Newer
Older
Juan Escobar
committed
if (LUseO) call mpi_isend(a%s(0,yoffset,1),blocklength, &
MPI_DOUBLE_PRECISION,neighbour_w_rank,sendtag, &
MPI_COMM_HORIZ, requests_ew(4), ierr)
Juan Escobar
committed
recvtag = 1011
if (LUseT) call mpi_isend(a%st(1,0,0),1, &
halo_wet(level,m),neighbour_w_rank,sendtag, &
MPI_COMM_HORIZ, requests_ewT(4), ierr)
! Wait for East <-> West communication to complete
if (halo_size == 1) then
! Wait for North <-> South communication to complete
Juan Escobar
committed
if (LUseO) call mpi_waitall(4,requests_ns, MPI_STATUSES_IGNORE, ierr)
if (LUseT) call mpi_waitall(4,requests_nsT, MPI_STATUSES_IGNORE, ierr)
Juan Escobar
committed
if (LUseO) call mpi_waitall(4,requests_ew, MPI_STATUSES_IGNORE, ierr)
if (LUseT) call mpi_waitall(4,requests_ewT, MPI_STATUSES_IGNORE, ierr)
end if
if (comm_measuretime) then
call finish_timer(t_haloswap(level,m))
end if
end if
Juan Escobar
committed
end subroutine haloswap_mnh
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
!==================================================================
subroutine haloswap(level,m, & ! multigrid- and processor- level
a) ! data field
implicit none
integer, intent(in) :: level
integer, intent(in) :: m
type(scalar3d), intent(inout) :: a
integer :: a_n ! horizontal grid size
integer :: nz ! vertical grid size
integer, dimension(2) :: p_horiz
integer :: stepsize
integer :: ierr, rank, sendtag, recvtag
integer :: stat(MPI_STATUS_SIZE)
integer :: halo_size
integer :: neighbour_n_rank
integer :: neighbour_s_rank
integer :: neighbour_e_rank
integer :: neighbour_w_rank
integer :: yoffset, blocklength
integer, dimension(4) :: requests_ns
integer, dimension(4) :: requests_ew
halo_size = comm_param%halo_size
! Do nothing if we are only using one processor
if (m > 0) then
if (comm_measuretime) then
call start_timer(t_haloswap(level,m))
end if
a_n = a%ix_max-a%ix_min+1
nz = a%grid_param%nz
stepsize = 2**(pproc-m)
! Work out rank, only execute on relevant processes
call mpi_comm_rank(MPI_COMM_HORIZ, rank, ierr)
call mpi_cart_coords(MPI_COMM_HORIZ,rank,dim_horiz,p_horiz,ierr)
! Work out ranks of neighbours
! W -> E
call mpi_cart_shift(MPI_COMM_HORIZ,1, stepsize, &
neighbour_w_rank,neighbour_e_rank,ierr)
! N -> S
call mpi_cart_shift(MPI_COMM_HORIZ,0, stepsize, &
neighbour_n_rank,neighbour_s_rank,ierr)
if ( (stepsize == 1) .or. &
( (mod(p_horiz(1),stepsize) == 0) .and. &
(mod(p_horiz(2),stepsize) == 0) ) ) then
if (halo_size == 1) then
! Do not include corners in send/recv
yoffset = 1
blocklength = a_n*(nz+2)*halo_size
else
yoffset = 1-halo_size
blocklength = (a_n+2*halo_size)*(nz+2)*halo_size
end if
! Receive from north
recvtag = 2
call mpi_irecv(a%s(0,0-(halo_size-1),1),1, &
halo_ns(level,m),neighbour_n_rank,recvtag, &
MPI_COMM_HORIZ, requests_ns(1), ierr)
! Receive from south
recvtag = 3
call mpi_irecv(a%s(0,a_n+1,1),1, &
halo_ns(level,m),neighbour_s_rank,recvtag, &
MPI_COMM_HORIZ, requests_ns(2), ierr)
! Send to south
sendtag = 2
call mpi_isend(a%s(0,a_n-(halo_size-1),1),1, &
halo_ns(level,m),neighbour_s_rank,sendtag, &
MPI_COMM_HORIZ, requests_ns(3), ierr)
! Send to north
sendtag = 3
call mpi_isend(a%s(0,1,1),1, &
halo_ns(level,m),neighbour_n_rank,sendtag, &
MPI_COMM_HORIZ, requests_ns(4), ierr)
if (halo_size > 1) then
! Wait for North <-> South communication to complete
call mpi_waitall(4,requests_ns, MPI_STATUSES_IGNORE, ierr)
end if
! Receive from west
recvtag = 0
call mpi_irecv(a%s(0,yoffset,0-(halo_size-1)),blocklength, &
MPI_DOUBLE_PRECISION,neighbour_w_rank,recvtag, &
