static int
qopqdp_gauge_zero(lua_State *L)
{
qassert(lua_gettop(L)==1);
gauge_t *g = qopqdp_gauge_check(L, -1);
for(int i=0; i<g->nd; i++) {
QDP_M_eq_zero(g->links[i], QDP_all_L(g->qlat));
}
return 0;
}
void
QOP_hisq_force_multi_fnmat2_qdp(QOP_info_t *info,
QOP_FermionLinksHisq *flh,
QDP_ColorMatrix *force[],
QOP_hisq_coeffs_t *hisq_coeff,
REAL *residues,
QDP_ColorVector *x[],
int *n_orders_naik)
{
#define NC QDP_get_nc(force[0])
double dtime = QOP_time();
QDP_ColorMatrix *deriv[4];
for(int mu=0; mu<4; mu++) {
deriv[mu] = QDP_create_M();
QDP_M_eq_zero(deriv[mu], QDP_all);
}
QOP_hisq_deriv_multi_fnmat2_qdp(info, flh, deriv, hisq_coeff, residues, x, n_orders_naik);
// contraction with the link in question should be done here,
// after contributions from all levels of smearing are taken into account
// Put antihermitian traceless part into momentum
// add force to momentum
QDP_ColorMatrix *mtmp = QDP_create_M();
for(int dir=0; dir<4; dir++) {
QDP_M_eq_M_times_Ma(mtmp, flh->U_links[dir], deriv[dir], QDP_all);
QDP_M_eq_antiherm_M(deriv[dir], mtmp, QDP_all);
QDP_M_peq_M(force[dir], deriv[dir], QDP_all);
}
info->final_flop += (4.*(198+24+18))*QDP_sites_on_node;
QDP_destroy_M(mtmp);
for(int mu=0; mu<4; mu++) {
QDP_destroy_M(deriv[mu]);
}
info->final_sec = QOP_time() - dtime;
//QOP_printf0("HISQ force flops = %g\n", info->final_flop);
#undef NC
}
void
QOPPC(symanzik_1loop_gauge_force1) (QOP_info_t *info, QOP_GaugeField *gauge,
QOP_Force *force, QOP_gauge_coeffs_t *coeffs, REAL eps)
{
REAL Plaq, Rect, Pgm ;
QDP_ColorMatrix *tempmom_qdp[4];
QDP_ColorMatrix *Amu[6]; // products of 2 links Unu(x)*Umu(x+nu)
QDP_ColorMatrix *tmpmat;
QDP_ColorMatrix *tmpmat1;
QDP_ColorMatrix *tmpmat2;
QDP_ColorMatrix *staples;
QDP_ColorMatrix *tmpmat3;
QDP_ColorMatrix *tmpmat4;
int i, k;
int mu, nu, sig;
double dtime;
//REAL eb3 = -eps*beta/3.0;
REAL eb3 = -eps/3.0;
int j[3][2] = {{1,2},
{0,2},
{0,1}};
// QOP_printf0("beta: %e, eb3: %e\n", beta, eb3);
dtime = -QOP_time();
for(mu=0; mu<4; mu++) {
tempmom_qdp[mu] = QDP_create_M();
QDP_M_eq_zero(tempmom_qdp[mu], QDP_all);
}
tmpmat = QDP_create_M();
for(i=0; i<QOP_common.ndim; i++) {
fblink[i] = gauge->links[i];
fblink[OPP_DIR(i)] = QDP_create_M();
QDP_M_eq_sM(tmpmat, fblink[i], QDP_neighbor[i], QDP_backward, QDP_all);
QDP_M_eq_Ma(fblink[OPP_DIR(i)], tmpmat, QDP_all);
}
for(i=0; i<6; i++) {
Amu[i] = QDP_create_M();
}
staples = QDP_create_M();
tmpmat1 = QDP_create_M();
tmpmat2 = QDP_create_M();
tmpmat3 = QDP_create_M();
tmpmat4 = QDP_create_M();
Plaq = coeffs->plaquette;
Rect = coeffs->rectangle;
Pgm = coeffs->parallelogram;
//Construct 3-staples and rectangles
for(mu=0; mu<4; mu++) {
i=0;
for(nu=0; nu<4; nu++) {
if(nu!=mu){
// tmpmat1 = Umu(x+nu)
QDP_M_eq_sM(tmpmat1, fblink[mu], QDP_neighbor[nu], QDP_forward, QDP_all);
QDP_M_eq_M_times_M(Amu[i], fblink[nu], tmpmat1, QDP_all);
//tmpmat2 = Umu(x-nu)
QDP_M_eq_sM(tmpmat2, fblink[mu], QDP_neighbor[nu], QDP_backward, QDP_all);
