1 | SUBROUTINE cpu_statistics |
---|
2 | |
---|
3 | !--------------------------------------------------------------------------------! |
---|
4 | ! This file is part of PALM. |
---|
5 | ! |
---|
6 | ! PALM is free software: you can redistribute it and/or modify it under the terms |
---|
7 | ! of the GNU General Public License as published by the Free Software Foundation, |
---|
8 | ! either version 3 of the License, or (at your option) any later version. |
---|
9 | ! |
---|
10 | ! PALM is distributed in the hope that it will be useful, but WITHOUT ANY |
---|
11 | ! WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR |
---|
12 | ! A PARTICULAR PURPOSE. See the GNU General Public License for more details. |
---|
13 | ! |
---|
14 | ! You should have received a copy of the GNU General Public License along with |
---|
15 | ! PALM. If not, see <http://www.gnu.org/licenses/>. |
---|
16 | ! |
---|
17 | ! Copyright 1997-2012 Leibniz University Hannover |
---|
18 | !--------------------------------------------------------------------------------! |
---|
19 | ! |
---|
20 | ! Current revisions: |
---|
21 | ! ----------------- |
---|
22 | ! |
---|
23 | ! |
---|
24 | ! Former revisions: |
---|
25 | ! ----------------- |
---|
26 | ! $Id: cpu_statistics.f90 1112 2013-03-09 00:34:37Z witha $ |
---|
27 | ! |
---|
28 | ! 1111 2013-03-08 23:54:10Z raasch |
---|
29 | ! output of grid point numbers and average CPU time per grid point and timestep |
---|
30 | ! |
---|
31 | ! 1092 2013-02-02 11:24:22Z raasch |
---|
32 | ! unused variables removed |
---|
33 | ! |
---|
34 | ! 1036 2012-10-22 13:43:42Z raasch |
---|
35 | ! code put under GPL (PALM 3.9) |
---|
36 | ! |
---|
37 | ! 1015 2012-09-27 09:23:24Z raasch |
---|
38 | ! output of accelerator board information |
---|
39 | ! |
---|
40 | ! 683 2011-02-09 14:25:15Z raasch |
---|
41 | ! output of handling of ghostpoint exchange |
---|
42 | ! |
---|
43 | ! 622 2010-12-10 08:08:13Z raasch |
---|
44 | ! output of handling of collective operations |
---|
45 | ! |
---|
46 | ! 222 2009-01-12 16:04:16Z letzel |
---|
47 | ! Bugfix for nonparallel execution |
---|
48 | ! |
---|
49 | ! 197 2008-09-16 15:29:03Z raasch |
---|
50 | ! Format adjustments in order to allow CPU# > 999, |
---|
51 | ! data are collected from PE0 in an ordered sequence which seems to avoid |
---|
52 | ! hanging of processes on SGI-ICE |
---|
53 | ! |
---|
54 | ! 82 2007-04-16 15:40:52Z raasch |
---|
55 | ! Preprocessor directives for old systems removed |
---|
56 | ! |
---|
57 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
58 | ! |
---|
59 | ! Revision 1.13 2006/04/26 12:10:51 raasch |
---|
60 | ! Output of number of threads per task, max = min in case of 1 PE |
---|
61 | ! |
---|
62 | ! Revision 1.1 1997/07/24 11:11:11 raasch |
---|
63 | ! Initial revision |
---|
64 | ! |
---|