MPI_COMM_HORIZ, requests_ew(1), ierr)
! Receive from east
sendtag = 1
call mpi_irecv(a%s(0,yoffset,a_n+1),blocklength, &
MPI_DOUBLE_PRECISION,neighbour_e_rank,recvtag, &
MPI_COMM_HORIZ, requests_ew(2), ierr)
! Send to east
sendtag = 0
call mpi_isend(a%s(0,yoffset,a_n-(halo_size-1)),blocklength, &
MPI_DOUBLE_PRECISION,neighbour_e_rank,sendtag, &
MPI_COMM_HORIZ, requests_ew(3), ierr)
! Send to west
recvtag = 1
call mpi_isend(a%s(0,yoffset,1),blocklength, &
MPI_DOUBLE_PRECISION,neighbour_w_rank,sendtag, &
MPI_COMM_HORIZ, requests_ew(4), ierr)
! Wait for East <-> West communication to complete
if (halo_size == 1) then
! Wait for North <-> South communication to complete
call mpi_waitall(4,requests_ns, MPI_STATUSES_IGNORE, ierr)
end if
call mpi_waitall(4,requests_ew, MPI_STATUSES_IGNORE, ierr)
end if
if (comm_measuretime) then
call finish_timer(t_haloswap(level,m))
end if
end if
end subroutine haloswap
!==================================================================
! Collect from a(level,m) and store on less processors
! in b(level,m-1)
!
! Example for pproc-m = 1, i.e. stepsize = 2:
!
! NW (0,0) <-- NE (0,2)
!
! ^ .
! ! .
! .
! SW (2,0) SE (2,2) [send to 0,0]
!
!==================================================================
subroutine collect(level,m, & ! multigrid and processor level
a, & ! IN: data on level (level,m)
b) ! OUT: data on level (level,m-1)
implicit none
integer, intent(in) :: level
integer, intent(in) :: m
type(scalar3d), intent(in) :: a
type(scalar3d), intent(inout) :: b
integer :: a_n, b_n ! horizontal grid sizes
integer :: nz ! vertical grid size
integer, dimension(2) :: p_horiz
integer :: stepsize
integer :: ierr, source_rank, dest_rank, rank, recv_tag, send_tag, iz
logical :: corner_nw, corner_ne, corner_sw, corner_se
integer :: recv_request(3)
call start_timer(t_collect(m))
stepsize = 2**(pproc-m)
a_n = a%ix_max-a%ix_min+1
b_n = b%ix_max-b%ix_min+1
nz = b%grid_param%nz
! Work out rank, only execute on relevant processes
call mpi_comm_rank(MPI_COMM_HORIZ, rank, ierr)
! Store position in process grid in in p_horiz
! Note we can NOT use cart_shift as we need diagonal neighburs as well
call mpi_cart_coords(MPI_COMM_HORIZ,rank,dim_horiz,p_horiz,ierr)
! Ignore all processes that do not participate at this level
if ( (stepsize .eq. 1) .or. ((mod(p_horiz(1),stepsize) == 0) .and. (mod(p_horiz(2),stepsize)) == 0)) then
! Determine position in local 2x2 block
if (stepsize .eq. 1) then
corner_nw = ((mod(p_horiz(1),2) == 0) .and. (mod(p_horiz(2),2) == 0))
corner_ne = ((mod(p_horiz(1),2) == 0) .and. (mod(p_horiz(2),2) == 1))
corner_sw = ((mod(p_horiz(1),2) == 1) .and. (mod(p_horiz(2),2) == 0))
corner_se = ((mod(p_horiz(1),2) == 1) .and. (mod(p_horiz(2),2) == 1))
else
corner_nw = ((mod(p_horiz(1)/stepsize,2) == 0) .and. (mod(p_horiz(2)/stepsize,2) == 0))
corner_ne = ((mod(p_horiz(1)/stepsize,2) == 0) .and. (mod(p_horiz(2)/stepsize,2) == 1))
corner_sw = ((mod(p_horiz(1)/stepsize,2) == 1) .and. (mod(p_horiz(2)/stepsize,2) == 0))
corner_se = ((mod(p_horiz(1)/stepsize,2) == 1) .and. (mod(p_horiz(2)/stepsize,2) == 1))
end if
! NW receives from the other three processes
if ( corner_nw ) then
! Receive from NE
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1),p_horiz(2)+stepsize/), &
source_rank, &
ierr)
recv_tag = 0
call mpi_irecv(b%s(0,1,b_n/2+1),1,sub_interior(level,m-1), source_rank, recv_tag, MPI_COMM_HORIZ, &
recv_request(1),ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Collect: receive from NE failed in mpi_irecv().")
