1 |
C $Header$ |
C $Header$ |
2 |
|
C $Name$ |
3 |
|
|
4 |
C-- File global_sum.F: Routines that perform global sum on an array |
C-- File global_sum.F: Routines that perform global sum on an array |
5 |
C of thread values. |
C of thread values. |
8 |
C o global_sum_r8 |
C o global_sum_r8 |
9 |
#include "CPP_EEOPTIONS.h" |
#include "CPP_EEOPTIONS.h" |
10 |
|
|
11 |
CStartOfInterface |
CBOP |
12 |
|
C !ROUTINE: GLOBAL_SUM_R4 |
13 |
|
|
14 |
|
C !INTERFACE: |
15 |
SUBROUTINE GLOBAL_SUM_R4( |
SUBROUTINE GLOBAL_SUM_R4( |
16 |
I phi, |
U sumPhi, |
|
O sumPhi, |
|
17 |
I myThid ) |
I myThid ) |
18 |
C /==========================================================\ |
IMPLICIT NONE |
19 |
C | SUBROUTINE GLOBAL_SUM_R4 | |
C !DESCRIPTION: |
20 |
C | o Handle sum for real*4 data. | |
C *==========================================================* |
21 |
C |==========================================================| |
C | SUBROUTINE GLOBAL_SUM_R4 |
22 |
C | Perform sum an array of one value per thread and then | |
C | o Handle sum for real*4 data. |
23 |
C | sum result of all the processes. | |
C *==========================================================* |
24 |
C | Notes | |
C | Perform sum an array of one value per thread and then |
25 |
C | ===== | |
C | sum result of all the processes. |
26 |
C | Within a process only one thread does the sum, each | |
C | Notes |
27 |
C | thread is assumed to have already summed its local data. | |
C | ===== |
28 |
C | The same thread also does the inter-process sum for | |
C | Within a process only one thread does the sum, each |
29 |
C | example with MPI and then writes the result into a shared| |
C | thread is assumed to have already summed its local data. |
30 |
C | location. All threads wait until the sum is avaiailable. | |
C | The same thread also does the inter-process sum for |
31 |
C \==========================================================/ |
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 |
C == Global data == |
C == Global data == |
37 |
#include "SIZE.h" |
#include "SIZE.h" |
38 |
#include "EEPARAMS.h" |
#include "EEPARAMS.h" |
39 |
#include "EESUPPORT.h" |
#include "EESUPPORT.h" |
40 |
|
#include "GLOBAL_SUM.h" |
41 |
|
|
42 |
|
C !INPUT/OUTPUT PARAMETERS: |
43 |
C == Routine arguments == |
C == Routine arguments == |
|
C phi - Array to be summed. |
|
44 |
C sumPhi - Result of sum. |
C sumPhi - Result of sum. |
45 |
C myThid - My thread id. |
C myThid - My thread id. |
|
Real*4 phi(lShare4,MAX_NO_THREADS) |
|
46 |
Real*4 sumPhi |
Real*4 sumPhi |
47 |
INTEGER myThid |
INTEGER myThid |
|
CEndOfInterface |
|
48 |
|
|
49 |
|
C !LOCAL VARIABLES: |
50 |
C == Local variables == |
C == Local variables == |
51 |
C I - Loop counters |
C I - Loop counters |
52 |
C mpiRC - MPI return code |
C mpiRC - MPI return code |
55 |
#ifdef ALLOW_USE_MPI |
#ifdef ALLOW_USE_MPI |
56 |
INTEGER mpiRC |
INTEGER mpiRC |
57 |
#endif /* ALLOW_USE_MPI */ |
#endif /* ALLOW_USE_MPI */ |
58 |
|
CEOP |
59 |
|
|
60 |
C-- Can't start until everyone is ready |
C-- write local sum into array |
