1 | MODULE solpcg |
---|
2 | !!====================================================================== |
---|
3 | !! *** MODULE solfet |
---|
4 | !! Ocean solver : preconditionned conjugate gradient solver |
---|
5 | !!===================================================================== |
---|
6 | |
---|
7 | !!---------------------------------------------------------------------- |
---|
8 | !! sol_pcg : preconditionned conjugate gradient solver |
---|
9 | !!---------------------------------------------------------------------- |
---|
10 | !! * Modules used |
---|
11 | USE oce ! ocean dynamics and tracers variables |
---|
12 | USE dom_oce ! ocean space and time domain variables |
---|
13 | USE sol_oce ! ocean solver variables |
---|
14 | USE lib_mpp ! distributed memory computing |
---|
15 | USE lbclnk ! ocean lateral boundary conditions (or mpp link) |
---|
16 | |
---|
17 | IMPLICIT NONE |
---|
18 | PRIVATE |
---|
19 | |
---|
20 | !! * Routine accessibility |
---|
21 | PUBLIC sol_pcg ! ??? |
---|
22 | |
---|
23 | !! * Substitutions |
---|
24 | # include "vectopt_loop_substitute.h90" |
---|
25 | !!---------------------------------------------------------------------- |
---|
26 | |
---|
27 | CONTAINS |
---|
28 | |
---|
29 | SUBROUTINE sol_pcg( kindic ) |
---|
30 | !!---------------------------------------------------------------------- |
---|
31 | !! *** ROUTINE sol_pcg *** |
---|
32 | !! |
---|
33 | !! ** Purpose : Solve the ellipic equation for the barotropic stream |
---|
34 | !! function system (default option) or the transport divergence |
---|
35 | !! system ("key_dynspg_fsc") using a diagonal preconditionned |
---|
36 | !! conjugate gradient method. |
---|
37 | !! In the former case, the barotropic stream function trend has a |
---|
38 | !! zero boundary condition along all coastlines (i.e. continent |
---|
39 | !! as well as islands) while in the latter the boundary condition |
---|
40 | !! specification is not required. |
---|
41 | !! |
---|
42 | !! ** Method : Diagonal preconditionned conjugate gradient method. |
---|
43 | !! the algorithm is multitasked. (case of 5 points matrix) |
---|
44 | !! define pa = q^-1 * a |
---|
45 | !! pgcb = q^-1 * gcb |
---|
46 | !! < . ; . >_q = ( . )^t q ( . ) |
---|
47 | !! where q is the preconditioning matrix = diagonal matrix of the |
---|
48 | !! diagonal elements of a |
---|
49 | !! Initialization: |
---|
50 | !! x(o) = gcx |
---|
51 | !! r(o) = d(o) = pgcb - pa.x(o) |
---|
52 | !! rr(o)= < r(o) , r(o) >_q |
---|
53 | !! Iteration n : |
---|
54 | !! z(n) = pa.d(n) |
---|
55 | !! alp(n) = rr(n) / < z(n) , d(n) >_q |
---|
56 | !! x(n+1) = x(n) + alp(n) d(n) |
---|
57 | !! r(n+1) = r(n) - alp(n) z(n) |
---|
58 | !! rr(n+1)= < r(n+1) , r(n+1) >_q |
---|
59 | !! bet(n) = rr(n+1) / rr(n) |
---|
60 | !! r(n+1) = r(n+1) + bet(n+1) d(n) |
---|
61 | !! Convergence test : |
---|
62 | !! rr(n+1) / < gcb , gcb >_q =< epsr |
---|
63 | !! |
---|
64 | !! ** Action : - niter : solver number of iteration done |
---|
65 | !! - res : solver residu reached |
---|
66 | !! - gcx() : solution of the elliptic system |
---|
67 | !! |
---|
68 | !! References : |
---|
69 | !! Madec et al. 1988, Ocean Modelling, issue 78, 1-6. |
---|
70 | !! |
---|
71 | !! History : |
---|
72 | !! ! 90-10 (G. Madec) Original code |
---|
73 | !! ! 91-11 (G. Madec) |
---|
74 | !! ! 93-04 (M. Guyon) loops and suppress pointers |
---|
75 | !! ! 95-09 (M. Imbard, J. Escobar) mpp exchange |
---|
76 | !! ! 96-05 (G. Madec) merge sor and pcg formulations |
---|
77 | !! ! 96-11 (A. Weaver) correction to preconditioning |
---|
78 | !! 8.5 ! 02-08 (G. Madec) F90: Free form |
---|
79 | !!---------------------------------------------------------------------- |