#endif
! Receive from SW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)+stepsize,p_horiz(2)/), &
source_rank, &
ierr)
recv_tag = 1
call mpi_irecv(b%s(0,b_n/2+1,1),1,sub_interior(level,m-1), source_rank, recv_tag, MPI_COMM_HORIZ, &
recv_request(2),ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Collect: receive from SW failed in mpi_irecv().")
#endif
! Receive from SE
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)+stepsize,p_horiz(2)+stepsize/), &
source_rank, &
ierr)
recv_tag = 2
call mpi_irecv(b%s(0,b_n/2+1,b_n/2+1),1,sub_interior(level,m-1), source_rank, recv_tag, MPI_COMM_HORIZ, &
recv_request(3),ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Collect: receive from SE failed in mpi_irecv().")
#endif
! Copy local data while waiting for data from other processes
do iz=0,nz+1
b%s(iz,1:a_n,1:a_n) = a%s(iz,1:a_n,1:a_n)
end do
! Wait for receives to complete before proceeding
call mpi_waitall(3,recv_request,MPI_STATUSES_IGNORE,ierr)
end if
if ( corner_ne ) then
! Send to NW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1),p_horiz(2)-stepsize/), &
dest_rank, &
ierr)
send_tag = 0
call mpi_send(a%s(0,1,1),1,interior(level,m),dest_rank,send_tag,MPI_COMM_HORIZ,ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Collect: send from NE failed in mpi_send().")
#endif
end if
if ( corner_sw ) then
! Send to NW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)-stepsize,p_horiz(2)/), &
dest_rank, &
ierr)
send_tag = 1
call mpi_send(a%s(0,1,1),1,interior(level,m),dest_rank,send_tag,MPI_COMM_HORIZ,ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Collect: send from SW failed in mpi_send().")
#endif
end if
if ( corner_se ) then
! send to NW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)-stepsize,p_horiz(2)-stepsize/), &
dest_rank, &
ierr)
send_tag = 2
call mpi_send(a%s(0,1,1),1,interior(level,m),dest_rank,send_tag,MPI_COMM_HORIZ,ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Collect: send from SE failed in mpi_send().")
#endif
end if
end if
call finish_timer(t_collect(m))
end subroutine collect
!==================================================================
! Distribute data in a(level,m-1) and store in b(level,m)
!
! Example for p-m = 1, i.e. stepsize = 2:
!
! NW (0,0) --> NE (2,0)
!
! ! .
! v .
! .
