[1] | 1 | MODULE poisfft_hybrid_mod |
---|
[808] | 2 | |
---|
[1036] | 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 | ! |
---|
[257] | 20 | ! Current revisions: |
---|
[1] | 21 | ! ----------------- |
---|
| 22 | ! |
---|
[1014] | 23 | ! |
---|
[1] | 24 | ! Former revisions: |
---|
| 25 | ! ----------------- |
---|
[3] | 26 | ! $Id: poisfft_hybrid.f90 1036 2012-10-22 13:43:42Z raasch $ |
---|
[392] | 27 | ! |
---|
[1014] | 28 | ! 1013 2012-09-21 07:03:55Z raasch |
---|
| 29 | ! FLOAT type conversion replaced by REAL |
---|
| 30 | ! |
---|
[810] | 31 | ! 809 2012-01-30 13:32:58Z maronga |
---|
| 32 | ! Bugfix: replaced .AND. and .NOT. with && and ! in the preprocessor directives |
---|
| 33 | ! |
---|
[808] | 34 | ! 807 2012-01-25 11:53:51Z maronga |
---|
| 35 | ! New cpp directive "__check" implemented which is used by check_namelist_files |
---|
| 36 | ! (most of the code is unneeded by check_namelist_files). |
---|
| 37 | ! |
---|
[668] | 38 | ! 667 2010-12-23 12:06:00Z suehring/gryschka |
---|
| 39 | ! ddzu replaced by ddzu_pres due to changes in zu(0) |
---|
| 40 | ! |
---|
[482] | 41 | ! 415 2009-12-15 10:26:23Z raasch |
---|
[807] | 42 | ! Dimension of array stat in cascade change to prevent type problems with___ |
---|
[482] | 43 | ! mpi2 libraries |
---|
| 44 | ! |
---|
[392] | 45 | ! 274 2009-03-26 15:11:21Z heinze |
---|
| 46 | ! Output of messages replaced by message handling routine. |
---|
| 47 | ! |
---|
| 48 | ! Feb. 2007 |
---|
[3] | 49 | ! RCS Log replace by Id keyword, revision history cleaned up |
---|
| 50 | ! |
---|
[1] | 51 | ! Revision 1.11 2004/04/30 12:43:14 raasch |
---|
| 52 | ! Renaming of fft routines, additional argument in calls of fft_y_m |
---|
| 53 | ! |
---|
| 54 | ! Revision 1.2 2002/12/19 16:08:31 raasch |
---|
| 55 | ! Preprocessor directive KKMP introduced (OMP does NOT work), |
---|
| 56 | ! array tri will be a shared array in OpenMP loop, to get better cache |
---|
| 57 | ! utilization, the i index (x-direction) will be executed in stride |
---|
| 58 | ! "istride" as outer loop and in a shorter inner loop, |
---|
| 59 | ! overlapping of computation and communication realized by new routine |
---|
| 60 | ! poisfft_hybrid_nodes, name of old routine poisfft_hybrid changed to |
---|
| 61 | ! poisfft_hybrid_omp, STOP statement replaced by call of subroutine local_stop |
---|
| 62 | ! |
---|
| 63 | ! |
---|
| 64 | ! Description: |
---|
| 65 | ! ------------ |
---|
| 66 | ! Solution of the Poisson equation with a 2D spectral method. |
---|
| 67 | ! Hybrid version for parallel computers using a 1D domain decomposition, |
---|
| 68 | ! realized with MPI, along x and parallelization with OPEN-MP along y |
---|
| 69 | ! (routine poisfft_hybrid_omp). In a second version (poisfft_hybrid_nodes), |
---|
| 70 | ! optimization is realized by overlapping of computation and communication |
---|
| 71 | ! and by simultaneously executing as many communication calls as switches |
---|
| 72 | ! per logical partition (LPAR) are available. This version comes into |
---|
| 73 | ! effect if more than one node is used and if the environment variable |
---|
| 74 | ! tasks_per_node is set in a way that it can be devided by switch_per_lpar |
---|
| 75 | ! without any rest. |
---|
| 76 | ! |
---|
| 77 | ! WARNING: In case of OpenMP, there are problems with allocating large |
---|
| 78 | ! arrays in parallel regions. |
---|
| 79 | ! |
---|
| 80 | ! Copyright Klaus Ketelsen / Siegfried Raasch May 2002 |
---|
| 81 | !------------------------------------------------------------------------------! |
---|
| 82 | |
---|
| 83 | USE fft_xy |
---|
| 84 | USE indices |
---|
| 85 | USE pegrid |
---|
| 86 | |
---|
| 87 | IMPLICIT NONE |
---|
| 88 | |
---|
| 89 | INTEGER, PARAMETER :: switch_per_lpar = 2 |
---|
| 90 | |
---|
| 91 | INTEGER, SAVE :: nxl_a, nxr_a, & ! total x dimension |
---|
| 92 | nxl_p, nxr_p, & ! partial x dimension |
---|
| 93 | nys_a, nyn_a, & ! total y dimension |
---|
| 94 | nys_p, nyn_p, & ! partial y dimension |
---|
| 95 | |
---|
| 96 | npe_s, & ! total number of PEs for solver |
---|
| 97 | nwords, & ! number of points to be exchanged |
---|
| 98 | ! with MPI_ALLTOALL |
---|
| 99 | n_omp_threads ! number of OpenMP threads |
---|
| 100 | |
---|
| 101 | ! |
---|
| 102 | !-- Variables for multi node version (cluster version) using routine |
---|
| 103 | !-- poisfft_hybrid_nodes |
---|
| 104 | INTEGER, SAVE :: comm_nodes, & ! communicater nodes |
---|
| 105 | comm_node_all, & ! communicater all PEs node version |
---|
| 106 | comm_tasks, & ! communicater tasks |
---|
| 107 | me, me_node, me_task,& ! identity of this PE |
---|
| 108 | nodes, & ! number of nodes |
---|
| 109 | tasks_per_logical_node = -1 ! default no cluster |
---|
| 110 | |
---|
[807] | 111 | |
---|
| 112 | PRIVATE |
---|
| 113 | |
---|
| 114 | |
---|
[809] | 115 | #if ! defined ( __check ) |
---|
[807] | 116 | PUBLIC poisfft_hybrid, poisfft_hybrid_ini |
---|
| 117 | |
---|
| 118 | |
---|
[1] | 119 | ! |
---|
| 120 | !-- Public interfaces |
