| 1 |
|
| 2 |
|
| 3 |
subroutine simul( |
| 4 |
I indic, |
| 5 |
I isforward, |
| 6 |
I mylen, |
| 7 |
O xx, |
| 8 |
O objf, |
| 9 |
O adxx |
| 10 |
& ) |
| 11 |
|
| 12 |
c ================================================================== |
| 13 |
c SUBROUTINE simul |
| 14 |
c ================================================================== |
| 15 |
c |
| 16 |
c o This routine is called by the large-scale optimization lsopt. |
| 17 |
c |
| 18 |
c Input : indic - Parameter for optimcycle |
| 19 |
c nmax - Number of control variables. |
| 20 |
c |
| 21 |
c Output : xx - Array of control variables. |
| 22 |
c objf - Value of objective function. |
| 23 |
c adxx - Gradients of objective function with respect |
| 24 |
c to the control variables. |
| 25 |
c |
| 26 |
c |
| 27 |
c started: Christian Eckert eckert@mit.edu 15-Feb-2000 |
| 28 |
c |
| 29 |
c changed: Christian Eckert eckert@mit.edu 10-Mar-2000 |
| 30 |
c |
| 31 |
c - Added ECCO layout. |
| 32 |
c |
| 33 |
c changed: Patrick Heimbach heimbach@mit.edu 19-Jun-2000 |
| 34 |
c - finished, revised and debugged |
| 35 |
c |
| 36 |
c ================================================================== |
| 37 |
c SUBROUTINE simul |
| 38 |
c ================================================================== |
| 39 |
|
| 40 |
implicit none |
| 41 |
|
| 42 |
c == global variables == |
| 43 |
|
| 44 |
#include "EEPARAMS.h" |
| 45 |
#include "SIZE.h" |
| 46 |
#include "ctrl.h" |
| 47 |
include 'mpif.h' |
| 48 |
|
| 49 |
c == routine arguments == |
| 50 |
|
| 51 |
logical isforward |
| 52 |
integer indic |
| 53 |
integer mylen |
| 54 |
_RL xx(mylen) |
| 55 |
_RL objf |
| 56 |
_RL adxx(mylen) |
| 57 |
integer nmax |
| 58 |
parameter( nmax = MAX_INDEPEND ) |
| 59 |
|
| 60 |
c == local variables == |
| 61 |
|
| 62 |
integer i |
| 63 |
_RL adobjf |
| 64 |
_RS,allocatable:: vv(:) |
| 65 |
_RS,allocatable:: tempvv(:) |
| 66 |
|
| 67 |
logical lheaderonly |
| 68 |
integer pidlen,myindx(2) |
| 69 |
integer status(MPI_STATUS_SIZE),ierr |
| 70 |
common /mpi_parm/myid, nprocs,mystart,myend |
| 71 |
integer myid,nprocs,mystart,myend |
| 72 |
|
| 73 |
c == end of interface == |
| 74 |
|
| 75 |
if(myid .eq. 0) |
| 76 |
& print *, 'pathei-lsopt in simul' |
| 77 |
|
| 78 |
c-- Call the combined modified forward model and the adjoint model. |
| 79 |
do i = 1,mylen |
| 80 |
xx(i) = 0.D0 |
| 81 |
adxx(i) = 0.D0 |
| 82 |
enddo |
| 83 |
|
| 84 |
adobjf = 1. |
| 85 |
c |
| 86 |
lheaderonly = .false. |
| 87 |
|
| 88 |
if (myid .eq. 0) then |
| 89 |
|
| 90 |
print *, 'pathei-lsopt vor optim_readdata' |
