src/external/thread_mpi/src/scatter.c

   1 /*
   2    This source code file is part of thread_mpi.
   3    Written by Sander Pronk, Erik Lindahl, and possibly others.
   4
   5    Copyright (c) 2009,2016, Sander Pronk, Erik Lindahl.
   6    All rights reserved.
   7
   8    Redistribution and use in source and binary forms, with or without
   9    modification, are permitted provided that the following conditions are met:
  10    1) Redistributions of source code must retain the above copyright
  11    notice, this list of conditions and the following disclaimer.
  12    2) Redistributions in binary form must reproduce the above copyright
  13    notice, this list of conditions and the following disclaimer in the
  14    documentation and/or other materials provided with the distribution.
  15    3) Neither the name of the copyright holders nor the
  16    names of its contributors may be used to endorse or promote products
  17    derived from this software without specific prior written permission.
  18
  19    THIS SOFTWARE IS PROVIDED BY US ''AS IS'' AND ANY
  20    EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  21    WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  22    DISCLAIMED. IN NO EVENT SHALL WE BE LIABLE FOR ANY
  23    DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
  24    (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  25    LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
  26    ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  28    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29
  30    If you want to redistribute modifications, please consider that
  31    scientific software is very special. Version control is crucial -
  32    bugs must be traceable. We will be happy to consider code for
  33    inclusion in the official distribution, but derived work should not
  34    be called official thread_mpi. Details are found in the README & COPYING
  35    files.
  36  */
  37
  38 #ifdef HAVE_TMPI_CONFIG_H
  39 #include "tmpi_config.h"
  40 #endif
  41
  42 #ifdef HAVE_CONFIG_H
  43 #include "config.h"
  44 #endif
  45
  46
  47 #ifdef HAVE_UNISTD_H
  48 #include <unistd.h>
  49 #endif
  50
  51 #include <errno.h>
  52 #include <stdlib.h>
  53 #include <stdio.h>
  54 #include <stdarg.h>
  55 #include <string.h>
  56
  57 #include "impl.h"
  58 #include "collective.h"
  59
  60
  61
  62 int tMPI_Scatter(const void* sendbuf, int sendcount, tMPI_Datatype sendtype,
  63                  void* recvbuf, int recvcount, tMPI_Datatype recvtype,
  64                  int root, tMPI_Comm comm)
  65 {
  66     int                 synct;
  67     struct coll_env    *cev;
  68     int                 myrank;
  69     int                 ret = TMPI_SUCCESS;
  70     struct tmpi_thread *cur = tMPI_Get_current();
  71
  72 #ifdef TMPI_PROFILE
  73     tMPI_Profile_count_start(cur);
  74 #endif
  75 #ifdef TMPI_TRACE
  76     tMPI_Trace_print("tMPI_Scatter(%p, %d, %p, %p, %d, %p, %d, %p)",
  77                      sendbuf, sendcount, sendtype,
  78                      recvbuf, recvcount, recvtype, root, comm);
  79 #endif
  80     if (!comm)
  81     {
  82         return tMPI_Error(TMPI_COMM_WORLD, TMPI_ERR_COMM);
  83     }
  84     myrank = tMPI_Comm_seek_rank(comm, cur);
  85
  86     /* we increase our counter, and determine which coll_env we get */
  87     cev = tMPI_Get_cev(comm, myrank, &synct);
  88
  89     if (myrank == root)
  90     {
  91         int       i;
  92         size_t    sendsize        = sendtype->size*sendcount;
  93         size_t    total_send_size = 0;
  94 #ifdef USE_COLLECTIVE_COPY_BUFFER
  95         tmpi_bool using_cb;
  96 #endif
  97
  98         if (!sendbuf) /* don't do pointer arithmetic on a NULL ptr */
  99         {
 100             return tMPI_Error(comm, TMPI_ERR_BUF);
 101         }
 102
 103         /* we set up multiple posts, so no Post_multi */
 104         cev->met[myrank].tag      = TMPI_SCATTER_TAG;
 105         cev->met[myrank].datatype = sendtype;
 106         tMPI_Atomic_memory_barrier_rel();
 107         tMPI_Atomic_set( &(cev->met[myrank].n_remaining), cev->N-1 );
 108         for (i = 0; i < comm->grp.N; i++)
 109         {
 110             total_send_size            += sendtype->size*sendcount;
