sp_channel bs_log::add_channel(const sp_channel &dest) {
BS_ERROR (dest, "bs_log::add_channel: dest is null");
channel_iterator_const_t itr = channel_map_.find(dest->get_name ());
if (itr == channel_map_.end()) {
channel_map_.insert (std::make_pair (dest->get_name (), dest));
this->fire_signal(bs_log::log_channel_added);
return dest;
}
return itr->second;
}
int cgs_solver::solve (sp_matrix_t matrix, spv_double sp_rhs, spv_double sp_sol)
{
#ifdef _DEBUG
BOSOUT (section::solvers, level::debug) << "CGS\n" << bs_end;
#endif
BS_ERROR (sp_rhs->size (), "cgs_solve");
BS_ERROR (sp_sol->size (), "cgs_solve");
BS_ERROR (prop, "cgs_solve");
t_double rho_1, rho_2 = 1, alpha = 1, beta, sigma;
int iter;
const double epsmac = 1e-24;
t_double r_norm, b_norm, den_norm;
//fp_type *x = solution;
t_double *rhs = &(*sp_rhs)[0];
t_double *sol = &(*sp_sol)[0];
const t_double one = 1.0;
//OMP_TIME_MEASURE_START (cgs_solve_timer);
t_double tol = prop->get_f (tol_idx);
tol *= tol;
//resid = prop->get_residuals ();
//convergence_rate = prop->get_convergence_rate ();
int max_iter = prop->get_i (max_iters_idx);
t_long n = matrix->get_n_rows () * matrix->get_n_block_size ();
BS_ASSERT (n == (t_long)sp_sol->size ());
t_double *p = &(*sp_p)[0];
t_double *phat = &(*sp_phat)[0];
t_double *v = &(*sp_v)[0];
t_double *tmp = &(*sp_tmp)[0];
t_double *q = &(*sp_q)[0];
t_double *u = &(*sp_u)[0];
t_double *d = &(*sp_d)[0];
t_double *dhat = &(*sp_dhat)[0];
t_double *r = &(*sp_r)[0];
t_double *rtilde = &(*sp_rtilde)[0];
//t_double *r_old = &(*sp_r_old)[0];
prop->set_b (success_idx, false);
// solution = {0}
//assign (solution, n, 0);
memset (sol, 0, sizeof (t_double) * n);
//solution.assign (n, 0);
sp_p->resize (n);
sp_phat->resize (n);
sp_v->resize (n);
sp_tmp->resize (n);
sp_q->resize (n);
sp_u->resize (n);
sp_d->resize (n);
sp_dhat->resize (n);
sp_r->resize (n);
sp_rtilde->resize (n);
sp_r_old->resize (n);
// r = {0}
//r.assign (n, 0);
memset (r, 0, sizeof (t_double) * n);
memset (tmp, 0, sizeof (t_double) * n);
memset (p, 0, sizeof (t_double) * n);
memset (v, 0, sizeof (t_double) * n);
memset (q, 0, sizeof (t_double) * n);
// TODO:paste
//tmp.assign (n, 0);
//p.assign (n, 0);
//v.assign (n, 0);
//q.assign (n, 0);
// p0 = u0 = r0;
//u.assign (r.begin (), r.end ());
memcpy (u, r, sizeof (t_double) * n);
// TODO:end
// r = Ax0 - b
matrix->calc_lin_comb (-1.0, 1.0, sp_sol, sp_rhs, sp_r);
//rtilde.assign (r.begin (), r.end ());
memcpy (rtilde, r, sizeof (t_double) * n);
//tools::save_seq_vector (tools::string_formater ("1_well_bhp.%s.txt", it->first).str).save (it->second);
r_norm = mv_vector_inner_product_n (r, r, n);
if (r_norm <= tol) // initial guess quite good
return 0;
rho_1 = r_norm;
b_norm = sqrt (mv_vector_inner_product_n (rhs, rhs, n));
// TODO:delete
//p.assign (r.begin (), r.end ());
//rtilde.assign (r.begin (), r.end ());
//v.assign (n, 0);
// TODO:end
if (b_norm > epsmac) // choose convergence criterion
{
// |r_i|/|b| <= eps if |b| > 0
tol *= b_norm;
den_norm = b_norm;
}
else // (r_norm > epsmac)
{
// |r_i|/|r0| <= eps if |b| = 0
