1 |
cnh |
1.10 |
C $Header: /u/gcmpack/models/MITgcmUV/eesupp/src/global_sum.F,v 1.9 2001/05/29 14:01:36 adcroft Exp $ |
2 |
adcroft |
1.9 |
C $Name: $ |
3 |
cnh |
1.1 |
|
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 |
|
|
#include "CPP_EEOPTIONS.h" |
10 |
|
|
|
11 |
cnh |
1.10 |
CBOP |
12 |
|
|
C !ROUTINE: GLOBAL_SUM_R4 |
13 |
|
|
|
14 |
|
|
C !INTERFACE: |
15 |
cnh |
1.1 |
SUBROUTINE GLOBAL_SUM_R4( |
16 |
adcroft |
1.6 |
U sumPhi, |
17 |
cnh |
1.1 |
I myThid ) |
18 |
adcroft |
1.6 |
IMPLICIT NONE |
19 |
cnh |
1.10 |
C !DESCRIPTION: |
20 |
|
|
C *==========================================================* |
21 |
|
|
C | SUBROUTINE GLOBAL_SUM_R4 |
22 |
|
|
C | o Handle sum for real*4 data. |
23 |
|
|
C *==========================================================* |
24 |
|
|
C | Perform sum an array of one value per thread and then |
25 |
|
|
C | sum result of all the processes. |
26 |
|
|
C | Notes |
27 |
|
|
C | ===== |
28 |
|
|
C | Within a process only one thread does the sum, each |
29 |
|
|
C | thread is assumed to have already summed its local data. |
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 |
cnh |
1.1 |
|
35 |
cnh |
1.10 |
C !USES: |
36 |
cnh |
1.1 |
C == Global data == |
37 |
|
|
#include "SIZE.h" |
38 |
|
|
#include "EEPARAMS.h" |
39 |
|
|
#include "EESUPPORT.h" |
40 |
adcroft |
1.9 |
#include "GLOBAL_SUM.h" |
41 |
cnh |
1.1 |
|
42 |
cnh |
1.10 |
C !INPUT/OUTPUT PARAMETERS: |
43 |
cnh |
1.1 |
C == Routine arguments == |
44 |
|
|
C sumPhi - Result of sum. |
45 |
|
|
C myThid - My thread id. |
46 |
|
|
Real*4 sumPhi |
47 |
|
|
INTEGER myThid |
48 |
|
|
|
49 |
cnh |
1.10 |
C !LOCAL VARIABLES: |
50 |
cnh |
1.1 |
C == Local variables == |
51 |
|
|
C I - Loop counters |
52 |
|
|
C mpiRC - MPI return code |
53 |
|
|
INTEGER I |
54 |
|
|
Real*4 tmp |
55 |
|
|
#ifdef ALLOW_USE_MPI |
56 |
|
|
INTEGER mpiRC |
57 |
|
|
#endif /* ALLOW_USE_MPI */ |
58 |
cnh |
1.10 |
CEOP |
59 |
cnh |
1.1 |
|
60 |
adcroft |
1.6 |
C-- write local sum into array |
61 |
adcroft |
1.9 |
CALL BAR2( myThid ) |
62 |
|
|
phiGSRS(1,myThid) = sumPhi |
63 |
adcroft |
1.6 |
|
64 |
cnh |
1.4 |
C-- Can not start until everyone is ready |
65 |
adcroft |
1.9 |
CALL BAR2( myThid ) |
66 |
cnh |
1.1 |
|
67 |
|
|
C-- Sum within the process first |
68 |
|
|
_BEGIN_MASTER( myThid ) |
69 |
|
|
tmp = 0. |
70 |
|
|
DO I=1,nThreads |
71 |
adcroft |
1.9 |
tmp = tmp + phiGSRS(1,I) |
72 |
cnh |
1.1 |
ENDDO |
73 |
|
|
sumPhi = tmp |
74 |
|
|
#ifdef ALLOW_USE_MPI |
75 |
|
|
#ifndef ALWAYS_USE_MPI |
76 |
|
|
IF ( usingMPI ) THEN |
77 |
|
|
#endif |
78 |
|
|
CALL MPI_Allreduce(tmp,sumPhi,1,MPI_REAL,MPI_SUM, |
79 |
adcroft |
1.6 |
& MPI_COMM_WORLD,mpiRC) |
80 |
cnh |
1.1 |
#ifndef ALWAYS_USE_MPI |
81 |
|
|
ENDIF |
82 |
|
|
#endif |
83 |
|
|
#endif /* ALLOW_USE_MPI */ |
84 |
adcroft |
1.6 |
C-- Write solution to place where all threads can see it |
85 |
adcroft |
1.9 |
phiGSRS(1,1) = sumPhi |
86 |
adcroft |
1.6 |
|
87 |
cnh |