61 |
_BARRIER |
CALL BAR2( myThid ) |
62 |
|
phiGSRS(1,myThid) = sumPhi |
63 |
|
|
64 |
|
C-- Can not start until everyone is ready |
65 |
|
CALL BAR2( myThid ) |
66 |
|
|
67 |
C-- Sum within the process first |
C-- Sum within the process first |
68 |
_BEGIN_MASTER( myThid ) |
_BEGIN_MASTER( myThid ) |
69 |
tmp = 0. |
tmp = 0. |
70 |
DO I=1,nThreads |
DO I=1,nThreads |
71 |
tmp = tmp + phi(1,I) |
tmp = tmp + phiGSRS(1,I) |
72 |
ENDDO |
ENDDO |
73 |
sumPhi = tmp |
sumPhi = tmp |
74 |
#ifdef ALLOW_USE_MPI |
#ifdef ALLOW_USE_MPI |
81 |
ENDIF |
ENDIF |
82 |
#endif |
#endif |
83 |
#endif /* ALLOW_USE_MPI */ |
#endif /* ALLOW_USE_MPI */ |
84 |
phi(1,1) = sumPhi |
C-- Write solution to place where all threads can see it |
85 |
|
phiGSRS(1,1) = sumPhi |
86 |
|
|
87 |
_END_MASTER( myThid ) |
_END_MASTER( myThid ) |
88 |
C-- |
C-- |
89 |
_BARRIER |
CALL BAR2( myThid ) |
90 |
C |
|
91 |
|
C-- set result for every process |
92 |
|
sumPhi = phiGSRS(1,1) |
93 |
|
CALL BAR2( myThid ) |
94 |
|
|
95 |
RETURN |
RETURN |
96 |
END |
END |
97 |
|
|
98 |
|
|
99 |
CStartOfInterface |
CBOP |
100 |
|
C !ROUTINE: GLOBAL_SUM_R8 |
101 |
|
|
102 |
|
C !INTERFACE: |
103 |
SUBROUTINE GLOBAL_SUM_R8( |
SUBROUTINE GLOBAL_SUM_R8( |
104 |
I phi, |
U sumPhi, |
|
O sumPhi, |
|
105 |
I myThid ) |
I myThid ) |
106 |
C /==========================================================\ |
IMPLICIT NONE |
107 |
C | SUBROUTINE GLOBAL_SUM_R8 | |
C !DESCRIPTION: |
108 |
C | o Handle sum for real*8 data. | |
C *==========================================================* |
109 |
C |==========================================================| |
C | SUBROUTINE GLOBAL_SUM_R8 |
110 |
C | Perform sum an array of one value per thread and then | |
C | o Handle sum for real*8 data. |
111 |
C | sum result of all the processes. | |
C *==========================================================* |
112 |
C | Notes | |
C | Perform sum an array of one value per thread and then |
113 |
C | ===== | |
C | sum result of all the processes. |
114 |
C | Within a process only one thread does the sum, each | |
C | Notes |
115 |
C | thread is assumed to have already summed its local data. | |
C | ===== |
116 |
C | The same thread also does the inter-process sum for | |
C | Within a process only one thread does the sum, each |
117 |
C | example with MPI and then writes the result into a shared| |
C | thread is assumed to have already summed its local data. |
118 |
C | location. All threads wait until the sum is avaiailable. | |
C | The same thread also does the inter-process sum for |
119 |
C \==========================================================/ |
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 === |
C === Global data === |
125 |
#include "SIZE.h" |
#include "SIZE.h" |
126 |
#include "EEPARAMS.h" |
#include "EEPARAMS.h" |
127 |
#include "EESUPPORT.h" |
#include "EESUPPORT.h" |
128 |
|
#include "GLOBAL_SUM.h" |
129 |
|
|
130 |
|
C !INPUT/OUTPUT PARAMETERS: |
131 |
C === Routine arguments === |
C === Routine arguments === |
|
C phi - Array to be summed. |
