void Msap_eo_old(spinor * const P, spinor * const Q, const int Ncy, const int Niter) {
int blk, ncy = 0, eo, vol, vols;
spinor * r, * a, * b, * c;
double nrm;
double musave = g_mu;
double kappasave = g_kappa;
spinor * b_even, * b_odd, * a_even, * a_odd;
spinor ** solver_field = NULL;
// also get space for mrblk! 6 = 3+3
const int nr_sf = 6;
if(kappa_dflgen > 0) {
g_kappa = kappa_dfl;
}
if(mu_dflgen > -10) {
g_mu = mu_dfl;
// make sure the sign is correct!
if(g_mu*musave < 0) g_mu *= -1.;
}
boundary(g_kappa);
/*
* here it would be probably better to get the working fields as a parameter
* from the calling function
*/
vols = block_list[0].volume/2+block_list[0].spinpad;
vol = block_list[0].volume/2;
init_solver_field(&solver_field, nb_blocks*2*vols, nr_sf);
r = solver_field[0];
a = solver_field[1];
b = solver_field[2];
for(ncy = 0; ncy < Ncy; ncy++) {
/* compute the global residue */
/* this can be done more efficiently */
/* here only a naive implementation */
for(eo = 0; eo < 2; eo++) {
D_psi(r, P);
diff(r, Q, r, VOLUME);
nrm = square_norm(r, VOLUME, 1);
if(g_proc_id == 0 && g_debug_level > 2 && eo == 0) {
printf("Msap_eo: %d %1.3e mu = %e\n", ncy, nrm, g_mu/2./g_kappa);
fflush(stdout);
}
/* choose the even (odd) block */
// rely on nested parallelism
//
#ifdef TM_USE_OMP
# pragma omp parallel for private (a_even, a_odd, b_even, b_odd, c)
#endif
for (blk = 0; blk < nb_blocks; blk++) {
b_even = b + blk*2*vols;
b_odd = b +blk*2*vols + vols;
a_even = a + blk*2*vols;
a_odd = a + blk*2*vols + vols;
c = solver_field[3] + blk*vols;
if(block_list[blk].evenodd == eo) {
/* get part of r corresponding to block blk into b_even and b_odd */
copy_global_to_block_eo(b_even, b_odd, r, blk);
if(g_c_sw > 0) {
assign_mul_one_sw_pm_imu_inv_block(EE, a_even, b_even, g_mu, &block_list[blk]);
Block_H_psi(&block_list[blk], a_odd, a_even, OE);
/* a_odd = b_odd - a_odd */
diff(a_odd, b_odd, a_odd, vol);
mrblk(b_odd, a_odd, solver_field[3] + blk*2*3*vols, Niter, 1.e-31, 1, vol, &Msw_plus_block_psi, blk);
Block_H_psi(&block_list[blk], b_even, b_odd, EO);
assign(c, b_even, vol);
assign_mul_one_sw_pm_imu_inv_block(EE, b_even, c, g_mu, &block_list[blk]);
}
else {
assign_mul_one_pm_imu_inv(a_even, b_even, +1., vol);
Block_H_psi(&block_list[blk], a_odd, a_even, OE);
/* a_odd = b_odd - a_odd */
diff(a_odd, b_odd, a_odd, vol);
mrblk(b_odd, a_odd, solver_field[3] + blk*2*3*vols, Niter, 1.e-31, 1, vol, &Mtm_plus_block_psi, blk);
Block_H_psi(&block_list[blk], b_even, b_odd, EO);
mul_one_pm_imu_inv(b_even, +1., vol);
}
/* a_even = a_even - b_even */
diff(a_even, a_even, b_even, vol);
/* add even and odd part up to full spinor P */
add_eo_block_to_global(P, a_even, b_odd, blk);
}
}
}
}
finalize_solver(solver_field, nr_sf);
g_mu = musave;
g_kappa = kappasave;
boundary(g_kappa);
return;
}
void Msap_eo(spinor * const P, spinor * const Q, const int Ncy) {
int blk, ncy = 0, eo, vol;
spinor * r, * a, * b;
double nrm;
spinor * b_even, * b_odd, * a_even, * a_odd;
spinor ** solver_field = NULL;
const int nr_sf = 3;
/*
* here it would be probably better to get the working fields as a parameter
* from the calling function
*/
init_solver_field(&solver_field, VOLUME, nr_sf);
r = solver_field[0];
a = solver_field[1];
b = solver_field[2];
vol = block_list[0].volume/2;
b_even = b;
b_odd = b + vol + 1;
a_even = a;
a_odd = a + vol + 1;
for(ncy = 0; ncy < Ncy; ncy++) {
/* compute the global residue */
/* this can be done more efficiently */
/* here only a naive implementation */
for(eo = 0; eo < 2; eo++) {
D_psi(r, P);
diff(r, Q, r, VOLUME);
nrm = square_norm(r, VOLUME, 1);
if(g_proc_id == 0 && g_debug_level > 1 && eo == 1) {
printf("Msap: %d %1.3e\n", ncy, nrm);
}
/* choose the even (odd) block */
for (blk = 0; blk < nb_blocks; blk++) {
if(block_list[blk].evenodd == eo) {
/* get part of r corresponding to block blk into b_even and b_odd */
copy_global_to_block_eo(b_even, b_odd, r, blk);
assign_mul_one_pm_imu_inv(a_even, b_even, +1., vol);
Block_H_psi(&block_list[blk], a_odd, a_even, OE);
/* a_odd = a_odd - b_odd */
assign_mul_add_r(a_odd, -1., b_odd, vol);
mrblk(b_odd, a_odd, 3, 1.e-31, 1, vol, &Mtm_plus_block_psi, blk);
Block_H_psi(&block_list[blk], b_even, b_odd, EO);
mul_one_pm_imu_inv(b_even, +1., vol);
/* a_even = a_even - b_even */
assign_add_mul_r(a_even, b_even, -1., vol);
/* add even and odd part up to full spinor P */
add_eo_block_to_global(P, a_even, b_odd, blk);
}
}
}
}
finalize_solver(solver_field, nr_sf);
return;
}