---|
80 | !! * Arguments |
---|
81 | INTEGER, INTENT( inout ) :: kindic ! solver indicator, < 0 if the conver- |
---|
82 | ! ! gence is not reached: the model is |
---|
83 | ! ! stopped in step |
---|
84 | ! ! set to zero before the call of solpcg |
---|
85 | |
---|
86 | !! * Local declarations |
---|
87 | INTEGER :: ji, jj, jn ! dummy loop indices |
---|
88 | REAL(wp) :: zgcad ! temporary scalars |
---|
89 | !!---------------------------------------------------------------------- |
---|
90 | |
---|
91 | ! !================ |
---|
92 | DO jn = 1, nmax ! Iterative loop |
---|
93 | ! !================ |
---|
94 | |
---|
95 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, |
---|
96 | |
---|
97 | IF( jn == 1 ) THEN |
---|
98 | |
---|
99 | ! 1.0 Initialization of the algorithm |
---|
100 | ! ----------------------------------- |
---|
101 | |
---|
102 | #if defined key_dynspg_fsc |
---|
103 | # if defined key_mpp |
---|
104 | ! Mpp: export boundary values to neighbouring processors |
---|
105 | CALL lbc_lnk( gcx, 'S', 1. ) |
---|
106 | # else |
---|
107 | ! mono- or macro-tasking: W-point, >0, 2D array, no slab |
---|
108 | CALL lbc_lnk( gcx, 'T', 1. ) |
---|
109 | # endif |
---|
110 | #else |
---|
111 | # if defined key_mpp |
---|
112 | ! ... Mpp: export boundary values to neighbouring processors |
---|
113 | CALL lbc_lnk( gcx, 'G', 1. ) |
---|
114 | # else |
---|
115 | ! ... mono- or macro-tasking: F-point, >0, 2D array, no slab |
---|
116 | CALL lbc_lnk( gcx, 'F', 1. ) |
---|
117 | # endif |
---|
118 | #endif |
---|
119 | |
---|
120 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro ,,,,,,,,,,,,,,,,,,,,,,, |
---|
121 | |
---|
122 | ! gcr = gcb-a.gcx |
---|
123 | ! gcdes = gsr |
---|
124 | |
---|
125 | DO jj = 2, jpjm1 |
---|
126 | DO ji = fs_2, fs_jpim1 ! vector opt. |
---|
127 | zgcad = bmask(ji,jj)*( & |
---|
128 | gcb(ji,jj ) - gcx(ji ,jj ) & |
---|
129 | - gcp(ji,jj,1)*gcx(ji ,jj-1) & |
---|
130 | - gcp(ji,jj,2)*gcx(ji-1,jj ) & |
---|
131 | - gcp(ji,jj,3)*gcx(ji+1,jj ) & |
---|
132 | - gcp(ji,jj,4)*gcx(ji ,jj+1) ) |
---|
133 | gcr (ji,jj) = zgcad |
---|
134 | gcdes(ji,jj) = zgcad |
---|
135 | END DO |
---|
136 | END DO |
---|
137 | |
---|
138 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro ,,,,,,,,,,,,,,,,,,,,,,, |
---|
139 | |
---|
140 | rnorme = SUM( gcr(:,:) * gcdmat(:,:) * gcr(:,:) ) |
---|
141 | |
---|
142 | #if defined key_mpp |
---|
143 | ! Mpp: sum over all the global domain |
---|
144 | CALL mpp_sum( rnorme ) |
---|
145 | #endif |
---|
146 | rr = rnorme |
---|
147 | |
---|
148 | ENDIF |
---|
149 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro ,,,,,,,,,,,,,,,,,,,,,,, |
---|
150 | |
---|
151 | |
---|
152 | ! 1.1 Algorithm |
---|
153 | ! ------------- |
---|
154 | |
---|
155 | ! boundary condition on gcdes (only cyclic bc are required) |
---|
156 | #if defined key_dynspg_fsc |
---|
157 | # if defined key_mpp |
---|
158 | ! Mpp: export boundary values to neighbouring processors |
---|
159 | CALL lbc_lnk( gcdes, 'S', 1. ) |
---|
160 | # else |
---|
161 | ! mono- or macro-tasking: W-point, >0, 2D array, no slab |
---|
162 | CALL lbc_lnk( gcdes, 'T', 1. ) |
---|
163 | # endif |
---|
164 | #else |
---|
165 | # if defined key_mpp |
---|
166 | ! Mpp: export boundary values to neighbouring processors |
---|
167 | CALL lbc_lnk( gcdes, 'G', 1. ) |
---|
168 | # else |
---|
169 | ! mono- or macro-tasking: F-point, >0, 2D array, no slab |
---|
170 | CALL lbc_lnk( gcdes, 'F', 1. ) |
---|
171 | # endif |
---|
172 | #endif |
---|
173 | |
---|
174 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro if macrotasking,,,,,,,,,,,,,,,,,,,,,,, |
---|
175 | |
---|
176 | ! ... gccd = matrix . gcdes |
---|
