void
qopWilsonSolve(Layout *l, real *x, real *u[8], real mass, real *y,
double rsq, char *sub)
{
QDP_ColorMatrix *qu[4];
QDP_DiracFermion *out, *in;
in = QDP_create_D();
out = QDP_create_D();
unpackD(l, in, y);
unpackD(l, out, x);
for(int i=0; i<4; i++) {
qu[i] = QDP_create_M();
unpackM(l, qu[i], u[2*i]);
QLA_Real two = 2;
QDP_M_eq_r_times_M(qu[i], &two, qu[i], QDP_all);
}
QOP_FermionLinksWilson *fla;
fla = QOP_wilson_create_L_from_qdp(qu, NULL);
QOP_evenodd_t eo=QOP_EVENODD;
if(sub[0]=='e') {
eo = QOP_EVEN;
}
if(sub[0]=='o') {
eo = QOP_ODD;
}
QOP_info_t info = QOP_INFO_ZERO;
QOP_invert_arg_t inv_arg = QOP_INVERT_ARG_DEFAULT;
QOP_resid_arg_t res_arg = QOP_RESID_ARG_DEFAULT;
res_arg.rsqmin = rsq;
inv_arg.max_iter = 1000;
inv_arg.restart = 500;
inv_arg.max_restarts = 5;
inv_arg.evenodd = eo;
inv_arg.mixed_rsq = 0;
QDP_D_eq_zero(out, QDP_even);
//QOP_verbose(3);
QOP_wilson_invert_qdp(&info, fla, &inv_arg, &res_arg, mass, out, in);
//QLA_Real n2;
//QDP_r_eq_norm2_D(&n2, (QDP_DiracFermion*)out, QDP_all);
printf0("QOP its: %i\n", res_arg.final_iter);
packD(l, x, out);
QDP_destroy_D(in);
QDP_destroy_D(out);
for(int i=0; i<4; i++) {
QDP_destroy_M(qu[i]);
}
}
void
qopWilsonDslash(Layout *l, real *x, real *u[8], real mass, int sign,
real *y, char *sub)
{
QDP_ColorMatrix *qu[4];
QDP_DiracFermion *out, *in;
in = QDP_create_D();
out = QDP_create_D();
unpackD(l, in, y);
unpackD(l, out, x);
for(int i=0; i<4; i++) {
qu[i] = QDP_create_M();
unpackM(l, qu[i], u[2*i]);
QLA_Real two = 2;
QDP_M_eq_r_times_M(qu[i], &two, qu[i], QDP_all);
}
QOP_FermionLinksWilson *fla;
fla = QOP_wilson_create_L_from_qdp(qu, NULL);
QOP_evenodd_t eoOut=QOP_EVENODD, eoIn=QOP_EVENODD;
if(sub[0]=='e') {
eoOut = QOP_EVEN;
eoIn = QOP_ODD;
}
if(sub[0]=='o') {
eoOut = QOP_ODD;
eoIn = QOP_EVEN;
}
real kappa = 0.5/(4+mass);
QOP_wilson_dslash_qdp(NULL, fla, kappa, sign, out, in, eoOut, eoIn);
QLA_Real n2;
QDP_r_eq_norm2_D(&n2, out, QDP_all);
printf0("out2: %g\n", n2);
packD(l, x, out);
QDP_destroy_D(in);
QDP_destroy_D(out);
for(int i=0; i<4; i++) {
QDP_destroy_M(qu[i]);
}
}
void
qopWilsonSolveMulti(Layout *l, real *x[], real *u[8], double masses[],
real *y, int nmasses, double rsq, char *sub)
{
QDP_ColorMatrix *qu[4];
QDP_DiracFermion *out[nmasses], *in, **outp;
outp = out;
in = QDP_create_D();
unpackD(l, in, y);
for(int i=0; i<nmasses; i++) {
out[i] = QDP_create_D();
unpackD(l, out[i], x[i]);
QDP_D_eq_zero(out[i], QDP_even);
}
for(int i=0; i<4; i++) {
qu[i] = QDP_create_M();
unpackM(l, qu[i], u[2*i]);
QLA_Real two = 2;
QDP_M_eq_r_times_M(qu[i], &two, qu[i], QDP_all);
}
QOP_FermionLinksWilson *fla;
fla = QOP_wilson_create_L_from_qdp(qu, NULL);
#if 0
QOP_evenodd_t eo = QOP_EVENODD;
if(sub[0]=='e') {
eo = QOP_EVEN;
}
if(sub[0]=='o') {
eo = QOP_ODD;
}
#endif
QOP_evenodd_t eo = QOP_EVEN;
QOP_info_t info = QOP_INFO_ZERO;
QOP_invert_arg_t inv_arg = QOP_INVERT_ARG_DEFAULT;
inv_arg.max_iter = 1000;
inv_arg.restart = 500;
inv_arg.max_restarts = 5;
inv_arg.evenodd = eo;
inv_arg.mixed_rsq = 0;
QOP_resid_arg_t res_arg = QOP_RESID_ARG_DEFAULT;
res_arg.rsqmin = rsq;
QOP_resid_arg_t *ra[nmasses];
QOP_resid_arg_t **rap = ra;
real mf[nmasses], *mfp;
mfp = mf;
for(int i=0; i<nmasses; i++) {
ra[i] = &res_arg;
mf[i] = masses[i];
}
//QOP_verbose(3);
QOP_wilson_invert_multi_qdp(&info, fla, &inv_arg, &rap,
&mfp, &nmasses, &outp, &in, 1);
//QLA_Real n2;
//QDP_r_eq_norm2_D(&n2, (QDP_DiracFermion*)out, QDP_all);
