Logo Search packages:      
Sourcecode: lam version File versions  Download package

noncontig_coll2.c

/* -*- Mode: C; c-basic-offset:4 ; -*- */
#include "mpi.h"
#include "mpio.h"  /* not necessary with MPICH 1.1.1 or HPMPI 1.4 */
#include <stdio.h>
#include <string.h>
#include <stdlib.h>

/* tests noncontiguous reads/writes using collective I/O */

/* this test is almost exactly like noncontig_coll.c with the following changes:
 *
 * . generalized file writing/reading to handle arbitrary number of processors
 * . provides the "cb_config_list" hint with several permutations of the
 *   avaliable processors.  
 *   [ makes use of code copied from ROMIO's ADIO code to collect the names of
 *   the processors ]
 */

/* we are going to muck with this later to make it evenly divisible by however many compute nodes we have */
#define STARTING_SIZE 5000

int test_file(char *filename, int mynod, int nprocs, char * cb_hosts, 
            char *msg, int verbose); 

static int cb_config_list_keyval = MPI_KEYVAL_INVALID;

#define ADIOI_Free free
#define ADIOI_Malloc malloc
#define FPRINTF fprintf
/* I have no idea what the "D" stands for; it's how things are done in adio.h 
 */ 
struct ADIO_cb_name_arrayD {   
       int refct;              
       int namect;
       char **names;
};  
typedef struct ADIO_cb_name_arrayD *ADIO_cb_name_array;



void handle_error(int errcode, char *str) 
{
      char msg[MPI_MAX_ERROR_STRING];
      int resultlen;
      MPI_Error_string(errcode, msg, &resultlen);
      fprintf(stderr, "%s: %s\n", str, msg);
      MPI_Abort(MPI_COMM_WORLD, 1);
}
 /* cb_copy_name_array() - attribute copy routine
 */
int cb_copy_name_array(MPI_Comm comm, 
                   int *keyval, 
                   void *extra, 
                   void *attr_in,
                   void **attr_out, 
                   int *flag)
{
    ADIO_cb_name_array array;

    array = (ADIO_cb_name_array) attr_in;
    array->refct++;

    *attr_out = attr_in;
    *flag = 1; /* make a copy in the new communicator */
    
    return MPI_SUCCESS;
}

/* cb_delete_name_array() - attribute destructor
 */
int cb_delete_name_array(MPI_Comm comm, 
                   int *keyval, 
                   void *attr_val, 
                   void *extra)
{
    int i;
    ADIO_cb_name_array array;

    array = (ADIO_cb_name_array) attr_val;
    array->refct--;

    if (array->refct <= 0) {
      /* time to free the structures (names, array of ptrs to names, struct) 
       */
      for (i=0; i < array->namect; i++) {
          ADIOI_Free(array->names[i]);
      }
      if (array->names != NULL) ADIOI_Free(array->names);
      ADIOI_Free(array);
    }

    return MPI_SUCCESS;
}   

/* cb_gather_name_array() - gather a list of processor names from all processes
 *                          in a communicator and store them on rank 0.
 *                   
 * This is a collective call on the communicator(s) passed in.
 *
 * Obtains a rank-ordered list of processor names from the processes in
 * "dupcomm".
 *
 * Returns 0 on success, -1 on failure.
 * 
 * NOTE: Needs some work to cleanly handle out of memory cases!  
 */
int cb_gather_name_array(MPI_Comm comm, ADIO_cb_name_array *arrayp)
{
    char my_procname[MPI_MAX_PROCESSOR_NAME], **procname = 0;
    int *procname_len = NULL, my_procname_len, *disp = NULL, i;
    int commsize, commrank, found;
    ADIO_cb_name_array array = NULL;

    if (cb_config_list_keyval == MPI_KEYVAL_INVALID) {
      MPI_Keyval_create((MPI_Copy_function *) cb_copy_name_array, 
                    (MPI_Delete_function *) cb_delete_name_array,
                    &cb_config_list_keyval, NULL);
    }
    else {
      MPI_Attr_get(comm, cb_config_list_keyval, (void *) &array, &found);
      if (found) {
          *arrayp = array;
          return 0;
      }
    }

    MPI_Comm_size(comm, &commsize);
    MPI_Comm_rank(comm, &commrank);