177 | DO jj = 2, jpjm1 |
---|
178 | DO ji = fs_2, fs_jpim1 ! vector opt. |
---|
179 | gccd(ji,jj) = bmask(ji,jj)*( & |
---|
180 | gcdes(ji,jj) & |
---|
181 | +gcp(ji,jj,1)*gcdes(ji,jj-1)+gcp(ji,jj,2)*gcdes(ji-1,jj) & |
---|
182 | +gcp(ji,jj,4)*gcdes(ji,jj+1)+gcp(ji,jj,3)*gcdes(ji+1,jj) & |
---|
183 | ) |
---|
184 | END DO |
---|
185 | END DO |
---|
186 | |
---|
187 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro if macrotasking,,,,,,,,,,,,,,,,,,,,,,, |
---|
188 | |
---|
189 | ! alph = (gcr,gcr)/(gcdes,gccd) |
---|
190 | |
---|
191 | radd = SUM( gcdes(:,:) * gcdmat(:,:) * gccd(:,:) ) |
---|
192 | |
---|
193 | #if defined key_mpp |
---|
194 | ! Mpp: sum over all the global domain |
---|
195 | CALL mpp_sum( radd ) |
---|
196 | #endif |
---|
197 | alph = rr / radd |
---|
198 | |
---|
199 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro if macrotasking,,,,,,,,,,,,,,,,,,,,,,, |
---|
200 | |
---|
201 | ! gcx = gcx + alph * gcdes |
---|
202 | ! gcr = gcr - alph * gccd |
---|
203 | DO jj = 2, jpjm1 |
---|
204 | DO ji = fs_2, fs_jpim1 ! vector opt. |
---|
205 | gcx(ji,jj) = bmask(ji,jj) * ( gcx(ji,jj) + alph * gcdes(ji,jj) ) |
---|
206 | gcr(ji,jj) = bmask(ji,jj) * ( gcr(ji,jj) - alph * gccd (ji,jj) ) |
---|
207 | END DO |
---|
208 | END DO |
---|
209 | |
---|
210 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro if macrotasking,,,,,,,,,,,,,,,,,,,,,,, |
---|
211 | |
---|
212 | ! rnorme = (gcr,gcr) |
---|
213 | |
---|
214 | rnorme = SUM( gcr(:,:) * gcdmat(:,:) * gcr(:,:) ) |
---|
215 | |
---|
216 | #if defined key_mpp |
---|
217 | ! Mpp: sum over all the global domain |
---|
218 | CALL mpp_sum( rnorme ) |
---|
219 | #endif |
---|
220 | |
---|
221 | ! test of convergence |
---|
222 | IF ( rnorme < epsr .OR. jn == nmax ) THEN |
---|
223 | res = SQRT( rnorme ) |
---|
224 | niter = jn |
---|
225 | ncut = 999 |
---|
226 | ENDIF |
---|
227 | |
---|
228 | ! beta = (rk+1,rk+1)/(rk,rk) |
---|
229 | beta = rnorme / rr |
---|
230 | rr = rnorme |
---|
231 | |
---|
232 | !,,,,,,,,,,,,,,,,,,,,,,,,synchro if macrotasking,,,,,,,,,,,,,,,,,,,,,,, |
---|
233 | |
---|
234 | ! indicator of non-convergence or explosion |
---|
235 | IF( jn == nmax .OR. sqrt(epsr)/eps > 1.e+20 ) kindic = -2 |
---|
236 | IF( ncut == 999 ) GOTO 999 |
---|
237 | |
---|
238 | ! gcdes = gcr + beta * gcdes |
---|
239 | DO jj = 2, jpjm1 |
---|
240 | DO ji = fs_2, fs_jpim1 ! vector opt. |
---|
241 | gcdes(ji,jj) = bmask(ji,jj)*( gcr(ji,jj) + beta * gcdes(ji,jj) ) |
---|
242 | END DO |
---|
243 | END DO |
---|
244 | |
---|
245 | ! !================ |
---|
246 | END DO ! End Loop |
---|
247 | ! !================ |
---|
248 | |
---|
249 | 999 CONTINUE |
---|
250 | |
---|
251 | |
---|
252 | ! 2. Output in gcx with lateral b.c. applied |
---|
253 | ! ------------------------------------------ |
---|
254 | |
---|
255 | ! boundary conditions !!bug ??? |
---|
256 | #if defined key_mpp |
---|
257 | ! Mpp: export boundary values to neighbouring processors |
---|
258 | # if defined key_dynspg_fsc |
---|
259 | CALL lbc_lnk( gcx, 'S', 1. ) |
---|
260 | # else |
---|
261 | CALL lbc_lnk( gcx, 'G', 1. ) |
---|
262 | # endif |
---|
263 | #else |
---|
264 | IF ( nperio /= 0 ) THEN |
---|
265 | # if defined key_dynspg_fsc |
---|
266 | ! mono- or macro-tasking: W-point, >0, 2D array, no slab |
---|
267 | CALL lbc_lnk( gcx, 'T', 1. ) |
---|
268 | # else |
---|
269 | ! mono- or macro-tasking: F-point, >0, 2D array, no slab |
---|
270 | CALL lbc_lnk( gcx, 'F', 1. ) |
---|
271 | # endif |
---|
272 | ENDIF |
---|
273 | #endif |
---|
274 | |
---|
275 | END SUBROUTINE sol_pcg |
---|
276 | |
---|
277 | !!===================================================================== |
---|
278 | END MODULE solpcg |
---|