65 | ! |
---|
66 | ! Description: |
---|
67 | ! ------------ |
---|
68 | ! Analysis and output of the cpu-times measured. All PE results are collected |
---|
69 | ! on PE0 in order to calculate the mean cpu-time over all PEs and other |
---|
70 | ! statistics. The output is sorted according to the amount of cpu-time consumed |
---|
71 | ! and output on PE0. |
---|
72 | !------------------------------------------------------------------------------! |
---|
73 | |
---|
74 | USE control_parameters |
---|
75 | USE cpulog |
---|
76 | USE indices, ONLY: nx, ny, nz |
---|
77 | USE pegrid |
---|
78 | |
---|
79 | IMPLICIT NONE |
---|
80 | |
---|
81 | INTEGER :: i, ii(1), iii, sender |
---|
82 | REAL :: average_cputime |
---|
83 | REAL, SAVE :: norm = 1.0 |
---|
84 | REAL, DIMENSION(:), ALLOCATABLE :: pe_max, pe_min, pe_rms, sum |
---|
85 | REAL, DIMENSION(:,:), ALLOCATABLE :: pe_log_points |
---|
86 | |
---|
87 | |
---|
88 | ! |
---|
89 | !-- Compute cpu-times in seconds |
---|
90 | log_point%mtime = log_point%mtime / norm |
---|
91 | log_point%sum = log_point%sum / norm |
---|
92 | log_point%vector = log_point%vector / norm |
---|
93 | WHERE ( log_point%counts /= 0 ) |
---|
94 | log_point%mean = log_point%sum / log_point%counts |
---|
95 | END WHERE |
---|
96 | |
---|
97 | |
---|
98 | ! |
---|
99 | !-- Collect cpu-times from all PEs and calculate statistics |
---|
100 | IF ( myid == 0 ) THEN |
---|
101 | ! |
---|
102 | !-- Allocate and initialize temporary arrays needed for statistics |
---|
103 | ALLOCATE( pe_max( SIZE( log_point ) ), pe_min( SIZE( log_point ) ), & |
---|
104 | pe_rms( SIZE( log_point ) ), & |
---|
105 | pe_log_points( SIZE( log_point ), 0:numprocs-1 ) ) |
---|
106 | pe_min = log_point%sum |
---|
107 | pe_max = log_point%sum ! need to be set in case of 1 PE |
---|
108 | pe_rms = 0.0 |
---|
109 | |
---|
110 | #if defined( __parallel ) |
---|
111 | ! |
---|
112 | !-- Receive data from all PEs |
---|
113 | DO i = 1, numprocs-1 |
---|
114 | CALL MPI_RECV( pe_max(1), SIZE( log_point ), MPI_REAL, & |
---|
115 | i, i, comm2d, status, ierr ) |
---|
116 | sender = status(MPI_SOURCE) |
---|
117 | pe_log_points(:,sender) = pe_max |
---|
118 | ENDDO |
---|
119 | pe_log_points(:,0) = log_point%sum ! Results from PE0 |
---|
120 | ! |
---|
121 | !-- Calculate mean of all PEs, store it on log_point%sum |
---|
122 | !-- and find minimum and maximum |
---|
123 | DO iii = 1, SIZE( log_point ) |
---|
124 | DO i = 1, numprocs-1 |
---|
125 | log_point(iii)%sum = log_point(iii)%sum + pe_log_points(iii,i) |
---|
126 | pe_min(iii) = MIN( pe_min(iii), pe_log_points(iii,i) ) |
---|
127 | pe_max(iii) = MAX( pe_max(iii), pe_log_points(iii,i) ) |
---|
128 | ENDDO |
---|
129 | log_point(iii)%sum = log_point(iii)%sum / numprocs |
---|
130 | ! |
---|
131 | !-- Calculate rms |
---|
132 | DO i = 0, numprocs-1 |
