main()

int main	(	int	argc,
		char *	argv[]
	)

Definition at line 243 of file pzgemrdrv.c.

{
  /* We initialize the data-block on the current processor, then redistribute
   * it, and perform the inverse redistribution  to compare the local memory
   * with the initial one. */
  /* Data file */
  FILE *fp;
  Int   nbre, nbremax;
  /* Data distribution 0 parameters */
  Int   p0,     /* # of rows in the processor grid */
        q0;     /* # of columns in the processor grid */
  /* Data distribution 1 parameters */
  Int   p1, q1;
  /* # of parameter to be read on the keyboard */
#define nbparameter 24
  /* General variables */
  Int   blocksize0;
  Int   mypnum, nprocs;
  Int   parameters[nbparameter], nberrors;
  Int   i;
  Int   ia, ja, ib, jb, m, n;
  Int   gcontext, context0, context1;
  Int   myprow1, myprow0, mypcol0, mypcol1;
  Int   dummy;
  MDESC ma, mb;
  dcomplex *ptrmyblock, *ptrsavemyblock, *ptrmyblockcopy, *ptrmyblockvide;
#ifdef UsingMpiBlacs
   MPI_Init(&argc, &argv);
#endif
  setvbuf(stdout, NULL, _IOLBF, 0);
  setvbuf(stderr, NULL, _IOLBF, 0);
#ifdef T3D
  free(malloc(14000000));
#endif
  initforpvm(argc, argv);
  /* Read physical parameters */
  Cblacs_pinfo(&mypnum, &nprocs);
  /* initialize BLACS for the parameter communication */
  Cblacs_get((Int)0, (Int)0, &gcontext);
  Cblacs_gridinit(&gcontext, "R", nprocs, (Int)1);
  Cblacs_gridinfo(gcontext, &dummy, &dummy, &mypnum, &dummy);
  if (mypnum == 0) {
    if ((fp = fopen("GEMR2D.dat", "r")) == NULL) {
      fprintf(stderr, "Can't open GEMR2D.dat\n");
      exit(1);
    };
    printf("\n// ZGEMR2D TESTER for COMPLEX*16 //\n");
    getparam(fp, &nbre, NULL);
    printf("////////// %d tests \n\n", nbre);
    parameters[0] = nbre;
    Cigebs2d(gcontext, "All", "H", (Int)1, (Int)1, parameters, (Int)1);
  } else {
    Cigebr2d(gcontext, "All", "H", (Int)1, (Int)1, parameters, (Int)1, (Int)0, (Int)0);
    nbre = parameters[0];
  };
  if (mypnum == 0) {
    printf("\n  m   n  m0  n0  sr0 sc0 i0  j0  p0  q0 nbr0 nbc0 \
m1  n1  sr1 sc1 i1  j1  p1  q1 nbr1 nbc1\n\n");
  };
  /****** TEST LOOP *****/
  /* Here we are in grip 1xnprocs */
  nbremax = nbre;
#ifdef DEBUG
  fprintf(stderr, "bonjour,je suis le noeud %d\n", mypnum);
#endif
  while (nbre-- != 0) { /* Loop on the serie of tests */
    /* All the processors read the parameters so we have to be in a 1xnprocs
     * grid at each iteration */
    /* Read processors grid and matrices parameters */
    if (mypnum == 0) {
      Int   u, d;
      getparam(fp,
               &m, &n,
               &ma.m, &ma.n, &ma.sprow, &ma.spcol,
               &ia, &ja, &p0, &q0, &ma.nbrow, &ma.nbcol,
               &mb.m, &mb.n, &mb.sprow, &mb.spcol,
               &ib, &jb, &p1, &q1, &mb.nbrow, &mb.nbcol,
               NULL);
      printf("\t\t************* TEST # %d **********\n",
             nbremax - nbre);
      printf(" %3d %3d %3d %3d %3d %3d %3d %3d \
%3d %3d %3d %3d %3d %3d %3d %3d %3d %3d %3d %3d %3d %3d",
             m, n,
             ma.m, ma.n, ma.sprow, ma.spcol,
             ia, ja, p0, q0, ma.nbrow, ma.nbcol,
             mb.m, mb.n, mb.sprow, mb.spcol,
