QOP_status_t initialize_qop(){
QOP_status_t status;
static int latsize[4];
static QOP_layout_t layout;
latsize[0] = nx;
latsize[1] = ny;
latsize[2] = nz;
latsize[3] = nt;
if(is_qop_initialized)return QOP_SUCCESS;
layout.node_number = milc_node_number;
layout.node_index = milc_node_index;
layout.latdim = 4;
layout.latsize = latsize;
layout.machdim = 4;
layout.machsize = (int *)get_logical_dimensions();
layout.this_node = this_node;
layout.sites_on_node = sites_on_node;
status = QOP_init(&layout);
if(status == QOP_SUCCESS)
is_qop_initialized = 1;
return status;
}
void
qopWilsonDslashSetup(Layout *l)
{
int nd = l->nDim;
int *ls = l->physGeom;
//printf("%i nd: %i\t", myrank, nd);
//for(int i=0; i<nd; i++) printf(" %i", ls[i]);
//printf("\n");
//QDP_finalize();
//exit(1);
QDP_set_latsize(nd, ls);
//QMP_barrier();
//TRACE_ALL;
QDP_create_layout();
//TRACE_ALL;
//QMP_barrier();
QOP_layout_t qoplayout = QOP_LAYOUT_ZERO;
qoplayout.latdim = nd;
qoplayout.latsize = (int *) malloc(nd*sizeof(int));
for(int i=0; i<nd; i++) {
qoplayout.latsize[i] = ls[i];
}
qoplayout.machdim = -1;
QOP_init(&qoplayout);
}
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();
}
void
start(void)
{
double mf, best_mf;
QLA_Real plaq;
QDP_ColorMatrix **u;
int i, bs, bsi, 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, 0.3);
plaq = get_plaq(u);
if(QDP_this_node==0) printf("plaquette = %g\n", plaq);
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;
if(QDP_this_node==0) { printf("begin init\n"); fflush(stdout); }
QOP_init(&qoplayout);
gauge = QOP_create_G_from_qdp(u);
QOP_Force *force;
QDP_ColorMatrix *cm[4];
for(i=0; i<4; i++) {
cm[i] = QDP_create_M();
QDP_M_eq_zero(cm[i], QDP_all);
}
QOP_gauge_coeffs_t gcoeffs = QOP_GAUGE_COEFFS_ZERO;
gcoeffs.plaquette = 0.2;
gcoeffs.rectangle = 0.2;
gcoeffs.parallelogram = 0.2;
gcoeffs.adjoint_plaquette = 0.2;
force = QOP_create_F_from_qdp(cm);
mf = bench_action(&gcoeffs, force);
QOP_destroy_F(force);
printf0("action: sec%7.4f mflops = %g\n", secs, mf);
if(QDP_this_node==0) { printf("begin force\n"); fflush(stdout); }
best_mf = 0;
best_bs = bsa[0];
for(bsi=0; bsi<bsn; bsi++) {
bs = bsa[bsi];
QDP_set_block_size(bs);
force = QOP_create_F_from_qdp(cm);
mf = bench_force(&gcoeffs, force);
QOP_destroy_F(force);
printf0("GF: bs%5i sec%7.4f mflops = %g\n", bs, secs, mf);
if(mf>best_mf) {
best_mf = mf;
best_bs = bs;
}
}
QDP_set_block_size(best_bs);
QDP_profcontrol(1);
force = QOP_create_F_from_qdp(cm);
mf = bench_force(&gcoeffs, force);
QDP_profcontrol(0);
printf0("prof: GF: bs%5i sec%7.4f mflops = %g\n", best_bs, secs, mf);
printf0("best: GF: bs%5i mflops = %g\n", best_bs, best_mf);
if(QDP_this_node==0) { printf("begin unload links\n"); fflush(stdout); }
//QOP_asqtad_invert_unload_links();
if(QDP_this_node==0) { printf("begin finalize\n"); fflush(stdout); }
QOP_finalize();
}
