| 1 | SUBROUTINE transpose_xy( f_in, work, f_out ) |
|---|
| 2 | |
|---|
| 3 | !------------------------------------------------------------------------------! |
|---|
| 4 | ! Current revisions: |
|---|
| 5 | ! ----------------- |
|---|
| 6 | ! |
|---|
| 7 | ! |
|---|
| 8 | ! Former revisions: |
|---|
| 9 | ! ----------------- |
|---|
| 10 | ! $Id: transpose.f90 684 2011-02-09 14:49:31Z suehring $ |
|---|
| 11 | ! |
|---|
| 12 | ! 683 2011-02-09 14:25:15Z raasch |
|---|
| 13 | ! openMP parallelization of transpositions for 2d-domain-decomposition |
|---|
| 14 | ! |
|---|
| 15 | ! 622 2010-12-10 08:08:13Z raasch |
|---|
| 16 | ! optional barriers included in order to speed up collective operations |
|---|
| 17 | ! |
|---|
| 18 | ! 164 2008-05-15 08:46:15Z raasch |
|---|
| 19 | ! f_inv changed from subroutine argument to automatic array in order to do |
|---|
| 20 | ! re-ordering from f_in to f_inv in one step, one array work is needed instead |
|---|
| 21 | ! of work1 and work2 |
|---|
| 22 | ! |
|---|
| 23 | ! February 2007 |
|---|
| 24 | ! RCS Log replace by Id keyword, revision history cleaned up |
|---|
| 25 | ! |
|---|
| 26 | ! Revision 1.2 2004/04/30 13:12:17 raasch |
|---|
| 27 | ! Switched from mpi_alltoallv to the simpler mpi_alltoall, |
|---|
| 28 | ! all former transpose-routine files collected in this file, enlarged |
|---|
| 29 | ! transposition arrays introduced |
|---|
| 30 | ! |
|---|
| 31 | ! Revision 1.1 2004/04/30 13:08:16 raasch |
|---|
| 32 | ! Initial revision (collection of former routines transpose_xy, transpose_xz, |
|---|
| 33 | ! transpose_yx, transpose_yz, transpose_zx, transpose_zy) |
|---|
| 34 | ! |
|---|
| 35 | ! Revision 1.1 1997/07/24 11:25:18 raasch |
|---|
| 36 | ! Initial revision |
|---|
| 37 | ! |
|---|
| 38 | ! |
|---|
| 39 | ! Description: |
|---|
| 40 | ! ------------ |
|---|
| 41 | ! Transposition of input array (f_in) from x to y. For the input array, all |
|---|
| 42 | ! elements along x reside on the same PE, while after transposition, all |
|---|
| 43 | ! elements along y reside on the same PE. |
|---|
| 44 | !------------------------------------------------------------------------------! |
|---|
| 45 | |
|---|
| 46 | USE cpulog |
|---|
| 47 | USE indices |
|---|
| 48 | USE interfaces |
|---|
| 49 | USE pegrid |
|---|
| 50 | USE transpose_indices |
|---|
| 51 | |
|---|
| 52 | IMPLICIT NONE |
|---|
| 53 | |
|---|
| 54 | INTEGER :: i, j, k, l, m, ys |
|---|
| 55 | |
|---|
| 56 | REAL :: f_in(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
|---|
| 57 | f_inv(nys_x:nyn_xa,nzb_x:nzt_xa,0:nxa), & |
|---|
| 58 | f_out(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
|---|
| 59 | work(nnx*nny*nnz) |
|---|
| 60 | |
|---|
| 61 | #if defined( __parallel ) |
|---|
| 62 | |
|---|
| 63 | ! |
|---|
| 64 | !-- Rearrange indices of input array in order to make data to be send |
|---|
| 65 | !-- by MPI contiguous |
|---|
| 66 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 67 | !$OMP DO |
|---|
| 68 | DO i = 0, nxa |
|---|
| 69 | DO k = nzb_x, nzt_xa |
|---|
| 70 | DO j = nys_x, nyn_xa |
|---|
| 71 | f_inv(j,k,i) = f_in(i,j,k) |
|---|
| 72 | ENDDO |
|---|
| 73 | ENDDO |
|---|
| 74 | ENDDO |
|---|
| 75 | !$OMP END PARALLEL |
|---|
| 76 | |
|---|
| 77 | ! |
|---|
| 78 | !-- Transpose array |
|---|
| 79 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 80 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 81 | CALL MPI_ALLTOALL( f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
|---|
