extern "C" magma_int_t
magma_scgs(
magma_s_matrix A, magma_s_matrix b, magma_s_matrix *x,
magma_s_solver_par *solver_par,
magma_queue_t queue )
{
magma_int_t info = MAGMA_NOTCONVERGED;
// prepare solver feedback
solver_par->solver = Magma_CGS;
solver_par->numiter = 0;
solver_par->spmv_count = 0;
// constants
const float c_zero = MAGMA_S_ZERO;
const float c_one = MAGMA_S_ONE;
const float c_neg_one = MAGMA_S_NEG_ONE;
// solver variables
float nom0, r0, res=0, nomb;
float rho, rho_l = c_one, alpha, beta;
magma_int_t dofs = A.num_rows* b.num_cols;
// GPU workspace
magma_s_matrix r={Magma_CSR}, rt={Magma_CSR}, r_tld={Magma_CSR},
p={Magma_CSR}, q={Magma_CSR}, u={Magma_CSR}, v={Magma_CSR}, t={Magma_CSR},
p_hat={Magma_CSR}, q_hat={Magma_CSR}, u_hat={Magma_CSR}, v_hat={Magma_CSR};
CHECK( magma_svinit( &r, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &rt,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &r_tld,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &p, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &p_hat, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &q, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &q_hat, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &u, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &u_hat, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &v, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &v_hat, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &t, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
// solver setup
CHECK( magma_sresidualvec( A, b, *x, &r, &nom0, queue));
magma_scopy( dofs, r.dval, 1, r_tld.dval, 1, queue );
solver_par->init_res = nom0;
nomb = magma_snrm2( dofs, b.dval, 1, queue );
if ( nomb == 0.0 ){
nomb=1.0;
}
if ( (r0 = nomb * solver_par->rtol) < ATOLERANCE ){
r0 = ATOLERANCE;
}
solver_par->final_res = solver_par->init_res;
solver_par->iter_res = solver_par->init_res;
if ( solver_par->verbose > 0 ) {
solver_par->res_vec[0] = (real_Double_t)nom0;
solver_par->timing[0] = 0.0;
}
if ( nom0 < r0 ) {
info = MAGMA_SUCCESS;
goto cleanup;
}
//Chronometry
real_Double_t tempo1, tempo2;
tempo1 = magma_sync_wtime( queue );
solver_par->numiter = 0;
solver_par->spmv_count = 0;
// start iteration
do
{
solver_par->numiter++;
rho = magma_sdot( dofs, r.dval, 1, r_tld.dval, 1, queue );
// rho = < r,r_tld>
if( magma_s_isnan_inf( rho ) ){
info = MAGMA_DIVERGENCE;
break;
}
if ( solver_par->numiter > 1 ) { // direction vectors
beta = rho / rho_l;
magma_scopy( dofs, r.dval, 1, u.dval, 1, queue ); // u = r
magma_saxpy( dofs, beta, q.dval, 1, u.dval, 1, queue ); // u = u + beta q
magma_sscal( dofs, beta, p.dval, 1, queue ); // p = beta*p
magma_saxpy( dofs, c_one, q.dval, 1, p.dval, 1, queue ); // p = q + beta*p
magma_sscal( dofs, beta, p.dval, 1, queue ); // p = beta*(q + beta*p)
magma_saxpy( dofs, c_one, u.dval, 1, p.dval, 1, queue ); // p = u + beta*(q + beta*p)
//u = r + beta*q;
//p = u + beta*( q + beta*p );
}
else{
magma_scopy( dofs, r.dval, 1, u.dval, 1, queue ); // u = r
magma_scopy( dofs, r.dval, 1, p.dval, 1, queue ); // p = r
}
CHECK( magma_s_spmv( c_one, A, p, c_zero, v_hat, queue )); // v = A p
solver_par->spmv_count++;
alpha = rho / magma_sdot( dofs, r_tld.dval, 1, v_hat.dval, 1, queue );
magma_scopy( dofs, u.dval, 1, q.dval, 1, queue ); // q = u
