You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
psblas3/cbind/test/gpu/pdegen3dc.c

510 lines
17 KiB
C

/*----------------------------------------------------------------------------------*/
/* Parallel Sparse BLAS v 3.5.0 */
/* (C) Copyright 2017 Salvatore Filippone */
/* */
/* Redistribution and use in source and binary forms, with or without */
/* modification, are permitted provided that the following conditions */
/* are met: */
/* 1. Redistributions of source code must retain the above copyright */
/* notice, this list of conditions and the following disclaimer. */
/* 2. Redistributions in binary form must reproduce the above copyright */
/* notice, this list of conditions, and the following disclaimer in the */
/* documentation and/or other materials provided with the distribution. */
/* 3. The name of the PSBLAS group or the names of its contributors may */
/* not be used to endorse or promote products derived from this */
/* software without specific written permission. */
/* */
/* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS */
/* ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED */
/* TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR */
/* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE PSBLAS GROUP OR ITS CONTRIBUTORS */
/* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR */
/* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF */
/* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS */
/* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN */
/* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) */
/* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
/* POSSIBILITY OF SUCH DAMAGE. */
/* */
/* */
/* File: pdegen3dc.c */
/* */
/* Program: pdegen3dc */
/* This sample program shows how to build and solve a sparse linear */
/* */
/* The program solves a linear system based on the partial differential */
/* equation */
/* */
/* */
/* */
/* The equation generated is */
/* */
/* b1 d d (u) b2 d d (u) a1 d (u)) a2 d (u))) */
/* - ------ - ------ + ----- + ------ + a3 u = 0 */
/* dx dx dy dy dx dy */
/* */
/* */
/* with Dirichlet boundary conditions on the unit cube */
/* */
/* 0<=x,y,z<=1 */
/* */
/* The equation is discretized with finite differences and uniform stepsize; */
/* the resulting discrete equation is */
/* */
/* ( u(x,y,z)(2b1+2b2+a1+a2)+u(x-1,y)(-b1-a1)+u(x,y-1)(-b2-a2)+ */
/* -u(x+1,y)b1-u(x,y+1)b2)*(1/h**2) */
/* */
/* Example adapted from: C.T.Kelley */
/* Iterative Methods for Linear and Nonlinear Equations */
/* SIAM 1995 */
/* */
/* */
/* In this sample program the index space of the discretized */
/* computational domain is first numbered sequentially in a standard way, */
/* then the corresponding vector is distributed according to an HPF BLOCK */
/* distribution directive. The discretization ensures there are IDIM */
/* *internal* points in each direction. */
/* */
/*----------------------------------------------------------------------------------*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include "psb_base_cbind.h"
#include "psb_prec_cbind.h"
#include "psb_linsolve_cbind.h"
#define LINEBUFSIZE 1024
#define NBMAX 20
#define DUMPMATRIX 0
double a1(double x, double y, double z)
{
return (1.0 / 80.0);
}
double a2(double x, double y, double z)
{
return (1.0 / 80.0);
}
double a3(double x, double y, double z)
{
return (1.0 / 80.0);
}
double c(double x, double y, double z)
{
return (0.0);
}
double b1(double x, double y, double z)
{
return (0.0 / sqrt(3.0));
}
double b2(double x, double y, double z)
{
return (0.0 / sqrt(3.0));
}
double b3(double x, double y, double z)
{
return (0.0 / sqrt(3.0));
}
double g(double x, double y, double z)
{
if (x == 1.0)
{
return (1.0);
}
else if (x == 0.0)
{
return (exp(-y * y - z * z));
}
else
{
return (0.0);
}
}
psb_c_i_t matgen(psb_c_ctxt cctxt, psb_c_i_t nl, psb_c_i_t idim, psb_c_l_t vl[],
psb_c_dspmat *ah, const char *afmt,
psb_c_descriptor *cdh, const char *cdfmt,
psb_c_dvector *xh, psb_c_dvector *bh, psb_c_dvector *rh)
{
psb_c_i_t iam, np;
psb_c_l_t ix, iy, iz, el, glob_row;
psb_c_i_t i, k, info, ret;
double x, y, z, deltah, sqdeltah, deltah2;
double val[10 * NBMAX], zt[NBMAX];
psb_c_l_t irow[10 * NBMAX], icol[10 * NBMAX];
info = 0;
psb_c_info(cctxt, &iam, &np);
deltah = (double)1.0 / (idim + 1);
sqdeltah = deltah * deltah;
deltah2 = 2.0 * deltah;
psb_c_set_index_base(0);
for (i = 0; i < nl; i++)
{
glob_row = vl[i];
// if ((i%100000 == 0)||(i<10)) fprintf(stderr,"%d: generation loop at %d %ld \n",iam,i,glob_row);
el = 0;
ix = glob_row / (idim * idim);
iy = (glob_row - ix * idim * idim) / idim;
iz = glob_row - ix * idim * idim - iy * idim;
x = (ix + 1) * deltah;
y = (iy + 1) * deltah;
z = (iz + 1) * deltah;
zt[0] = 0.0;
/* internal point: build discretization */
/* term depending on (x-1,y,z) */
val[el] = -a1(x, y, z) / sqdeltah - b1(x, y, z) / deltah2;
if (ix == 0)
{
zt[0] += g(0.0, y, z) * (-val[el]);
}
else
{
icol[el] = (ix - 1) * idim * idim + (iy)*idim + (iz);
el = el + 1;
}
/* term depending on (x,y-1,z) */
val[el] = -a2(x, y, z) / sqdeltah - b2(x, y, z) / deltah2;
if (iy == 0)
{
zt[0] += g(x, 0.0, z) * (-val[el]);
}
else
{
icol[el] = (ix)*idim * idim + (iy - 1) * idim + (iz);
el = el + 1;
}
/* term depending on (x,y,z-1)*/
val[el] = -a3(x, y, z) / sqdeltah - b3(x, y, z) / deltah2;
if (iz == 0)
{
zt[0] += g(x, y, 0.0) * (-val[el]);
}
else
{
icol[el] = (ix)*idim * idim + (iy)*idim + (iz - 1);
el = el + 1;
}
/* term depending on (x,y,z)*/
val[el] = 2.0 * (a1(x, y, z) + a2(x, y, z) + a3(x, y, z)) / sqdeltah + c(x, y, z);
icol[el] = (ix)*idim * idim + (iy)*idim + (iz);
el = el + 1;
/* term depending on (x,y,z+1) */
val[el] = -a3(x, y, z) / sqdeltah + b3(x, y, z) / deltah2;
if (iz == idim - 1)
{
zt[0] += g(x, y, 1.0) * (-val[el]);
}
else
{
icol[el] = (ix)*idim * idim + (iy)*idim + (iz + 1);
el = el + 1;
}
/* term depending on (x,y+1,z) */
val[el] = -a2(x, y, z) / sqdeltah + b2(x, y, z) / deltah2;
if (iy == idim - 1)
{
zt[0] += g(x, 1.0, z) * (-val[el]);
}
else
{
icol[el] = (ix)*idim * idim + (iy + 1) * idim + (iz);
el = el + 1;
}
/* term depending on (x+1,y,z) */
val[el] = -a1(x, y, z) / sqdeltah + b1(x, y, z) / deltah2;
if (ix == idim - 1)
{
zt[0] += g(1.0, y, z) * (-val[el]);
}
else
{
icol[el] = (ix + 1) * idim * idim + (iy)*idim + (iz);
el = el + 1;
}
for (k = 0; k < el; k++)
irow[k] = glob_row;
if ((ret = psb_c_dspins(el, irow, icol, val, ah, cdh)) != 0)
fprintf(stderr, "From psb_c_dspins: %d\n", ret);
irow[0] = glob_row;
psb_c_dgeins(1, irow, zt, bh, cdh);
zt[0] = 0.0;
psb_c_dgeins(1, irow, zt, xh, cdh);
}
#ifdef PSB_HAVE_CUDA
info = psb_c_cdasb_format(cdh, cdfmt);
if (info != 0)
return (info);
info = psb_c_dspasb_opt(ah, cdh, afmt, PSB_UPD_DFLT, PSB_DUPL_DEF);
if (info != 0)
return (info);
info = psb_c_dgeasb_options_format(xh, cdh, PSB_DUPL_ADD, cdfmt);
if (info != 0)
return (info);