| 82 | work(1), sendrecvcount_xy, MPI_REAL, & |
|---|
| 83 | comm1dy, ierr ) |
|---|
| 84 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 85 | |
|---|
| 86 | ! |
|---|
| 87 | !-- Reorder transposed array |
|---|
| 88 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, ys ) |
|---|
| 89 | !$OMP DO |
|---|
| 90 | DO l = 0, pdims(2) - 1 |
|---|
| 91 | m = l * ( nxr_ya - nxl_y + 1 ) * ( nzt_ya - nzb_y + 1 ) * & |
|---|
| 92 | ( nyn_xa - nys_x + 1 ) |
|---|
| 93 | ys = 0 + l * ( nyn_xa - nys_x + 1 ) |
|---|
| 94 | DO i = nxl_y, nxr_ya |
|---|
| 95 | DO k = nzb_y, nzt_ya |
|---|
| 96 | DO j = ys, ys + nyn_xa - nys_x |
|---|
| 97 | m = m + 1 |
|---|
| 98 | f_out(j,i,k) = work(m) |
|---|
| 99 | ENDDO |
|---|
| 100 | ENDDO |
|---|
| 101 | ENDDO |
|---|
| 102 | ENDDO |
|---|
| 103 | !$OMP END PARALLEL |
|---|
| 104 | |
|---|
| 105 | #endif |
|---|
| 106 | |
|---|
| 107 | END SUBROUTINE transpose_xy |
|---|
| 108 | |
|---|
| 109 | |
|---|
| 110 | SUBROUTINE transpose_xz( f_in, work, f_out ) |
|---|
| 111 | |
|---|
| 112 | !------------------------------------------------------------------------------! |
|---|
| 113 | ! Description: |
|---|
| 114 | ! ------------ |
|---|
| 115 | ! Transposition of input array (f_in) from x to z. For the input array, all |
|---|
| 116 | ! elements along x reside on the same PE, while after transposition, all |
|---|
| 117 | ! elements along z reside on the same PE. |
|---|
| 118 | !------------------------------------------------------------------------------! |
|---|
| 119 | |
|---|
| 120 | USE cpulog |
|---|
| 121 | USE indices |
|---|
| 122 | USE interfaces |
|---|
| 123 | USE pegrid |
|---|
| 124 | USE transpose_indices |
|---|
| 125 | |
|---|
| 126 | IMPLICIT NONE |
|---|
| 127 | |
|---|
| 128 | INTEGER :: i, j, k, l, m, xs |
|---|
| 129 | |
|---|
| 130 | REAL :: f_in(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
|---|
| 131 | f_inv(nys:nyna,nxl:nxra,1:nza), & |
|---|
| 132 | f_out(1:nza,nys:nyna,nxl:nxra), & |
|---|
| 133 | work(nnx*nny*nnz) |
|---|
| 134 | |
|---|
| 135 | #if defined( __parallel ) |
|---|
| 136 | |
|---|
| 137 | ! |
|---|
| 138 | !-- If the PE grid is one-dimensional along y, the array has only to be |
|---|
| 139 | !-- reordered locally and therefore no transposition has to be done. |
|---|
| 140 | IF ( pdims(1) /= 1 ) THEN |
|---|
| 141 | ! |
|---|
| 142 | !-- Reorder input array for transposition |
|---|
| 143 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, xs ) |
|---|
| 144 | !$OMP DO |
|---|
| 145 | DO l = 0, pdims(1) - 1 |
|---|
| 146 | m = l * ( nzt_xa - nzb_x + 1 ) * nnx * ( nyn_xa - nys_x + 1 ) |
|---|
| 147 | xs = 0 + l * nnx |
|---|
| 148 | DO k = nzb_x, nzt_xa |
|---|
| 149 | DO i = xs, xs + nnx - 1 |
|---|
| 150 | DO j = nys_x, nyn_xa |
|---|
| 151 | m = m + 1 |
|---|
| 152 | work(m) = f_in(i,j,k) |
|---|
| 153 | ENDDO |
|---|
| 154 | ENDDO |
|---|
| 155 | ENDDO |
|---|
| 156 | ENDDO |
|---|
| 157 | !$OMP END PARALLEL |
|---|
| 158 | |
|---|
| 159 | ! |
|---|
| 160 | !-- Transpose array |
|---|
| 161 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 162 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 163 | CALL MPI_ALLTOALL( work(1), sendrecvcount_zx, MPI_REAL, & |
|---|
| 164 | f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
|---|
| 165 | comm1dx, ierr ) |
|---|
| 166 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 167 | |
|---|
| 168 | ! |
|---|
| 169 | !-- Reorder transposed array in a way that the z index is in first position |
|---|