             ib, jb, p1, q1, mb.nbrow, mb.nbcol);
      printf("\n");
      if (p0 * q0 > nprocs || p1 * q1 > nprocs) {
        fprintf(stderr, "not enough nodes:%d processors required\n",
                max(p0 * q0, p1 * q1));
        exit(1);
      }
      parameters[0] = p0;
      parameters[1] = q0;
      parameters[2] = ma.nbrow;
      parameters[3] = ma.nbcol;
      parameters[4] = p1;
      parameters[5] = q1;
      parameters[6] = mb.nbrow;
      parameters[7] = mb.nbcol;
      parameters[8] = ma.m;
      parameters[9] = ma.n;
      parameters[10] = ma.sprow;
      parameters[11] = ma.spcol;
      parameters[12] = mb.sprow;
      parameters[13] = mb.spcol;
      parameters[14] = ia;
      parameters[15] = ja;
      parameters[16] = ib;
      parameters[17] = jb;
      parameters[18] = m;
      parameters[19] = n;
      parameters[20] = mb.m;
      parameters[21] = mb.n;
      Cigebs2d(gcontext, "All", "H", (Int)1, nbparameter, parameters, (Int)1);
    } else {
      Cigebr2d(gcontext, "All", "H", (Int)1, nbparameter, parameters, (Int)1, (Int)0, (Int)0);
      p0 = parameters[0];
      q0 = parameters[1];
      ma.nbrow = parameters[2];
      ma.nbcol = parameters[3];
      p1 = parameters[4];
      q1 = parameters[5];
      mb.nbrow = parameters[6];
      mb.nbcol = parameters[7];
      ma.m = parameters[8];
      ma.n = parameters[9];
      ma.sprow = parameters[10];
      ma.spcol = parameters[11];
      mb.sprow = parameters[12];
      mb.spcol = parameters[13];
      ia = parameters[14];
      ja = parameters[15];
      ib = parameters[16];
      jb = parameters[17];
      m = parameters[18];
      n = parameters[19];
      mb.m = parameters[20];
      mb.n = parameters[21];
      ma.desctype = BLOCK_CYCLIC_2D;
      mb.desctype = BLOCK_CYCLIC_2D;
    };
    Cblacs_get((Int)0, (Int)0, &context0);
    Cblacs_gridinit(&context0, "R", p0, q0);
    Cblacs_get((Int)0, (Int)0, &context1);
    Cblacs_gridinit(&context1, "R", p1, q1);
    Cblacs_gridinfo(context0, &dummy, &dummy, &myprow0, &mypcol0);
    if (myprow0 >= p0 || mypcol0 >= q0)
      myprow0 = mypcol0 = -1;
    Cblacs_gridinfo(context1, &dummy, &dummy, &myprow1, &mypcol1);
    if (myprow1 >= p1 || mypcol1 >= q1)
      myprow1 = mypcol1 = -1;
    assert((myprow0 < p0 && mypcol0 < q0) || (myprow0 == -1 && mypcol0 == -1));
    assert((myprow1 < p1 && mypcol1 < q1) || (myprow1 == -1 && mypcol1 == -1));
    ma.ctxt = context0;
    mb.ctxt = context1;
    /* From here, we are not assuming that only the processors working in the
     * redistribution are calling  xxMR2D, but the ones not concerned will do
     * nothing. */
    /* We compute the exact size of the local memory block for the memory
     * allocations */
    if (myprow0 >= 0 && mypcol0 >= 0) {
      blocksize0 = memoryblocksize(&ma);
      ma.lda = localsize(SHIFT(myprow0, ma.sprow, p0), p0, ma.nbrow, ma.m);
      setmemory(&ptrmyblock, blocksize0);
      initblock(ptrmyblock, 1, blocksize0);
      setmemory(&ptrmyblockcopy, blocksize0);
      memcpy((char *) ptrmyblockcopy, (char *) ptrmyblock,
