1 |
SUBROUTINE CG2D |
2 |
C /==========================================================\ |
3 |
C | SUBROUTINE CG2D | |
4 |
C | o Two-dimensional grid problem conjugate-gradient | |
5 |
C | inverter (with preconditioner). | |
6 |
C |==========================================================| |
7 |
C | Con. grad is an iterative procedure for solving Ax = b. | |
8 |
C | It requires the A be symmetric. | |
9 |
C | This implementation assumes A is a five-diagonal | |
10 |
C | matrix of the form that arises in the discrete | |
11 |
C | representation of the del^2 operator in a | |
12 |
C | two-dimensional space. | |
13 |
C | Notes: | |
14 |
C | ====== | |
15 |
C | This implementation can support shared-memory | |
16 |
C | multi-threaded execution. In order to do this COMMON | |
17 |
C | blocks are used for many of the arrays - even ones that | |
18 |
C | are only used for intermedaite results. This design is | |
19 |
C | OK if you want to all the threads to collaborate on | |
20 |
C | solving the same problem. On the other hand if you want | |
21 |
C | the threads to solve several different problems | |
22 |
C | concurrently this implementation will not work. | |
23 |
C \==========================================================/ |
24 |
IMPLICIT NONE |
25 |
|
26 |
C === Global data === |
27 |
#include "SIZE.h" |
28 |
#include "EEPARAMS.h" |
29 |
#include "PARAMS.h" |
30 |
#include "CG2D.h" |
31 |
|
32 |
C === Routine arguments === |
33 |
C myThid - Thread on which I am working. |
34 |
INTEGER myThid |
35 |
|
36 |
C === Local variables ==== |
37 |
C actualIts - Number of iterations taken |
38 |
C actualResidual - residual |
39 |
C bi - Block index in X and Y. |
40 |
C bj |
41 |
C etaN - Used in computing search directions |
42 |
C etaNM1 suffix N and NM1 denote current and |
43 |
C beta previous iterations respectively. |
44 |
C alpha |
45 |
C sumRHS - Sum of right-hand-side. Sometimes this is a |
46 |
C useful debuggin/trouble shooting diagnostic. |
47 |
C For neumann problems sumRHS needs to be ~0. |
48 |
C or they converge at a non-zero residual. |
49 |
C err - Measure of residual of Ax - b, usually the norm. |
50 |
C I, J, N - Loop counters ( N counts CG iterations ) |
51 |
INTEGER actualIts |
52 |
REAL actualResidual |
53 |
INTEGER bi, bj |
54 |
INTEGER I, J, N |
55 |
REAL err |
56 |
REAL errSum |
57 |
REAL etaN |
58 |
REAL etaNM1 |
59 |
REAL etaNSum |
60 |
REAL beta |
61 |
REAL alpha |
62 |
REAL alphaSum |
63 |
REAL sumRHS |
64 |
REAL temp |
65 |
|
66 |
C-- Initialise inverter |
67 |
etaNM1 = 1. D0 |
68 |
|
69 |
C-- Initial residual calculation |
70 |
err = 0. _d 0 |
71 |
sumRHS = 0. _d 0 |
72 |
DO J=1,sNy |
73 |
DO I=1,sNx |
74 |
cg2d_s(I,J) = 0. |
75 |
cg2d_r(I,J) = cg2d_b(I,J) - |
76 |
& ( aW2d(I ,J )*cg2d_x(I-1,J )+aW2d(I+1,J )*cg2d_x(I+1,J ) |
77 |
& +aS2d(I ,J )*cg2d_x(I ,J-1)+aS2d(I ,J+1)*cg2d_x(I ,J+1) |
78 |
& -aW2d(I ,J )*cg2d_x(I ,J )-aW2d(I+1,J )*cg2d_x(I ,J ) |
79 |
& -aS2d(I ,J )*cg2d_x(I ,J )-aS2d(I ,J+1)*cg2d_x(I ,J ) |
80 |
& ) |
81 |
err = err + cg2d_r(I,J)*cg2d_r(I,J) |
82 |
sumRHS = sumRHS + cg2d_b(I,J) |
83 |
ENDDO |
84 |
ENDDO |
85 |
CALL EXCH_XY_R8( cg2d_r ) |
86 |
CALL EXCH_XY_R8( cg2d_s ) |
87 |
CALL GSUM_R8( temp, err ) |
88 |
err = temp |
89 |
CALL GSUM_R8( temp, sumRHS ) |
90 |
sumRHS = temp |
91 |
|
92 |
actualIts = 0 |
93 |
actualResidual = SQRT(err) |
94 |
WRITE(6,*) ' CG2D iters, err = ', actualIts, actualResidual |
95 |
IF ( actualResidual .EQ. 0. ) STOP 'ABNORMAL END: RESIDUAL 0' |
96 |
|
97 |
C >>>>>>>>>>>>>>> BEGIN SOLVER <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< |
98 |
DO 10 N=1, cg2dMaxIters |
99 |
C-- Solve preconditioning equation and update |
100 |
C-- conjugate direction vector "s". |
101 |
etaN = 0. _d 0 |
102 |
DO J=1,sNy |
103 |
DO I=1,sNx |
104 |
cg2d_q(I,J) = |
105 |
& pW(I ,J )*cg2d_r(I-1,J )+pW(I+1,J )*cg2d_r(I+1,J ) |
106 |
& +pS(I ,J )*cg2d_r(I ,J-1)+pS(I ,J+1)*cg2d_r(I ,J+1) |
107 |
& +pC(I ,J )*cg2d_r(I ,J ) |
108 |
etaN = etaN+cg2d_q(I,J)*cg2d_r(I,J) |
109 |
ENDDO |
110 |
ENDDO |
111 |
|
112 |
CALL GSUM_R8( temp, etaN ) |
113 |
etaN = temp |
114 |
beta = etaN/etaNM1 |
115 |
etaNM1 = etaN |
116 |
|
117 |
DO J=1,sNy |
118 |
DO I=1,sNx |
119 |
cg2d_s(I,J) = cg2d_q(I,J) + beta*cg2d_s(I,J) |
120 |
ENDDO |
121 |
ENDDO |
122 |
|
123 |
C-- Do exchanges that require messages i.e. between |
124 |
C-- processes. |
125 |
CALL EXCH_XY_R8( cg2d_s ) |
126 |
|
127 |
C-- Ten extra exchanges |
128 |
#ifdef TEN_EXTRA_EXCHS |
129 |
CALL EXCH_XY_R8( cg2d_s ) |
130 |
CALL EXCH_XY_R8( cg2d_s ) |
131 |
CALL EXCH_XY_R8( cg2d_s ) |
132 |
CALL EXCH_XY_R8( cg2d_s ) |
133 |
CALL EXCH_XY_R8( cg2d_s ) |
134 |
CALL EXCH_XY_R8( cg2d_s ) |
135 |
CALL EXCH_XY_R8( cg2d_s ) |
136 |
CALL EXCH_XY_R8( cg2d_s ) |
137 |
CALL EXCH_XY_R8( cg2d_s ) |
138 |
CALL EXCH_XY_R8( cg2d_s ) |
139 |
#endif |
140 |
|
141 |
C== Evaluate laplace operator on conjugate gradient vector |
142 |
C== q = A.s |
143 |
alpha = 0. _d 0 |
144 |
DO J=1,sNy |
145 |
DO I=1,sNx |
146 |
cg2d_q(I,J) = |
147 |
& aW2d(I ,J )*cg2d_s(I-1,J )+aW2d(I+1,J )*cg2d_s(I+1,J ) |
148 |
& +aS2d(I ,J )*cg2d_s(I ,J-1)+aS2d(I ,J+1)*cg2d_s(I ,J+1) |
149 |
& -aW2d(I ,J )*cg2d_s(I ,J )-aW2d(I+1,J )*cg2d_s(I ,J ) |
150 |
& -aS2d(I ,J )*cg2d_s(I ,J )-aS2d(I ,J+1)*cg2d_s(I ,J ) |
151 |
alpha = alpha+cg2d_s(I,J)*cg2d_q(I,J) |
152 |
ENDDO |
153 |
ENDDO |
154 |
CALL GSUM_R8( temp, alpha ) |
155 |
|
156 |
#ifdef HUNDRED_EXTRA_SUMS |
157 |
C-- Hundred extra global sums |
158 |
CALL GSUM_R8( temp, alpha ) |
159 |
CALL GSUM_R8( temp, alpha ) |
160 |
CALL GSUM_R8( temp, alpha ) |
161 |
CALL GSUM_R8( temp, alpha ) |
162 |
CALL GSUM_R8( temp, alpha ) |
163 |
CALL GSUM_R8( temp, alpha ) |
164 |
CALL GSUM_R8( temp, alpha ) |
165 |
CALL GSUM_R8( temp, alpha ) |
166 |
CALL GSUM_R8( temp, alpha ) |
167 |
CALL GSUM_R8( temp, alpha ) |
168 |
CALL GSUM_R8( temp, alpha ) |
169 |
CALL GSUM_R8( temp, alpha ) |
170 |
CALL GSUM_R8( temp, alpha ) |
171 |
CALL GSUM_R8( temp, alpha ) |
172 |
CALL GSUM_R8( temp, alpha ) |
173 |
CALL GSUM_R8( temp, alpha ) |
174 |
CALL GSUM_R8( temp, alpha ) |
175 |
CALL GSUM_R8( temp, alpha ) |
176 |
CALL GSUM_R8( temp, alpha ) |
177 |
CALL GSUM_R8( temp, alpha ) |
178 |
CALL GSUM_R8( temp, alpha ) |
179 |
CALL GSUM_R8( temp, alpha ) |
180 |
CALL GSUM_R8( temp, alpha ) |
181 |
CALL GSUM_R8( temp, alpha ) |
182 |
CALL GSUM_R8( temp, alpha ) |
183 |
CALL GSUM_R8( temp, alpha ) |
184 |
CALL GSUM_R8( temp, alpha ) |
185 |
CALL GSUM_R8( temp, alpha ) |
186 |
CALL GSUM_R8( temp, alpha ) |
187 |
CALL GSUM_R8( temp, alpha ) |
188 |
CALL GSUM_R8( temp, alpha ) |
189 |
CALL GSUM_R8( temp, alpha ) |
190 |
CALL GSUM_R8( temp, alpha ) |
191 |
CALL GSUM_R8( temp, alpha ) |
192 |
CALL GSUM_R8( temp, alpha ) |
193 |
CALL GSUM_R8( temp, alpha ) |
194 |
CALL GSUM_R8( temp, alpha ) |
195 |
CALL GSUM_R8( temp, alpha ) |
196 |
CALL GSUM_R8( temp, alpha ) |
197 |
CALL GSUM_R8( temp, alpha ) |
198 |
CALL GSUM_R8( temp, alpha ) |
199 |
CALL GSUM_R8( temp, alpha ) |
200 |
CALL GSUM_R8( temp, alpha ) |
201 |
CALL GSUM_R8( temp, alpha ) |
202 |
CALL GSUM_R8( temp, alpha ) |
203 |
CALL GSUM_R8( temp, alpha ) |
204 |
CALL GSUM_R8( temp, alpha ) |
205 |
CALL GSUM_R8( temp, alpha ) |
206 |
CALL GSUM_R8( temp, alpha ) |
207 |
CALL GSUM_R8( temp, alpha ) |
208 |
CALL GSUM_R8( temp, alpha ) |
209 |
CALL GSUM_R8( temp, alpha ) |
210 |
CALL GSUM_R8( temp, alpha ) |
211 |
CALL GSUM_R8( temp, alpha ) |
212 |
CALL GSUM_R8( temp, alpha ) |
213 |
CALL GSUM_R8( temp, alpha ) |
214 |
CALL GSUM_R8( temp, alpha ) |
215 |
CALL GSUM_R8( temp, alpha ) |
216 |
CALL GSUM_R8( temp, alpha ) |
217 |
CALL GSUM_R8( temp, alpha ) |
218 |
CALL GSUM_R8( temp, alpha ) |
219 |
CALL GSUM_R8( temp, alpha ) |
220 |
CALL GSUM_R8( temp, alpha ) |
221 |
CALL GSUM_R8( temp, alpha ) |
222 |
CALL GSUM_R8( temp, alpha ) |
223 |
CALL GSUM_R8( temp, alpha ) |
224 |
CALL GSUM_R8( temp, alpha ) |