| 170 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 171 | !$OMP DO |
|---|
| 172 | DO k = 1, nza |
|---|
| 173 | DO i = nxl, nxra |
|---|
| 174 | DO j = nys, nyna |
|---|
| 175 | f_out(k,j,i) = f_inv(j,i,k) |
|---|
| 176 | ENDDO |
|---|
| 177 | ENDDO |
|---|
| 178 | ENDDO |
|---|
| 179 | !$OMP END PARALLEL |
|---|
| 180 | ELSE |
|---|
| 181 | ! |
|---|
| 182 | !-- Reorder the array in a way that the z index is in first position |
|---|
| 183 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 184 | !$OMP DO |
|---|
| 185 | DO i = nxl, nxra |
|---|
| 186 | DO j = nys, nyna |
|---|
| 187 | DO k = 1, nza |
|---|
| 188 | f_inv(j,i,k) = f_in(i,j,k) |
|---|
| 189 | ENDDO |
|---|
| 190 | ENDDO |
|---|
| 191 | ENDDO |
|---|
| 192 | !$OMP END PARALLEL |
|---|
| 193 | |
|---|
| 194 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 195 | !$OMP DO |
|---|
| 196 | DO k = 1, nza |
|---|
| 197 | DO i = nxl, nxra |
|---|
| 198 | DO j = nys, nyna |
|---|
| 199 | f_out(k,j,i) = f_inv(j,i,k) |
|---|
| 200 | ENDDO |
|---|
| 201 | ENDDO |
|---|
| 202 | ENDDO |
|---|
| 203 | !$OMP END PARALLEL |
|---|
| 204 | |
|---|
| 205 | ENDIF |
|---|
| 206 | |
|---|
| 207 | |
|---|
| 208 | #endif |
|---|
| 209 | |
|---|
| 210 | END SUBROUTINE transpose_xz |
|---|
| 211 | |
|---|
| 212 | |
|---|
| 213 | SUBROUTINE transpose_yx( f_in, work, f_out ) |
|---|
| 214 | |
|---|
| 215 | !------------------------------------------------------------------------------! |
|---|
| 216 | ! Description: |
|---|
| 217 | ! ------------ |
|---|
| 218 | ! Transposition of input array (f_in) from y to x. For the input array, all |
|---|
| 219 | ! elements along y reside on the same PE, while after transposition, all |
|---|
| 220 | ! elements along x reside on the same PE. |
|---|
| 221 | !------------------------------------------------------------------------------! |
|---|
| 222 | |
|---|
| 223 | USE cpulog |
|---|
| 224 | USE indices |
|---|
| 225 | USE interfaces |
|---|
| 226 | USE pegrid |
|---|
| 227 | USE transpose_indices |
|---|
| 228 | |
|---|
| 229 | IMPLICIT NONE |
|---|
| 230 | |
|---|
| 231 | INTEGER :: i, j, k, l, m, ys |
|---|
| 232 | |
|---|
| 233 | REAL :: f_in(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
|---|
| 234 | f_inv(nys_x:nyn_xa,nzb_x:nzt_xa,0:nxa), & |
|---|
| 235 | f_out(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
|---|
| 236 | work(nnx*nny*nnz) |
|---|
| 237 | |
|---|
| 238 | #if defined( __parallel ) |
|---|
| 239 | |
|---|
| 240 | ! |
|---|
| 241 | !-- Reorder input array for transposition |
|---|
| 242 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, ys ) |
|---|
| 243 | !$OMP DO |
|---|
| 244 | DO l = 0, pdims(2) - 1 |
|---|
| 245 | m = l * ( nxr_ya - nxl_y + 1 ) * ( nzt_ya - nzb_y + 1 ) * & |
|---|
| 246 | ( nyn_xa - nys_x + 1 ) |
|---|
| 247 | ys = 0 + l * ( nyn_xa - nys_x + 1 ) |
|---|
| 248 | DO i = nxl_y, nxr_ya |
|---|
| 249 | DO k = nzb_y, nzt_ya |
|---|
| 250 | DO j = ys, ys + nyn_xa - nys_x |
|---|
| 251 | m = m + 1 |
|---|
| 252 | work(m) = f_in(j,i,k) |
|---|
| 253 | ENDDO |
|---|
| 254 | ENDDO |
|---|
| 255 | ENDDO |
|---|
| 256 | ENDDO |
|---|
| 257 | !$OMP END PARALLEL |
|---|
| 258 | |
|---|
| 259 | ! |
|---|
| 260 | !-- Transpose array |
|---|
| 261 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 262 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 263 | CALL MPI_ALLTOALL( work(1), sendrecvcount_xy, MPI_REAL, & |
|---|
| 264 | f_inv(nys_x,nzb_x,0), sendrecvcount_xy, MPI_REAL, & |
|---|
| 265 | comm1dy, ierr ) |
|---|