magma_saxpy( dofs, -alpha, v_hat.dval, 1, q.dval, 1, queue ); // q = u - alpha v_hat
magma_scopy( dofs, u.dval, 1, t.dval, 1, queue ); // t = q
magma_saxpy( dofs, c_one, q.dval, 1, t.dval, 1, queue ); // t = u + q
CHECK( magma_s_spmv( c_one, A, t, c_zero, rt, queue )); // t = A u_hat
solver_par->spmv_count++;
magma_saxpy( dofs, c_neg_one*alpha, rt.dval, 1, r.dval, 1, queue ); // r = r -alpha*A u_hat
magma_saxpy( dofs, alpha, t.dval, 1, x->dval, 1, queue ); // x = x + alpha u_hat
rho_l = rho;
res = magma_snrm2( dofs, r.dval, 1, queue );
if ( solver_par->verbose > 0 ) {
tempo2 = magma_sync_wtime( queue );
if ( (solver_par->numiter)%solver_par->verbose == 0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
if ( res/nomb <= solver_par->rtol || res <= solver_par->atol ){
break;
}
}
while ( solver_par->numiter+1 <= solver_par->maxiter );
tempo2 = magma_sync_wtime( queue );
solver_par->runtime = (real_Double_t) tempo2-tempo1;
float residual;
CHECK( magma_sresidualvec( A, b, *x, &r, &residual, queue));
solver_par->iter_res = res;
solver_par->final_res = residual;
if ( solver_par->numiter < solver_par->maxiter && info == MAGMA_SUCCESS ) {
info = MAGMA_SUCCESS;
} else if ( solver_par->init_res > solver_par->final_res ) {
if ( solver_par->verbose > 0 ) {
if ( (solver_par->numiter)%solver_par->verbose == 0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
info = MAGMA_SLOW_CONVERGENCE;
if( solver_par->iter_res < solver_par->rtol*solver_par->init_res ||
solver_par->iter_res < solver_par->atol ) {
info = MAGMA_SUCCESS;
}
}
else {
if ( solver_par->verbose > 0 ) {
if ( (solver_par->numiter)%solver_par->verbose == 0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
info = MAGMA_DIVERGENCE;
}
cleanup:
magma_smfree(&r, queue );
magma_smfree(&rt, queue );
magma_smfree(&r_tld, queue );
magma_smfree(&p, queue );
magma_smfree(&q, queue );
magma_smfree(&u, queue );
magma_smfree(&v, queue );
magma_smfree(&t, queue );
magma_smfree(&p_hat, queue );
magma_smfree(&q_hat, queue );
magma_smfree(&u_hat, queue );
magma_smfree(&v_hat, queue );
solver_par->info = info;
return info;
} /* magma_scgs */
extern "C" magma_int_t
magma_sbicg(
magma_s_matrix A, magma_s_matrix b, magma_s_matrix *x,
magma_s_solver_par *solver_par,
magma_queue_t queue )
{
magma_int_t info = MAGMA_NOTCONVERGED;
// prepare solver feedback
solver_par->solver = Magma_BICG;
solver_par->numiter = 0;
solver_par->spmv_count = 0;
// some useful variables
float c_zero = MAGMA_S_ZERO;
float c_one = MAGMA_S_ONE;
float c_neg_one = MAGMA_S_NEG_ONE;
magma_int_t dofs = A.num_rows * b.num_cols;
// workspace
magma_s_matrix r={Magma_CSR}, rt={Magma_CSR}, p={Magma_CSR}, pt={Magma_CSR},
z={Magma_CSR}, zt={Magma_CSR}, q={Magma_CSR}, y={Magma_CSR},
yt={Magma_CSR}, qt={Magma_CSR};
// need to transpose the matrix
magma_s_matrix AT={Magma_CSR}, Ah1={Magma_CSR}, Ah2={Magma_CSR};
CHECK( magma_svinit( &r, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &rt,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &p, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &pt,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &q, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &qt,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &y, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &yt,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &z, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &zt,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