---|
| 121 | INTERFACE poisfft_hybrid_ini |
---|
| 122 | MODULE PROCEDURE poisfft_hybrid_ini |
---|
| 123 | END INTERFACE poisfft_hybrid_ini |
---|
| 124 | |
---|
| 125 | INTERFACE poisfft_hybrid |
---|
| 126 | MODULE PROCEDURE poisfft_hybrid |
---|
| 127 | END INTERFACE poisfft_hybrid |
---|
| 128 | |
---|
| 129 | ! |
---|
| 130 | !-- Private interfaces |
---|
| 131 | INTERFACE poisfft_hybrid_omp |
---|
| 132 | MODULE PROCEDURE poisfft_hybrid_omp |
---|
| 133 | END INTERFACE poisfft_hybrid_omp |
---|
| 134 | |
---|
| 135 | INTERFACE poisfft_hybrid_omp_vec |
---|
| 136 | MODULE PROCEDURE poisfft_hybrid_omp_vec |
---|
| 137 | END INTERFACE poisfft_hybrid_omp_vec |
---|
| 138 | |
---|
| 139 | INTERFACE poisfft_hybrid_nodes |
---|
| 140 | MODULE PROCEDURE poisfft_hybrid_nodes |
---|
| 141 | END INTERFACE poisfft_hybrid_nodes |
---|
| 142 | |
---|
| 143 | INTERFACE tridia_hybrid |
---|
| 144 | MODULE PROCEDURE tridia_hybrid |
---|
| 145 | END INTERFACE tridia_hybrid |
---|
| 146 | |
---|
| 147 | INTERFACE cascade |
---|
| 148 | MODULE PROCEDURE cascade |
---|
| 149 | END INTERFACE cascade |
---|
[807] | 150 | #else |
---|
| 151 | PUBLIC poisfft_hybrid_ini |
---|
[1] | 152 | |
---|
[807] | 153 | ! |
---|
| 154 | !-- Public interfaces |
---|
| 155 | INTERFACE poisfft_hybrid_ini |
---|
| 156 | MODULE PROCEDURE poisfft_hybrid_ini |
---|
| 157 | END INTERFACE poisfft_hybrid_ini |
---|
| 158 | #endif |
---|
| 159 | |
---|
[1] | 160 | CONTAINS |
---|
| 161 | |
---|
[807] | 162 | |
---|
[1] | 163 | SUBROUTINE poisfft_hybrid_ini |
---|
| 164 | |
---|
| 165 | USE control_parameters |
---|
| 166 | USE pegrid |
---|
| 167 | |
---|
| 168 | IMPLICIT NONE |
---|
| 169 | |
---|
| 170 | CHARACTER(LEN=8) :: cdummy |
---|
| 171 | INTEGER :: idummy, istat |
---|
| 172 | INTEGER, DIMENSION(2) :: coords, dims |
---|
| 173 | |
---|
| 174 | LOGICAL, DIMENSION(2) :: period = .false., re_dims |
---|
| 175 | |
---|
| 176 | |
---|
| 177 | ! |
---|
| 178 | !-- Set the internal index values for the hybrid solver |
---|
| 179 | #if defined( __parallel ) |
---|
| 180 | npe_s = pdims(1) |
---|
| 181 | #else |
---|
| 182 | npe_s = 1 |
---|
| 183 | #endif |
---|
| 184 | nxl_a = 0 |
---|
| 185 | nxr_a = nx |
---|
| 186 | nxl_p = 0 |
---|
| 187 | nxr_p = ( ( nx+1 ) / npe_s ) - 1 |
---|
| 188 | nys_a = nys |
---|
| 189 | nyn_a = nyn |
---|
| 190 | nys_p = 0 |
---|
| 191 | nyn_p = ( ( ny+1 ) / npe_s ) - 1 |
---|
| 192 | |
---|
| 193 | nwords = ( nxr_p-nxl_p+1 ) * nz * ( nyn_p-nys_p+1 ) |
---|
| 194 | |
---|
[809] | 195 | #if defined( __KKMP ) && ! defined ( __check ) |
---|
[1] | 196 | CALL LOCAL_GETENV( 'OMP_NUM_THREADS', 15, cdummy, idummy ) |
---|
| 197 | READ ( cdummy, '(I8)' ) n_omp_threads |
---|
[257] | 198 | IF ( n_omp_threads > 1 ) THEN |
---|
| 199 | WRITE( message_string, * ) 'Number of OpenMP threads = ', & |
---|
| 200 | n_omp_threads |
---|
| 201 | CALL message( 'poisfft_hybrid_ini', 'PA0280', 0, 0, 0, 6, 0 ) |
---|
[1] | 202 | ENDIF |
---|
| 203 | #else |
---|
| 204 | n_omp_threads = 1 |
---|
| 205 | #endif |
---|
| 206 | ! |
---|
| 207 | !-- Initialize the one-dimensional FFT routines |
---|
| 208 | CALL fft_init |
---|
| 209 | |
---|
| 210 | ! |
---|
| 211 | !-- Setup for multi node version (poisfft_hybrid_nodes) |
---|
| 212 | IF ( n_omp_threads == 1 .AND. & |
---|
| 213 | ( host(1:4) == 'ibmh' .OR. host(1:4) == 'ibmb' ) ) THEN |
---|
| 214 | |
---|
| 215 | IF ( tasks_per_node /= -9999 ) THEN |
---|
| 216 | ! |
---|
| 217 | !-- Multi node version requires that the available number of |
---|
| 218 | !-- switches per logical partition must be an integral divisor |
---|
| 219 | !-- of the chosen number of tasks per node |
---|
| 220 | IF ( MOD( tasks_per_node, switch_per_lpar ) == 0 ) THEN |
---|
| 221 | ! |
---|
| 222 | !-- Set the switch which decides about usage of the multi node |
---|
| 223 | !-- version |
---|
| 224 | IF ( tasks_per_node / switch_per_lpar > 1 .AND. & |
---|
| 225 | numprocs > tasks_per_node ) THEN |
---|
| 226 | tasks_per_logical_node = tasks_per_node / switch_per_lpar |
---|
| 227 | ENDIF |
---|
| 228 | |
---|
[257] | 229 | IF ( tasks_per_logical_node > -1 ) THEN |
---|
| 230 | |
---|
| 231 | WRITE( message_string, * ) 'running optimized ', & |
---|
| 232 | 'multinode version', & |
---|
| 233 | '&switch_per_lpar = ', & |
---|
| 234 | switch_per_lpar, & |
---|
| 235 | '&tasks_per_lpar = ', & |
---|
| 236 | tasks_per_node, & |
---|
| 237 | 'tasks_per_logical_node = ', & |
---|
| 238 | tasks_per_logical_node |
---|
[274] | 239 | CALL message( 'poisfft_hybrid_ini', 'PA0281', 0, 0, 0, 6, 0 ) |
---|
[257] | 240 | |
---|
[1] | 241 | ENDIF |
---|
| 242 | |
---|
| 243 | ENDIF |
---|
| 244 | ENDIF |
---|
| 245 | ENDIF |
---|
| 246 | |
---|
| 247 | ! |
---|
| 248 | !-- Determine sub-topologies for multi node version |
---|
| 249 | IF ( tasks_per_logical_node >= 2 ) THEN |
---|
| 250 | |
---|
[809] | 251 | #if defined( __parallel ) && ! defined ( __check ) |
---|
[1] | 252 | nodes = ( numprocs + tasks_per_logical_node - 1 ) / & |
