |
45 |
C == Routine arguments == |
46 |
C sumPhi - Result of sum. |
47 |
C myThid - My thread id. |
48 |
Real*4 sumPhi |
49 |
INTEGER myThid |
50 |
|
51 |
C !LOCAL VARIABLES: |
52 |
C == Local variables == |
53 |
C I - Loop counters |
54 |
C mpiRC - MPI return code |
55 |
INTEGER I |
56 |
Real*4 tmp |
57 |
#ifdef ALLOW_USE_MPI |
58 |
INTEGER mpiRC |
59 |
#endif /* ALLOW_USE_MPI */ |
60 |
CEOP |
61 |
|
62 |
C-- write local sum into array |
63 |
CALL BAR2( myThid ) |
64 |
phiGSRS(1,myThid) = sumPhi |
65 |
|
66 |
C-- Can not start until everyone is ready |
67 |
CALL BAR2( myThid ) |
68 |
|
69 |
C-- Sum within the process first |
70 |
_BEGIN_MASTER( myThid ) |
71 |
tmp = 0. |
72 |
DO I=1,nThreads |
73 |
tmp = tmp + phiGSRS(1,I) |
74 |
ENDDO |
75 |
sumPhi = tmp |
76 |
#ifdef ALLOW_USE_MPI |
77 |
#ifndef ALWAYS_USE_MPI |
78 |
IF ( usingMPI ) THEN |
79 |
#endif |
80 |
CALL MPI_Allreduce(tmp,sumPhi,1,MPI_REAL,MPI_SUM, |
81 |
& MPI_COMM_MODEL,mpiRC) |
82 |
#ifndef ALWAYS_USE_MPI |
83 |
ENDIF |
84 |
#endif |
85 |
#endif /* ALLOW_USE_MPI */ |
86 |
C-- Write solution to place where all threads can see it |
87 |
phiGSRS(1,1) = sumPhi |
88 |
|
89 |
_END_MASTER( myThid ) |
90 |
C-- |
91 |
CALL BAR2( myThid ) |
92 |
|
93 |
C-- set result for every process |
94 |
sumPhi = phiGSRS(1,1) |
95 |
CALL BAR2( myThid ) |
96 |
|
97 |
RETURN |
98 |
END |
99 |
|
100 |
|
101 |
CBOP |
102 |
C !ROUTINE: GLOBAL_SUM_R8 |
103 |
|
104 |
C !INTERFACE: |
105 |
SUBROUTINE GLOBAL_SUM_R8( |
106 |
U sumPhi, |
107 |
I myThid ) |
108 |
IMPLICIT NONE |
109 |
C !DESCRIPTION: |
110 |
C *==========================================================* |
111 |
C | SUBROUTINE GLOBAL\_SUM\_R8 |
112 |
C | o Handle sum for real*8 data. |
113 |
C *==========================================================* |
114 |
C | Perform sum an array of one value per thread and then |
115 |
C | sum result of all the processes. |
116 |
C | Notes: |
117 |
C | Within a process only one thread does the sum. |
118 |
C | The same thread also does the inter-process sum for |
119 |
C | example with MPI and then writes the result into a shared |
120 |
C | location. All threads wait until the sum is avaiailable. |
121 |
C *==========================================================* |
122 |
|
123 |
C !USES: |
124 |
C === Global data === |
125 |
#include "SIZE.h" |
126 |
#include "EEPARAMS.h" |
127 |
#include "EESUPPORT.h" |
128 |
#include "GLOBAL_SUM.h" |
129 |
|
130 |
C !INPUT/OUTPUT PARAMETERS: |
131 |
C === Routine arguments === |
132 |
C sumPhi - Result of sum. |
133 |
C myThid - My thread id. |
134 |
Real*8 sumPhi |
135 |
INTEGER myThid |
136 |
|
137 |
C !LOCAL VARIABLES: |
138 |
C === Local variables === |
139 |
C I - Loop counters |
140 |
C mpiRC - MPI return code |
141 |
INTEGER I |
142 |
Real*8 tmp |
143 |
#ifdef ALLOW_USE_MPI |
144 |
INTEGER mpiRC |
145 |
#endif /* ALLOW_USE_MPI */ |
146 |
CEOP |
147 |
|
148 |
CALL BAR2( myThid ) |
149 |
C-- write local sum into array |
150 |
phiGSRL(1,myThid) = sumPhi |
151 |
|
152 |
C-- Can not start until everyone is ready |
153 |
C CALL FOOL_THE_COMPILER( phiGSRL ) |
154 |
C CALL MS |
155 |
CALL BAR2( myThid ) |
156 |
C _BARRIER |
157 |
C _BARRIER |
158 |
C CALL FOOL_THE_COMPILER( phiGSRL ) |
159 |
|
160 |
C-- Sum within the process first |
161 |
_BEGIN_MASTER( myThid ) |
162 |
tmp = 0. _d 0 |
163 |
DO I=1,nThreads |
164 |
tmp = tmp + phiGSRL(1,I) |
165 |
ENDDO |
166 |
sumPhi = tmp |
167 |
#ifdef ALLOW_USE_MPI |
168 |
#ifndef ALWAYS_USE_MPI |
169 |
IF ( usingMPI ) THEN |
170 |
#endif |
171 |
CALL MPI_Allreduce(tmp,sumPhi,1,MPI_DOUBLE_PRECISION,MPI_SUM, |
172 |
& MPI_COMM_MODEL,mpiRC) |
173 |
#ifndef ALWAYS_USE_MPI |
174 |
ENDIF |
175 |
#endif |
176 |
#endif /* ALLOW_USE_MPI */ |
177 |
C-- Write solution to place where all threads can see it |
178 |
phiGSRL(1,1) = sumPhi |
179 |
_END_MASTER( myThid ) |
180 |
|
181 |
C-- Do not leave until we are sure that the sum is done |
182 |
C CALL FOOL_THE_COMPILER( phiGSRL ) |
183 |
C CALL MS |
184 |
C _BARRIER |
185 |
C _BARRIER |
186 |
CALL BAR2( myThid ) |
187 |
C CALL FOOL_THE_COMPILER( phiGSRL ) |
188 |
|
189 |
C-- set result for every process |
190 |
sumPhi = phiGSRL(1,1) |
191 |
CALL BAR2( myThid ) |
192 |
|
193 |
RETURN |
194 |
END |
195 |
|
196 |
CBOP |
197 |
C !ROUTINE: GLOBAL_SUM_INT |
198 |
C !INTERFACE: |
199 |
SUBROUTINE GLOBAL_SUM_INT( |
200 |
U sumPhi, |
201 |
I myThid ) |
202 |
IMPLICIT NONE |
203 |
C !DESCRIPTION: |
204 |
C *==========================================================* |
205 |
C | SUBROUTINE GLOBAL\_SUM\_INT |
206 |
C | o Handle sum for integer data. |
207 |
C *==========================================================* |
208 |
C | Perform sum an array of one value per thread and then |
209 |
C | sum result of all the processes. |
210 |
C | Notes: |
211 |
C | Within a process only one thread does the sum. |
212 |
C | The same thread also does the inter-process sum for |
213 |
C | example with MPI and then writes the result into a shared |
214 |
C | location. All threads wait until the sum is avaiailable. |
215 |
C *==========================================================* |
216 |
|
217 |
C !USES: |
218 |
C === Global data === |
219 |
#include "SIZE.h" |
220 |
#include "EEPARAMS.h" |
221 |
#include "EESUPPORT.h" |
222 |
#include "GLOBAL_SUM.h" |
223 |
|
224 |
C !INPUT/OUTPUT PARAMETERS: |
225 |
C === Routine arguments === |
226 |
C sumPhi - Result of sum. |
227 |
C myThid - My thread id. |
228 |
INTEGER sumPhi |
229 |
INTEGER myThid |
230 |
|
231 |
C !LOCAL VARIABLES: |
232 |
C === Local variables === |
233 |
C I - Loop counters |
234 |
C mpiRC - MPI return code |
235 |
INTEGER I |
236 |
INTEGER tmp |
237 |
#ifdef ALLOW_USE_MPI |
238 |
INTEGER mpiRC |
239 |
#endif /* ALLOW_USE_MPI */ |
240 |
CEOP |
241 |
|
242 |
_BARRIER |
243 |
C-- write local sum into array |
244 |
phiGSI(1,myThid) = sumPhi |
245 |
|
246 |
C-- Can not start until everyone is ready |
247 |
_BARRIER |
248 |
|
249 |
C-- Sum within the process first |
250 |
_BEGIN_MASTER( myThid ) |
251 |
tmp = 0 |
252 |
DO I=1,nThreads |
253 |
tmp = tmp + phiGSI(1,I) |
254 |
ENDDO |
255 |
sumPhi = tmp |
256 |
#ifdef ALLOW_USE_MPI |
257 |
#ifndef ALWAYS_USE_MPI |
258 |
IF ( usingMPI ) THEN |
259 |
#endif |
260 |
CALL MPI_Allreduce(tmp,sumPhi,1,MPI_INTEGER,MPI_SUM, |
261 |
& MPI_COMM_MODEL,mpiRC) |
262 |
#ifndef ALWAYS_USE_MPI |
263 |
ENDIF |
264 |
#endif |
265 |
#endif /* ALLOW_USE_MPI */ |
266 |
C-- Write solution to place where all threads can see it |
267 |
phiGSI(1,1) = sumPhi |
268 |
_END_MASTER( myThid ) |
269 |
|
270 |
C-- Do not leave until we are sure that the sum is done |
271 |
_BARRIER |
272 |
|
273 |
C-- set result for every process |
274 |
sumPhi = phiGSI(1,1) |
275 |
_BARRIER |
276 |
|
277 |
RETURN |
278 |
END |