| 266 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 267 | |
|---|
| 268 | ! |
|---|
| 269 | !-- Reorder transposed array in a way that the x index is in first position |
|---|
| 270 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 271 | !$OMP DO |
|---|
| 272 | DO i = 0, nxa |
|---|
| 273 | DO k = nzb_x, nzt_xa |
|---|
| 274 | DO j = nys_x, nyn_xa |
|---|
| 275 | f_out(i,j,k) = f_inv(j,k,i) |
|---|
| 276 | ENDDO |
|---|
| 277 | ENDDO |
|---|
| 278 | ENDDO |
|---|
| 279 | !$OMP END PARALLEL |
|---|
| 280 | |
|---|
| 281 | #endif |
|---|
| 282 | |
|---|
| 283 | END SUBROUTINE transpose_yx |
|---|
| 284 | |
|---|
| 285 | |
|---|
| 286 | SUBROUTINE transpose_yxd( f_in, work, f_out ) |
|---|
| 287 | |
|---|
| 288 | !------------------------------------------------------------------------------! |
|---|
| 289 | ! Description: |
|---|
| 290 | ! ------------ |
|---|
| 291 | ! Transposition of input array (f_in) from y to x. For the input array, all |
|---|
| 292 | ! elements along y reside on the same PE, while after transposition, all |
|---|
| 293 | ! elements along x reside on the same PE. |
|---|
| 294 | ! This is a direct transposition for arrays with indices in regular order |
|---|
| 295 | ! (k,j,i) (cf. transpose_yx). |
|---|
| 296 | !------------------------------------------------------------------------------! |
|---|
| 297 | |
|---|
| 298 | USE cpulog |
|---|
| 299 | USE indices |
|---|
| 300 | USE interfaces |
|---|
| 301 | USE pegrid |
|---|
| 302 | USE transpose_indices |
|---|
| 303 | |
|---|
| 304 | IMPLICIT NONE |
|---|
| 305 | |
|---|
| 306 | INTEGER :: i, j, k, l, m, recvcount_yx, sendcount_yx, xs |
|---|
| 307 | |
|---|
| 308 | REAL :: f_in(1:nza,nys:nyna,nxl:nxra), f_inv(nxl:nxra,1:nza,nys:nyna), & |
|---|
| 309 | f_out(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
|---|
| 310 | work(nnx*nny*nnz) |
|---|
| 311 | |
|---|
| 312 | #if defined( __parallel ) |
|---|
| 313 | |
|---|
| 314 | ! |
|---|
| 315 | !-- Rearrange indices of input array in order to make data to be send |
|---|
| 316 | !-- by MPI contiguous |
|---|
| 317 | DO k = 1, nza |
|---|
| 318 | DO j = nys, nyna |
|---|
| 319 | DO i = nxl, nxra |
|---|
| 320 | f_inv(i,k,j) = f_in(k,j,i) |
|---|
| 321 | ENDDO |
|---|
| 322 | ENDDO |
|---|
| 323 | ENDDO |
|---|
| 324 | |
|---|
| 325 | ! |
|---|
| 326 | !-- Transpose array |
|---|
| 327 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 328 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 329 | CALL MPI_ALLTOALL( f_inv(nxl,1,nys), sendrecvcount_xy, MPI_REAL, & |
|---|
| 330 | work(1), sendrecvcount_xy, MPI_REAL, & |
|---|
| 331 | comm1dx, ierr ) |
|---|
| 332 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 333 | |
|---|
| 334 | ! |
|---|
| 335 | !-- Reorder transposed array |
|---|
| 336 | m = 0 |
|---|
| 337 | DO l = 0, pdims(1) - 1 |
|---|
| 338 | xs = 0 + l * nnx |
|---|
| 339 | DO j = nys_x, nyn_xa |
|---|
| 340 | DO k = 1, nza |
|---|
| 341 | DO i = xs, xs + nnx - 1 |
|---|
| 342 | m = m + 1 |
|---|
| 343 | f_out(i,j,k) = work(m) |
|---|
| 344 | ENDDO |
|---|
| 345 | ENDDO |
|---|
| 346 | ENDDO |
|---|
| 347 | ENDDO |
|---|
| 348 | |
|---|
| 349 | #endif |
|---|
| 350 | |
|---|
| 351 | END SUBROUTINE transpose_yxd |
|---|
| 352 | |
|---|
| 353 | |
|---|
| 354 | SUBROUTINE transpose_yz( f_in, work, f_out ) |
|---|
| 355 | |
|---|
| 356 | !------------------------------------------------------------------------------! |
|---|
| 357 | ! Description: |
|---|
| 358 | ! ------------ |
|---|