---|
133 | pe_rms(iii) = pe_rms(iii) + ( & |
---|
134 | pe_log_points(iii,i) - log_point(iii)%sum & |
---|
135 | )**2 |
---|
136 | ENDDO |
---|
137 | pe_rms(iii) = SQRT( pe_rms(iii) / numprocs ) |
---|
138 | ENDDO |
---|
139 | ELSE |
---|
140 | ! |
---|
141 | !-- Send data to PE0 (pe_max is used as temporary storage to send |
---|
142 | !-- the data in order to avoid sending the data type log) |
---|
143 | ALLOCATE( pe_max( SIZE( log_point ) ) ) |
---|
144 | pe_max = log_point%sum |
---|
145 | CALL MPI_SEND( pe_max(1), SIZE( log_point ), MPI_REAL, 0, myid, comm2d, & |
---|
146 | ierr ) |
---|
147 | #endif |
---|
148 | |
---|
149 | ENDIF |
---|
150 | |
---|
151 | ! |
---|
152 | !-- Write cpu-times |
---|
153 | IF ( myid == 0 ) THEN |
---|
154 | ! |
---|
155 | !-- Re-store sums |
---|
156 | ALLOCATE( sum( SIZE( log_point ) ) ) |
---|
157 | WHERE ( log_point%counts /= 0 ) |
---|
158 | sum = log_point%sum |
---|
159 | ELSEWHERE |
---|
160 | sum = -1.0 |
---|
161 | ENDWHERE |
---|
162 | |
---|
163 | ! |
---|
164 | !-- Get total time in order to calculate CPU-time per gridpoint and timestep |
---|
165 | IF ( nr_timesteps_this_run /= 0 ) THEN |
---|
166 | average_cputime = log_point(1)%sum / REAL( (nx+1) * (ny+1) * nz ) / & |
---|
167 | REAL( nr_timesteps_this_run ) * 1E6 ! in micro-sec |
---|
168 | ELSE |
---|
169 | average_cputime = -1.0 |
---|
170 | ENDIF |
---|
171 | |
---|
172 | ! |
---|
173 | !-- Write cpu-times sorted by size |
---|
174 | CALL check_open( 18 ) |
---|
175 | #if defined( __parallel ) |
---|
176 | WRITE ( 18, 100 ) TRIM( run_description_header ), & |
---|
177 | numprocs * threads_per_task, pdims(1), pdims(2), & |
---|
178 | threads_per_task, nx+1, ny+1, nz, nr_timesteps_this_run, & |
---|
179 | average_cputime |
---|
180 | |
---|
181 | IF ( num_acc_per_node /= 0 ) WRITE ( 18, 108 ) num_acc_per_node |
---|
182 | WRITE ( 18, 110 ) |
---|
183 | #else |
---|
184 | WRITE ( 18, 100 ) TRIM( run_description_header ), & |
---|
185 | numprocs * threads_per_task, 1, 1, & |
---|
186 | threads_per_task, nx+1, ny+1, nz, nr_timesteps_this_run, & |
---|
187 | average_cputime |
---|
188 | |
---|
189 | IF ( num_acc_per_node /= 0 ) WRITE ( 18, 109 ) num_acc_per_node |
---|
190 | WRITE ( 18, 110 ) |
---|
191 | #endif |
---|
192 | DO |
---|
193 | ii = MAXLOC( sum ) |
---|
194 | i = ii(1) |
---|
195 | IF ( sum(i) /= -1.0 ) THEN |
---|
196 | WRITE ( 18, 102 ) & |
---|
197 | log_point(i)%place, log_point(i)%sum, & |
---|
198 | log_point(i)%sum / log_point(1)%sum * 100.0, & |
---|
199 | log_point(i)%counts, pe_min(i), pe_max(i), pe_rms(i) |
---|
200 | sum(i) = -1.0 |
---|
201 | ELSE |
---|
202 | EXIT |
---|
203 | ENDIF |
---|
204 | ENDDO |
---|
205 | ENDIF |
---|
206 | |
---|
207 | |
---|
208 | ! |
---|
209 | !-- The same procedure again for the individual measurements. |
---|