tol *= r_norm;
den_norm = r_norm;
}
// set up initial norm and convergense factor
//prop->set_relative_factor (den_norm);
// main loop
for (iter = 0; iter < max_iter; ++iter)
{
//printf ("CGS iteration: %d, resid = %le\n", iter, r_norm);
//fflush (stdout);
// TODO: paste
if (iter)
{
//rho_1 = (r,rtilde)
rho_1 = mv_vector_inner_product_n (r, rtilde, n); //in first iter equals to r_norm
if (rho_1 == 0) // failure
{
if (den_norm > epsmac)
prop->set_f (final_res_idx, r_norm / den_norm);
else
prop->set_f (final_res_idx, r_norm);
bs_throw_exception ("CGS: Failure - rho_1 == 0");
}
}
beta = rho_1/rho_2; // beta = rho_n/rho_n-1
rho_2 = rho_1;
// u = r + beta*q
sum_vector_n (r, one, q, beta, u, n);
// tmp = q+beta*p_old
sum_vector_n (q, one, p, beta, tmp, n);
// p_new = u + beta*tmp
sum_vector_n (u, one, tmp, beta, p, n);
//temp_p.assign (p.begin (), p.end ());
if (prec)
{
if (prec->solve_prec (matrix, sp_p, sp_phat))
{
bs_throw_exception ("CGS: Preconditioner failed");
}
}
else // no precondition (preconditioner=identity_matrix)
{
memcpy (phat, p, sizeof (t_double) * n);
//phat.assign (p.begin (), p.end ());
}
// v = A * phat = A * p, if no precondition;
//v.assign (n, 0);
memset (v, 0, sizeof (t_double) * n);
matrix->matrix_vector_product (sp_phat, sp_v);
// sigma = (v,rtilde)
sigma = mv_vector_inner_product_n (rtilde, v, n);
if (sigma > epsmac || sigma < -epsmac)
// alpha = rho_1/sigma
alpha = rho_1 / sigma;
else // failure
{
if (den_norm > epsmac)
prop->set_f (final_res_idx, r_norm / den_norm);
else
prop->set_f (final_res_idx, r_norm);
bs_throw_exception ("CGS: Failure - sigma == 0");
}
// q = u - alpha*v
sum_vector_n (u, one, v, -alpha, q, n);
// d = u + q
sum_vector_n (u, one, q, one, d, n);
// dhat = M^(-1) * d;
//temp_d.assign (d.begin (), d.end ());
if (prec)
{
if(prec->solve_prec (matrix, sp_d, sp_dhat))
{
bs_throw_exception ("CGS: Preconditioner failed");
}
}
else // no precondition (preconditioner=identity_matrix)
{
//dhat.assign (d.begin (), d.end ());
memcpy (dhat, d, sizeof (t_double) * n);
}
//tmp.assign (n, 0);
memset (tmp, 0, sizeof (t_double) * n);
// tmp = A*d
matrix->matrix_vector_product (sp_dhat, sp_tmp);
// r = r - alpha*tmp
sum_vector_n (r, one, tmp, -alpha, r, n);
// x = x + alpha*dhat
sum_vector_n (sol, one, dhat, alpha, sol, n);
r_norm = mv_vector_inner_product_n (r, r, n);
if (r_norm <= tol) // && check_resid_for_matbalance (n_rows, nb, r, matb_tol))
break;
}
//tools::save_seq_vector ("solution.txt").save(solution);
//TODO: end
prop->set_i (iters_idx, iter + 1);
prop->set_b (success_idx, true);
/*
//additional checking convergence
mv_calc_lin_comb (matrix, -1.0, 1.0, solution, rhs, r);
r_norm = mv_vector_inner_product (r, r, n);
*/
if (den_norm > epsmac)
prop->set_f (final_res_idx, r_norm / den_norm);
else
prop->set_f (final_res_idx, r_norm);
//printf ("CGS OK! iters = %d, resid = %le\n", lprop->iters, lprop->final_resid);
//OMP_TIME_MEASURE_END (bicgstab_solve_timer);
return 0;
}
int tfqmr_solver::solve (sp_matrix_t matrix, spv_double sp_rhs, spv_double sp_sol)
{