| 359 | ! Transposition of input array (f_in) from y to z. For the input array, all |
|---|
| 360 | ! elements along y reside on the same PE, while after transposition, all |
|---|
| 361 | ! elements along z reside on the same PE. |
|---|
| 362 | !------------------------------------------------------------------------------! |
|---|
| 363 | |
|---|
| 364 | USE cpulog |
|---|
| 365 | USE indices |
|---|
| 366 | USE interfaces |
|---|
| 367 | USE pegrid |
|---|
| 368 | USE transpose_indices |
|---|
| 369 | |
|---|
| 370 | IMPLICIT NONE |
|---|
| 371 | |
|---|
| 372 | INTEGER :: i, j, k, l, m, zs |
|---|
| 373 | |
|---|
| 374 | REAL :: f_in(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
|---|
| 375 | f_inv(nxl_y:nxr_ya,nzb_y:nzt_ya,0:nya), & |
|---|
| 376 | f_out(nxl_z:nxr_za,nys_z:nyn_za,1:nza), & |
|---|
| 377 | work(nnx*nny*nnz) |
|---|
| 378 | |
|---|
| 379 | #if defined( __parallel ) |
|---|
| 380 | |
|---|
| 381 | ! |
|---|
| 382 | !-- Rearrange indices of input array in order to make data to be send |
|---|
| 383 | !-- by MPI contiguous |
|---|
| 384 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 385 | !$OMP DO |
|---|
| 386 | DO j = 0, nya |
|---|
| 387 | DO k = nzb_y, nzt_ya |
|---|
| 388 | DO i = nxl_y, nxr_ya |
|---|
| 389 | f_inv(i,k,j) = f_in(j,i,k) |
|---|
| 390 | ENDDO |
|---|
| 391 | ENDDO |
|---|
| 392 | ENDDO |
|---|
| 393 | !$OMP END PARALLEL |
|---|
| 394 | |
|---|
| 395 | ! |
|---|
| 396 | !-- Move data to different array, because memory location of work1 is |
|---|
| 397 | !-- needed further below (work1 = work2). |
|---|
| 398 | !-- If the PE grid is one-dimensional along y, only local reordering |
|---|
| 399 | !-- of the data is necessary and no transposition has to be done. |
|---|
| 400 | IF ( pdims(1) == 1 ) THEN |
|---|
| 401 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 402 | !$OMP DO |
|---|
| 403 | DO j = 0, nya |
|---|
| 404 | DO k = nzb_y, nzt_ya |
|---|
| 405 | DO i = nxl_y, nxr_ya |
|---|
| 406 | f_out(i,j,k) = f_inv(i,k,j) |
|---|
| 407 | ENDDO |
|---|
| 408 | ENDDO |
|---|
| 409 | ENDDO |
|---|
| 410 | !$OMP END PARALLEL |
|---|
| 411 | RETURN |
|---|
| 412 | ENDIF |
|---|
| 413 | |
|---|
| 414 | ! |
|---|
| 415 | !-- Transpose array |
|---|
| 416 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 417 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 418 | CALL MPI_ALLTOALL( f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
|---|
| 419 | work(1), sendrecvcount_yz, MPI_REAL, & |
|---|
| 420 | comm1dx, ierr ) |
|---|
| 421 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 422 | |
|---|
| 423 | ! |
|---|
| 424 | !-- Reorder transposed array |
|---|
| 425 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, zs ) |
|---|
| 426 | !$OMP DO |
|---|
| 427 | DO l = 0, pdims(1) - 1 |
|---|
| 428 | m = l * ( nyn_za - nys_z + 1 ) * ( nzt_ya - nzb_y + 1 ) * & |
|---|
| 429 | ( nxr_za - nxl_z + 1 ) |
|---|
| 430 | zs = 1 + l * ( nzt_ya - nzb_y + 1 ) |
|---|
| 431 | DO j = nys_z, nyn_za |
|---|
| 432 | DO k = zs, zs + nzt_ya - nzb_y |
|---|
| 433 | DO i = nxl_z, nxr_za |
|---|
| 434 | m = m + 1 |
|---|
| 435 | f_out(i,j,k) = work(m) |
|---|
| 436 | ENDDO |
|---|
| 437 | ENDDO |
|---|
| 438 | ENDDO |
|---|
| 439 | ENDDO |
|---|
| 440 | !$OMP END PARALLEL |
|---|
| 441 | |
|---|
| 442 | #endif |
|---|
| 443 | |
|---|
| 444 | END SUBROUTINE transpose_yz |
|---|
| 445 | |
|---|
| 446 | |
|---|
| 447 | SUBROUTINE transpose_zx( f_in, work, f_out ) |
|---|
| 448 | |
|---|
| 449 | !------------------------------------------------------------------------------! |