    MPI_Get_processor_name(my_procname, &my_procname_len);

    /* allocate space for everything */
    array = (ADIO_cb_name_array) malloc(sizeof(*array));
    if (array == NULL) {
      return -1;
    }
    array->refct = 1; 

    if (commrank == 0) {
      /* process 0 keeps the real list */
      array->namect = commsize;

      array->names = (char **) ADIOI_Malloc(sizeof(char *) * commsize);
      if (array->names == NULL) {
          return -1;
      }
      procname = array->names; /* simpler to read */

      procname_len = (int *) ADIOI_Malloc(commsize * sizeof(int));
      if (procname_len == NULL) { 
          return -1;
      }
    }
    else {
      /* everyone else just keeps an empty list as a placeholder */
      array->namect = 0;
      array->names = NULL;
    }
    /* gather lengths first */
    MPI_Gather(&my_procname_len, 1, MPI_INT, 
             procname_len, 1, MPI_INT, 0, comm);

    if (commrank == 0) {
#ifdef CB_CONFIG_LIST_DEBUG
      for (i=0; i < commsize; i++) {
          FPRINTF(stderr, "len[%d] = %d\n", i, procname_len[i]);
      }
#endif

      for (i=0; i < commsize; i++) {
          /* add one to the lengths because we need to count the
           * terminator, and we are going to use this list of lengths
           * again in the gatherv.  
           */
          procname_len[i]++;
          procname[i] = malloc(procname_len[i]);
          if (procname[i] == NULL) {
            return -1;
          }
      }
      
      /* create our list of displacements for the gatherv.  we're going
       * to do everything relative to the start of the region allocated
       * for procname[0]
       *
       * I suppose it is theoretically possible that the distance between 
       * malloc'd regions could be more than will fit in an int.  We don't
       * cover that case.
       */
      disp = malloc(commsize * sizeof(int));
      disp[0] = 0;
      for (i=1; i < commsize; i++) {
          disp[i] = (int) (procname[i] - procname[0]);
      }

    }

    /* now gather strings */
    if (commrank == 0) {
      MPI_Gatherv(my_procname, my_procname_len + 1, MPI_CHAR, 
                procname[0], procname_len, disp, MPI_CHAR,
                0, comm);
    }
    else {
      /* if we didn't do this, we would need to allocate procname[]
       * on all processes...which seems a little silly.
       */
      MPI_Gatherv(my_procname, my_procname_len + 1, MPI_CHAR, 
                NULL, NULL, NULL, MPI_CHAR, 0, comm);
    }

    if (commrank == 0) {
      /* no longer need the displacements or lengths */
      free(disp);
      free(procname_len);

#ifdef CB_CONFIG_LIST_DEBUG
      for (i=0; i < commsize; i++) {
          fprintf(stderr, "name[%d] = %s\n", i, procname[i]);
      }
#endif
    }

    /* store the attribute; we want to store SOMETHING on all processes
     * so that they can all tell if we have gone through this procedure 
     * or not for the given communicator.
     */
    MPI_Attr_put(comm, cb_config_list_keyval, array);
    *arrayp = array;
    return 0;
}

void default_str(int mynod, int len, ADIO_cb_name_array array, char *dest)
{
      char *ptr;
      int i, p;
      if (!mynod) {
          ptr = dest;
          for (i=0; i<array->namect; i++ ) {
#if 1
                /* JMS Not all systems have snprintf */
                strncpy(ptr, len, array->names[i]);
                ptr[len - 1] = '\0';
#else
                p = snprintf(ptr, len, "%s,", array->names[i]);
#endif
                ptr += p;
          }
      }
      /* chop off that last comma */
      dest[strlen(dest) - 1] = '\0';
      MPI_Bcast(dest, len, MPI_CHAR, 0, MPI_COMM_WORLD);
}
void reverse_str(int mynod, int len, ADIO_cb_name_array array, char *dest) 
{
      char *ptr;
      int i, p;
      if (!mynod) {
          ptr = dest;
          for (i=(array->namect - 1); i >= 0; i-- ) {
                p = snprintf(ptr, len, "%s,", array->names[i]);
                ptr += p;
          }
      }
      dest[strlen(dest) - 1] = '\0';
      MPI_Bcast(dest, len, MPI_CHAR, 0, MPI_COMM_WORLD);
}