BOSOUT (section::solvers, level::debug) << "TFQMR\n" << bs_end;
BS_ERROR (matrix, "tfqmr_solve");
BS_ERROR (sp_rhs->size (), "tfqmr_solve");
BS_ERROR (sp_sol->size (), "tfqmr_solve");
BS_ERROR (prop, "tfqmr_solve");
t_double rho_1, rho_2 = 1, alpha = 1, beta, sigma;
int iter;
const double epsmac = 1e-24;
t_double r_norm, b_norm, den_norm, w_norm, eta, nu, tau, c;
//fp_type *x = solution;
//OMP_TIME_MEASURE_START (tfqmr_solve_timer);
t_double *rhs = &(*sp_rhs)[0];
t_double *sol = &(*sp_sol)[0];
t_double tol = prop->get_f (tol_idx);
tol *= tol;
//resid = prop->get_residuals ();
//convergence_rate = prop->get_convergence_rate ();
int max_iter = prop->get_i (max_iters_idx);
t_long n = matrix->get_n_rows () * matrix->get_n_block_size ();
sp_p->resize (n);
sp_v->resize (n);
sp_w->resize (n);
sp_u->resize (n);
sp_q->resize (n);
sp_d->resize (n);
sp_res->resize (n);
sp_r->resize (n);
sp_rtilde->resize (n);
sp_tmp->resize (n);
sp_rhat->resize (n);
sp_y->resize (n);
//x_cgs = y + n;
prop->set_b (success_idx, false);
// solution = {0}
sp_sol->assign (0);
// r = {0}
sp_r->assign (0);
// TODO:paste
sp_tmp->assign (0);
sp_p->assign (0);
sp_v->assign (0);
sp_q->assign (0);
sp_d->assign (0);
t_double *r = &(*sp_r)[0];
t_double *p = &(*sp_p)[0];
t_double *v = &(*sp_v)[0];
t_double *w = &(*sp_w)[0];
t_double *u = &(*sp_u)[0];
t_double *q = &(*sp_q)[0];
t_double *d = &(*sp_d)[0];
t_double *res = &(*sp_res)[0];
t_double *rtilde = &(*sp_rtilde)[0];
t_double *tmp = &(*sp_tmp)[0];
t_double *rhat = &(*sp_rhat)[0];
t_double *y = &(*sp_y)[0];
// r = Ax0 - b
matrix->calc_lin_comb (-1.0, 1.0, sp_sol, sp_rhs, sp_r);
memcpy (rtilde, r, sizeof (t_double) * n);
//rtilde.assign (r.begin (), r.end ());
// p0 = u0 = r0;
//memcpy (p, r, n * sizeof (double));
memcpy (u, r, sizeof (t_double) * n);
memcpy (p, r, sizeof (t_double) * n);
memcpy (w, r, sizeof (t_double) * n);
//u.assign (r.begin (), r.end ());
//p.assign (r.begin (), r.end ());
//w.assign (r.begin (), r.end ());
// tmp = M^(-1) * u;
if (prec)
{
if (prec->solve_prec (matrix, sp_u, sp_tmp))
{
bs_throw_exception ("TFQMR: Preconditioner failed");
}
memcpy (u, tmp, sizeof (t_double) * n);
memcpy (p, tmp, sizeof (t_double) * n);
//u.assign (tmp.begin (), tmp.end ());
//p.assign (u.begin (), u.end ());
}
matrix->matrix_vector_product (sp_p, sp_v);
//tools::save_seq_vector (tools::string_formater ("1_well_bhp.%s.txt", it->first).str).save (it->second);
r_norm = mv_vector_inner_product_n (r, r, n);
if (r_norm <= tol) // initial guess quite good
return 0;
tau = sqrt (r_norm);
rho_1 = r_norm;
rho_2 = r_norm;
b_norm = sqrt (mv_vector_inner_product_n (rhs, rhs, n));
if (b_norm > epsmac) // choose convergence criterion
{
// |r_i|/|b| <= eps if |b| > 0
tol *= b_norm;
den_norm = b_norm;
}
else // (r_norm > epsmac)
{
// |r_i|/|r0| <= eps if |b| = 0
tol *= r_norm;
den_norm = r_norm;
}
// set up initial norm and convergense factor
//prop->set_relative_factor (den_norm);
int m, count;
// main loop
for (iter = 0; iter < max_iter; ++iter)
{
//printf ("TFQMR iteration: %d, resid = %le\n", iter, r_norm);
//fflush (stdout);