 111             cev->met[myrank].bufsize[i] = sendsize;
 112             cev->met[myrank].buf[i]     = (char*)sendbuf+sendsize*i;
 113         }
 114 #ifdef USE_COLLECTIVE_COPY_BUFFER
 115         /* we must copy our own data too, unfortunately (otherwise there's
 116            a hole) */
 117         using_cb                  = (total_send_size < (size_t)((cev->N)*COPY_BUFFER_SIZE));
 118         cev->met[myrank].using_cb = using_cb;
 119         if (using_cb)
 120         {
 121             /* we set cpbuf stuff to NULL initially */
 122             for (i = 0; i < cev->N; i++)
 123             {
 124                 /*cev->met[myrank].cpbuf[i]=NULL;*/
 125                 tMPI_Atomic_ptr_set(&(cev->met[myrank].cpbuf[i]), NULL);
 126             }
 127             tMPI_Atomic_set(&(cev->met[myrank].buf_readcount), 0);
 128         }
 129 #endif
 130
 131         /* post availability */
 132         for (i = 0; i < cev->N; i++)
 133         {
 134             if (i != myrank)
 135             {
 136                 tMPI_Event_signal( &(cev->met[i].recv_ev) );
 137             }
 138         }
 139
 140
 141
 142
 143 #ifdef USE_COLLECTIVE_COPY_BUFFER
 144         /* do the copy_buffer thing */
 145         if (using_cb)
 146         {
 147             /* copy the buffer locally. First allocate */
 148             cev->met[myrank].cb = tMPI_Copy_buffer_list_get(
 149                         &(tMPI_Get_thread(comm, myrank)->cbl_multi));
 150             if (cev->met[myrank].cb->size < total_send_size)
 151             {
 152                 return tMPI_Error(comm, TMPI_ERR_COPY_BUFFER_SIZE);
 153             }
 154             /* copy to the new buf */
 155             memcpy(cev->met[myrank].cb->buf, sendbuf, total_send_size);
 156
 157             /* post the new buf */
 158             for (i = 0; i < cev->N; i++)
 159             {
 160                 tMPI_Atomic_memory_barrier_rel();
 161                 tMPI_Atomic_ptr_set(&(cev->met[myrank].cpbuf[i]),
 162                                     (char*)cev->met[myrank].cb->buf+sendsize*i);
 163                 /*cev->met[myrank].cpbuf[i] = (char*)cev->met[myrank].cb->buf +
 164                                             sendsize*i ;*/
 165             }
 166         }
 167 #endif
 168
 169         /* do root transfer */
 170         if (recvbuf != TMPI_IN_PLACE)
 171         {
 172             tMPI_Coll_root_xfer(comm, sendtype, recvtype,
 173                                 sendsize, recvtype->size*recvcount,
 174                                 (char*)sendbuf+sendsize*myrank,
 175                                 recvbuf, &ret);
 176         }
 177
 178         /* and wait until everybody is done copying */
 179         tMPI_Wait_for_others(cev, myrank);
 180     }
 181     else
 182     {
 183         /* get the root cev */
 184         size_t bufsize = recvcount*recvtype->size;
 185         /* wait until root becomes available */
 186         tMPI_Wait_for_data(cur, cev, myrank);
 187         tMPI_Mult_recv(comm, cev, root, myrank, TMPI_SCATTER_TAG, recvtype,
 188                        bufsize, recvbuf, &ret);
 189     }
 190 #ifdef TMPI_PROFILE
 191     tMPI_Profile_count_stop(cur, TMPIFN_Scatter);
 192 #endif
 193     return ret;
 194 }
 195
 196
 197
 198 int tMPI_Scatterv(const void* sendbuf, int *sendcounts, int *displs,
 199                   tMPI_Datatype sendtype, void* recvbuf, int recvcount,
 200                   tMPI_Datatype recvtype, int root, tMPI_Comm comm)
 201 {
 202     int                 synct;
 203     struct coll_env    *cev;
 204     int                 myrank;
 205     int                 ret = TMPI_SUCCESS;
 206     struct tmpi_thread *cur = tMPI_Get_current();
 207
 208 #ifdef TMPI_TRACE
 209     tMPI_Trace_print("tMPI_Scatterv(%p, %p, %p, %p, %p, %d, %p, %d, %p)",
 210                      sendbuf, sendcounts, displs, sendtype, recvbuf,
 211                      recvcount, recvtype, root, comm);
 212 #endif
 213 #ifdef TMPI_PROFILE
 214     tMPI_Profile_count_start(cur);
 215 #endif
 216
 217
 218     if (!comm)
 219     {
 220         return tMPI_Error(TMPI_COMM_WORLD, TMPI_ERR_COMM);
 221     }
 222     myrank = tMPI_Comm_seek_rank(comm, cur);
 223
 224     /* we increase our counter, and determine which coll_env we get */