1.1 |
_END_MASTER( myThid ) |
88 |
|
|
C-- |
89 |
adcroft |
1.9 |
CALL BAR2( myThid ) |
90 |
adcroft |
1.6 |
|
91 |
|
|
C-- set result for every process |
92 |
adcroft |
1.9 |
sumPhi = phiGSRS(1,1) |
93 |
|
|
CALL BAR2( myThid ) |
94 |
adcroft |
1.6 |
|
95 |
cnh |
1.1 |
RETURN |
96 |
|
|
END |
97 |
|
|
|
98 |
|
|
|
99 |
cnh |
1.10 |
CBOP |
100 |
|
|
C !ROUTINE: GLOBAL_SUM_R8 |
101 |
|
|
|
102 |
|
|
C !INTERFACE: |
103 |
cnh |
1.1 |
SUBROUTINE GLOBAL_SUM_R8( |
104 |
adcroft |
1.6 |
U sumPhi, |
105 |
cnh |
1.1 |
I myThid ) |
106 |
cnh |
1.10 |
IMPLICIT NONE |
107 |
|
|
C !DESCRIPTION: |
108 |
|
|
C *==========================================================* |
109 |
|
|
C | SUBROUTINE GLOBAL_SUM_R8 |
110 |
|
|
C | o Handle sum for real*8 data. |
111 |
|
|
C *==========================================================* |
112 |
|
|
C | Perform sum an array of one value per thread and then |
113 |
|
|
C | sum result of all the processes. |
114 |
|
|
C | Notes |
115 |
|
|
C | ===== |
116 |
|
|
C | Within a process only one thread does the sum, each |
117 |
|
|
C | thread is assumed to have already summed its local data. |
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 |
cnh |
1.1 |
|
123 |
cnh |
1.10 |
C !USES: |
124 |
cnh |
1.1 |
C === Global data === |
125 |
|
|
#include "SIZE.h" |
126 |
|
|
#include "EEPARAMS.h" |
127 |
|
|
#include "EESUPPORT.h" |
128 |
adcroft |
1.9 |
#include "GLOBAL_SUM.h" |
129 |
cnh |
1.1 |
|
130 |
cnh |
1.10 |
C !INPUT/OUTPUT PARAMETERS: |
131 |
cnh |
1.1 |
C === Routine arguments === |
132 |
|
|
C sumPhi - Result of sum. |
133 |
|
|
C myThid - My thread id. |
134 |
|
|
Real*8 sumPhi |
135 |
|
|
INTEGER myThid |
136 |
|
|
|
137 |
cnh |
1.10 |
C !LOCAL VARIABLES: |
138 |
cnh |
1.1 |
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 |
cnh |
1.10 |
CEOP |
147 |
cnh |
1.1 |
|
148 |
adcroft |
1.9 |
CALL BAR2( myThid ) |
149 |
adcroft |
1.6 |
C-- write local sum into array |
150 |
adcroft |
1.9 |
phiGSRL(1,myThid) = sumPhi |
151 |
adcroft |
1.6 |
|
152 |
cnh |
1.4 |
C-- Can not start until everyone is ready |
153 |
adcroft |
1.9 |
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 |
cnh |
1.1 |
|
160 |
|
|
C-- Sum within the process first |
161 |
|
|
_BEGIN_MASTER( myThid ) |
162 |
|
|
tmp = 0. _d 0 |
163 |
|
|
DO I=1,nThreads |
164 |
adcroft |
1.9 |
tmp = tmp + phiGSRL(1,I) |
165 |
cnh |
1.1 |
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 |
heimbach |
1.7 |
& MPI_COMM_WORLD,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 |
adcroft |
1.9 |
phiGSRL(1,1) = sumPhi |
179 |
heimbach |
1.7 |
_END_MASTER( myThid ) |
180 |
|
|
|
181 |
|
|
C-- Do not leave until we are sure that the sum is done |
182 |
adcroft |
1.9 |
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 |
heimbach |
1.7 |
|
189 |
|
|
C-- set result for every process |
190 |
adcroft |
1.9 |
sumPhi = phiGSRL(1,1) |
191 |
|
|
CALL BAR2( myThid ) |
192 |
heimbach |
1.7 |
|
193 |
|
|
RETURN |