225 |
CALL GSUM_R8( temp, alpha ) |
226 |
CALL GSUM_R8( temp, alpha ) |
227 |
CALL GSUM_R8( temp, alpha ) |
228 |
CALL GSUM_R8( temp, alpha ) |
229 |
CALL GSUM_R8( temp, alpha ) |
230 |
CALL GSUM_R8( temp, alpha ) |
231 |
CALL GSUM_R8( temp, alpha ) |
232 |
CALL GSUM_R8( temp, alpha ) |
233 |
CALL GSUM_R8( temp, alpha ) |
234 |
CALL GSUM_R8( temp, alpha ) |
235 |
CALL GSUM_R8( temp, alpha ) |
236 |
CALL GSUM_R8( temp, alpha ) |
237 |
CALL GSUM_R8( temp, alpha ) |
238 |
CALL GSUM_R8( temp, alpha ) |
239 |
CALL GSUM_R8( temp, alpha ) |
240 |
CALL GSUM_R8( temp, alpha ) |
241 |
CALL GSUM_R8( temp, alpha ) |
242 |
CALL GSUM_R8( temp, alpha ) |
243 |
CALL GSUM_R8( temp, alpha ) |
244 |
CALL GSUM_R8( temp, alpha ) |
245 |
CALL GSUM_R8( temp, alpha ) |
246 |
CALL GSUM_R8( temp, alpha ) |
247 |
CALL GSUM_R8( temp, alpha ) |
248 |
CALL GSUM_R8( temp, alpha ) |
249 |
CALL GSUM_R8( temp, alpha ) |
250 |
CALL GSUM_R8( temp, alpha ) |
251 |
CALL GSUM_R8( temp, alpha ) |
252 |
CALL GSUM_R8( temp, alpha ) |
253 |
CALL GSUM_R8( temp, alpha ) |
254 |
CALL GSUM_R8( temp, alpha ) |
255 |
CALL GSUM_R8( temp, alpha ) |
256 |
CALL GSUM_R8( temp, alpha ) |
257 |
CALL GSUM_R8( temp, alpha ) |
258 |
#endif |
259 |
|
260 |
alpha = temp |
261 |
alpha = etaN/alpha |
262 |
|
263 |
C== Update solution and residual vectors |
264 |
C Now compute "interior" points. |
265 |
err = 0. _d 0 |
266 |
DO J=1,sNy |
267 |
DO I=1,sNx |
268 |
cg2d_x(I,J)=cg2d_x(I,J)+alpha*cg2d_s(I,J) |
269 |
cg2d_r(I,J)=cg2d_r(I,J)-alpha*cg2d_q(I,J) |
270 |
err = err+cg2d_r(I,J)*cg2d_r(I,J) |
271 |
ENDDO |
272 |
ENDDO |
273 |
|
274 |
CALL GSUM_R8( temp, err ) |
275 |
|
276 |
err = temp |
277 |
err = SQRT(err) |
278 |
actualIts = N |
279 |
actualResidual = err |
280 |
CcnhDebugStarts |
281 |
C WRITE(6,*) ' CG2D iters, err = ', actualIts, actualResidual |
282 |
CcnhDebugEnds |
283 |
IF ( err .LT. cg2dTargetResidual ) GOTO 11 |
284 |
CALL EXCH_XY_R8(cg2d_r ) |
285 |
10 CONTINUE |
286 |
11 CONTINUE |
287 |
CALL EXCH_XY_R8(cg2d_x ) |
288 |
WRITE(6,*) ' CG2D iters, err = ', actualIts, actualResidual |
289 |
|
290 |
C Calc Ax to check result |
291 |
DO J=1,sNy |
292 |
DO I=1,sNx |
293 |
cg2d_Ax(I,J) = |
294 |
& ( aW2d(I ,J )*cg2d_x(I-1,J )+aW2d(I+1,J )*cg2d_x(I+1,J ) |
295 |
& +aS2d(I ,J )*cg2d_x(I ,J-1)+aS2d(I ,J+1)*cg2d_x(I ,J+1) |
296 |
& -aW2d(I ,J )*cg2d_x(I ,J )-aW2d(I+1,J )*cg2d_x(I ,J ) |
297 |
& -aS2d(I ,J )*cg2d_x(I ,J )-aS2d(I ,J+1)*cg2d_x(I ,J ) |
298 |
& ) |
299 |
cg2d_r(I,J) = cg2d_b(I,J)-cg2d_Ax(I,J) |
300 |
ENDDO |
301 |
ENDDO |
302 |
CALL EXCH_XY_R8(cg2d_Ax ) |
303 |
CALL EXCH_XY_R8(cg2d_r ) |
304 |
|
305 |
END |