 225     cev = tMPI_Get_cev(comm, myrank, &synct);
 226
 227     if (myrank == root)
 228     {
 229         int       i;
 230         size_t    total_send_size = 0;
 231 #ifdef USE_COLLECTIVE_COPY_BUFFER
 232         tmpi_bool using_cb;
 233 #endif
 234
 235         if (!sendbuf) /* don't do pointer arithmetic on a NULL ptr */
 236         {
 237             return tMPI_Error(comm, TMPI_ERR_BUF);
 238         }
 239
 240         /* we set up multiple posts, so no Post_multi */
 241         cev->met[myrank].tag      = TMPI_SCATTERV_TAG;
 242         cev->met[myrank].datatype = sendtype;
 243         tMPI_Atomic_memory_barrier_rel();
 244         tMPI_Atomic_set( &(cev->met[myrank].n_remaining), cev->N-1 );
 245         for (i = 0; i < cev->N; i++)
 246         {
 247             total_send_size            += sendtype->size*sendcounts[i];
 248             cev->met[myrank].bufsize[i] = sendtype->size*sendcounts[i];
 249             cev->met[myrank].buf[i]     = (char*)sendbuf+sendtype->size*displs[i];
 250         }
 251 #ifdef USE_COLLECTIVE_COPY_BUFFER
 252         /* we must copy our own data too, unfortunately (otherwise there's
 253            a hole) */
 254         using_cb                  = (total_send_size < (size_t)((cev->N)*COPY_BUFFER_SIZE));
 255         cev->met[myrank].using_cb = using_cb;
 256         if (using_cb)
 257         {
 258             /* we set cpbuf stuff to NULL initially */
 259             for (i = 0; i < cev->N; i++)
 260             {
 261                 /*cev->met[myrank].cpbuf[i]=NULL;*/
 262                 tMPI_Atomic_ptr_set(&(cev->met[myrank].cpbuf[i]), NULL);
 263             }
 264             tMPI_Atomic_set(&(cev->met[myrank].buf_readcount), 0);
 265         }
 266 #endif
 267
 268         /* post availability */
 269         for (i = 0; i < cev->N; i++)
 270         {
 271             if (i != myrank)
 272             {
 273                 tMPI_Event_signal( &(cev->met[i].recv_ev) );
 274             }
 275         }
 276
 277
 278
 279 #ifdef USE_COLLECTIVE_COPY_BUFFER
 280         /* do the copy_buffer thing */
 281         if (using_cb)
 282         {
 283             /* copy the buffer locally. First allocate */
 284             cev->met[myrank].cb = tMPI_Copy_buffer_list_get(
 285                         &(tMPI_Get_thread(comm, myrank)->cbl_multi));
 286             if (cev->met[myrank].cb->size < total_send_size)
 287             {
 288                 return tMPI_Error(comm, TMPI_ERR_COPY_BUFFER_SIZE);
 289             }
 290             /* copy to the new buf */
 291             memcpy(cev->met[myrank].cb->buf, sendbuf, total_send_size);
 292             /* post the new buf */
 293             for (i = 0; i < cev->N; i++)
 294             {
 295                 tMPI_Atomic_memory_barrier_rel();
 296                 tMPI_Atomic_ptr_set(&(cev->met[myrank].cpbuf[i]),
 297                                     (char*)cev->met[myrank].cb->buf +
 298                                     sendtype->size*displs[i]);
 299                 /*cev->met[myrank].cpbuf[i]=(char*)cev->met[myrank].cb->buf +
 300                           sendtype->size*displs[i];*/
 301             }
 302         }
 303 #endif
 304
 305         /* do root transfer */
 306         if (recvbuf != TMPI_IN_PLACE)
 307         {
 308             tMPI_Coll_root_xfer(comm, sendtype, recvtype,
 309                                 sendtype->size*sendcounts[myrank],
 310                                 recvtype->size*recvcount,
 311                                 (char*)sendbuf+sendtype->size*displs[myrank],
 312                                 recvbuf,
 313                                 &ret);
 314         }
 315
 316         /* and wait until everybody is done copying */
 317         tMPI_Wait_for_others(cev, myrank);
 318     }
 319     else
 320     {
 321         /* get the root cev */
 322         size_t bufsize = recvcount*recvtype->size;
 323         /* wait until root becomes available */
 324         tMPI_Wait_for_data(cur, cev, myrank);
 325         tMPI_Mult_recv(comm, cev, root, myrank, TMPI_SCATTERV_TAG,
 326                        recvtype, bufsize, recvbuf, &ret);
 327     }
 328 #ifdef TMPI_PROFILE
 329     tMPI_Profile_count_stop(cur, TMPIFN_Scatterv);
 330 #endif
 331     return ret;
 332 }