// solver variables
float alpha, rho, beta, rho_new, ptq;
float res, nomb, nom0, r0;
// transpose the matrix
magma_smtransfer( A, &Ah1, Magma_DEV, Magma_CPU, queue );
magma_smconvert( Ah1, &Ah2, A.storage_type, Magma_CSR, queue );
magma_smfree(&Ah1, queue );
magma_smtransposeconjugate( Ah2, &Ah1, queue );
magma_smfree(&Ah2, queue );
Ah2.blocksize = A.blocksize;
Ah2.alignment = A.alignment;
magma_smconvert( Ah1, &Ah2, Magma_CSR, A.storage_type, queue );
magma_smfree(&Ah1, queue );
magma_smtransfer( Ah2, &AT, Magma_CPU, Magma_DEV, queue );
magma_smfree(&Ah2, queue );
// solver setup
CHECK( magma_sresidualvec( A, b, *x, &r, &nom0, queue));
res = nom0;
solver_par->init_res = nom0;
magma_scopy( dofs, r.dval, 1, rt.dval, 1, queue ); // rr = r
rho_new = magma_sdot( dofs, rt.dval, 1, r.dval, 1, queue ); // rho=<rr,r>
rho = alpha = MAGMA_S_MAKE( 1.0, 0. );
nomb = magma_snrm2( dofs, b.dval, 1, queue );
if ( nomb == 0.0 ){
nomb=1.0;
}
if ( (r0 = nomb * solver_par->rtol) < ATOLERANCE ){
r0 = ATOLERANCE;
}
solver_par->final_res = solver_par->init_res;
solver_par->iter_res = solver_par->init_res;
if ( solver_par->verbose > 0 ) {
solver_par->res_vec[0] = nom0;
solver_par->timing[0] = 0.0;
}
if ( nom0 < r0 ) {
info = MAGMA_SUCCESS;
goto cleanup;
}
//Chronometry
real_Double_t tempo1, tempo2;
tempo1 = magma_sync_wtime( queue );
solver_par->numiter = 0;
solver_par->spmv_count = 0;
// start iteration
do
{
solver_par->numiter++;
magma_scopy( dofs, r.dval, 1 , y.dval, 1, queue ); // y=r
magma_scopy( dofs, y.dval, 1 , z.dval, 1, queue ); // z=y
magma_scopy( dofs, rt.dval, 1 , yt.dval, 1, queue ); // yt=rt
magma_scopy( dofs, yt.dval, 1 , zt.dval, 1, queue ); // zt=yt
rho= rho_new;
rho_new = magma_sdot( dofs, rt.dval, 1, z.dval, 1, queue ); // rho=<rt,z>
if( magma_s_isnan_inf( rho_new ) ){
info = MAGMA_DIVERGENCE;
break;
}
if( solver_par->numiter==1 ){
magma_scopy( dofs, z.dval, 1 , p.dval, 1, queue ); // yt=rt
magma_scopy( dofs, zt.dval, 1 , pt.dval, 1, queue ); // zt=yt
} else {
beta = rho_new/rho;
magma_sscal( dofs, beta, p.dval, 1, queue ); // p = beta*p
magma_saxpy( dofs, c_one , z.dval, 1 , p.dval, 1, queue ); // p = z+beta*p
magma_sscal( dofs, MAGMA_S_CONJ(beta), pt.dval, 1, queue ); // pt = beta*pt
magma_saxpy( dofs, c_one , zt.dval, 1 , pt.dval, 1, queue ); // pt = zt+beta*pt
}
CHECK( magma_s_spmv( c_one, A, p, c_zero, q, queue )); // v = Ap
CHECK( magma_s_spmv( c_one, AT, pt, c_zero, qt, queue )); // v = Ap
solver_par->spmv_count++;
solver_par->spmv_count++;
ptq = magma_sdot( dofs, pt.dval, 1, q.dval, 1, queue );
alpha = rho_new /ptq;
magma_saxpy( dofs, alpha, p.dval, 1 , x->dval, 1, queue ); // x=x+alpha*p
magma_saxpy( dofs, c_neg_one * alpha, q.dval, 1 , r.dval, 1, queue ); // r=r+alpha*q
magma_saxpy( dofs, c_neg_one * MAGMA_S_CONJ(alpha), qt.dval, 1 , rt.dval, 1, queue ); // r=r+alpha*q
res = magma_snrm2( dofs, r.dval, 1, queue );
if ( solver_par->verbose > 0 ) {
tempo2 = magma_sync_wtime( queue );
if ( (solver_par->numiter)%solver_par->verbose==0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