QDP_M_eq_M_times_M(Amu[i+3], fblink[OPP_DIR(nu)], tmpmat2, QDP_all);
//tmpmat = U_{nu}(x+mu)
QDP_M_eq_sM(tmpmat, fblink[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_M_times_Ma(staples, Amu[i], tmpmat, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[mu], &Plaq, staples, QDP_all);
//tmpmat = U_{-nu}(x+mu)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(nu)], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_Ma_times_M(tmpmat3, fblink[OPP_DIR(nu)], staples, QDP_all);
QDP_M_eq_M_times_M(tmpmat4, tmpmat3, tmpmat, QDP_all);
QDP_M_eq_sM(tmpmat, tmpmat4, QDP_neighbor[nu], QDP_forward, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[mu], &Rect, tmpmat, QDP_all);
QDP_M_eq_Ma_times_M(tmpmat4, tmpmat2, tmpmat3, QDP_all);
QDP_M_eq_sM(tmpmat, tmpmat4, QDP_neighbor[nu], QDP_forward, QDP_all);
QDP_M_eq_sM(tmpmat3, tmpmat, QDP_neighbor[mu], QDP_backward, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[nu], &Rect, tmpmat3, QDP_all);
//tmpmat = U_{-nu}(x+mu)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(nu)], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_M_times_Ma(tmpmat3, tmpmat2, tmpmat, QDP_all);
QDP_M_eq_M_times_Ma(tmpmat, tmpmat3, staples, QDP_all);
QDP_M_eq_sM(tmpmat3, tmpmat, QDP_neighbor[nu], QDP_forward, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[nu], &Rect, tmpmat3, QDP_all);
//tmpmat = U_{-nu}(x+mu)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(nu)], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_M_times_Ma(staples, Amu[i+3], tmpmat, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[mu], &Plaq, staples, QDP_all);
QDP_M_eq_Ma_times_M(tmpmat3, fblink[nu], staples, QDP_all);
QDP_M_eq_sM(tmpmat, fblink[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_M_times_M(tmpmat4, tmpmat3, tmpmat, QDP_all);
QDP_M_eq_sM(tmpmat, tmpmat4, QDP_neighbor[nu], QDP_backward, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[mu], &Rect, tmpmat, QDP_all);
QDP_M_eq_Ma_times_M(tmpmat, tmpmat3, tmpmat1, QDP_all);
QDP_M_eq_sM(tmpmat4, tmpmat, QDP_neighbor[mu], QDP_backward, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[nu], &Rect, tmpmat4, QDP_all);
QDP_M_eq_sM(tmpmat, fblink[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_M_times_M(tmpmat3, staples, tmpmat, QDP_all);
QDP_M_eq_M_times_Ma(tmpmat4, tmpmat3, tmpmat1, QDP_all);
QDP_M_peq_r_times_M(tempmom_qdp[nu], &Rect, tmpmat4, QDP_all);
i++;
}
}
// Construct the pgm staples and add them to force
QDP_M_eq_zero(staples, QDP_all);
i=0;
for(nu=0; nu<4; nu++){
if(nu!=mu){
k=0;
for(sig=0; sig<4;sig ++){
if(sig!=mu && nu!=sig){
// the nu_sig_mu ... staple and 3 reflections
//tmpmat = Amu["sig"](x+nu)
QDP_M_eq_sM(tmpmat, Amu[j[i][k]], QDP_neighbor[nu], QDP_forward, QDP_all);
//tmpmat1 = Unu(x)*Amu["sig"](x+nu)
QDP_M_eq_M_times_M(tmpmat1, fblink[nu], tmpmat, QDP_all);
//tmpmat3 = Unu(x+mu+sig)
QDP_M_eq_sM(tmpmat, fblink[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_sM(tmpmat3, tmpmat, QDP_neighbor[sig], QDP_forward, QDP_all); // HERE?