printf0("QOP its: %i\n", res_arg.final_iter);
QDP_destroy_D(in);
for(int i=0; i<nmasses; i++) {
packD(l, x[i], out[i]);
QDP_destroy_D(out[i]);
}
for(int i=0; i<4; i++) {
QDP_destroy_M(qu[i]);
}
}
void
start(void)
{
double mf, best_mf;
QLA_Real plaq;
QDP_ColorMatrix **u;
QDP_DiracFermion *out, *in;
int i, st, ns, nm, bs, sti, nsi, nmi, bsi,
best_st, best_ns, best_nm, best_bs;
u = (QDP_ColorMatrix **) malloc(ndim*sizeof(QDP_ColorMatrix *));
for(i=0; i<ndim; i++) u[i] = QDP_create_M();
get_random_links(u, ndim, 0.2);
plaq = get_plaq(u);
if(QDP_this_node==0) printf("plaquette = %g\n", plaq);
out = QDP_create_D();
in = QDP_create_D();
QDP_D_eq_gaussian_S(in, rs, QDP_all);
QOP_layout_t qoplayout = QOP_LAYOUT_ZERO;
qoplayout.latdim = ndim;
qoplayout.latsize = (int *) malloc(ndim*sizeof(int));
for(i=0; i<ndim; i++) {
qoplayout.latsize[i] = lattice_size[i];
}
qoplayout.machdim = -1;
QOP_info_t info = QOP_INFO_ZERO;
QOP_invert_arg_t inv_arg = QOP_INVERT_ARG_DEFAULT;
QOP_resid_arg_t res_arg = QOP_RESID_ARG_DEFAULT;
res_arg.rsqmin = rsqmin;
inv_arg.max_iter = 600;
inv_arg.restart = 200;
inv_arg.evenodd = QOP_EVEN;
if(QDP_this_node==0) { printf("begin init\n"); fflush(stdout); }
QOP_init(&qoplayout);
if(QDP_this_node==0) { printf("begin load links\n"); fflush(stdout); }
//flw = QOP_wilson_create_L_from_qdp(u, NULL);
if(QDP_this_node==0) { printf("begin invert\n"); fflush(stdout); }
if(cgtype>=0) {
QOP_opt_t optcg;
optcg.tag = "cg";
optcg.value = cgtype;
QOP_wilson_invert_set_opts(&optcg, 1);
}
best_mf = 0;
best_st = sta[0];
best_ns = nsa[0];
best_nm = nma[0];
best_bs = bsa[0];
QOP_opt_t optst;
optst.tag = "st";
QOP_opt_t optns;
optns.tag = "ns";
QOP_opt_t optnm;
optnm.tag = "nm";
for(sti=0; sti<stn; sti++) {
if((style>=0)&&(sti!=style)) continue;
st = sta[sti];
optst.value = st;
if(QOP_wilson_invert_set_opts(&optst, 1)==QOP_FAIL) continue;
for(nsi=0; nsi<nsn; nsi++) {
ns = nsa[nsi];
optns.value = ns;
if(QOP_wilson_invert_set_opts(&optns, 1)==QOP_FAIL) continue;
for(nmi=0; nmi<nmn; nmi++) {
nm = nma[nmi];
if(nm==0) nm = ns;
optnm.value = nm;
if(QOP_wilson_invert_set_opts(&optnm, 1)==QOP_FAIL) continue;
for(bsi=0; bsi<bsn; bsi++) {
bs = bsa[bsi];
QDP_set_block_size(bs);
flw = QOP_wilson_create_L_from_qdp(u, NULL);
mf = bench_inv(&info, &inv_arg, &res_arg, out, in);
QOP_wilson_destroy_L(flw);
printf0("CONGRAD: st%2i ns%2i nm%2i bs%5i iter%5i sec%7.4f mflops = %g\n", st,
ns, nm, bs, res_arg.final_iter, info.final_sec, mf);
if(mf>best_mf) {
best_mf = mf;
best_st = st;
best_ns = ns;
best_nm = nm;
best_bs = bs;
}
}
}
}
}
flw = QOP_wilson_create_L_from_qdp(u, NULL);
optst.value = best_st;
optns.value = best_ns;
optnm.value = best_nm;
QOP_wilson_invert_set_opts(&optst, 1);
QOP_wilson_invert_set_opts(&optns, 1);
QOP_wilson_invert_set_opts(&optnm, 1);
QDP_set_block_size(best_bs);
QDP_profcontrol(1);
mf = bench_inv(&info, &inv_arg, &res_arg, out, in);
QDP_profcontrol(0);
printf0("prof: CONGRAD: st%2i ns%2i nm%2i bs%5i iter%5i sec%7.4f mflops = %g\n",
best_st, best_ns, best_nm, best_bs,
res_arg.final_iter, info.final_sec, mf);
printf0("best: CONGRAD: st%2i ns%2i nm%2i bs%5i mflops = %g\n",
best_st, best_ns, best_nm, best_bs, best_mf);
if(QDP_this_node==0) { printf("begin unload links\n"); fflush(stdout); }
//QOP_wilson_invert_unload_links();
if(QDP_this_node==0) { printf("begin finalize\n"); fflush(stdout); }
QOP_finalize();
}