210 | ! |
---|
211 | !-- Compute cpu-times in seconds |
---|
212 | log_point_s%mtime = log_point_s%mtime / norm |
---|
213 | log_point_s%sum = log_point_s%sum / norm |
---|
214 | log_point_s%vector = log_point_s%vector / norm |
---|
215 | WHERE ( log_point_s%counts /= 0 ) |
---|
216 | log_point_s%mean = log_point_s%sum / log_point_s%counts |
---|
217 | END WHERE |
---|
218 | |
---|
219 | ! |
---|
220 | !-- Collect cpu-times from all PEs and calculate statistics |
---|
221 | #if defined( __parallel ) |
---|
222 | ! |
---|
223 | !-- Set barrier in order to avoid that PE0 receives log_point_s-data |
---|
224 | !-- while still busy with receiving log_point-data (see above) |
---|
225 | CALL MPI_BARRIER( comm2d, ierr ) |
---|
226 | #endif |
---|
227 | IF ( myid == 0 ) THEN |
---|
228 | ! |
---|
229 | !-- Initialize temporary arrays needed for statistics |
---|
230 | pe_min = log_point_s%sum |
---|
231 | pe_max = log_point_s%sum ! need to be set in case of 1 PE |
---|
232 | pe_rms = 0.0 |
---|
233 | |
---|
234 | #if defined( __parallel ) |
---|
235 | ! |
---|
236 | !-- Receive data from all PEs |
---|
237 | DO i = 1, numprocs-1 |
---|
238 | CALL MPI_RECV( pe_max(1), SIZE( log_point ), MPI_REAL, & |
---|
239 | MPI_ANY_SOURCE, MPI_ANY_TAG, comm2d, status, ierr ) |
---|
240 | sender = status(MPI_SOURCE) |
---|
241 | pe_log_points(:,sender) = pe_max |
---|
242 | ENDDO |
---|
243 | pe_log_points(:,0) = log_point_s%sum ! Results from PE0 |
---|
244 | ! |
---|
245 | !-- Calculate mean of all PEs, store it on log_point_s%sum |
---|
246 | !-- and find minimum and maximum |
---|
247 | DO iii = 1, SIZE( log_point ) |
---|
248 | DO i = 1, numprocs-1 |
---|
249 | log_point_s(iii)%sum = log_point_s(iii)%sum + pe_log_points(iii,i) |
---|
250 | pe_min(iii) = MIN( pe_min(iii), pe_log_points(iii,i) ) |
---|
251 | pe_max(iii) = MAX( pe_max(iii), pe_log_points(iii,i) ) |
---|
252 | ENDDO |
---|
253 | log_point_s(iii)%sum = log_point_s(iii)%sum / numprocs |
---|
254 | ! |
---|
255 | !-- Calculate rms |
---|
256 | DO i = 0, numprocs-1 |
---|
257 | pe_rms(iii) = pe_rms(iii) + ( & |
---|
258 | pe_log_points(iii,i) - log_point_s(iii)%sum & |
---|
259 | )**2 |
---|
260 | ENDDO |
---|
261 | pe_rms(iii) = SQRT( pe_rms(iii) / numprocs ) |
---|
262 | ENDDO |
---|
263 | ELSE |
---|
264 | ! |
---|
265 | !-- Send data to PE0 (pe_max is used as temporary storage to send |
---|
266 | !-- the data in order to avoid sending the data type log) |
---|
267 | pe_max = log_point_s%sum |
---|
268 | CALL MPI_SEND( pe_max(1), SIZE( log_point ), MPI_REAL, 0, 0, comm2d, & |
---|
269 | ierr ) |
---|
270 | #endif |
---|
271 | |
---|
272 | ENDIF |
---|
273 | |
---|
274 | ! |
---|
275 | !-- Write cpu-times |
---|
276 | IF ( myid == 0 ) THEN |
---|
277 | ! |
---|
278 | !-- Re-store sums |
---|
279 | WHERE ( log_point_s%counts /= 0 ) |
---|