             blocksize0 * sizeof(dcomplex));
      setmemory(&ptrmyblockvide, blocksize0);
      for (i = 0; i < blocksize0; i++)
        ptrmyblockvide[i].r = -1;
    };  /* if (mypnum < p0 * q0) */
    if (myprow1 >= 0 && mypcol1 >= 0) {
      setmemory(&ptrsavemyblock, memoryblocksize(&mb));
      mb.lda = localsize(SHIFT(myprow1, mb.sprow, p1), p1, mb.nbrow, mb.m);
    };  /* if (mypnum < p1 * q1)  */
    /* Redistribute the matrix from grid 0 to grid 1 (memory location
     * ptrmyblock to ptrsavemyblock) */
    Cpzgemr2d(m, n,
              ptrmyblock, ia, ja, &ma,
              ptrsavemyblock, ib, jb, &mb, gcontext);
    /* Perform the inverse redistribution of the matrix from grid 1 to grid 0
     * (memory location ptrsavemyblock to ptrmyblockvide) */
    Cpzgemr2d(m, n,
              ptrsavemyblock, ib, jb, &mb,
              ptrmyblockvide, ia, ja, &ma, gcontext);
    /* Check the differences */
    nberrors = 0;
    if (myprow0 >= 0 && mypcol0 >= 0) {
      /* only for the processors that do have data at the begining */
      for (i = 0; i < blocksize0; i++) {
        Int   li, lj, gi, gj;
        Int   in;
        in = 1;
        li = i % ma.lda;
        lj = i / ma.lda;
        gi = (li / ma.nbrow) * p0 * ma.nbrow +
              SHIFT(myprow0, ma.sprow, p0) * ma.nbrow + li % ma.nbrow;
        gj = (lj / ma.nbcol) * q0 * ma.nbcol +
              SHIFT(mypcol0, ma.spcol, q0) * ma.nbcol + lj % ma.nbcol;
        assert(gi < ma.m && gj < ma.n);
        gi -= (ia - 1);
        gj -= (ja - 1);
        if (gi < 0 || gj < 0 || gi >= m || gj >= n)
          in = 0;
        if (!in) {
          ptrmyblockcopy[i].r = -1;
        }
        if (ptrmyblockvide[i].r != ptrmyblockcopy[i].r) {
          nberrors++;
          printf("Proc %d : Error element number %d, value = %f , initvalue =%f \n"
                 ,mypnum, i,
                 ptrmyblockvide[i].r, ptrmyblockcopy[i].r);
        };
      };
      if (nberrors > 0) {
        printf("Processor %d, has tested  %d COMPLEX*16 elements,\
Number of redistribution errors = %d \n",
               mypnum, blocksize0, nberrors);
      }
    }
    /* Look at the errors on all the processors at this point. */
    Cigsum2d(gcontext, "All", "H", (Int)1, (Int)1, &nberrors, (Int)1, (Int)0, (Int)0);
    if (mypnum == 0)
      if (nberrors)
        printf("  => Total number of redistribution errors = %d \n",
               nberrors);
      else
        printf("TEST PASSED OK\n");
    /* release memory for the next iteration */
    if (myprow0 >= 0 && mypcol0 >= 0) {
      freememory((char *) ptrmyblock);
      freememory((char *) ptrmyblockvide);
      freememory((char *) ptrmyblockcopy);
    };  /* if (mypnum < p0 * q0) */
    /* release memory for the next iteration */
    if (myprow1 >= 0 && mypcol1 >= 0) {
      freememory((char *) ptrsavemyblock);
    };
    if (myprow0 >= 0)
      Cblacs_gridexit(context0);
    if (myprow1 >= 0)
      Cblacs_gridexit(context1);
  };    /* while nbre != 0 */
  if (mypnum == 0) {
    fclose(fp);
  };
  Cblacs_exit((Int)0);
  return 0;
}/* main */

Here is the call graph for this function: