int main (int argc, char **argv)
{
cholmod_common Common, *cc ;
cholmod_sparse *A ;
cholmod_dense *X, *B, *Residual ;
double rnorm, one [2] = {1,0}, minusone [2] = {-1,0} ;
int mtype ;
// start CHOLMOD
cc = &Common ;
cholmod_l_start (cc) ;
// load A
A = (cholmod_sparse *) cholmod_l_read_matrix (stdin, 1, &mtype, cc) ;
// B = ones (size (A,1),1)
B = cholmod_l_ones (A->nrow, 1, A->xtype, cc) ;
// X = A\B
X = SuiteSparseQR <double> (A, B, cc) ;
// rnorm = norm (B-A*X)
Residual = cholmod_l_copy_dense (B, cc) ;
cholmod_l_sdmult (A, 0, minusone, one, X, Residual, cc) ;
rnorm = cholmod_l_norm_dense (Residual, 2, cc) ;
printf ("2-norm of residual: %8.1e\n", rnorm) ;
printf ("rank %ld\n", cc->SPQR_istat [4]) ;
// free everything and finish CHOLMOD
cholmod_l_free_dense (&Residual, cc) ;
cholmod_l_free_sparse (&A, cc) ;
cholmod_l_free_dense (&X, cc) ;
cholmod_l_free_dense (&B, cc) ;
cholmod_l_finish (cc) ;
return (0) ;
void check_residual
(
cholmod_sparse *A,
cholmod_dense *X,
cholmod_dense *B,
cholmod_common *cc
)
{
Long m = A->nrow ;
Long n = A->ncol ;
Long rnk ;
double rnorm, anorm, xnorm, atrnorm ;
double one [2] = {1,0}, minusone [2] = {-1,0}, zero [2] = {0,0} ;
cholmod_dense *r, *atr ;
// get the rank(A) estimate
rnk = cc->SPQR_istat [4] ;
// anorm = norm (A,1) ;
anorm = cholmod_l_norm_sparse (A, 1, cc) ;
// rnorm = norm (A*X-B)
r = cholmod_l_copy_dense (B, cc) ;
cholmod_l_sdmult (A, 0, one, minusone, X, r, cc) ;
rnorm = cholmod_l_norm_dense (r, 2, cc) ;
// xnorm = norm (X)
xnorm = cholmod_l_norm_dense (X, 2, cc) ;
// atrnorm = norm (A'*r)
atr = cholmod_l_zeros (n, 1, r->xtype, cc) ; // atr = zeros (n,1)
cholmod_l_sdmult (A, 1, one, zero, r, atr, cc) ; // atr = A'*r
atrnorm = cholmod_l_norm_dense (atr, 2, cc) ; // atrnorm = norm (atr)
if (anorm > 0) atrnorm /= anorm ;
if (m <= n && anorm > 0 && xnorm > 0)
{
// find the relative residual, except for least-squares systems
rnorm /= (anorm * xnorm) ;
}
printf ("relative norm(Ax-b): %8.1e rank: %6ld "
"rel. norm(A'(Ax-b)) %8.1e\n", rnorm, rnk, atrnorm) ;
cholmod_l_free_dense (&r, cc) ;
cholmod_l_free_dense (&atr, cc) ;
}
double DenseMatrix :: norm( void )
// returns the maximum magnitude of any entry
{
return cholmod_l_norm_dense( **this, 0, context );
}
int main() {
/* Time Recording Variables */
int now_s, now_u;
struct rusage usage;
double time_now, time_past;
/* x, y, z and global indices */
int i, j, k, m, n, nodes;
/* Connectivity enties */
int m_above, m_below, m_left, m_right, m_front, m_back;
double c_above, c_below, c_left, c_right, c_front, c_back, c_self;
/* Creat cholmod object */
cholmod_common Common, *cc;
/* Compress column form sparse matrix */
cholmod_sparse *S, *ST;
/* forcing function on input; voltages after solving */
cholmod_dense *b, *v0, *r;
size_t *Si, *Sp, *Snz, *bi, *bj;
double *Sx, *bx, *v0x;
/* store the sparse matrix */
FILE *store;