// TODO: paste
if (iter)
{
//rho_1 = mv_vector_inner_product (r, rtilde, n);//in first iter equals to r_norm
if (rho_1 == 0) // failure
{
if (den_norm > epsmac)
prop->set_f (final_res_idx, r_norm / den_norm);
else
prop->set_f (final_res_idx, r_norm);
bs_throw_exception ("TFQMR: Failure - rho_1 == 0");
}
sum_vector_n (u, (t_double)1., res, beta, p, n); //p[n] = u[n]+beta*res
//v.assign (n, 0);
memset (v, 0, sizeof (t_double) * n);
matrix->matrix_vector_product (sp_p, sp_v); //v[n]=Ap[n]
}
sigma = mv_vector_inner_product_n (rtilde, v, n); //sigma=(rtilde,v[n-1])
alpha = rho_1/sigma;
// tmp = M^(-1)*v
if (prec)
{
if (prec->solve_prec (matrix, sp_v, sp_tmp))
{
bs_throw_exception ("TFQMR: Preconditioner failed");
}
memcpy (v, tmp, sizeof (t_double) * n);
//v.assign (tmp.begin (), tmp.end ());
}
sum_vector_n (u, (t_double)1., v, -alpha, q, n); //q[n] = u[n-1]-alpha*v[n-1]
sum_vector_n (u, (t_double)1., q, (t_double)1., res, n); //res = u[n-1]+q[n]
//tmp.assign (n, 0);
memset (tmp, 0, sizeof (t_double) * n);
matrix->matrix_vector_product (sp_res, sp_tmp);// tmp=A*res
sum_vector_n (r, (t_double)1., tmp, -alpha, r, n);// r=r-alpha*res
//r_norm_old = r_norm;
r_norm = mv_vector_inner_product_n (r, r, n);
for (m = 1; m <= 2 ; m++)
{
if (m == 1) // m is odd
{
memcpy (y, u, sizeof (t_double) * n);
//y.assign (u.begin (), u.end ());
w_norm = sqrt(r_norm * r_norm);
}
else // m is even
{
memcpy (y, q, sizeof (t_double) * n);
//y.assign (q.begin (), q.end ());
w_norm = sqrt(r_norm);
}
sum_vector_n (y, (t_double)1., d, eta*nu*nu/alpha, d, n); //d[m] = y[m] + (eta[m-1]*nu[m-1]^2/alpha[n-1])*d[m-1]
nu = w_norm/tau; //nu[m]=||w[m+1]||/tau[m-1]
c = 1./sqrt (1. + nu*nu);
tau = tau*c*nu; //tau[m]=tau[m-1]nu[m]c[m]
eta = c*c*alpha; //eta[m]=c[m]^2*alpha[n-1]
//SUM_VECTOR(x,d,1,alpha,x_cgs,k,n); //x_cgs[n] = x[2n-1]+alpha[n-1]*d[2n]
sum_vector_n (sol, (t_double)1., d, eta, sol, n); //x[m] = x[m-1]+eta[m]*d[m]
if (r_norm <= tol)
{
count = 1;
break;
}
}
if (r_norm <= tol)
break;
rho_1 = mv_vector_inner_product_n (r, rtilde, n);//in first iter equals to r_norm
beta = rho_1 / rho_2;
// rhat = M^(-1) * r;
if (prec)
{
if (prec->solve_prec (matrix, sp_r, sp_rhat))
{
bs_throw_exception ("TFQMR: Preconditioner failed");
}
}
else // no precondition (preconditioner=identity_matrix)
{
memcpy (rhat, r, sizeof (t_double) * n);
//rhat.assign (r.begin (), r.end ());
}
sum_vector_n (rhat, (t_double)1., q, beta, u, n); //u[n] = r[n]+beta*q[n]
sum_vector_n (q, (t_double)1., p, beta, res, n); //res = q[n]+beta*p[n-1]
rho_2 = rho_1;
}
//TODO: end
prop->set_i (iters_idx, iter + 1);
prop->set_b (success_idx, true);
/*
//additional checking convergence
mv_calc_lin_comb (matrix, -1.0, 1.0, solution, rhs, r);
r_norm = mv_vector_inner_product (r, r, n);
*/
if (den_norm > epsmac)
prop->set_f (final_res_idx, r_norm / den_norm);
else
prop->set_f (final_res_idx, r_norm);
//printf ("TFQMR OK! iters = %d, resid = %le\n", lprop->iters, lprop->final_resid);
//OMP_TIME_MEASURE_END (tfqmr_solve_timer);
return 0;
}
/*
* Initialize nonvolatile variable table from sprom.