void reverse_alternating_str(int mynod, int len, ADIO_cb_name_array array, char *dest)
{
      char *ptr;
      int i, p;
      if (!mynod) {
          ptr = dest;
          /* evens */
          for (i=(array->namect - 1); i>= 0; i-=2 ) {
                p = snprintf(ptr, len, "%s,", array->names[i]);
                ptr += p;
          }
          /* odds */
          for (i=(array->namect - 2); i > 0; i-=2 ) {
                p = snprintf(ptr, len, "%s,", array->names[i]);
                ptr += p;
          }
    }
      dest[strlen(dest) - 1] = '\0';
    MPI_Bcast(dest, len, MPI_CHAR, 0, MPI_COMM_WORLD);
}

void simple_shuffle_str(int mynod, int len, ADIO_cb_name_array array, char *dest)
{
      char *ptr;
      int i, p;
      if (!mynod) {
          ptr = dest;
          for (i=(array->namect / 2 ); i < array->namect; i++) {
                p = snprintf(ptr, len, "%s,", array->names[i]);
                ptr += p;
          }
          for (i=0; i < (array->namect / 2); i++ ) {
                p = snprintf(ptr, len, "%s,", array->names[i]);
                ptr += p;
          }
      }
      dest[strlen(dest) - 1] = '\0';
      MPI_Bcast(dest, len, MPI_CHAR, 0, MPI_COMM_WORLD);
}

int main(int argc, char **argv)
{
    int i, mynod, nprocs, len, errs=0, sum_errs=0, verbose=0;
    char *filename;
    char * cb_config_string;
    int cb_config_len;
    ADIO_cb_name_array array;


    MPI_Init(&argc,&argv);
    MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
    MPI_Comm_rank(MPI_COMM_WORLD, &mynod); 
    
    /* process 0 takes the file name as a command-line argument and 
   broadcasts it to other processes */
    if (!mynod) {
      i = 1;
      /* TODO: at some point, accept -v for verbose */
      while ((i < argc) && strcmp("-fname", *argv)) {
          i++;
          argv++;
      }
      if (i >= argc) {
          fprintf(stderr, "\n*#  Usage: noncontig_coll -fname filename\n\n");
          MPI_Abort(MPI_COMM_WORLD, 1);
      }
      argv++;
      len = strlen(*argv);
      filename = (char *) malloc(len+1);
      strcpy(filename, *argv);
      MPI_Bcast(&len, 1, MPI_INT, 0, MPI_COMM_WORLD);
      MPI_Bcast(filename, len+1, MPI_CHAR, 0, MPI_COMM_WORLD);
    }
    else {
      MPI_Bcast(&len, 1, MPI_INT, 0, MPI_COMM_WORLD);
      filename = (char *) malloc(len+1);
      MPI_Bcast(filename, len+1, MPI_CHAR, 0, MPI_COMM_WORLD);
    }

    /* want to hint the cb_config_list, but do so in a non-sequential way */
    cb_gather_name_array(MPI_COMM_WORLD,  &array);

    /* sanity check */
    if (!mynod) {
          if (array->namect < 2 ) {
                fprintf(stderr, "Run this test on two or more hosts\n");
                MPI_Abort(MPI_COMM_WORLD, 1);
          }
    }
    /* get space for the permuted cb_config_string */
    if (!mynod) {
          cb_config_len = 0;
          for (i=0; i < array->namect; i++) {
                /* +1: space for either a , or \0 if last */
                cb_config_len += strlen(array->names[i]) + 1;
          }
          ++cb_config_len;
    }
    MPI_Bcast(&cb_config_len, 1, MPI_INT, 0, MPI_COMM_WORLD);
    if ( (cb_config_string = malloc(cb_config_len)) == NULL ) {
          perror("malloc");
          MPI_Abort(MPI_COMM_WORLD, 1);
    }

    /* first, no hinting */
    errs += test_file(filename, mynod, nprocs, NULL, "collective w/o hinting", verbose);

    /* hint, but no change in order */
    default_str(mynod, cb_config_len, array, cb_config_string);
    errs += test_file(filename, mynod, nprocs, cb_config_string, "collective w/ hinting: default order", verbose);