|---|
| 450 | ! Description: |
|---|
| 451 | ! ------------ |
|---|
| 452 | ! Transposition of input array (f_in) from z to x. For the input array, all |
|---|
| 453 | ! elements along z reside on the same PE, while after transposition, all |
|---|
| 454 | ! elements along x reside on the same PE. |
|---|
| 455 | !------------------------------------------------------------------------------! |
|---|
| 456 | |
|---|
| 457 | USE cpulog |
|---|
| 458 | USE indices |
|---|
| 459 | USE interfaces |
|---|
| 460 | USE pegrid |
|---|
| 461 | USE transpose_indices |
|---|
| 462 | |
|---|
| 463 | IMPLICIT NONE |
|---|
| 464 | |
|---|
| 465 | INTEGER :: i, j, k, l, m, xs |
|---|
| 466 | |
|---|
| 467 | REAL :: f_in(1:nza,nys:nyna,nxl:nxra), f_inv(nys:nyna,nxl:nxra,1:nza), & |
|---|
| 468 | f_out(0:nxa,nys_x:nyn_xa,nzb_x:nzt_xa), & |
|---|
| 469 | work(nnx*nny*nnz) |
|---|
| 470 | |
|---|
| 471 | #if defined( __parallel ) |
|---|
| 472 | |
|---|
| 473 | ! |
|---|
| 474 | !-- Rearrange indices of input array in order to make data to be send |
|---|
| 475 | !-- by MPI contiguous |
|---|
| 476 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 477 | !$OMP DO |
|---|
| 478 | DO k = 1,nza |
|---|
| 479 | DO i = nxl, nxra |
|---|
| 480 | DO j = nys, nyna |
|---|
| 481 | f_inv(j,i,k) = f_in(k,j,i) |
|---|
| 482 | ENDDO |
|---|
| 483 | ENDDO |
|---|
| 484 | ENDDO |
|---|
| 485 | !$OMP END PARALLEL |
|---|
| 486 | |
|---|
| 487 | ! |
|---|
| 488 | !-- Move data to different array, because memory location of work1 is |
|---|
| 489 | !-- needed further below (work1 = work2). |
|---|
| 490 | !-- If the PE grid is one-dimensional along y, only local reordering |
|---|
| 491 | !-- of the data is necessary and no transposition has to be done. |
|---|
| 492 | IF ( pdims(1) == 1 ) THEN |
|---|
| 493 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 494 | !$OMP DO |
|---|
| 495 | DO k = 1, nza |
|---|
| 496 | DO i = nxl, nxra |
|---|
| 497 | DO j = nys, nyna |
|---|
| 498 | f_out(i,j,k) = f_inv(j,i,k) |
|---|
| 499 | ENDDO |
|---|
| 500 | ENDDO |
|---|
| 501 | ENDDO |
|---|
| 502 | !$OMP END PARALLEL |
|---|
| 503 | RETURN |
|---|
| 504 | ENDIF |
|---|
| 505 | |
|---|
| 506 | ! |
|---|
| 507 | !-- Transpose array |
|---|
| 508 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 509 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 510 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zx, MPI_REAL, & |
|---|
| 511 | work(1), sendrecvcount_zx, MPI_REAL, & |
|---|
| 512 | comm1dx, ierr ) |
|---|
| 513 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 514 | |
|---|
| 515 | ! |
|---|
| 516 | !-- Reorder transposed array |
|---|
| 517 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, xs ) |
|---|
| 518 | !$OMP DO |
|---|
| 519 | DO l = 0, pdims(1) - 1 |
|---|
| 520 | m = l * ( nzt_xa - nzb_x + 1 ) * nnx * ( nyn_xa - nys_x + 1 ) |
|---|
| 521 | xs = 0 + l * nnx |
|---|
| 522 | DO k = nzb_x, nzt_xa |
|---|
| 523 | DO i = xs, xs + nnx - 1 |
|---|
| 524 | DO j = nys_x, nyn_xa |
|---|
| 525 | m = m + 1 |
|---|
| 526 | f_out(i,j,k) = work(m) |
|---|
| 527 | ENDDO |
|---|
| 528 | ENDDO |
|---|
| 529 | ENDDO |
|---|
| 530 | ENDDO |
|---|
| 531 | !$OMP END PARALLEL |
|---|
| 532 | |
|---|
| 533 | #endif |
|---|
| 534 | |
|---|
| 535 | END SUBROUTINE transpose_zx |
|---|
| 536 | |
|---|
| 537 | |
|---|
| 538 | SUBROUTINE transpose_zy( f_in, work, f_out ) |
|---|
| 539 | |
|---|
| 540 | !------------------------------------------------------------------------------! |