info = psb_c_dgeasb_options_format(bh, cdh, PSB_DUPL_ADD, cdfmt);
if (info != 0)
return (info);
info = psb_c_dgeasb_options_format(rh, cdh, PSB_DUPL_ADD, cdfmt);
if (info != 0)
return (info);
#else
if ((info = psb_c_cdasb(cdh)) != 0)
return (info);
if ((info = psb_c_dspasb(ah, cdh)) != 0)
return (info);
if ((info = psb_c_dgeasb(xh, cdh)) != 0)
return (info);
if ((info = psb_c_dgeasb(bh, cdh)) != 0)
return (info);
if ((info = psb_c_dgeasb(rh, cdh)) != 0)
return (info);
#endif
return (info);
}
int main(int argc, char *argv[])
{
psb_c_ctxt *cctxt;
psb_c_i_t iam, np;
char methd[40], ptype[20], afmt[8], cdfmt[8], buffer[LINEBUFSIZE + 1];
psb_c_i_t nparms;
psb_c_i_t idim, info, istop, itmax, itrace, irst, iter, ret;
psb_c_dprec *ph;
psb_c_dspmat *ah;
psb_c_dvector *bh, *xh, *rh;
psb_c_i_t nb, nlr, nl;
psb_c_l_t i, ng, *vl, k;
double t1, t2, eps, err;
double *xv, *bv, *rv;
double one = 1.0, zero = 0.0, res2;
psb_c_SolverOptions options;
psb_c_descriptor *cdh;
FILE *vectfile;
cctxt = psb_c_new_ctxt();
psb_c_init(cctxt);
#ifdef PSB_HAVE_CUDA
psb_c_cuda_init(cctxt);
#endif
psb_c_info(*cctxt, &iam, &np);
#ifdef PSB_HAVE_CUDA
if (iam == 0)
printf("Example compiled with GPU support.\n");
#endif
psb_c_barrier(*cctxt);
if (iam == 0)
{
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%d ", &nparms);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%s", methd);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%s", ptype);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%s", afmt);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%s", cdfmt);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%d", &idim);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%d", &istop);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%d", &itmax);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%d", &itrace);
fgets(buffer, LINEBUFSIZE, stdin);
sscanf(buffer, "%d", &irst);
/* Echo input parameters */
printf("Number of processes: %d\n", np);
printf("Number of parameters read: %d\n", nparms);
printf("Krylov method: %s\n", methd);
printf("Preconditioner type: %s\n", ptype);
printf("Sparse matrix format: %s\n", afmt);
printf("Descriptor format: %s\n", cdfmt);
printf("Problem dimension (internal points per direction): %d\n", idim);
printf("Stopping criterion (1:backward error 2: ||r||_2/||b||_2): %d\n", istop);
printf("Maximum number of iterations: %d\n", itmax);
printf("Info message frequency (every itrace iterations): %d\n", itrace);
printf("Restart depth for RGMRES or BiCGSTAB(L): %d\n", irst);
}
/* Now broadcast the values, and check they're OK */
psb_c_ibcast(*cctxt, 1, &nparms, 0);
psb_c_hbcast(*cctxt, methd, 0);
psb_c_hbcast(*cctxt, ptype, 0);
psb_c_hbcast(*cctxt, afmt, 0);
psb_c_hbcast(*cctxt, cdfmt, 0);
psb_c_ibcast(*cctxt, 1, &idim, 0);
psb_c_ibcast(*cctxt, 1, &istop, 0);
psb_c_ibcast(*cctxt, 1, &itmax, 0);
psb_c_ibcast(*cctxt, 1, &itrace, 0);
psb_c_ibcast(*cctxt, 1, &irst, 0);
fflush(stderr);
psb_c_barrier(*cctxt);
cdh = psb_c_new_descriptor();
psb_c_set_index_base(0);
/* Simple minded BLOCK data distribution */
ng = ((psb_c_l_t)idim) * idim * idim;
nb = (ng + np - 1) / np;
nl = nb;
if ((ng - iam * nb) < nl)
nl = ng - iam * nb;
if ((vl = malloc(nb * sizeof(psb_c_l_t))) == NULL)
{
fprintf(stderr, "On %d: malloc failure\n", iam);
psb_c_abort(*cctxt);
}
i = ((psb_c_l_t)iam) * nb;
for (k = 0; k < nl; k++)
vl[k] = i + k;
if ((info = psb_c_cdall_vl(nl, vl, *cctxt, cdh)) != 0)