|
132 |
C sumPhi - Result of sum. |
C sumPhi - Result of sum. |
133 |
C myThid - My thread id. |
C myThid - My thread id. |
|
Real*8 phi(lShare8,MAX_NO_THREADS) |
|
134 |
Real*8 sumPhi |
Real*8 sumPhi |
135 |
INTEGER myThid |
INTEGER myThid |
|
CEndOfInterface |
|
136 |
|
|
137 |
|
C !LOCAL VARIABLES: |
138 |
C === Local variables === |
C === Local variables === |
139 |
C I - Loop counters |
C I - Loop counters |
140 |
C mpiRC - MPI return code |
C mpiRC - MPI return code |
143 |
#ifdef ALLOW_USE_MPI |
#ifdef ALLOW_USE_MPI |
144 |
INTEGER mpiRC |
INTEGER mpiRC |
145 |
#endif /* ALLOW_USE_MPI */ |
#endif /* ALLOW_USE_MPI */ |
146 |
|
CEOP |
147 |
|
|
148 |
C-- Can't start until everyone is ready |
CALL BAR2( myThid ) |
149 |
_BARRIER |
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 |
C-- Sum within the process first |
161 |
_BEGIN_MASTER( myThid ) |
_BEGIN_MASTER( myThid ) |
162 |
tmp = 0. _d 0 |
tmp = 0. _d 0 |
163 |
DO I=1,nThreads |
DO I=1,nThreads |
164 |
tmp = tmp + phi(1,I) |
tmp = tmp + phiGSRL(1,I) |
165 |
ENDDO |
ENDDO |
166 |
sumPhi = tmp |
sumPhi = tmp |
167 |
#ifdef ALLOW_USE_MPI |
#ifdef ALLOW_USE_MPI |
175 |
#endif |
#endif |
176 |
#endif /* ALLOW_USE_MPI */ |
#endif /* ALLOW_USE_MPI */ |
177 |
C-- Write solution to place where all threads can see it |
C-- Write solution to place where all threads can see it |
178 |
phi(1,1) = sumPhi |
phiGSRL(1,1) = sumPhi |
179 |
_END_MASTER( myThid ) |
_END_MASTER( myThid ) |
180 |
|
|
181 |
C-- Don't leave until we are sure that the sum is done |
C-- Do not leave until we are sure that the sum is done |
182 |
_BARRIER |
C CALL FOOL_THE_COMPILER( phiGSRL ) |
183 |
C |
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 |
RETURN |
194 |
END |
END |
195 |
|
|
196 |
C $Id$ |
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 | ===== |
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 |
|
|
219 |
|
C !USES: |
220 |
|
C === Global data === |
221 |
|
#include "SIZE.h" |
222 |
|
#include "EEPARAMS.h" |
223 |
|
#include "EESUPPORT.h" |
224 |
|
#include "GLOBAL_SUM.h" |
225 |
|
|
226 |
|
C !INPUT/OUTPUT PARAMETERS: |
227 |
|
C === Routine arguments === |
228 |
|
C sumPhi - Result of sum. |
229 |
|
C myThid - My thread id. |
230 |
|
INTEGER sumPhi |
231 |
|
INTEGER myThid |
232 |
|
|
233 |
|
C !LOCAL VARIABLES: |
234 |
|
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 |
|
CEOP |
243 |
|
|
244 |
|
_BARRIER |
245 |
|
C-- write local sum into array |
246 |
|
phiGSI(1,myThid) = sumPhi |
247 |
|
|
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 |
|
tmp = tmp + phiGSI(1,I) |
256 |
|
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 |
|
& MPI_COMM_WORLD,mpiRC) |
264 |
|
#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 |
|
phiGSI(1,1) = sumPhi |
270 |
|
_END_MASTER( myThid ) |
271 |
|
|
272 |
|
C-- Do not leave until we are sure that the sum is done |
273 |
|
_BARRIER |
274 |
|
|
275 |
|
C-- set result for every process |
276 |
|
sumPhi = phiGSI(1,1) |
277 |
|
_BARRIER |
278 |
|
|
279 |
|
RETURN |
280 |
|
END |