    /*  reverse order */
    reverse_str(mynod, cb_config_len, array, cb_config_string); 
    errs += test_file(filename, mynod, nprocs, cb_config_string, "collective w/ hinting: reverse order", verbose);

    /* reverse, every other */
    reverse_alternating_str(mynod, cb_config_len, array, cb_config_string);
    errs += test_file(filename, mynod, nprocs, cb_config_string,"collective w/ hinting: permutation1", verbose);

    /* second half, first half */
    simple_shuffle_str(mynod, cb_config_len, array, cb_config_string);
    errs += test_file(filename, mynod, nprocs, cb_config_string, "collective w/ hinting: permutation2", verbose);

    MPI_Allreduce(&errs, &sum_errs, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
       
    if (!mynod) {
          if (sum_errs) fprintf(stderr, "Found %d error cases\n", sum_errs);
          else printf("No errors.\n");
    }
    free(filename);
    free(cb_config_string);
    MPI_Finalize();
    return 0;
}

#define SEEDER(x,y,z) ((x)*1000000 + (y) + (x)*(z))

int test_file(char *filename, int mynod, int nprocs, char * cb_hosts, char *msg, int verbose) 
{
    MPI_Datatype typevec, newtype, t[3];
    int *buf, i, b[3], errcode, errors=0;
    MPI_File fh;
    MPI_Aint d[3];
    MPI_Status status;
    int SIZE = (STARTING_SIZE/nprocs)*nprocs;
    MPI_Info info;

    buf = (int *) malloc(SIZE*sizeof(int));
    if (verbose) fprintf(stderr, "[%d/%d] caller buffer: %p\n",mynod, nprocs, buf);


    if (cb_hosts != NULL ) {
          MPI_Info_create(&info);
          MPI_Info_set(info, "cb_config_list", cb_hosts);
    } else {
          info = MPI_INFO_NULL;
    }

    MPI_Type_vector(SIZE/nprocs, 1, nprocs, MPI_INT, &typevec);

    b[0] = b[1] = b[2] = 1;
    d[0] = 0;
    d[1] = mynod*sizeof(int);
    d[2] = SIZE*sizeof(int);
    t[0] = MPI_LB;
    t[1] = typevec;
    t[2] = MPI_UB;

    MPI_Type_struct(3, b, d, t, &newtype);
    MPI_Type_commit(&newtype);
    MPI_Type_free(&typevec);

    if (!mynod) {
      if(verbose) fprintf(stderr, "\ntesting noncontiguous in memory, noncontiguous in file using collective I/O\n");
      MPI_File_delete(filename, info);
    }
    MPI_Barrier(MPI_COMM_WORLD);

    errcode = MPI_File_open(MPI_COMM_WORLD, filename, 
                MPI_MODE_CREATE | MPI_MODE_RDWR, info, &fh);
    if (errcode != MPI_SUCCESS) {
          handle_error(errcode, "MPI_File_open");
    }

    MPI_File_set_view(fh, 0, MPI_INT, newtype, "native", info);

    for (i=0; i<SIZE; i++) buf[i] = SEEDER(mynod,i,SIZE);
    MPI_File_write_all(fh, buf, 1, newtype, &status);

    MPI_Barrier(MPI_COMM_WORLD);

    for (i=0; i<SIZE; i++) buf[i] = -1;

    MPI_File_read_at_all(fh, 0, buf, 1, newtype, &status);

    /* the verification for N compute nodes is tricky. Say we have 3
     * processors.  
     * process 0 sees: 0 -1 -1 3 -1 -1 ...
     * process 1 sees: -1 34 -1 -1 37 -1 ...
     * process 2 sees: -1 -1 68 -1 -1 71 ... */