---|
| 253 | tasks_per_logical_node |
---|
| 254 | dims(1) = nodes |
---|
| 255 | dims(2) = tasks_per_logical_node |
---|
| 256 | |
---|
| 257 | CALL MPI_CART_CREATE( comm2d, 2, dims, period, .FALSE., & |
---|
| 258 | comm_node_all, istat ) |
---|
| 259 | CALL MPI_COMM_RANK( comm_node_all, me, istat ) |
---|
| 260 | |
---|
| 261 | re_dims(1) = .TRUE. |
---|
| 262 | re_dims(2) = .FALSE. |
---|
| 263 | CALL MPI_CART_SUB( comm_node_all, re_dims, comm_nodes, istat ) |
---|
| 264 | CALL MPI_COMM_RANK( comm_nodes, me_node, istat ) |
---|
| 265 | |
---|
| 266 | re_dims(1) = .FALSE. |
---|
| 267 | re_dims(2) = .TRUE. |
---|
| 268 | CALL MPI_CART_SUB( comm_node_all, re_dims, comm_tasks, istat ) |
---|
| 269 | CALL MPI_COMM_RANK( comm_tasks, me_task, istat ) |
---|
| 270 | |
---|
| 271 | ! write(0,*) 'who am i',myid,me,me_node,me_task,nodes,& |
---|
| 272 | ! tasks_per_logical_node |
---|
[809] | 273 | #elif ! defined( __parallel ) |
---|
[257] | 274 | message_string = 'parallel environment (MPI) required' |
---|
| 275 | CALL message( 'poisfft_hybrid_ini', 'PA0282', 1, 2, 0, 6, 0 ) |
---|
[1] | 276 | #endif |
---|
| 277 | ENDIF |
---|
| 278 | |
---|
| 279 | END SUBROUTINE poisfft_hybrid_ini |
---|
| 280 | |
---|
[809] | 281 | #if ! defined ( __check ) |
---|
[1] | 282 | SUBROUTINE poisfft_hybrid( ar ) |
---|
| 283 | |
---|
| 284 | USE control_parameters |
---|
| 285 | USE interfaces |
---|
| 286 | |
---|
| 287 | IMPLICIT NONE |
---|
| 288 | |
---|
| 289 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
| 290 | |
---|
| 291 | IF ( host(1:3) == 'nec' ) THEN |
---|
| 292 | CALL poisfft_hybrid_omp_vec( ar ) |
---|
| 293 | ELSE |
---|
| 294 | IF ( tasks_per_logical_node == -1 ) THEN |
---|
| 295 | CALL poisfft_hybrid_omp( ar ) |
---|
| 296 | ELSE |
---|
| 297 | CALL poisfft_hybrid_nodes( ar ) |
---|
| 298 | ENDIF |
---|
| 299 | ENDIF |
---|
| 300 | |
---|
| 301 | END SUBROUTINE poisfft_hybrid |
---|
| 302 | |
---|
| 303 | |
---|
| 304 | SUBROUTINE poisfft_hybrid_omp ( ar ) |
---|
| 305 | |
---|
| 306 | USE cpulog |
---|
| 307 | USE interfaces |
---|
| 308 | |
---|
| 309 | IMPLICIT NONE |
---|
| 310 | |
---|
| 311 | INTEGER, PARAMETER :: istride = 4 ! stride of i loop |
---|
| 312 | INTEGER :: i, ii, ir, iei, iouter, istat, j, jj, k, m, n, jthread |
---|
| 313 | |
---|
| 314 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
| 315 | |
---|
| 316 | REAL, DIMENSION(0:nx) :: fftx_ar |
---|
| 317 | REAL, DIMENSION(0:ny,istride) :: ffty_ar |
---|
| 318 | |
---|
| 319 | REAL, DIMENSION(0:nx,nz) :: tri_ar |
---|
| 320 | |
---|
| 321 | REAL, DIMENSION(nxl_p:nxr_p,nz,nys_p:nyn_p,npe_s) :: work1, work2 |
---|
| 322 | #if defined( __KKMP ) |
---|
| 323 | INTEGER :: omp_get_thread_num |
---|
| 324 | REAL, DIMENSION(:,:,:,:), ALLOCATABLE :: tri |
---|
| 325 | ALLOCATE( tri(5,0:nx,0:nz-1,n_omp_threads ) ) |
---|
| 326 | #else |
---|
| 327 | REAL, DIMENSION(5,0:nx,0:nz-1,1) :: tri |
---|
| 328 | #endif |
---|
| 329 | |
---|
| 330 | |
---|
| 331 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_omp', 'start' ) |
---|
| 332 | |
---|
| 333 | CALL cpu_log( log_point_s(7), 'fft_y', 'start' ) |
---|
| 334 | |
---|
| 335 | !$OMP PARALLEL PRIVATE (i,iouter,ii,ir,iei,j,k,m,n,ffty_ar) |
---|
| 336 | !$OMP DO |
---|
| 337 | ! |
---|
| 338 | !-- Store grid points to be transformed on a 1d-array, do the fft |
---|
| 339 | !-- and sample the results on a 4d-array |
---|
| 340 | DO iouter = nxl_p, nxr_p, istride ! stride loop, better cache |
---|
| 341 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
| 342 | DO k = 1, nz |
---|
| 343 | |
---|
| 344 | DO i = iouter, iei |
---|
| 345 | ii = nxl + i |
---|
| 346 | ir = i - iouter + 1 |
---|
| 347 | |
---|
| 348 | DO j = nys_a, nyn_a |
---|
| 349 | ffty_ar(j,ir) = ar(k,j,ii) |
---|
| 350 | ENDDO |
---|
| 351 | |
---|
| 352 | CALL fft_y( ffty_ar(:,ir), 'forward' ) |
---|
| 353 | ENDDO |
---|
| 354 | |
---|
| 355 | m = nys_a |
---|
| 356 | DO n = 1, npe_s |
---|
| 357 | DO j = nys_p, nyn_p |
---|
| 358 | DO i = iouter, iei |
---|
| 359 | ir = i - iouter + 1 |
---|
| 360 | work1(i,k,j,n) = ffty_ar(m,ir) |
---|
| 361 | ENDDO |
---|
| 362 | m = m+1 |
---|
| 363 | ENDDO |
---|
| 364 | ENDDO |
---|
| 365 | |
---|
| 366 | ENDDO |
---|
| 367 | ENDDO |
---|
| 368 | !$OMP END PARALLEL |
---|
| 369 | |
---|
| 370 | CALL cpu_log( log_point_s(7), 'fft_y', 'pause' ) |
---|
| 371 | |
---|
| 372 | #if defined( __parallel ) |
---|
| 373 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
| 374 | |
---|
| 375 | CALL MPI_ALLTOALL( work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 376 | work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 377 | comm2d, istat ) |
---|
| 378 | |
---|
| 379 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
| 380 | #else |
---|
| 381 | work2 = work1 |
---|
| 382 | #endif |
---|
| 383 | |
---|
| 384 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'start' ) |
---|
| 385 | |
---|
| 386 | #if defined( __KKMP ) |
---|
| 387 | !$OMP PARALLEL PRIVATE (i,j,jj,k,m,n,fftx_ar,tri_ar,jthread) |
