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