{
fprintf(stderr, "From cdall: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
bh = psb_c_new_dvector();
xh = psb_c_new_dvector();
rh = psb_c_new_dvector();
ah = psb_c_new_dspmat();
// fprintf(stderr,"From psb_c_new_dspmat: %p\n",ah);
/* Allocate mem space for sparse matrix and vectors */
ret = psb_c_dspall(ah, cdh);
// fprintf(stderr,"From psb_c_dspall: %d\n",ret);
psb_c_dgeall(bh, cdh);
psb_c_dgeall(xh, cdh);
psb_c_dgeall(rh, cdh);
/* Matrix generation */
if (matgen(*cctxt, nl, idim, vl, ah, afmt, cdh, cdfmt, xh, bh, rh) != 0)
{
fprintf(stderr, "Error during matrix build loop\n");
psb_c_abort(*cctxt);
}
psb_c_barrier(*cctxt);
/* Set up the preconditioner */
ph = psb_c_new_dprec();
psb_c_dprecinit(*cctxt, ph, ptype);
ret = psb_c_dprecbld(ah, cdh, ph);
// fprintf(stderr,"From psb_c_dprecbld: %d\n",ret);
/* Set up the solver options */
psb_c_DefaultSolverOptions(&options);
options.eps = 1.e-9;
options.itmax = itmax;
options.irst = irst;
options.itrace = 1;
options.istop = istop;
psb_c_seterraction_ret();
t1 = psb_c_wtime();
ret = psb_c_dkrylov(methd, ah, ph, bh, xh, cdh, &options);
t2 = psb_c_wtime();
iter = options.iter;
err = options.err;
// fprintf(stderr,"From krylov: %d %lf, %d %d\n",iter,err,ret,psb_c_get_errstatus());
if (psb_c_get_errstatus() != 0)
{
psb_c_print_errmsg();
}
// fprintf(stderr,"After cleanup %d\n",psb_c_get_errstatus());
/* Check 2-norm of residual on exit */
psb_c_dgeaxpby(one, bh, zero, rh, cdh);
psb_c_dspmm(-one, ah, xh, one, rh, cdh);
res2 = psb_c_dgenrm2(rh, cdh);
if (iam == 0)
{
fprintf(stdout, "Time: %lf\n", (t2 - t1));
fprintf(stdout, "Iter: %d\n", iter);
fprintf(stdout, "Err: %lg\n", err);
fprintf(stdout, "||r||_2: %lg\n", res2);
}
#if DUMPATRIX
psb_c_dmat_name_print(ah, "cbindmat.mtx");
nlr = psb_c_cd_get_local_rows(cdh);
bv = psb_c_dvect_get_cpy(bh);
vectfile = fopen("cbindb.mtx", "w");
for (i = 0; i < nlr; i++)
fprintf(vectfile, "%lf\n", bv[i]);
fclose(vectfile);
xv = psb_c_dvect_get_cpy(xh);
nlr = psb_c_cd_get_local_rows(cdh);
for (i = 0; i < nlr; i++)
fprintf(stdout, "SOL: %d %d %lf\n", iam, i, xv[i]);
rv = psb_c_dvect_get_cpy(rh);
nlr = psb_c_cd_get_local_rows(cdh);
for (i = 0; i < nlr; i++)
fprintf(stdout, "RES: %d %d %lf\n", iam, i, rv[i]);
#endif
/* Clean up memory */
if ((info = psb_c_dprecfree(ph)) != 0)
{
fprintf(stderr, "From dprecfree: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
if ((info = psb_c_dgefree(xh, cdh)) != 0)
{
fprintf(stderr, "From dgefree: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
if ((info = psb_c_dgefree(bh, cdh)) != 0)
{
fprintf(stderr, "From dgefree: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
if ((info = psb_c_dgefree(rh, cdh)) != 0)
{
fprintf(stderr, "From dgefree: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
if ((info = psb_c_dspfree(ah, cdh)) != 0)
{
fprintf(stderr, "From dspfree: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
if ((info = psb_c_cdfree(cdh)) != 0)
{
fprintf(stderr, "From cdfree: %d\nBailing out\n", info);
psb_c_abort(*cctxt);
}
// fprintf(stderr,"pointer from cdfree: %p\n",cdh->descriptor);
/* Clean up object handles */
free(ph);
free(xh);
free(bh);
free(rh);
free(ah);
free(cdh);
/* further cleanup */
free(vl);
// if (iam == 0) fprintf(stderr,"program completed successfully\n");
psb_c_barrier(*cctxt);
#ifdef PSB_HAVE_CUDA
psb_c_cuda_exit();
#endif
/* Finalize PSBLAS context */
psb_c_exit(*cctxt);
free(cctxt);
}