* Return 0 on success, nonzero on error.
*/
static int
initvars_srom_pci(void *sbh, void *curmap, char **vars, uint *count)
{
uint16 w, *b;
uint8 sromrev = 0;
struct ether_addr ea;
char eabuf[32];
uint32 w32;
int woff, i;
char *vp, *base;
osl_t *osh = sb_osh(sbh);
bool flash = FALSE;
char name[SB_DEVPATH_BUFSZ+16], *value;
char devpath[SB_DEVPATH_BUFSZ];
int err;
/*
* Apply CRC over SROM content regardless SROM is present or not,
* and use variable <devpath>sromrev's existance in flash to decide
* if we should return an error when CRC fails or read SROM variables
* from flash.
*/
b = MALLOC(osh, SROM_MAX);
ASSERT(b);
if (!b)
return -2;
err = sprom_read_pci(osh, (void*)((int8*)curmap + PCI_BAR0_SPROM_OFFSET), 0, b,
64, TRUE);
if (b[SROM4_SIGN] == SROM4_SIGNATURE) {
/* sromrev >= 4, read more */
err = sprom_read_pci(osh, (void*)((int8*)curmap + PCI_BAR0_SPROM_OFFSET), 0, b, SROM4_WORDS, TRUE);
sromrev = b[SROM4_WORDS - 1] & 0xff;
} else if (err == 0) {
/* srom is good and is rev < 4 */
/* top word of sprom contains version and crc8 */
sromrev = b[63] & 0xff;
/* bcm4401 sroms misprogrammed */
if (sromrev == 0x10)
sromrev = 1;
}
if (err) {
#ifdef WLTEST
BS_ERROR(("SROM Crc Error, so see if we could use a default\n"));
w32 = OSL_PCI_READ_CONFIG(osh, PCI_SPROM_CONTROL, sizeof(uint32));
if (w32 & SPROM_OTPIN_USE) {
BS_ERROR(("srom crc failed with OTP, use default vars....\n"));
vp = base = mfgsromvars;
if (sb_chip(sbh) == BCM4311_CHIP_ID) {
BS_ERROR(("setting the devid to be 4311\n"));
vp += sprintf(vp, "devid=0x4311");
vp++;
}
bcopy(defaultsromvars, vp, MFGSROM_DEFVARSLEN);
vp += MFGSROM_DEFVARSLEN;
goto varsdone;
} else {
BS_ERROR(("srom crc failed with SPROM....\n"));
#endif /* WLTEST */
if ((err = sb_devpath(sbh, devpath, sizeof(devpath))))
return err;
sprintf(name, "%ssromrev", devpath);
if (!(value = getvar(NULL, name)))
return (-1);
sromrev = (uint8)bcm_strtoul(value, NULL, 0);
flash = TRUE;
#ifdef WLTEST
}
#endif /* WLTEST */
}
/* srom version check */
if (sromrev > 4)
return (-2);
ASSERT(vars);
ASSERT(count);
base = vp = MALLOC(osh, VARS_MAX);
ASSERT(vp);
if (!vp)
return -2;
/* read variables from flash */
if (flash) {
if ((err = initvars_flash(osh, &vp, VARS_MAX, devpath)))
goto err;
goto varsdone;
}
vp += sprintf(vp, "sromrev=%d", sromrev);
vp++;
if (sromrev >= 4) {
uint path, pathbase;
const uint pathbases[MAX_PATH] = {SROM4_PATH0, SROM4_PATH1,
SROM4_PATH2, SROM4_PATH3};
vp += sprintf(vp, "boardrev=%d", b[SROM4_BREV]);
vp++;
vp += sprintf(vp, "boardflags=%d", (b[SROM4_BFL1] << 16) | b[SROM4_BFL0]);
vp++;
vp += sprintf(vp, "boardflags2=%d", (b[SROM4_BFL3] << 16) | b[SROM4_BFL2]);