    /* verify those leading -1s exist if they should */
    for (i=0; i<mynod; i++ ) {
          if ( buf[i] != -1 ) {
                if(verbose) fprintf(stderr, "Process %d: buf is %d, should be -1\n", mynod, buf[i]);
                errors++;
          }
    }
    /* now the modulo games are hairy.  processor 0 sees real data in the 0th,
     * 3rd, 6th... elements of the buffer (assuming nprocs==3 ).  proc 1 sees
     * the data in 1st, 4th, 7th..., and proc 2 sees it in 2nd, 5th, 8th */

    for(/* 'i' set in above loop */; i<SIZE; i++) {
          if ( ((i-mynod)%nprocs) && buf[i] != -1)  {
                if(verbose) fprintf(stderr, "Process %d: buf %d is %d, should be -1\n", 
                            mynod, i, buf[i]);
                errors++;
          }
          if ( !((i-mynod)%nprocs) && buf[i] != SEEDER(mynod,i,SIZE) ) {
                if(verbose) fprintf(stderr, "Process %d: buf %d is %d, should be %d\n",
                            mynod, i, buf[i], SEEDER(mynod,i,SIZE));
                errors++;
          }
    }
    MPI_File_close(&fh);

    MPI_Barrier(MPI_COMM_WORLD);

    if (!mynod) {
      if(verbose) fprintf(stderr, "\ntesting noncontiguous in memory, contiguous in file using collective I/O\n");
      MPI_File_delete(filename, info);
    }
    MPI_Barrier(MPI_COMM_WORLD);

    MPI_File_open(MPI_COMM_WORLD, filename, MPI_MODE_CREATE | MPI_MODE_RDWR,
                  info, &fh);

    for (i=0; i<SIZE; i++) buf[i] = SEEDER(mynod,i,SIZE);
    MPI_File_write_at_all(fh, mynod*(SIZE/nprocs)*sizeof(int), buf, 1, newtype, &status);

    MPI_Barrier(MPI_COMM_WORLD);

    for (i=0; i<SIZE; i++) buf[i] = -1;

    MPI_File_read_at_all(fh, mynod*(SIZE/nprocs)*sizeof(int), buf, 1, newtype, &status);

    /* just like as above */
    for (i=0; i<mynod; i++ ) {
          if ( buf[i] != -1 ) {
                if(verbose) fprintf(stderr, "Process %d: buf is %d, should be -1\n", mynod, buf[i]);
                errors++;
          }
    }
    for(/* i set in above loop */; i<SIZE; i++) {
          if ( ((i-mynod)%nprocs) && buf[i] != -1)  {
                if(verbose) fprintf(stderr, "Process %d: buf %d is %d, should be -1\n", 
                            mynod, i, buf[i]);
                errors++;
          }
          if ( !((i-mynod)%nprocs) && buf[i] != SEEDER(mynod,i,SIZE)) {
                if(verbose) fprintf(stderr, "Process %d: buf %d is %d, should be %d\n",
                            mynod, i, buf[i], SEEDER(mynod,i,SIZE) );
                errors++;
          }
    }

    MPI_File_close(&fh);

    MPI_Barrier(MPI_COMM_WORLD);

    if (!mynod) {
      if(verbose) fprintf(stderr, "\ntesting contiguous in memory, noncontiguous in file using collective I/O\n");
      MPI_File_delete(filename, info);
    }
    MPI_Barrier(MPI_COMM_WORLD);

    MPI_File_open(MPI_COMM_WORLD, filename, MPI_MODE_CREATE | MPI_MODE_RDWR,
                  info, &fh);

    MPI_File_set_view(fh, 0, MPI_INT, newtype, "native", info);

    for (i=0; i<SIZE; i++) buf[i] = SEEDER(mynod, i, SIZE);
    MPI_File_write_all(fh, buf, SIZE, MPI_INT, &status);

    MPI_Barrier(MPI_COMM_WORLD);

    for (i=0; i<SIZE; i++) buf[i] = -1;

    MPI_File_read_at_all(fh, 0, buf, SIZE, MPI_INT, &status);

    /* same crazy checking */
    for (i=0; i<SIZE; i++) {
          if (buf[i] != SEEDER(mynod, i, SIZE)) {
            if(verbose) fprintf(stderr, "Process %d: buf %d is %d, should be %d\n", mynod, i, buf[i], SEEDER(mynod, i, SIZE));
            errors++;
          }
    }

    MPI_File_close(&fh);

    MPI_Type_free(&newtype);
    free(buf);
    if (info != MPI_INFO_NULL) MPI_Info_free(&info);
    return errors;
}

Generated by  Doxygen 1.6.0   Back to index