---|
| 388 | !$OMP DO |
---|
| 389 | DO j = nys_p, nyn_p |
---|
| 390 | jthread = omp_get_thread_num() + 1 |
---|
| 391 | #else |
---|
| 392 | DO j = nys_p, nyn_p |
---|
| 393 | jthread = 1 |
---|
| 394 | #endif |
---|
| 395 | DO k = 1, nz |
---|
| 396 | |
---|
| 397 | m = nxl_a |
---|
| 398 | DO n = 1, npe_s |
---|
| 399 | DO i = nxl_p, nxr_p |
---|
| 400 | fftx_ar(m) = work2(i,k,j,n) |
---|
| 401 | m = m+1 |
---|
| 402 | ENDDO |
---|
| 403 | ENDDO |
---|
| 404 | |
---|
| 405 | CALL fft_x( fftx_ar, 'forward' ) |
---|
| 406 | |
---|
| 407 | DO i = nxl_a, nxr_a |
---|
| 408 | tri_ar(i,k) = fftx_ar(i) |
---|
| 409 | ENDDO |
---|
| 410 | |
---|
| 411 | ENDDO |
---|
| 412 | |
---|
| 413 | jj = myid * (nyn_p-nys_p+1) + j |
---|
| 414 | CALL tridia_hybrid( jj, tri_ar, tri(:,:,:,jthread)) |
---|
| 415 | |
---|
| 416 | DO k = 1, nz |
---|
| 417 | DO i = nxl_a, nxr_a |
---|
| 418 | fftx_ar(i) = tri_ar (i,k) |
---|
| 419 | ENDDO |
---|
| 420 | |
---|
| 421 | CALL fft_x( fftx_ar, 'backward' ) |
---|
| 422 | |
---|
| 423 | m = nxl_a |
---|
| 424 | DO n = 1, npe_s |
---|
| 425 | DO i = nxl_p, nxr_p |
---|
| 426 | work2(i,k,j,n) = fftx_ar(m) |
---|
| 427 | m = m+1 |
---|
| 428 | ENDDO |
---|
| 429 | ENDDO |
---|
| 430 | |
---|
| 431 | ENDDO |
---|
| 432 | ENDDO |
---|
| 433 | #if defined( __KKMP ) |
---|
| 434 | !$OMP END PARALLEL |
---|
| 435 | #endif |
---|
| 436 | |
---|
| 437 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'stop' ) |
---|
| 438 | |
---|
| 439 | #if defined( __parallel ) |
---|
| 440 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
| 441 | nwords = (nxr_p-nxl_p+1) * nz * (nyn_p-nys_p+1) |
---|
| 442 | |
---|
| 443 | CALL MPI_ALLTOALL( work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 444 | work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 445 | comm2d, istat ) |
---|
| 446 | |
---|
| 447 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
| 448 | #else |
---|
| 449 | work1 = work2 |
---|
| 450 | #endif |
---|
| 451 | |
---|
| 452 | CALL cpu_log( log_point_s(7), 'fft_y', 'continue' ) |
---|
| 453 | |
---|
| 454 | !$OMP PARALLEL PRIVATE (i,iouter,ii,ir,iei,j,k,m,n,ffty_ar) |
---|
| 455 | !$OMP DO |
---|
| 456 | DO iouter = nxl_p, nxr_p, istride |
---|
| 457 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
| 458 | DO k = 1, nz |
---|
| 459 | |
---|
| 460 | m = nys_a |
---|
| 461 | DO n = 1, npe_s |
---|
| 462 | DO j = nys_p, nyn_p |
---|
| 463 | DO i = iouter, iei |
---|
| 464 | ir = i - iouter + 1 |
---|
| 465 | ffty_ar(m,ir) = work1 (i,k,j,n) |
---|
| 466 | ENDDO |
---|
| 467 | m = m+1 |
---|
| 468 | ENDDO |
---|
| 469 | ENDDO |
---|
| 470 | |
---|
| 471 | DO i = iouter, iei |
---|
| 472 | ii = nxl + i |
---|
| 473 | ir = i - iouter + 1 |
---|
| 474 | CALL fft_y( ffty_ar(:,ir), 'backward' ) |
---|
| 475 | |
---|
| 476 | DO j = nys_a, nyn_a |
---|
| 477 | ar(k,j,ii) = ffty_ar(j,ir) |
---|
| 478 | ENDDO |
---|
| 479 | ENDDO |
---|
| 480 | |
---|
| 481 | ENDDO |
---|
| 482 | ENDDO |
---|
| 483 | !$OMP END PARALLEL |
---|
| 484 | |
---|
| 485 | CALL cpu_log( log_point_s(7), 'fft_y', 'stop' ) |
---|
| 486 | |
---|
| 487 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_omp', 'stop' ) |
---|
| 488 | |
---|
| 489 | #if defined( __KKMP ) |
---|
| 490 | DEALLOCATE( tri ) |
---|
| 491 | #endif |
---|
| 492 | |
---|
| 493 | END SUBROUTINE poisfft_hybrid_omp |
---|
| 494 | |
---|
| 495 | |
---|
| 496 | SUBROUTINE poisfft_hybrid_omp_vec ( ar ) |
---|
| 497 | |
---|
| 498 | USE cpulog |
---|
| 499 | USE interfaces |
---|
| 500 | |
---|
| 501 | IMPLICIT NONE |
---|
| 502 | |
---|
| 503 | INTEGER, PARAMETER :: istride = 4 ! stride of i loop |
---|
| 504 | INTEGER :: i, ii, ir, iei, iouter, istat, j, jj, k, m, n, jthread |
---|
| 505 | |
---|
| 506 | REAL, DIMENSION(0:nx,nz) :: tri_ar |
---|
| 507 | |
---|
| 508 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
| 509 | |
---|
| 510 | REAL, DIMENSION(0:ny+3,nz,nxl_p:nxr_p) :: ffty_ar3 |
---|
| 511 | REAL, DIMENSION(0:nx+3,nz,nys_p:nyn_p) :: fftx_ar3 |
---|
| 512 | |
---|
| 513 | REAL, DIMENSION(nxl_p:nxr_p,nz,nys_p:nyn_p,npe_s) :: work1, work2 |
---|
| 514 | #if defined( __KKMP ) |
---|
| 515 | INTEGER :: omp_get_thread_num |
---|
| 516 | REAL, DIMENSION(:,:,:,:), ALLOCATABLE :: tri |
---|
| 517 | ALLOCATE( tri(5,0:nx,0:nz-1,n_omp_threads ) ) |
---|
| 518 | #else |
---|
| 519 | REAL, DIMENSION(5,0:nx,0:nz-1,1) :: tri |
---|
| 520 | #endif |
---|
| 521 | |
---|
| 522 | |
---|
| 523 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_vec', 'start' ) |
---|
| 524 | |
---|
| 525 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'start' ) |
---|
| 526 | |
---|
| 527 | !$OMP PARALLEL PRIVATE (i,j,k,m,n) |
---|
| 528 | !$OMP DO |
---|
| 529 | ! |
---|
| 530 | !-- Store grid points to be transformed on a 1d-array, do the fft |
---|
| 531 | !-- and sample the results on a 4d-array |
---|
| 532 | DO i = nxl_p, nxr_p |
---|
| 533 | |
---|
| 534 | DO j = nys_a, nyn_a |
---|
| 535 | DO k = 1, nz |
---|
| 536 | ffty_ar3(j,k,i) = ar(k,j,i+nxl) |