char filename[100];
/* start CHOLMOD */
cc = &Common;
cholmod_l_start(cc);
/* initialize timing variables*/
getrusage(0, &usage);
now_s = usage.ru_utime.tv_sec;
now_u = usage.ru_utime.tv_usec;
time_past = now_s + now_u / 1.e6;
time_now = time_past;
/* total nodes in the grid */
nodes = IMAX * JMAX * KMAX;
/* Allocate space for connectivity matrix and forcing vector. */
S = cholmod_l_allocate_sparse(nodes, nodes, 7 * nodes,
0, 0, 0, CHOLMOD_REAL, cc);
b = cholmod_l_allocate_dense(nodes, 1, nodes, CHOLMOD_REAL, cc);
bx = b->x;
v0 = cholmod_l_allocate_dense(nodes, 1, nodes, CHOLMOD_REAL, cc);
v0x = v0->x;
/*================================================================*/
/*=============== make connectivity matrix ======================*/
/*================================================================*/
Si = (size_t *) S->i;
Sp = (size_t *) S->p;
Sx = (double *) S->x;
Snz = (size_t *) S->nz;
n = 0;
Sp[0] = 0;
for (k = 0; k < KMAX; k++) {
for (j = 0; j < JMAX; j++) {
for (i = 0; i < IMAX; i++) {
/* Global index in x-fastest order*/
m = k * IMAX * JMAX + j * IMAX + i;
Sp[m + 1] = Sp[m];
v0x[m] = (k + 1.0) / (KMAX + 1.0);
/* Six coef.s of every node */
c_below = (mepr(i - 0.5, j - 0.5, k - 0.5) +
mepr(i + 0.5, j - 0.5, k - 0.5) +
mepr(i - 0.5, j + 0.5, k - 0.5) +
mepr(i + 0.5, j + 0.5, k - 0.5)) / 4.0;
c_front = (mepr(i - 0.5, j - 0.5, k - 0.5) +
mepr(i + 0.5, j - 0.5, k - 0.5) +
mepr(i - 0.5, j - 0.5, k + 0.5) +
mepr(i + 0.5, j - 0.5, k + 0.5)) / 4.0;
c_left = (mepr(i - 0.5, j + 0.5, k - 0.5) +
mepr(i - 0.5, j - 0.5, k - 0.5) +
mepr(i - 0.5, j + 0.5, k + 0.5) +
mepr(i - 0.5, j - 0.5, k + 0.5)) / 4.0;
c_right = (mepr(i + 0.5, j + 0.5, k - 0.5) +
mepr(i + 0.5, j - 0.5, k - 0.5) +
mepr(i + 0.5, j + 0.5, k + 0.5) +
mepr(i + 0.5, j - 0.5, k + 0.5)) / 4.0;
c_back = (mepr(i - 0.5, j + 0.5, k - 0.5) +
mepr(i + 0.5, j + 0.5, k - 0.5) +
mepr(i - 0.5, j + 0.5, k + 0.5) +
mepr(i + 0.5, j + 0.5, k + 0.5)) / 4.0;
c_above = (mepr(i - 0.5, j - 0.5, k + 0.5) +
mepr(i + 0.5, j - 0.5, k + 0.5) +
mepr(i - 0.5, j + 0.5, k + 0.5) +
mepr(i + 0.5, j + 0.5, k + 0.5))/4.0;
/* Self term. */
c_self = -(c_above + c_below + c_left + c_right + c_front + c_back);
Si[n] = m; Sx[n] = c_self; n++; Sp[m + 1]++;
/* Node below. Ensure not on bottom face */
if (k != 0) {
m_below = m - IMAX * JMAX;
Si[n] = m_below; Sx[n] = c_below; n++; Sp[m + 1]++;
} else {
bx[m] = -c_below * VBOT;
}
/* Node front. Ensure not on front face. */
if (j != 0) {
m_front = m - IMAX;
} else {
m_front = m + (JMAX - 1) * IMAX;
}
Si[n] = m_front; Sx[n] = c_front; n++; Sp[m + 1]++;
/* Node to left. Ensure not on left face. */
if (i != 0) {
m_left = m - 1;
} else {
m_left = m + IMAX -1;
}
Si[n] = m_left; Sx[n] = c_left; n++; Sp[m + 1]++;
/* Node to right. Ensure not on right face. */
if (i != IMAX - 1) {
m_right = m + 1;
} else{
m_right = m - IMAX + 1;
}