! SW (0,2) SE (2,2) [receive from to 0,0]
!==================================================================
subroutine distribute(level,m, & ! multigrid and processor level
a, & ! IN: Data on level (level,m-1)
b) ! OUT: Data on level (level,m)
implicit none
integer, intent(in) :: level
integer, intent(in) :: m
type(scalar3d), intent(in) :: a
type(scalar3d), intent(inout) :: b
integer :: a_n, b_n ! horizontal grid sizes
integer :: nz ! vertical grid size
integer, dimension(2) :: p_horiz
integer :: stepsize
integer :: ierr, source_rank, dest_rank, send_tag, recv_tag, rank, iz
integer :: stat(MPI_STATUS_SIZE)
integer :: send_request(3)
logical :: corner_nw, corner_ne, corner_sw, corner_se
call start_timer(t_distribute(m))
stepsize = 2**(pproc-m)
a_n = a%ix_max-a%ix_min+1
b_n = b%ix_max-b%ix_min+1
nz = a%grid_param%nz
! Work out rank, only execute on relevant processes
call mpi_comm_rank(MPI_COMM_HORIZ, rank, ierr)
call mpi_cart_coords(MPI_COMM_HORIZ,rank,dim_horiz,p_horiz,ierr)
! Ignore all processes that do not participate at this level
if ( (stepsize .eq. 1) .or. ((mod(p_horiz(1),stepsize) == 0) .and. (mod(p_horiz(2),stepsize)) == 0)) then
! Work out coordinates in local 2 x 2 block
if (stepsize .eq. 1) then
corner_nw = ((mod(p_horiz(1),2) == 0) .and. (mod(p_horiz(2),2) == 0))
corner_ne = ((mod(p_horiz(1),2) == 0) .and. (mod(p_horiz(2),2) == 1))
corner_sw = ((mod(p_horiz(1),2) == 1) .and. (mod(p_horiz(2),2) == 0))
corner_se = ((mod(p_horiz(1),2) == 1) .and. (mod(p_horiz(2),2) == 1))
else
corner_nw = ((mod(p_horiz(1)/stepsize,2) == 0) .and. (mod(p_horiz(2)/stepsize,2) == 0))
corner_ne = ((mod(p_horiz(1)/stepsize,2) == 0) .and. (mod(p_horiz(2)/stepsize,2) == 1))
corner_sw = ((mod(p_horiz(1)/stepsize,2) == 1) .and. (mod(p_horiz(2)/stepsize,2) == 0))
corner_se = ((mod(p_horiz(1)/stepsize,2) == 1) .and. (mod(p_horiz(2)/stepsize,2) == 1))
end if
if ( corner_nw ) then
! (Asynchronous) send to NE
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1),p_horiz(2)+stepsize/), &
dest_rank, &
ierr)
send_tag = 0
call mpi_isend(a%s(0,1,a_n/2+1), 1,sub_interior(level,m-1),dest_rank, send_tag, &
MPI_COMM_HORIZ,send_request(1),ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Distribute: send to NE failed in mpi_isend().")
#endif
! (Asynchronous) send to SW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)+stepsize,p_horiz(2)/), &
dest_rank, &
ierr)
send_tag = 1
call mpi_isend(a%s(0,a_n/2+1,1),1,sub_interior(level,m-1), dest_rank, send_tag, &
MPI_COMM_HORIZ, send_request(2), ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Distribute: send to SW failed in mpi_isend().")
#endif
! (Asynchronous) send to SE
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)+stepsize,p_horiz(2)+stepsize/), &
dest_rank, &
ierr)
send_tag = 2
call mpi_isend(a%s(0,a_n/2+1,a_n/2+1),1,sub_interior(level,m-1), dest_rank, send_tag, &
MPI_COMM_HORIZ, send_request(3), ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Distribute: send to SE failed in mpi_isend().")
#endif
! While sending, copy local data
do iz=0,nz+1
b%s(iz,1:b_n,1:b_n) = a%s(iz,1:b_n,1:b_n)
end do
! Only proceed when async sends to complete
call mpi_waitall(3, send_request, MPI_STATUSES_IGNORE, ierr)
end if
if ( corner_ne ) then
! Receive from NW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1),p_horiz(2)-stepsize/), &
source_rank, &
ierr)
recv_tag = 0
call mpi_recv(b%s(0,1,1),1,interior(level,m),source_rank,recv_tag,MPI_COMM_HORIZ,stat,ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Distribute: receive on NE failed in mpi_recv().")
#endif
end if
if ( corner_sw ) then
! Receive from NW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)-stepsize,p_horiz(2)/), &
source_rank, &
ierr)
recv_tag = 1
call mpi_recv(b%s(0,1,1),1,interior(level,m),source_rank,recv_tag,MPI_COMM_HORIZ,stat,ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Distribute: receive on SW failed in mpi_recv().")
#endif
end if
if ( corner_se ) then
! Receive from NW
call mpi_cart_rank(MPI_COMM_HORIZ, &
(/p_horiz(1)-stepsize,p_horiz(2)-stepsize/), &
source_rank, &
ierr)
recv_tag = 2
call mpi_recv(b%s(0,1,1),1,interior(level,m),source_rank,recv_tag,MPI_COMM_HORIZ,stat,ierr)
#ifndef NDEBUG
if (ierr .ne. 0) &
call fatalerror("Distribute: receive on NW failed in mpi_recv().")
#endif
end if
end if
call finish_timer(t_distribute(m))
end subroutine distribute
end module communication