280 | sum = log_point_s%sum |
---|
281 | ELSEWHERE |
---|
282 | sum = -1.0 |
---|
283 | ENDWHERE |
---|
284 | |
---|
285 | ! |
---|
286 | !-- Write cpu-times sorted by size |
---|
287 | WRITE ( 18, 101 ) |
---|
288 | DO |
---|
289 | ii = MAXLOC( sum ) |
---|
290 | i = ii(1) |
---|
291 | IF ( sum(i) /= -1.0 ) THEN |
---|
292 | WRITE ( 18, 102 ) & |
---|
293 | log_point_s(i)%place, log_point_s(i)%sum, & |
---|
294 | log_point_s(i)%sum / log_point(1)%sum * 100.0, & |
---|
295 | log_point_s(i)%counts, pe_min(i), pe_max(i), pe_rms(i) |
---|
296 | sum(i) = -1.0 |
---|
297 | ELSE |
---|
298 | EXIT |
---|
299 | ENDIF |
---|
300 | ENDDO |
---|
301 | |
---|
302 | ! |
---|
303 | !-- Output of handling of MPI operations |
---|
304 | IF ( collective_wait ) THEN |
---|
305 | WRITE ( 18, 103 ) |
---|
306 | ELSE |
---|
307 | WRITE ( 18, 104 ) |
---|
308 | ENDIF |
---|
309 | IF ( synchronous_exchange ) THEN |
---|
310 | WRITE ( 18, 105 ) |
---|
311 | ELSE |
---|
312 | WRITE ( 18, 106 ) |
---|
313 | ENDIF |
---|
314 | |
---|
315 | ! |
---|
316 | !-- Empty lines in order to create a gap to the results of the model |
---|
317 | !-- continuation runs |
---|
318 | WRITE ( 18, 107 ) |
---|
319 | |
---|
320 | ! |
---|
321 | !-- Unit 18 is not needed anymore |
---|
322 | CALL close_file( 18 ) |
---|
323 | |
---|
324 | ENDIF |
---|
325 | |
---|
326 | |
---|
327 | 100 FORMAT (A/11('-')//'CPU measures for ',I5,' PEs (',I5,'(x) * ',I5,'(y', & |
---|
328 | &') tasks *',I5,' threads):'// & |
---|
329 | 'gridpoints (x/y/z): ',20X,I5,' * ',I5,' * ',I5/ & |
---|
330 | 'nr of timesteps: ',22X,I6/ & |
---|
331 | 'cpu time per grid point and timestep: ',5X,F8.5,' * 10**-6 s') |
---|
332 | |
---|
333 | 101 FORMAT (/'special measures:'/ & |
---|
334 | &'-----------------------------------------------------------', & |
---|
335 | &'--------------------') |
---|
336 | |
---|
337 | 102 FORMAT (A20,2X,F9.3,2X,F7.2,1X,I7,3(1X,F9.3)) |
---|
338 | 103 FORMAT (/'Barriers are set in front of collective operations') |
---|
339 | 104 FORMAT (/'No barriers are set in front of collective operations') |
---|
340 | 105 FORMAT (/'Exchange of ghostpoints via MPI_SENDRCV') |
---|
341 | 106 FORMAT (/'Exchange of ghostpoints via MPI_ISEND/MPI_IRECV') |
---|
342 | 107 FORMAT (//) |
---|
343 | 108 FORMAT ('Accelerator boards per node: ',14X,I2) |
---|
344 | 109 FORMAT ('Accelerator boards: ',23X,I2) |
---|
345 | 110 FORMAT ('----------------------------------------------------------', & |
---|
346 | &'------------'//& |
---|
347 | &'place: mean counts min ', & |
---|
348 | &' max rms'/ & |
---|
349 | &' sec. % sec. ', & |
---|
350 | &' sec. sec.'/ & |
---|
351 | &'-----------------------------------------------------------', & |
---|
352 | &'-------------------') |
---|
353 | |
---|
354 | END SUBROUTINE cpu_statistics |
---|
355 | |
---|