Si[n] = m_right; Sx[n] = c_right; n++; Sp[m + 1]++;
/* Node back. Ensure not on back face. */
if (j != JMAX - 1) {
m_back = m + IMAX;
} else {
m_back = m - (JMAX - 1) * IMAX;
}
Si[n] = m_back; Sx[n] = c_back; n++; Sp[m + 1]++;
/* Node top. Ensure not on top face */
if (k != KMAX - 1) {
m_above = m + IMAX * JMAX;
Si[n] = m_above; Sx[n] = c_above; n++; Sp[m + 1]++;
} else {
bx[m] = -c_above * VTOP;
}
Snz[m] = Sp[m + 1] - Sp[m];
}
}
}
/*====================================================================*/
/*===================Done creating connectivity matrix.===============*/
/*====================================================================*/
/* report time*/
getrusage(0, &usage);
now_s = usage.ru_utime.tv_sec;
now_u = usage.ru_utime.tv_usec;
time_now = now_s + now_u / 1.e6;
printf("\nFinished creating connectivity matrix\n"
" Incremental time %f\n"
" Running time %f\n", time_now - time_past, time_now);
time_past = time_now;
/* Print all three matrixes */
cholmod_l_print_sparse(S, "S", cc);
cholmod_l_print_dense(b, "b", cc);
cholmod_l_print_dense(v0, "v0", cc);
/* Allocate residual vector */
r = cholmod_l_zeros(nodes, 1, CHOLMOD_REAL, cc);
/* The Preconditioned Conjugate Gradient Method */
/* Calculate the first residual value */
double one[2], zero[2], minusone[2];
zero[0] = 0.0;
zero[1] = 0.0;
one[0] = 1.0;
one[1] = 0.0;
minusone[0] = -1.0;
minusone[1] = 0.0;
cholmod_l_copy_dense2(b, r, cc);
cholmod_l_sdmult(S, 0, minusone, one, v0, r, cc);
printf("Initial 2-norm = %f\n", cholmod_l_norm_dense(r, 2, cc));
printf("Initial 1-norm = %f\n", cholmod_l_norm_dense(r, 1, cc));
printf("Initial 0-norm = %f\n", cholmod_l_norm_dense(r, 0, cc));
/* The iteration */
double rho1, rho0, beta = 0.0, alpha = 0.0;
cholmod_dense *p1, *p0, *q;
double *p1x, *p0x, *qx, *rx;
/*
p1 = cholmod_l_allocate_dense(nodes, 1, nodes, CHOLMOD_REAL, cc);
p0 = cholmod_l_allocate_dense(nodes, 1, nodes, CHOLMOD_REAL, cc);
q = cholmod_l_allocate_dense(nodes, 1, nodes, CHOLMOD_REAL, cc);
*/
p1 = cholmod_l_zeros(nodes, 1, CHOLMOD_REAL, cc);
p0 = cholmod_l_zeros(nodes, 1, CHOLMOD_REAL, cc);
q = cholmod_l_zeros(nodes, 1, CHOLMOD_REAL, cc);
p1x = (double *) p1->x;
p0x = (double *) p0->x;
qx = (double *) q->x;
rx = (double *) r->x;
int iter;
for (iter = 0; iter < MAX_ITER; iter++) {
if (cholmod_l_norm_dense(r, 0, cc) < 1e-10)
break;
rho1 = cholmod_l_norm_dense(r, 2, cc);
rho1 = rho1 * rho1;
if (iter == 0) {
cholmod_l_copy_dense2(r, p1, cc);
} else {
beta = rho1 / rho0;
for (i = 0; i < nodes; i++) {
p1x[i] = rx[i] + beta * p0x[i];
}
}
cholmod_l_sdmult(S, 0, one, zero, p1, q, cc);
alpha = 0;
for (i = 0; i < nodes; i++) {
alpha += p1x[i] * qx[i];
}
alpha = rho1 / alpha;
for (i = 0; i < nodes; i++) {
v0x[i] += alpha * p1x[i];
rx[i] -= alpha * qx[i];
}
/*
printf("iter = %d\n: rho1 = %f, rho0 = %f, alpha = %f, beta = %f\n",
iter, rho1, rho0, alpha, beta);
*/