---|
| 537 | ENDDO |
---|
| 538 | ENDDO |
---|
| 539 | |
---|
| 540 | CALL fft_y_m( ffty_ar3(:,:,i), ny+3, 'forward' ) |
---|
| 541 | ENDDO |
---|
| 542 | |
---|
| 543 | !$OMP DO |
---|
| 544 | DO k = 1, nz |
---|
| 545 | m = nys_a |
---|
| 546 | DO n = 1, npe_s |
---|
| 547 | DO j = nys_p, nyn_p |
---|
| 548 | DO i = nxl_p, nxr_p |
---|
| 549 | work1(i,k,j,n) = ffty_ar3(m,k,i) |
---|
| 550 | ENDDO |
---|
| 551 | m = m+1 |
---|
| 552 | ENDDO |
---|
| 553 | ENDDO |
---|
| 554 | ENDDO |
---|
| 555 | !$OMP END PARALLEL |
---|
| 556 | |
---|
| 557 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'pause' ) |
---|
| 558 | |
---|
| 559 | #if defined( __parallel ) |
---|
| 560 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
| 561 | CALL MPI_ALLTOALL( work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 562 | work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 563 | comm2d, istat ) |
---|
| 564 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
| 565 | #else |
---|
| 566 | work2 = work1 |
---|
| 567 | #endif |
---|
| 568 | |
---|
| 569 | CALL cpu_log( log_point_s(33), 'fft_x_m + tridia', 'start' ) |
---|
| 570 | |
---|
| 571 | #if defined( __KKMP ) |
---|
| 572 | !$OMP PARALLEL PRIVATE (i,j,jj,k,m,n,tri_ar,jthread) |
---|
| 573 | !$OMP DO |
---|
| 574 | DO j = nys_p, nyn_p |
---|
| 575 | jthread = omp_get_thread_num() + 1 |
---|
| 576 | #else |
---|
| 577 | DO j = nys_p, nyn_p |
---|
| 578 | jthread = 1 |
---|
| 579 | #endif |
---|
| 580 | DO k = 1, nz |
---|
| 581 | |
---|
| 582 | m = nxl_a |
---|
| 583 | DO n = 1, npe_s |
---|
| 584 | DO i = nxl_p, nxr_p |
---|
| 585 | fftx_ar3(m,k,j) = work2(i,k,j,n) |
---|
| 586 | m = m+1 |
---|
| 587 | ENDDO |
---|
| 588 | ENDDO |
---|
| 589 | ENDDO |
---|
| 590 | |
---|
| 591 | CALL fft_x_m( fftx_ar3(:,:,j), 'forward' ) |
---|
| 592 | |
---|
| 593 | DO k = 1, nz |
---|
| 594 | DO i = nxl_a, nxr_a |
---|
| 595 | tri_ar(i,k) = fftx_ar3(i,k,j) |
---|
| 596 | ENDDO |
---|
| 597 | ENDDO |
---|
| 598 | |
---|
| 599 | jj = myid * (nyn_p-nys_p+1) + j |
---|
| 600 | CALL tridia_hybrid( jj, tri_ar, tri(:,:,:,jthread)) |
---|
| 601 | |
---|
| 602 | DO k = 1, nz |
---|
| 603 | DO i = nxl_a, nxr_a |
---|
| 604 | fftx_ar3(i,k,j) = tri_ar (i,k) |
---|
| 605 | ENDDO |
---|
| 606 | ENDDO |
---|
| 607 | |
---|
| 608 | CALL fft_x_m( fftx_ar3(:,:,j), 'backward' ) |
---|
| 609 | |
---|
| 610 | DO k = 1, nz |
---|
| 611 | m = nxl_a |
---|
| 612 | DO n = 1, npe_s |
---|
| 613 | DO i = nxl_p, nxr_p |
---|
| 614 | work2(i,k,j,n) = fftx_ar3(m,k,j) |
---|
| 615 | m = m+1 |
---|
| 616 | ENDDO |
---|
| 617 | ENDDO |
---|
| 618 | ENDDO |
---|
| 619 | |
---|
| 620 | ENDDO |
---|
| 621 | #if defined( __KKMP ) |
---|
| 622 | !$OMP END PARALLEL |
---|
| 623 | #endif |
---|
| 624 | |
---|
| 625 | CALL cpu_log( log_point_s(33), 'fft_x_m + tridia', 'stop' ) |
---|
| 626 | |
---|
| 627 | #if defined( __parallel ) |
---|
| 628 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
---|
| 629 | nwords = (nxr_p-nxl_p+1) * nz * (nyn_p-nys_p+1) |
---|
| 630 | CALL MPI_ALLTOALL( work2(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 631 | work1(nxl_p,1,nys_p,1), nwords, MPI_REAL, & |
---|
| 632 | comm2d, istat ) |
---|
| 633 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
---|
| 634 | #else |
---|
| 635 | work1 = work2 |
---|
| 636 | #endif |
---|
| 637 | |
---|
| 638 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'continue' ) |
---|
| 639 | |
---|
| 640 | !$OMP PARALLEL PRIVATE (i,iouter,ii,ir,iei,j,k,m,n) |
---|
| 641 | !$OMP DO |
---|
| 642 | DO k = 1, nz |
---|
| 643 | m = nys_a |
---|
| 644 | DO n = 1, npe_s |
---|
| 645 | DO j = nys_p, nyn_p |
---|
| 646 | DO i = nxl_p, nxr_p |
---|
| 647 | ffty_ar3(m,k,i) = work1(i,k,j,n) |
---|
| 648 | ENDDO |
---|
| 649 | m = m+1 |
---|
| 650 | ENDDO |
---|
| 651 | ENDDO |
---|
| 652 | ENDDO |
---|
| 653 | |
---|
| 654 | !$OMP DO |
---|
| 655 | DO i = nxl_p, nxr_p |
---|
| 656 | CALL fft_y_m( ffty_ar3(:,:,i), ny+3, 'backward' ) |
---|
| 657 | DO j = nys_a, nyn_a |
---|
| 658 | DO k = 1, nz |
---|
| 659 | ar(k,j,i+nxl) = ffty_ar3(j,k,i) |
---|
| 660 | ENDDO |
---|
| 661 | ENDDO |
---|
| 662 | ENDDO |
---|
| 663 | !$OMP END PARALLEL |
---|
| 664 | |
---|
| 665 | CALL cpu_log( log_point_s(7), 'fft_y_m', 'stop' ) |
---|
| 666 | |
---|
| 667 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_vec', 'stop' ) |
---|
| 668 | |
---|
| 669 | #if defined( __KKMP ) |
---|
| 670 | DEALLOCATE( tri ) |
---|
| 671 | #endif |
---|
| 672 | |
---|
| 673 | END SUBROUTINE poisfft_hybrid_omp_vec |
---|
| 674 | |
---|
| 675 | |
---|
| 676 | SUBROUTINE poisfft_hybrid_nodes ( ar ) |
---|
| 677 | |
---|
| 678 | USE cpulog |
---|
| 679 | USE interfaces |
---|
| 680 | |
---|
| 681 | IMPLICIT NONE |
---|
| 682 | |
---|
| 683 | INTEGER, PARAMETER :: istride = 4 ! stride of i loop |
---|
| 684 | INTEGER :: i, iei, ii, iouter, ir, istat, j, jj, k, m, & |
---|
| 685 | n, nn, nt, nw1, nw2 |
---|
| 686 | |
---|
| 687 | REAL, DIMENSION(1:nz,nys:nyn,nxl:nxr) :: ar |
---|