194 |
|
|
END |
195 |
cnh |
1.10 |
|
196 |
|
|
CBOP |
197 |
|
|
C !ROUTINE: GLOBAL_SUM_INT |
198 |
|
|
C !INTERFACE: |
199 |
heimbach |
1.7 |
SUBROUTINE GLOBAL_SUM_INT( |
200 |
|
|
U sumPhi, |
201 |
|
|
I myThid ) |
202 |
|
|
IMPLICIT NONE |
203 |
cnh |
1.10 |
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 | ===== |
212 |
|
|
C | Within a process only one thread does the sum, each |
213 |
|
|
C | thread is assumed to have already summed its local data. |
214 |
|
|
C | The same thread also does the inter-process sum for |
215 |
|
|
C | example with MPI and then writes the result into a shared |
216 |
|
|
C | location. All threads wait until the sum is avaiailable. |
217 |
|
|
C *==========================================================* |
218 |
heimbach |
1.7 |
|
219 |
cnh |
1.10 |
C !USES: |
220 |
heimbach |
1.7 |
C === Global data === |
221 |
|
|
#include "SIZE.h" |
222 |
|
|
#include "EEPARAMS.h" |
223 |
|
|
#include "EESUPPORT.h" |
224 |
adcroft |
1.9 |
#include "GLOBAL_SUM.h" |
225 |
heimbach |
1.7 |
|
226 |
cnh |
1.10 |
C !INPUT/OUTPUT PARAMETERS: |
227 |
heimbach |
1.7 |
C === Routine arguments === |
228 |
|
|
C sumPhi - Result of sum. |
229 |
|
|
C myThid - My thread id. |
230 |
|
|
INTEGER sumPhi |
231 |
|
|
INTEGER myThid |
232 |
|
|
|
233 |
cnh |
1.10 |
C !LOCAL VARIABLES: |
234 |
heimbach |
1.7 |
C === Local variables === |
235 |
|
|
C I - Loop counters |
236 |
|
|
C mpiRC - MPI return code |
237 |
|
|
INTEGER I |
238 |
|
|
INTEGER tmp |
239 |
|
|
#ifdef ALLOW_USE_MPI |
240 |
|
|
INTEGER mpiRC |
241 |
|
|
#endif /* ALLOW_USE_MPI */ |
242 |
cnh |
1.10 |
CEOP |
243 |
heimbach |
1.7 |
|
244 |
adcroft |
1.9 |
_BARRIER |
245 |
heimbach |
1.7 |
C-- write local sum into array |
246 |
adcroft |
1.9 |
phiGSI(1,myThid) = sumPhi |
247 |
heimbach |
1.7 |
|
248 |
|
|
C-- Can not start until everyone is ready |
249 |
|
|
_BARRIER |
250 |
|
|
|
251 |
|
|
C-- Sum within the process first |
252 |
|
|
_BEGIN_MASTER( myThid ) |
253 |
|
|
tmp = 0. _d 0 |
254 |
|
|
DO I=1,nThreads |
255 |
adcroft |
1.9 |
tmp = tmp + phiGSI(1,I) |
256 |
heimbach |
1.7 |
ENDDO |
257 |
|
|
sumPhi = tmp |
258 |
|
|
#ifdef ALLOW_USE_MPI |
259 |
|
|
#ifndef ALWAYS_USE_MPI |
260 |
|
|
IF ( usingMPI ) THEN |
261 |
|
|
#endif |
262 |
|
|
CALL MPI_Allreduce(tmp,sumPhi,1,MPI_INTEGER,MPI_SUM, |
263 |
adcroft |
1.6 |
& MPI_COMM_WORLD,mpiRC) |
264 |
cnh |
1.1 |
#ifndef ALWAYS_USE_MPI |
265 |
|
|
ENDIF |
266 |
|
|
#endif |
267 |
|
|
#endif /* ALLOW_USE_MPI */ |
268 |
|
|
C-- Write solution to place where all threads can see it |
269 |
adcroft |
1.9 |
phiGSI(1,1) = sumPhi |
270 |
cnh |
1.1 |
_END_MASTER( myThid ) |
271 |
|
|
|
272 |
cnh |
1.4 |
C-- Do not leave until we are sure that the sum is done |
273 |
cnh |
1.1 |
_BARRIER |
274 |
adcroft |
1.6 |
|
275 |
|
|
C-- set result for every process |
276 |
adcroft |
1.9 |
sumPhi = phiGSI(1,1) |
277 |
|
|
_BARRIER |
278 |
adcroft |
1.6 |
|
279 |
cnh |
1.1 |
RETURN |
280 |
|
|
END |