1 |
jahn |
1.1 |
C $Header$ |
2 |
|
|
C $Name$ |
3 |
|
|
|
4 |
|
|
C-- File global_sum_singlecpu.F: Routines that perform global sum |
5 |
|
|
C on a single CPU |
6 |
|
|
#include "CPP_EEOPTIONS.h" |
7 |
|
|
|
8 |
|
|
CBOP |
9 |
|
|
C !ROUTINE: GLOBAL_SUM_SINGLECPU_RL |
10 |
|
|
|
11 |
|
|
C !INTERFACE: |
12 |
|
|
SUBROUTINE GLOBAL_SUM_SINGLECPU_RL( |
13 |
|
|
I phiLocal, |
14 |
|
|
O sumPhi, |
15 |
|
|
I myThid ) |
16 |
|
|
IMPLICIT NONE |
17 |
|
|
C !DESCRIPTION: |
18 |
|
|
C *==========================================================* |
19 |
|
|
C | SUBROUTINE GLOBAL\_SUM\_SINGLECPU\_RL |
20 |
|
|
C | o Handle sum for _RL data. |
21 |
|
|
C *==========================================================* |
22 |
|
|
C | Global sum of 2d array |
23 |
|
|
C | independent of tiling as sum is performed on a single CPU |
24 |
|
|
C | sum is performed in REAL*8 |
25 |
|
|
C *==========================================================* |
26 |
|
|
|
27 |
|
|
C !USES: |
28 |
|
|
C == Global data == |
29 |
|
|
#include "SIZE.h" |
30 |
|
|
#include "EEPARAMS.h" |
31 |
|
|
#include "EESUPPORT.h" |
32 |
|
|
#include "GLOBAL_SUM.h" |
33 |
|
|
|
34 |
|
|
C !INPUT/OUTPUT PARAMETERS: |
35 |
|
|
C == Routine arguments == |
36 |
|
|
C phiLocal :: local input array without overlap regions. |
37 |
|
|
C sumPhi :: Result of sum. |
38 |
|
|
C myThid :: My thread id. |
39 |
|
|
_RL phiLocal(1:sNx,1:sNy,nSx,nSy) |
40 |
|
|
_RL sumPhi |
41 |
|
|
INTEGER myThid |
42 |
|
|
|
43 |
|
|
C !LOCAL VARIABLES: |
44 |
|
|
C == Local variables == |
45 |
|
|
C- type declaration of: sumAll, globalBuf : |
46 |
|
|
C sumAll needs to have the same length as MPI_DOUBLE_PRECISION |
47 |
|
|
INTEGER I,J |
48 |
|
|
Real*8 globalBuf(Nx,Ny) |
49 |
|
|
Real*8 sumAll |
50 |
|
|
#ifdef ALLOW_USE_MPI |
51 |
|
|
INTEGER npe |
52 |
|
|
INTEGER lbuff, idest, itag |
53 |
|
|
INTEGER istatus(MPI_STATUS_SIZE), ierr |
54 |
|
|
#endif /* ALLOW_USE_MPI */ |
55 |
|
|
CEOP |
56 |
|
|
|
57 |
|
|
C- note(jmc): do not see why we need this Barrier here + the one at the end ; |
58 |
|
|
C however a) both are present in the standard version of global_sum ; |
59 |
|
|
C b) with OpenMP on hugo, does not work if both are commented out |
60 |
|
|
CALL BAR2( myThid ) |
61 |
|
|
|
62 |
|
|
C-- Gather local arrays |
63 |
|
|
CALL GATHER_2D( globalBuf, phiLocal, myThid ) |
64 |
|
|
|
65 |
|
|
C-- Master thread does the communications and the global sum |
66 |
|
|
C-- Master thread cannot start until everyone is ready: |
67 |
|
|
CALL BAR2( myThid ) |
68 |
|
|
_BEGIN_MASTER( myThid ) |
69 |
|
|
|
70 |
|
|
#ifdef ALLOW_USE_MPI |
71 |
|
|
idest = 0 |
72 |
|
|
itag = 0 |
73 |
|
|
|
74 |
|
|
IF ( mpiMyId.EQ.0 ) THEN |
75 |
|
|
#endif |
76 |
|
|
|
77 |
|
|
C-- Process 0 sums the global array |
78 |
|
|
sumAll = 0.d0 |
79 |
|
|
DO J=1,Ny |
80 |
|
|
DO I=1,Nx |
81 |
|
|
sumAll = sumAll + globalBuf(I,J) |
82 |
|
|
ENDDO |
83 |
|
|
ENDDO |
84 |
|
|
|
85 |
|
|
#ifdef ALLOW_USE_MPI |
86 |
|
|
C-- Process 0 sends result to all other processes |
87 |
|
|
lbuff = 1 |
88 |
|
|
DO npe = 1, numberOfProcs-1 |
89 |
|
|
CALL MPI_SEND (sumAll, 1, MPI_DOUBLE_PRECISION, |
90 |
|
|
& npe, itag, MPI_COMM_MODEL, ierr) |
91 |
|
|
ENDDO |
92 |
|
|
|
93 |
|
|
ELSE |
94 |
|
|
|
95 |
|
|
C-- All proceses except 0 receive result from process 0 |
96 |
|
|
CALL MPI_RECV (sumAll, 1, MPI_DOUBLE_PRECISION, |
97 |
|
|
& idest, itag, MPI_COMM_MODEL, istatus, ierr) |
98 |
|
|
|
99 |
|
|
ENDIF |
100 |
|
|
#endif /* not ALLOW_USE_MPI */ |
101 |
|
|
|
102 |
|
|
C-- Write solution to shared buffer (all threads can see it) |
103 |
|
|
shareBufGSR8(1,1) = sumAll |
104 |
|
|
|
105 |
|
|
_END_MASTER( myThid ) |
106 |
|
|
C-- Everyone wait for Master thread to be ready |
107 |
|
|
CALL BAR2( myThid ) |
108 |
|
|
|
109 |
|
|
C-- set result for every threads |
110 |
|
|
sumPhi = shareBufGSR8(1,1) |
111 |
|
|
|
112 |
|
|
C- note(jmc): do not see why we need this Barrier here (see comment @ the top) |
113 |
|
|
CALL BAR2( myThid ) |
114 |
|
|
|
115 |
|
|
RETURN |
116 |
|
|
END |