| 688 | |
---|
| 689 | REAL, DIMENSION(0:nx) :: fftx_ar |
---|
| 690 | REAL, DIMENSION(0:ny,istride) :: ffty_ar |
---|
| 691 | |
---|
| 692 | REAL, DIMENSION(0:nx,nz) :: tri_ar |
---|
| 693 | |
---|
| 694 | REAL, DIMENSION(nxl_p:nxr_p,nz,tasks_per_logical_node, & |
---|
| 695 | nodes,nys_p:nyn_p) :: work1,work2 |
---|
| 696 | REAL, DIMENSION(5,0:nx,0:nz-1) :: tri |
---|
| 697 | |
---|
| 698 | |
---|
| 699 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_nodes', 'start' ) |
---|
| 700 | |
---|
| 701 | CALL cpu_log( log_point_s(7), 'fft_y', 'start' ) |
---|
| 702 | |
---|
| 703 | ! |
---|
| 704 | !-- Store grid points to be transformed on a 1d-array, do the fft |
---|
| 705 | !-- and sample the results on a 4d-array |
---|
| 706 | DO iouter = nxl_p, nxr_p, istride ! stride loop, better cache |
---|
| 707 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
| 708 | DO k = 1, nz |
---|
| 709 | |
---|
| 710 | DO i = iouter, iei |
---|
| 711 | ii = nxl + i |
---|
| 712 | ir = i - iouter + 1 |
---|
| 713 | |
---|
| 714 | DO j = nys_a, nyn_a |
---|
| 715 | ffty_ar(j,ir) = ar(k,j,ii) |
---|
| 716 | ENDDO |
---|
| 717 | |
---|
| 718 | CALL fft_y( ffty_ar(:,ir), 'forward' ) |
---|
| 719 | ENDDO |
---|
| 720 | |
---|
| 721 | m = nys_a |
---|
| 722 | DO nn = 1, nodes |
---|
| 723 | DO nt = 1, tasks_per_logical_node |
---|
| 724 | DO j = nys_p, nyn_p |
---|
| 725 | DO i = iouter, iei |
---|
| 726 | ir = i - iouter + 1 |
---|
| 727 | work1(i,k,nt,nn,j) = ffty_ar(m,ir) |
---|
| 728 | ENDDO |
---|
| 729 | m = m+1 |
---|
| 730 | ENDDO |
---|
| 731 | ENDDO |
---|
| 732 | ENDDO |
---|
| 733 | |
---|
| 734 | ENDDO |
---|
| 735 | ENDDO |
---|
| 736 | |
---|
| 737 | CALL cpu_log( log_point_s(7), 'fft_y', 'pause' ) |
---|
| 738 | |
---|
| 739 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'start' ) |
---|
| 740 | nw1 = SIZE( work1, 1 ) * SIZE( work1, 2 ) |
---|
| 741 | DO nn = 1, nodes |
---|
| 742 | DO j = nys_p, nyn_p |
---|
| 743 | #if defined( __parallel ) |
---|
| 744 | CALL MPI_ALLTOALL( work1(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
| 745 | work2(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
| 746 | comm_tasks, istat ) |
---|
| 747 | #endif |
---|
| 748 | ENDDO |
---|
| 749 | ENDDO |
---|
| 750 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'stop' ) |
---|
| 751 | |
---|
| 752 | |
---|
| 753 | DO j = nys_p, nyn_p |
---|
| 754 | |
---|
| 755 | CALL cascade( 1, j, nys_p, nyn_p ) |
---|
| 756 | nw2 = nw1 * SIZE( work1, 3 ) |
---|
| 757 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'start' ) |
---|
| 758 | #if defined( __parallel ) |
---|
| 759 | CALL MPI_ALLTOALL( work2(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
| 760 | work1(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
| 761 | comm_nodes, istat ) |
---|
| 762 | #endif |
---|
| 763 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'pause' ) |
---|
| 764 | CALL cascade( 2, j, nys_p, nyn_p ) |
---|
| 765 | |
---|
| 766 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'start' ) |
---|
| 767 | DO k = 1, nz |
---|
| 768 | |
---|
| 769 | m = nxl_a |
---|
| 770 | DO nn = 1, nodes |
---|
| 771 | DO nt = 1, tasks_per_logical_node |
---|
| 772 | DO i = nxl_p, nxr_p |
---|
| 773 | fftx_ar(m) = work1(i,k,nt,nn,j) |
---|
| 774 | m = m+1 |
---|
| 775 | ENDDO |
---|
| 776 | ENDDO |
---|
| 777 | ENDDO |
---|
| 778 | |
---|
| 779 | CALL fft_x( fftx_ar, 'forward' ) |
---|
| 780 | |
---|
| 781 | DO i = nxl_a, nxr_a |
---|
| 782 | tri_ar(i,k) = fftx_ar(i) |
---|
| 783 | ENDDO |
---|
| 784 | |
---|
| 785 | ENDDO |
---|
| 786 | |
---|
| 787 | jj = myid * (nyn_p-nys_p+1) + j |
---|
| 788 | CALL tridia_hybrid( jj, tri_ar, tri(:,:,:) ) |
---|
| 789 | |
---|
| 790 | DO k = 1, nz |
---|
| 791 | DO i = nxl_a, nxr_a |
---|
| 792 | fftx_ar(i) = tri_ar(i,k) |
---|
| 793 | ENDDO |
---|
| 794 | |
---|
| 795 | CALL fft_x( fftx_ar, 'backward' ) |
---|
| 796 | |
---|
| 797 | m = nxl_a |
---|
| 798 | DO nn = 1, nodes |
---|
| 799 | DO nt = 1, tasks_per_logical_node |
---|
| 800 | DO i = nxl_p, nxr_p |
---|
| 801 | work1(i,k,nt,nn,j) = fftx_ar(m) |
---|
| 802 | m = m+1 |
---|
| 803 | ENDDO |
---|
| 804 | ENDDO |
---|
| 805 | ENDDO |
---|
| 806 | ENDDO |
---|
| 807 | |
---|
| 808 | CALL cpu_log( log_point_s(33), 'fft_x + tridia', 'stop' ) |
---|
| 809 | nw2 = nw1 * SIZE( work1, 3 ) |
---|
| 810 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'continue' ) |
---|
| 811 | #if defined( __parallel ) |
---|
| 812 | CALL MPI_ALLTOALL( work1(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
| 813 | work2(nxl_p,1,1,1,j), nw2, MPI_REAL, & |
---|
| 814 | comm_nodes, istat ) |
---|
| 815 | #endif |
---|
| 816 | CALL cpu_log( log_point_s(37), 'alltoall_node', 'stop' ) |
---|
| 817 | |
---|
| 818 | ENDDO |
---|
| 819 | |
---|
| 820 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'start' ) |
---|
| 821 | DO nn = 1, nodes |
---|
| 822 | DO j = nys_p, nyn_p |
---|
| 823 | #if defined( __parallel ) |
---|
| 824 | CALL MPI_ALLTOALL( work2(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