|---|
| 541 | ! Description: |
|---|
| 542 | ! ------------ |
|---|
| 543 | ! Transposition of input array (f_in) from z to y. For the input array, all |
|---|
| 544 | ! elements along z reside on the same PE, while after transposition, all |
|---|
| 545 | ! elements along y reside on the same PE. |
|---|
| 546 | !------------------------------------------------------------------------------! |
|---|
| 547 | |
|---|
| 548 | USE cpulog |
|---|
| 549 | USE indices |
|---|
| 550 | USE interfaces |
|---|
| 551 | USE pegrid |
|---|
| 552 | USE transpose_indices |
|---|
| 553 | |
|---|
| 554 | IMPLICIT NONE |
|---|
| 555 | |
|---|
| 556 | INTEGER :: i, j, k, l, m, zs |
|---|
| 557 | |
|---|
| 558 | REAL :: f_in(nxl_z:nxr_za,nys_z:nyn_za,1:nza), & |
|---|
| 559 | f_inv(nxl_y:nxr_ya,nzb_y:nzt_ya,0:nya), & |
|---|
| 560 | f_out(0:nya,nxl_y:nxr_ya,nzb_y:nzt_ya), & |
|---|
| 561 | work(nnx*nny*nnz) |
|---|
| 562 | |
|---|
| 563 | #if defined( __parallel ) |
|---|
| 564 | |
|---|
| 565 | ! |
|---|
| 566 | !-- If the PE grid is one-dimensional along y, the array has only to be |
|---|
| 567 | !-- reordered locally and therefore no transposition has to be done. |
|---|
| 568 | IF ( pdims(1) /= 1 ) THEN |
|---|
| 569 | ! |
|---|
| 570 | !-- Reorder input array for transposition |
|---|
| 571 | !$OMP PARALLEL PRIVATE ( i, j, k, l, m, zs ) |
|---|
| 572 | !$OMP DO |
|---|
| 573 | DO l = 0, pdims(1) - 1 |
|---|
| 574 | m = l * ( nyn_za - nys_z + 1 ) * ( nzt_ya - nzb_y + 1 ) * & |
|---|
| 575 | ( nxr_za - nxl_z + 1 ) |
|---|
| 576 | zs = 1 + l * ( nzt_ya - nzb_y + 1 ) |
|---|
| 577 | DO j = nys_z, nyn_za |
|---|
| 578 | DO k = zs, zs + nzt_ya - nzb_y |
|---|
| 579 | DO i = nxl_z, nxr_za |
|---|
| 580 | m = m + 1 |
|---|
| 581 | work(m) = f_in(i,j,k) |
|---|
| 582 | ENDDO |
|---|
| 583 | ENDDO |
|---|
| 584 | ENDDO |
|---|
| 585 | ENDDO |
|---|
| 586 | !$OMP END PARALLEL |
|---|
| 587 | |
|---|
| 588 | ! |
|---|
| 589 | !-- Transpose array |
|---|
| 590 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 591 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 592 | CALL MPI_ALLTOALL( work(1), sendrecvcount_yz, MPI_REAL, & |
|---|
| 593 | f_inv(nxl_y,nzb_y,0), sendrecvcount_yz, MPI_REAL, & |
|---|
| 594 | comm1dx, ierr ) |
|---|
| 595 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 596 | |
|---|
| 597 | ! |
|---|
| 598 | !-- Reorder transposed array in a way that the y index is in first position |
|---|
| 599 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 600 | !$OMP DO |
|---|
| 601 | DO j = 0, nya |
|---|
| 602 | DO k = nzb_y, nzt_ya |
|---|
| 603 | DO i = nxl_y, nxr_ya |
|---|
| 604 | f_out(j,i,k) = f_inv(i,k,j) |
|---|
| 605 | ENDDO |
|---|
| 606 | ENDDO |
|---|
| 607 | ENDDO |
|---|
| 608 | !$OMP END PARALLEL |
|---|
| 609 | ELSE |
|---|
| 610 | ! |
|---|
| 611 | !-- Reorder the array in a way that the y index is in first position |
|---|
| 612 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 613 | !$OMP DO |
|---|
| 614 | DO k = nzb_y, nzt_ya |
|---|
| 615 | DO j = 0, nya |
|---|
| 616 | DO i = nxl_y, nxr_ya |
|---|
| 617 | f_inv(i,k,j) = f_in(i,j,k) |
|---|
| 618 | ENDDO |
|---|
| 619 | ENDDO |
|---|
| 620 | ENDDO |
|---|
| 621 | !$OMP END PARALLEL |
|---|
| 622 | ! |
|---|
| 623 | !-- Move data to output array |
|---|
| 624 | !$OMP PARALLEL PRIVATE ( i, j, k ) |
|---|
| 625 | !$OMP DO |
|---|
| 626 | DO k = nzb_y, nzt_ya |
|---|
| 627 | DO i = nxl_y, nxr_ya |
|---|
| 628 | DO j = 0, nya |
|---|
| 629 | f_out(j,i,k) = f_inv(i,k,j) |
|---|