vp++;
/* The macaddr */
ea.octet[0] = (b[SROM4_MACHI] >> 8) & 0xff;
ea.octet[1] = b[SROM4_MACHI] & 0xff;
ea.octet[2] = (b[SROM4_MACMID] >> 8) & 0xff;
ea.octet[3] = b[SROM4_MACMID] & 0xff;
ea.octet[4] = (b[SROM4_MACLO] >> 8) & 0xff;
ea.octet[5] = b[SROM4_MACLO] & 0xff;
bcm_ether_ntoa(&ea, eabuf);
vp += sprintf(vp, "macaddr=%s", eabuf);
vp++;
w = b[SROM4_CCODE];
if (w == 0)
vp += sprintf(vp, "ccode=");
else
vp += sprintf(vp, "ccode=%c%c", (w >> 8), (w & 0xff));
vp++;
vp += sprintf(vp, "regrev=%d", b[SROM4_REGREV]);
vp++;
w = b[SROM4_LEDBH10];
if ((w != 0) && (w != 0xffff)) {
/* ledbh0 */
vp += sprintf(vp, "ledbh0=%d", (w & 0xff));
vp++;
/* ledbh1 */
vp += sprintf(vp, "ledbh1=%d", (w >> 8) & 0xff);
vp++;
}
int gs_solver::solve (sp_matrix_t matrix, spv_double sp_rhs, spv_double sp_sol)
{
BS_ERROR (matrix, "gs_solve");
BS_ERROR (sp_rhs->size (), "gs_solve");
BS_ERROR (sp_sol->size (), "gs_solve");
BS_ERROR (prop, "gs_solve");
int iter;
const double epsmac = 1e-24;
t_double r_norm, b_norm, den_norm;
t_double *rhs = &(*sp_rhs)[0];
//t_double *sol = &(*sp_sol)[0];
sp_bcsr_t bcsr;
if (!dynamic_cast<bcsr_t *> (matrix.lock ()))
{
bcsr = matrix;
BS_ASSERT (bcsr);
}
t_long n = matrix->get_n_rows () * matrix->get_n_block_size ();
spv_long flags;
t_double tol = prop->get_f (tol_idx);
tol *= tol;
int max_iter = prop->get_i (max_iters_idx);
prop->reset_i (cf_type_idx);
prop->set_b (success_idx, false);
matrix->init_vector (sp_r);
t_double *r = &(*sp_r)[0];
matrix->calc_lin_comb (-1.0, 1.0, sp_sol, sp_rhs, sp_r);
r_norm = mv_vector_inner_product_n (r, r, n);
b_norm = sqrt (mv_vector_inner_product_n (rhs, rhs, n));
if (b_norm > epsmac) // choose convergence criterion
{
// |r_i|/|b| <= eps if |b| > 0
tol *= b_norm;
den_norm = b_norm;
}
else // (r_norm > epsmac)
{
// |r_i|/|r0| <= eps if |b| = 0
tol *= r_norm;
den_norm = r_norm;
}
// main loop
for (iter = 0; iter < max_iter; ++iter)
{
smooth (bcsr, flags, 1, sp_rhs, sp_sol);
matrix->calc_lin_comb (-1.0, 1.0, sp_sol, sp_rhs, sp_r);
r_norm = mv_vector_inner_product_n (r, r, n);
if (r_norm <= tol) // initial guess quite good
break;
} // end of main loop
if (iter < max_iter)
{
prop->set_i (iters_idx, iter + 1);
prop->set_b (success_idx, true);
}
else
{
prop->set_i (iters_idx, iter + 1);
prop->set_b (success_idx, false);
}
if (den_norm > epsmac)
prop->set_f(final_res_idx, r_norm / den_norm);
else
prop->set_f (final_res_idx, r_norm);
BOSOUT (section::solvers, level::low) << "r_norm = " << r_norm << " r_norm / den_norm = " << r_norm / den_norm << " iter = " << iter << bs_end;
return 0;
}