if ( res/nomb <= solver_par->rtol || res <= solver_par->atol ){
break;
}
}
while ( solver_par->numiter+1 <= solver_par->maxiter );
tempo2 = magma_sync_wtime( queue );
solver_par->runtime = (real_Double_t) tempo2-tempo1;
float residual;
CHECK( magma_sresidualvec( A, b, *x, &r, &residual, queue));
solver_par->iter_res = res;
solver_par->final_res = residual;
if ( solver_par->numiter < solver_par->maxiter ) {
info = MAGMA_SUCCESS;
} else if ( solver_par->init_res > solver_par->final_res ) {
if ( solver_par->verbose > 0 ) {
if ( (solver_par->numiter)%solver_par->verbose==0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
info = MAGMA_SLOW_CONVERGENCE;
if( solver_par->iter_res < solver_par->rtol*solver_par->init_res ||
solver_par->iter_res < solver_par->atol ) {
info = MAGMA_SUCCESS;
}
}
else {
if ( solver_par->verbose > 0 ) {
if ( (solver_par->numiter)%solver_par->verbose==0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
info = MAGMA_DIVERGENCE;
}
cleanup:
magma_smfree(&r, queue );
magma_smfree(&rt, queue );
magma_smfree(&p, queue );
magma_smfree(&pt, queue );
magma_smfree(&q, queue );
magma_smfree(&qt, queue );
magma_smfree(&y, queue );
magma_smfree(&yt, queue );
magma_smfree(&z, queue );
magma_smfree(&zt, queue );
magma_smfree(&AT, queue );
magma_smfree(&Ah1, queue );
magma_smfree(&Ah2, queue );
solver_par->info = info;
return info;
} /* magma_sbicg */
extern "C" magma_int_t
magma_sbicgstab_merge(
magma_s_matrix A, magma_s_matrix b, magma_s_matrix *x,
magma_s_solver_par *solver_par,
magma_queue_t queue )
{
magma_int_t info = MAGMA_NOTCONVERGED;
// prepare solver feedback
solver_par->solver = Magma_BICGSTAB;
solver_par->numiter = 0;
solver_par->spmv_count = 0;
// some useful variables
float c_zero = MAGMA_S_ZERO;
float c_one = MAGMA_S_ONE;
magma_int_t dofs = A.num_rows * b.num_cols;
// workspace
magma_s_matrix r={Magma_CSR}, rr={Magma_CSR}, p={Magma_CSR}, v={Magma_CSR},
s={Magma_CSR}, t={Magma_CSR}, d1={Magma_CSR}, d2={Magma_CSR};
CHECK( magma_svinit( &r, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &rr,Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &p, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &v, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &s, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &t, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &d1, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
CHECK( magma_svinit( &d2, Magma_DEV, A.num_rows, b.num_cols, c_zero, queue ));
// solver variables
float alpha, beta, omega, rho_old, rho_new;
float betanom, nom0, r0, res, nomb;
res=0;
//float nom;
//float den;
// solver setup
CHECK( magma_sresidualvec( A, b, *x, &r, &nom0, queue));
magma_scopy( dofs, r.dval, 1, rr.dval, 1, queue ); // rr = r
betanom = nom0;
//nom = nom0*nom0;
rho_new = magma_sdot( dofs, r.dval, 1, r.dval, 1, queue ); // rho=<rr,r>
rho_old = omega = alpha = MAGMA_S_MAKE( 1.0, 0. );
solver_par->init_res = nom0;
CHECK( magma_s_spmv( c_one, A, r, c_zero, v, queue )); // z = A r
//den = MAGMA_S_REAL( magma_sdot( dofs, v.dval, 1, r.dval, 1), queue ); // den = z' * r
nomb = magma_snrm2( dofs, b.dval, 1, queue );
if ( nomb == 0.0 ){
nomb=1.0;
}
if ( (r0 = nomb * solver_par->rtol) < ATOLERANCE ){
r0 = ATOLERANCE;
}
solver_par->final_res = solver_par->init_res;