| 825 | work1(nxl_p,1,1,nn,j), nw1, MPI_REAL, & |
---|
| 826 | comm_tasks, istat ) |
---|
| 827 | #endif |
---|
| 828 | ENDDO |
---|
| 829 | ENDDO |
---|
| 830 | CALL cpu_log( log_point_s(32), 'alltoall_task', 'stop' ) |
---|
| 831 | |
---|
| 832 | CALL cpu_log( log_point_s(7), 'fft_y', 'continue' ) |
---|
| 833 | |
---|
| 834 | DO iouter = nxl_p, nxr_p, istride |
---|
| 835 | iei = MIN( iouter+istride-1, nxr_p ) |
---|
| 836 | DO k = 1, nz |
---|
| 837 | |
---|
| 838 | m = nys_a |
---|
| 839 | DO nn = 1, nodes |
---|
| 840 | DO nt = 1, tasks_per_logical_node |
---|
| 841 | DO j = nys_p, nyn_p |
---|
| 842 | DO i = iouter, iei |
---|
| 843 | ir = i - iouter + 1 |
---|
| 844 | ffty_ar(m,ir) = work1(i,k,nt,nn,j) |
---|
| 845 | ENDDO |
---|
| 846 | m = m+1 |
---|
| 847 | ENDDO |
---|
| 848 | ENDDO |
---|
| 849 | ENDDO |
---|
| 850 | |
---|
| 851 | DO i = iouter, iei |
---|
| 852 | ii = nxl + i |
---|
| 853 | ir = i - iouter + 1 |
---|
| 854 | CALL fft_y( ffty_ar(:,ir), 'backward' ) |
---|
| 855 | |
---|
| 856 | DO j = nys_a, nyn_a |
---|
| 857 | ar(k,j,ii) = ffty_ar(j,ir) |
---|
| 858 | ENDDO |
---|
| 859 | ENDDO |
---|
| 860 | |
---|
| 861 | ENDDO |
---|
| 862 | ENDDO |
---|
| 863 | |
---|
| 864 | CALL cpu_log( log_point_s(7), 'fft_y', 'stop' ) |
---|
| 865 | |
---|
| 866 | CALL cpu_log( log_point_s(30), 'poisfft_hybrid_nodes', 'stop' ) |
---|
| 867 | |
---|
| 868 | END SUBROUTINE poisfft_hybrid_nodes |
---|
| 869 | |
---|
| 870 | |
---|
| 871 | |
---|
| 872 | SUBROUTINE tridia_hybrid( j, ar, tri ) |
---|
| 873 | |
---|
| 874 | USE arrays_3d |
---|
| 875 | USE control_parameters |
---|
| 876 | USE grid_variables |
---|
| 877 | |
---|
| 878 | IMPLICIT NONE |
---|
| 879 | |
---|
| 880 | INTEGER :: i, j, k, nnyh |
---|
| 881 | REAL, DIMENSION(0:nx,nz) :: ar |
---|
| 882 | REAL, DIMENSION(0:nx,0:nz-1) :: ar1 |
---|
| 883 | REAL, DIMENSION(5,0:nx,0:nz-1) :: tri |
---|
| 884 | |
---|
| 885 | nnyh = (ny+1) / 2 |
---|
| 886 | |
---|
| 887 | tri = 0.0 |
---|
| 888 | ! |
---|
| 889 | !-- Define constant elements of the tridiagonal matrix. |
---|
| 890 | DO k = 0, nz-1 |
---|
| 891 | DO i = 0,nx |
---|
[667] | 892 | tri(2,i,k) = ddzu_pres(k+1) * ddzw(k+1) |
---|
| 893 | tri(3,i,k) = ddzu_pres(k+2) * ddzw(k+1) |
---|
[1] | 894 | ENDDO |
---|
| 895 | ENDDO |
---|
| 896 | |
---|
| 897 | IF ( j <= nnyh ) THEN |
---|
| 898 | CALL maketri_hybrid( j ) |
---|
| 899 | ELSE |
---|
| 900 | CALL maketri_hybrid( ny+1-j) |
---|
| 901 | ENDIF |
---|
| 902 | CALL zerleg_hybrid |
---|
| 903 | CALL substi_hybrid( ar, tri ) |
---|
| 904 | |
---|
| 905 | CONTAINS |
---|
| 906 | |
---|
| 907 | SUBROUTINE maketri_hybrid( j ) |
---|
| 908 | |
---|
| 909 | !----------------------------------------------------------------------! |
---|
| 910 | ! maketri ! |
---|
| 911 | ! ! |
---|
| 912 | ! computes the i- and j-dependent component of the matrix ! |
---|
| 913 | !----------------------------------------------------------------------! |
---|
| 914 | |
---|
| 915 | USE constants |
---|
| 916 | |
---|
| 917 | IMPLICIT NONE |
---|
| 918 | |
---|
| 919 | INTEGER :: i, j, k, nnxh |
---|
| 920 | REAL :: a, c |
---|
| 921 | |
---|
| 922 | REAL, DIMENSION(0:nx) :: l |
---|
| 923 | |
---|
| 924 | |
---|
| 925 | nnxh = (nx+1) / 2 |
---|
| 926 | ! |
---|
| 927 | !-- Provide the tridiagonal matrix for solution of the Poisson equation |
---|
| 928 | !-- in Fourier space. The coefficients are computed following the method |
---|
| 929 | !-- of Schmidt et al. (DFVLR-Mitteilung 84-15) --> departs from Stephan |
---|
| 930 | !-- Siano's original version. |
---|
| 931 | DO i = 0,nx |
---|
| 932 | IF ( i >= 0 .AND. i < nnxh ) THEN |
---|
| 933 | l(i) = 2.0 * ( 1.0 - COS( ( 2.0 * pi * i ) / & |
---|
[1013] | 934 | REAL( nx+1 ) ) ) / ( dx * dx ) + & |
---|
[1] | 935 | 2.0 * ( 1.0 - COS( ( 2.0 * pi * j ) / & |
---|
[1013] | 936 | REAL( ny+1 ) ) ) / ( dy * dy ) |
---|
[1] | 937 | ELSEIF ( i == nnxh ) THEN |
---|
| 938 | l(i) = 2.0 * ( 1.0 - COS( ( 2.0 * pi * ( nx+1-i ) ) / & |
---|
[1013] | 939 | REAL( nx+1 ) ) ) / ( dx * dx ) + & |
---|
[1] | 940 | 2.0 * ( 1.0 - COS( ( 2.0 * pi * j ) / & |
---|
[1013] | 941 | REAL(ny+1) ) ) / ( dy * dy ) |
---|
[1] | 942 | ELSE |
---|
| 943 | l(i) = 2.0 * ( 1.0 - COS( ( 2.0 * pi * ( nx+1-i ) ) / & |
---|
[1013] | 944 | REAL( nx+1 ) ) ) / ( dx * dx ) + & |
---|
[1] | 945 | 2.0 * ( 1.0 - COS( ( 2.0 * pi * j ) / & |
---|
[1013] | 946 | REAL( ny+1 ) ) ) / ( dy * dy ) |
---|
[1] | 947 | ENDIF |
---|
| 948 | ENDDO |
---|
| 949 | |
---|
| 950 | DO k = 0,nz-1 |
---|
| 951 | DO i = 0, nx |
---|
[667] | 952 | a = -1.0 * ddzu_pres(k+2) * ddzw(k+1) |
---|
| 953 | c = -1.0 * ddzu_pres(k+1) * ddzw(k+1) |
---|
[1] | 954 | tri(1,i,k) = a + c - l(i) |
---|
| 955 | ENDDO |
---|
| 956 | ENDDO |
---|
| 957 | IF ( ibc_p_b == 1 .OR. ibc_p_b == 2 ) THEN |
---|
| 958 | DO i = 0,nx |
---|
| 959 | tri(1,i,0) = tri(1,i,0) + tri(2,i,0) |
---|
| 960 | ENDDO |
---|
| 961 | ENDIF |
---|
| 962 | IF ( ibc_p_t == 1 ) THEN |
---|
| 963 | DO i = 0,nx |