| 630 | ENDDO |
|---|
| 631 | ENDDO |
|---|
| 632 | ENDDO |
|---|
| 633 | !$OMP END PARALLEL |
|---|
| 634 | |
|---|
| 635 | ENDIF |
|---|
| 636 | |
|---|
| 637 | #endif |
|---|
| 638 | |
|---|
| 639 | END SUBROUTINE transpose_zy |
|---|
| 640 | |
|---|
| 641 | |
|---|
| 642 | SUBROUTINE transpose_zyd( f_in, work, f_out ) |
|---|
| 643 | |
|---|
| 644 | !------------------------------------------------------------------------------! |
|---|
| 645 | ! Description: |
|---|
| 646 | ! ------------ |
|---|
| 647 | ! Transposition of input array (f_in) from z to y. For the input array, all |
|---|
| 648 | ! elements along z reside on the same PE, while after transposition, all |
|---|
| 649 | ! elements along y reside on the same PE. |
|---|
| 650 | ! This is a direct transposition for arrays with indices in regular order |
|---|
| 651 | ! (k,j,i) (cf. transpose_zy). |
|---|
| 652 | !------------------------------------------------------------------------------! |
|---|
| 653 | |
|---|
| 654 | USE cpulog |
|---|
| 655 | USE indices |
|---|
| 656 | USE interfaces |
|---|
| 657 | USE pegrid |
|---|
| 658 | USE transpose_indices |
|---|
| 659 | |
|---|
| 660 | IMPLICIT NONE |
|---|
| 661 | |
|---|
| 662 | INTEGER :: i, j, k, l, m, ys |
|---|
| 663 | |
|---|
| 664 | REAL :: f_in(1:nza,nys:nyna,nxl:nxra), f_inv(nys:nyna,nxl:nxra,1:nza), & |
|---|
| 665 | f_out(0:nya,nxl_yd:nxr_yda,nzb_yd:nzt_yda), & |
|---|
| 666 | work(nnx*nny*nnz) |
|---|
| 667 | |
|---|
| 668 | #if defined( __parallel ) |
|---|
| 669 | |
|---|
| 670 | ! |
|---|
| 671 | !-- Rearrange indices of input array in order to make data to be send |
|---|
| 672 | !-- by MPI contiguous |
|---|
| 673 | DO i = nxl, nxra |
|---|
| 674 | DO j = nys, nyna |
|---|
| 675 | DO k = 1, nza |
|---|
| 676 | f_inv(j,i,k) = f_in(k,j,i) |
|---|
| 677 | ENDDO |
|---|
| 678 | ENDDO |
|---|
| 679 | ENDDO |
|---|
| 680 | |
|---|
| 681 | ! |
|---|
| 682 | !-- Move data to different array, because memory location of work1 is |
|---|
| 683 | !-- needed further below (work1 = work2). |
|---|
| 684 | !-- If the PE grid is one-dimensional along x, only local reordering |
|---|
| 685 | !-- of the data is necessary and no transposition has to be done. |
|---|
| 686 | IF ( pdims(2) == 1 ) THEN |
|---|
| 687 | DO k = 1, nza |
|---|
| 688 | DO i = nxl, nxra |
|---|
| 689 | DO j = nys, nyna |
|---|
| 690 | f_out(j,i,k) = f_inv(j,i,k) |
|---|
| 691 | ENDDO |
|---|
| 692 | ENDDO |
|---|
| 693 | ENDDO |
|---|
| 694 | RETURN |
|---|
| 695 | ENDIF |
|---|
| 696 | |
|---|
| 697 | ! |
|---|
| 698 | !-- Transpose array |
|---|
| 699 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'start' ) |
|---|
| 700 | IF ( collective_wait ) CALL MPI_BARRIER( comm2d, ierr ) |
|---|
| 701 | CALL MPI_ALLTOALL( f_inv(nys,nxl,1), sendrecvcount_zyd, MPI_REAL, & |
|---|
| 702 | work(1), sendrecvcount_zyd, MPI_REAL, & |
|---|
| 703 | comm1dy, ierr ) |
|---|
| 704 | CALL cpu_log( log_point_s(32), 'mpi_alltoall', 'stop' ) |
|---|
| 705 | |
|---|
| 706 | ! |
|---|
| 707 | !-- Reorder transposed array |
|---|
| 708 | m = 0 |
|---|
| 709 | DO l = 0, pdims(2) - 1 |
|---|
| 710 | ys = 0 + l * nny |
|---|
| 711 | DO k = nzb_yd, nzt_yda |
|---|
| 712 | DO i = nxl_yd, nxr_yda |
|---|
| 713 | DO j = ys, ys + nny - 1 |
|---|
| 714 | m = m + 1 |
|---|
| 715 | f_out(j,i,k) = work(m) |
|---|
| 716 | ENDDO |
|---|
| 717 | ENDDO |
|---|
| 718 | ENDDO |
|---|
| 719 | ENDDO |
|---|
| 720 | |
|---|
| 721 | #endif |
|---|
| 722 | |
|---|
| 723 | END SUBROUTINE transpose_zyd |
|---|