solver_par->iter_res = solver_par->init_res;
if ( solver_par->verbose > 0 ) {
solver_par->res_vec[0] = nom0;
solver_par->timing[0] = 0.0;
}
if ( nomb < r0 ) {
info = MAGMA_SUCCESS;
goto cleanup;
}
//Chronometry
real_Double_t tempo1, tempo2;
tempo1 = magma_sync_wtime( queue );
solver_par->numiter = 0;
solver_par->spmv_count = 0;
// start iteration
do
{
solver_par->numiter++;
rho_old = rho_new; // rho_old=rho
rho_new = magma_sdot( dofs, rr.dval, 1, r.dval, 1, queue ); // rho=<rr,r>
beta = rho_new/rho_old * alpha/omega; // beta=rho/rho_old *alpha/omega
if( magma_s_isnan_inf( beta ) ){
info = MAGMA_DIVERGENCE;
break;
}
// p = r + beta * ( p - omega * v )
magma_sbicgstab_1(
r.num_rows,
r.num_cols,
beta,
omega,
r.dval,
v.dval,
p.dval,
queue );
CHECK( magma_s_spmv( c_one, A, p, c_zero, v, queue )); // v = Ap
solver_par->spmv_count++;
//alpha = rho_new / tmpval;
alpha = rho_new /magma_sdot( dofs, rr.dval, 1, v.dval, 1, queue );
if( magma_s_isnan_inf( alpha ) ){
info = MAGMA_DIVERGENCE;
break;
}
// s = r - alpha v
magma_sbicgstab_2(
r.num_rows,
r.num_cols,
alpha,
r.dval,
v.dval,
s.dval,
queue );
CHECK( magma_s_spmv( c_one, A, s, c_zero, t, queue )); // t=As
solver_par->spmv_count++;
omega = magma_sdot( dofs, t.dval, 1, s.dval, 1, queue ) // omega = <s,t>/<t,t>
/ magma_sdot( dofs, t.dval, 1, t.dval, 1, queue );
// x = x + alpha * p + omega * s
// r = s - omega * t
magma_sbicgstab_3(
r.num_rows,
r.num_cols,
alpha,
omega,
p.dval,
s.dval,
t.dval,
x->dval,
r.dval,
queue );
res = betanom = magma_snrm2( dofs, r.dval, 1, queue );
//nom = betanom*betanom;
if ( solver_par->verbose > 0 ) {
tempo2 = magma_sync_wtime( queue );
if ( (solver_par->numiter)%solver_par->verbose==0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) res;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
if ( res/nomb <= solver_par->rtol || res <= solver_par->atol ){
break;
}
}
while ( solver_par->numiter+1 <= solver_par->maxiter );
tempo2 = magma_sync_wtime( queue );
solver_par->runtime = (real_Double_t) tempo2-tempo1;
float residual;
CHECK( magma_sresidualvec( A, b, *x, &r, &residual, queue));
solver_par->iter_res = res;
solver_par->final_res = residual;
if ( solver_par->numiter < solver_par->maxiter && info == MAGMA_SUCCESS ) {
info = MAGMA_SUCCESS;
} else if ( solver_par->init_res > solver_par->final_res ) {
if ( solver_par->verbose > 0 ) {
if ( (solver_par->numiter)%solver_par->verbose==0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) betanom;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
info = MAGMA_SLOW_CONVERGENCE;
if( solver_par->iter_res < solver_par->rtol*solver_par->init_res ||
solver_par->iter_res < solver_par->atol ) {
info = MAGMA_SUCCESS;
}
}
else {
if ( solver_par->verbose > 0 ) {
if ( (solver_par->numiter)%solver_par->verbose==0 ) {
solver_par->res_vec[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) betanom;
solver_par->timing[(solver_par->numiter)/solver_par->verbose]
= (real_Double_t) tempo2-tempo1;
}
}
info = MAGMA_DIVERGENCE;
}
cleanup:
magma_smfree(&r, queue );
magma_smfree(&rr, queue );
magma_smfree(&p, queue );
magma_smfree(&v, queue );
magma_smfree(&s, queue );
magma_smfree(&t, queue );
magma_smfree(&d1, queue );
magma_smfree(&d2, queue );
solver_par->info = info;
return info;
} /* magma_sbicgstab_merge */