---|
| 964 | tri(1,i,nz-1) = tri(1,i,nz-1) + tri(3,i,nz-1) |
---|
| 965 | ENDDO |
---|
| 966 | ENDIF |
---|
| 967 | |
---|
| 968 | END SUBROUTINE maketri_hybrid |
---|
| 969 | |
---|
| 970 | |
---|
| 971 | SUBROUTINE zerleg_hybrid |
---|
| 972 | |
---|
| 973 | !----------------------------------------------------------------------! |
---|
| 974 | ! zerleg ! |
---|
| 975 | ! ! |
---|
| 976 | ! Splitting of the tridiagonal matrix (Thomas algorithm) ! |
---|
| 977 | !----------------------------------------------------------------------! |
---|
| 978 | |
---|
| 979 | USE indices |
---|
| 980 | |
---|
| 981 | IMPLICIT NONE |
---|
| 982 | |
---|
| 983 | INTEGER :: i, k |
---|
| 984 | |
---|
| 985 | ! |
---|
| 986 | !-- Splitting |
---|
| 987 | DO i = 0, nx |
---|
| 988 | tri(4,i,0) = tri(1,i,0) |
---|
| 989 | ENDDO |
---|
| 990 | DO k = 1, nz-1 |
---|
| 991 | DO i = 0,nx |
---|
| 992 | tri(5,i,k) = tri(2,i,k) / tri(4,i,k-1) |
---|
| 993 | tri(4,i,k) = tri(1,i,k) - tri(3,i,k-1) * tri(5,i,k) |
---|
| 994 | ENDDO |
---|
| 995 | ENDDO |
---|
| 996 | |
---|
| 997 | END SUBROUTINE zerleg_hybrid |
---|
| 998 | |
---|
| 999 | SUBROUTINE substi_hybrid( ar, tri ) |
---|
| 1000 | |
---|
| 1001 | !----------------------------------------------------------------------! |
---|
| 1002 | ! substi ! |
---|
| 1003 | ! ! |
---|
| 1004 | ! Substitution (Forward and Backward) (Thomas algorithm) ! |
---|
| 1005 | !----------------------------------------------------------------------! |
---|
| 1006 | |
---|
| 1007 | IMPLICIT NONE |
---|
| 1008 | |
---|
| 1009 | INTEGER :: i, j, k |
---|
| 1010 | REAL, DIMENSION(0:nx,nz) :: ar |
---|
| 1011 | REAL, DIMENSION(0:nx,0:nz-1) :: ar1 |
---|
| 1012 | REAL, DIMENSION(5,0:nx,0:nz-1) :: tri |
---|
| 1013 | |
---|
| 1014 | ! |
---|
| 1015 | !-- Forward substitution |
---|
| 1016 | DO i = 0, nx |
---|
| 1017 | ar1(i,0) = ar(i,1) |
---|
| 1018 | ENDDO |
---|
| 1019 | DO k = 1, nz - 1 |
---|
| 1020 | DO i = 0,nx |
---|
| 1021 | ar1(i,k) = ar(i,k+1) - tri(5,i,k) * ar1(i,k-1) |
---|
| 1022 | ENDDO |
---|
| 1023 | ENDDO |
---|
| 1024 | |
---|
| 1025 | ! |
---|
| 1026 | !-- Backward substitution |
---|
| 1027 | DO i = 0,nx |
---|
| 1028 | ar(i,nz) = ar1(i,nz-1) / tri(4,i,nz-1) |
---|
| 1029 | ENDDO |
---|
| 1030 | DO k = nz-2, 0, -1 |
---|
| 1031 | DO i = 0,nx |
---|
| 1032 | ar(i,k+1) = ( ar1(i,k) - tri(3,i,k) * ar(i,k+2) ) & |
---|
| 1033 | / tri(4,i,k) |
---|
| 1034 | ENDDO |
---|
| 1035 | ENDDO |
---|
| 1036 | |
---|
| 1037 | END SUBROUTINE substi_hybrid |
---|
| 1038 | |
---|
| 1039 | END SUBROUTINE tridia_hybrid |
---|
| 1040 | |
---|
| 1041 | |
---|
| 1042 | SUBROUTINE cascade( loca, j, nys_p, nyn_p ) |
---|
| 1043 | |
---|
| 1044 | USE cpulog |
---|
| 1045 | |
---|
| 1046 | IMPLICIT NONE |
---|
| 1047 | |
---|
| 1048 | INTEGER :: ier, j, loca, nyn_p, nys_p, req, reqa(1) |
---|
| 1049 | INTEGER, SAVE :: tag = 10 |
---|
| 1050 | #if defined( __parallel ) |
---|
[415] | 1051 | INTEGER, DIMENSION(MPI_STATUS_SIZE) :: stat |
---|
| 1052 | INTEGER, DIMENSION(MPI_STATUS_SIZE,1) :: stata |
---|
[1] | 1053 | #endif |
---|
| 1054 | |
---|
| 1055 | REAL :: buf, buf1 |
---|
| 1056 | |
---|
| 1057 | |
---|
| 1058 | buf = 1.0 |
---|
| 1059 | buf1 = 1.1 |
---|
| 1060 | IF ( me_node == 0 ) THEN ! first node only |
---|
| 1061 | |
---|
| 1062 | SELECT CASE ( loca ) |
---|
| 1063 | |
---|
| 1064 | CASE ( 1 ) ! before alltoall |
---|
| 1065 | |
---|
| 1066 | IF( me_task > 0 ) THEN ! first task does not wait |
---|
| 1067 | #if defined( __parallel ) |
---|
| 1068 | CALL MPI_SENDRECV( buf, 1, MPI_REAL, me_task-1, 0, & |
---|
| 1069 | buf1, 1, MPI_REAL, me_task-1, 0, & |
---|
[415] | 1070 | comm_tasks, stat, ierr ) |
---|
[1] | 1071 | #endif |
---|
| 1072 | ELSEIF ( j > nys_p ) THEN |
---|
| 1073 | req = 0 |
---|
| 1074 | tag = MOD( tag-10, 10 ) + 10 |
---|
| 1075 | #if defined( __parallel ) |
---|
| 1076 | CALL MPI_IRECV( buf, 1, MPI_REAL, tasks_per_logical_node-1,& |
---|
| 1077 | tag, comm_tasks, req, ierr ) |
---|
| 1078 | reqa = req |
---|
[415] | 1079 | CALL MPI_WAITALL( 1, reqa, stata, ierr ) |
---|
[1] | 1080 | #endif |
---|
| 1081 | ENDIF |
---|
| 1082 | |
---|
| 1083 | CASE ( 2 ) ! after alltoall |
---|
| 1084 | |
---|
| 1085 | IF ( me_task < tasks_per_logical_node-1 ) THEN ! last task |
---|
| 1086 | #if defined( __parallel ) |
---|
| 1087 | CALL MPI_SENDRECV( buf, 1, MPI_REAL, me_task+1, 0, & |
---|
| 1088 | buf1, 1, MPI_REAL, me_task+1, 0, & |
---|
| 1089 | comm_tasks, stat, ierr) |
---|
| 1090 | #endif |
---|
| 1091 | ELSEIF ( j < nyn_p ) THEN |
---|
| 1092 | req = 0 |
---|
| 1093 | tag = MOD( tag-10, 10 ) + 10 |
---|
| 1094 | #if defined( __parallel ) |
---|
| 1095 | CALL MPI_ISEND( buf, 1, MPI_REAL, 0, tag, comm_tasks, req, & |
---|
| 1096 | ierr ) |
---|
| 1097 | #endif |
---|
| 1098 | ENDIF |
---|
| 1099 | |
---|
| 1100 | END SELECT |
---|
| 1101 | |
---|
| 1102 | ENDIF |
---|
| 1103 | |
---|
| 1104 | END SUBROUTINE cascade |
---|
[807] | 1105 | #endif |
---|
[1] | 1106 | END MODULE poisfft_hybrid_mod |
---|