cholmod_l_copy_dense2(p1, p0, cc);
rho0 = rho1;
}
cholmod_l_copy_dense2(b, r, cc);
cholmod_l_sdmult(S, 0, minusone, one, v0, r, cc);
printf("After %d iterations:\n", iter);
printf("Final 2-norm: %f\n", cholmod_l_norm_dense(r, 2, cc));
printf("Final 1-norm: %f\n", cholmod_l_norm_dense(r, 1, cc));
printf("Final 0-norm: %f\n", cholmod_l_norm_dense(r, 0, cc));
/* sort sparse matrix and store it
cholmod_l_sort(S, cc);
S->stype = 0;
*/
/* Check if S is symmetric
ST = cholmod_l_transpose(S, 2, cc);
store = fopen("LpT.dat", "w");
cholmod_l_write_sparse(store, ST, NULL, NULL, cc);
fclose(store);
cholmod_l_free_sparse(&ST, cc);
*/
/* store the sparse matrix */
/*
store = fopen("Laplace.dat", "w");
cholmod_l_write_sparse(store, S, NULL, NULL, cc);
fclose(store);
cholmod_l_free_sparse(&S, cc);
*/
/* store the force vector */
/*
store = fopen("Force.dat", "w");
cholmod_l_write_dense(store, b, NULL, cc);
fclose(store);
cholmod_l_free_dense(&b,cc);
*/
/* store the guess vector */
store = fopen("Voltage.dat", "w");
cholmod_l_write_dense(store, v0, NULL, cc);
fclose(store);
cholmod_l_free_dense(&v0, cc);
/* store the residual vector */
store = fopen("Residual.dat", "w");
cholmod_l_write_dense(store, r, NULL, cc);
fclose(store);
cholmod_l_free_dense(&r, cc);
/* report time*/
getrusage(0, &usage);
now_s = usage.ru_utime.tv_sec;
now_u = usage.ru_utime.tv_usec;
time_now = now_s + now_u / 1.e6;
printf("\nFinished writing matrix\n"
" Incremental time %f\n"
" Running time %f\n", time_now - time_past, time_now);
cholmod_l_finish(cc);
return 0;
}
double Dense :: norm( void )
// returns the infinity norm
{
return cholmod_l_norm_dense( **this, 0, common );
}
int main (int argc, char **argv)
{
cholmod_sparse *A ;
cholmod_dense *X, *B, *r, *atr ;
double anorm, xnorm, rnorm, one [2] = {1,0}, minusone [2] = {-1,0}, t ;
double zero [2] = {0,0}, atrnorm ;
int mtype ;
long m, n, rnk ;
size_t total_mem, available_mem ;
// start CHOLMOD
cholmod_common *cc, Common ;
cc = &Common ;
cholmod_l_start (cc) ;
// warmup the GPU. This can take some time, but only needs
// to be done once
cc->useGPU = true ;
t = SuiteSparse_time ( ) ;
cholmod_l_gpu_memorysize (&total_mem, &available_mem, cc) ;
cc->gpuMemorySize = available_mem ;
t = SuiteSparse_time ( ) - t ;
if (cc->gpuMemorySize <= 1)
{
printf ("no GPU available\n") ;
}
printf ("available GPU memory: %g MB, warmup time: %g\n",
(double) (cc->gpuMemorySize) / (1024 * 1024), t) ;
// A = mread (stdin) ; read in the sparse matrix A
const char *filename = (argc < 2 ? "Problems/2.mtx" : argv[1]);
FILE *file = fopen(filename, "r");
A = (cholmod_sparse *) cholmod_l_read_matrix (file, 1, &mtype, cc) ;
fclose(file);
if (mtype != CHOLMOD_SPARSE)
{
printf ("input matrix must be sparse\n") ;
exit (1) ;
}
// [m n] = size (A) ;
m = A->nrow ;
n = A->ncol ;
long ordering = (argc < 3 ? SPQR_ORDERING_DEFAULT : atoi(argv[2]));
#if 1