| 91 |
|
| 92 |
allocate(vv(nmax)) |
| 93 |
call optim_readdata(indic, nmax, ctrlname, lheaderonly, objf, vv) |
| 94 |
endif |
| 95 |
|
| 96 |
allocate(tempvv(mylen)) |
| 97 |
tempvv = 0. |
| 98 |
|
| 99 |
! As a master, sent out vectors of length pidlen to myid |
| 100 |
if(myid.eq.0) then |
| 101 |
do i=1,nprocs-1 |
| 102 |
call MPI_RECV(myindx,2,MPI_INTEGER,i,10,MPI_COMM_WORLD, |
| 103 |
& status,ierr) |
| 104 |
pidlen = myindx(2)-myindx(1)+1 |
| 105 |
call MPI_SEND(vv(myindx(1):myindx(2)),pidlen, |
| 106 |
& MPI_FLOAT,i,11,MPI_COMM_WORLD,ierr) |
| 107 |
enddo |
| 108 |
! |
| 109 |
xx(1:mylen) = vv(mystart:myend) |
| 110 |
! |
| 111 |
else |
| 112 |
! |
| 113 |
call MPI_SEND((/mystart,myend/),2,MPI_INTEGER,0,10, |
| 114 |
& MPI_COMM_WORLD,ierr) |
| 115 |
call MPI_RECV(tempvv,myend-mystart+1,MPI_FLOAT,0,11, |
| 116 |
& MPI_COMM_WORLD,status,ierr) |
| 117 |
! |
| 118 |
xx(1:mylen) = tempvv |
| 119 |
endif |
| 120 |
|
| 121 |
|
| 122 |
if (.not. isforward) then |
| 123 |
|
| 124 |
if (myid .eq. 0 ) then |
| 125 |
vv = 0. |
| 126 |
call optim_readdata(indic, nmax, costname, lheaderonly, objf, vv ) |
| 127 |
endif |
| 128 |
tempvv = 0. |
| 129 |
|
| 130 |
if(myid.eq.0) then |
| 131 |
do i=1,nprocs-1 |
| 132 |
call MPI_RECV(myindx,2,MPI_INTEGER,i,10,MPI_COMM_WORLD, |
| 133 |
& status,ierr) |
| 134 |
pidlen = myindx(2)-myindx(1)+1 |
| 135 |
call MPI_SEND(vv(myindx(1):myindx(2)),pidlen, |
| 136 |
& MPI_FLOAT,i,11,MPI_COMM_WORLD,ierr) |
| 137 |
enddo |
| 138 |
! |
| 139 |
adxx(1:mylen) = vv(mystart:myend) |
| 140 |
! |
| 141 |
else |
| 142 |
! |
| 143 |
call MPI_SEND((/mystart,myend/),2,MPI_INTEGER,0,10, |
| 144 |
& MPI_COMM_WORLD,ierr) |
| 145 |
call MPI_RECV(tempvv,myend-mystart+1,MPI_FLOAT,0,11, |
| 146 |
& MPI_COMM_WORLD,status,ierr) |
| 147 |
! |
| 148 |
adxx(1:mylen) = tempvv |
| 149 |
endif |
| 150 |
|
| 151 |
endif !end isforward |
| 152 |
|
| 153 |
call MPI_BCAST(objf,1,MPI_DOUBLE_PRECISION,0,MPI_COMM_WORLD,ierr) |
| 154 |
|
| 155 |
if(myid .eq. 0) then |
| 156 |
cph( |
| 157 |
print *, ' leaving simul with nn, objf = ', nmax, objf |
| 158 |
print *, ' leaving simul with xx = ', xx(1), xx(2) |
| 159 |
print *, ' leaving simul with adxx = ', adxx(1), adxx(2) |
| 160 |
|
| 161 |
deallocate(vv) |
| 162 |
endif |
| 163 |
deallocate(tempvv) |
| 164 |
|
| 165 |
do i=1,mylen |
| 166 |
c if (xx(i).EQ.'NaN') then |
| 167 |
c print *, 'pathei - out: i = ', i |
| 168 |
c end if |
| 169 |
end do |
| 170 |
cph) |
| 171 |
|
| 172 |
return |
| 173 |
end |
Parent Directory
| 
Revision Log
| 
Revision Graph