printf ("Matrix %6ld-by-%-6ld nnz: %6ld\n",
m, n, cholmod_l_nnz (A, cc)) ;
#endif
// anorm = norm (A,1) ;
anorm = cholmod_l_norm_sparse (A, 1, cc) ;
// B = ones (m,1), a dense right-hand-side of the same type as A
B = cholmod_l_ones (m, 1, A->xtype, cc) ;
// X = A\B ; with default ordering and default column 2-norm tolerance
if (A->xtype == CHOLMOD_REAL)
{
// A, X, and B are all real
X = SuiteSparseQR <double>(ordering, SPQR_NO_TOL, A, B, cc) ;
}
else
{
#if SUPPORTS_COMPLEX
// A, X, and B are all complex
X = SuiteSparseQR < std::complex<double> >
(SPQR_ORDERING_DEFAULT, SPQR_NO_TOL, A, B, cc) ;
#else
printf("Code doesn't support std::complex<?> types.\n");
#endif
}
// get the rank(A) estimate
rnk = cc->SPQR_istat [4] ;
// compute the residual r, and A'*r, and their norms
r = cholmod_l_copy_dense (B, cc) ; // r = B
cholmod_l_sdmult (A, 0, one, minusone, X, r, cc) ; // r = A*X-r = A*x-b
rnorm = cholmod_l_norm_dense (r, 2, cc) ; // rnorm = norm (r)
atr = cholmod_l_zeros (n, 1, CHOLMOD_REAL, cc) ; // atr = zeros (n,1)
cholmod_l_sdmult (A, 1, one, zero, r, atr, cc) ; // atr = A'*r
atrnorm = cholmod_l_norm_dense (atr, 2, cc) ; // atrnorm = norm (atr)
// xnorm = norm (X)
xnorm = cholmod_l_norm_dense (X, 2, cc) ;
// write out X to a file
FILE *f = fopen ("X.mtx", "w") ;
cholmod_l_write_dense (f, X, NULL, cc) ;
fclose (f) ;
if (m <= n && anorm > 0 && xnorm > 0)
{
// find the relative residual, except for least-squares systems
rnorm /= (anorm * xnorm) ;
}
printf ("\nnorm(Ax-b): %8.1e\n", rnorm) ;
printf ("norm(A'(Ax-b)) %8.1e rank: %ld of %ld\n",
atrnorm, rnk, (m < n) ? m:n) ;
/* Write an info file. */
FILE *info = fopen("gpu_results.txt", "w");
fprintf(info, "%ld\n", cc->SPQR_istat[7]); // ordering method
fprintf(info, "%ld\n", cc->memory_usage); // memory usage (bytes)
fprintf(info, "%30.16e\n", cc->SPQR_flopcount); // flop count
fprintf(info, "%lf\n", cc->SPQR_analyze_time); // analyze time
fprintf(info, "%lf\n", cc->SPQR_factorize_time); // factorize time
fprintf(info, "-1\n") ; // cpu memory (bytes)
fprintf(info, "-1\n") ; // gpu memory (bytes)
fprintf(info, "%32.16e\n", rnorm); // residual
fprintf(info, "%ld\n", cholmod_l_nnz (A, cc)); // nnz(A)
fprintf(info, "%ld\n", cc->SPQR_istat [0]); // nnz(R)
fprintf(info, "%ld\n", cc->SPQR_istat [2]); // # of frontal matrices
fprintf(info, "%ld\n", cc->SPQR_istat [3]); // ntasks, for now
fprintf(info, "%lf\n", cc->gpuKernelTime); // kernel time (ms)
fprintf(info, "%ld\n", cc->gpuFlops); // "actual" gpu flops
fprintf(info, "%d\n", cc->gpuNumKernelLaunches); // # of kernel launches
fprintf(info, "%32.16e\n", atrnorm) ; // norm (A'*(Ax-b))
fclose(info);
// free everything
cholmod_l_free_dense (&r, cc) ;
cholmod_l_free_dense (&atr, cc) ;
cholmod_l_free_sparse (&A, cc) ;
cholmod_l_free_dense (&X, cc) ;
cholmod_l_free_dense (&B, cc) ;
cholmod_l_finish (cc) ;
return (0) ;
}