//tmpmat2 = Unu(x)*Amu["sig"](x+nu)*adj(Unu(x+mu+sig))
QDP_M_eq_M_times_Ma(tmpmat2, tmpmat1, tmpmat3, QDP_all);
//tmpmat = Usig(x+mu)
QDP_M_eq_sM(tmpmat, fblink[sig], QDP_neighbor[mu], QDP_forward, QDP_all);
//tmpmat1 = Unu(x)*Amu["sig"](x+nu)*adj(Unu(x+mu+sig))*adj(Usig(x+mu))
QDP_M_eq_M_times_Ma(tmpmat1, tmpmat2, tmpmat, QDP_all);
QDP_M_peq_M(staples, tmpmat1, QDP_all);
//tmpmat = Amu["sig"](x-nu)
QDP_M_eq_sM(tmpmat, Amu[j[i][k]], QDP_neighbor[nu], QDP_backward, QDP_all);
//tmpmat1 = U_{-nu}(x)*Amu["sig"](x-nu)
QDP_M_eq_M_times_M(tmpmat1, fblink[OPP_DIR(nu)], tmpmat, QDP_all);
//tmpmat3 = U_{-nu}(x+mu+sig)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(nu)], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_sM(tmpmat3, tmpmat, QDP_neighbor[sig], QDP_forward, QDP_all); // HERE?
//tmpmat2 = U_{-nu}nu(x)*Amu["sig"](x-nu)*adj(Unu(x+mu+sig))
QDP_M_eq_M_times_Ma(tmpmat2, tmpmat1, tmpmat3, QDP_all);
//tmpmat = Usig(x+mu)
QDP_M_eq_sM(tmpmat, fblink[sig], QDP_neighbor[mu], QDP_forward, QDP_all);
//tmpmat1 = U_{-nu}(x)*Amu["sig"](x-nu)*adj(Unu(x+mu+sig))*adj(Usig(x+mu))
QDP_M_eq_M_times_Ma(tmpmat1, tmpmat2, tmpmat, QDP_all);
QDP_M_peq_M(staples, tmpmat1, QDP_all);
//tmpmat = Amu["-sig"](x-nu)
QDP_M_eq_sM(tmpmat, Amu[j[i][k]+3], QDP_neighbor[nu], QDP_backward, QDP_all);
//tmpmat1 = U_{-nu}(x)*Amu["-sig"](x-nu)
QDP_M_eq_M_times_M(tmpmat1, fblink[OPP_DIR(nu)], tmpmat, QDP_all);
//tmpmat = U_{-nu}(x+mu-sig)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(nu)], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_sM(tmpmat3, tmpmat, QDP_neighbor[sig], QDP_backward, QDP_all); // HERE?
//tmpmat2 = U_{-nu}nu(x)*Amu["-sig"](x-nu)*adj(Unu(x+mu-sig))
QDP_M_eq_M_times_Ma(tmpmat2, tmpmat1, tmpmat3, QDP_all);
//tmpmat = U_{-sig}(x+mu)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(sig)], QDP_neighbor[mu], QDP_forward, QDP_all);
//tmpmat1 = U_{-nu}(x)*Amu["-sig"](x-nu)*adj(Unu(x+mu-sig))*adj(U_{-sig}(x+mu))
QDP_M_eq_M_times_Ma(tmpmat1, tmpmat2, tmpmat, QDP_all);
QDP_M_peq_M(staples, tmpmat1, QDP_all);
//tmpmat = Amu["-sig"](x+nu)
QDP_M_eq_sM(tmpmat, Amu[j[i][k]+3], QDP_neighbor[nu], QDP_forward, QDP_all);
//tmpmat1 = Unu(x)*Amu["-sig"](x+nu)
QDP_M_eq_M_times_M(tmpmat1, fblink[nu], tmpmat, QDP_all);
//tmpmat3 = Unu(x+mu-sig)
QDP_M_eq_sM(tmpmat, fblink[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_sM(tmpmat3, tmpmat, QDP_neighbor[sig], QDP_backward, QDP_all); // HERE?
//tmpmat2 = Unu(x)*Amu["-sig"](x+nu)*adj(Unu(x+mu-sig))
QDP_M_eq_M_times_Ma(tmpmat2, tmpmat1, tmpmat3, QDP_all);
//tmpmat = U_{-sig}(x+mu)
QDP_M_eq_sM(tmpmat, fblink[OPP_DIR(sig)], QDP_neighbor[mu], QDP_forward, QDP_all);
//tmpmat1 = Unu(x)*Amu["sig"](x+nu)*adj(Unu(x+mu+sig))*adj(Usig(x+mu))
QDP_M_eq_M_times_Ma(tmpmat1, tmpmat2, tmpmat, QDP_all);
QDP_M_peq_M(staples, tmpmat1, QDP_all);
k++;
}//close if sig!=nu ...
}//close sig loop
i++;
}// close if nu!=mu
}//close the pgm nu loop
QDP_M_peq_r_times_M(tempmom_qdp[mu], &Pgm, staples, QDP_all);
}// closes the mu loop
#ifdef CHKSUM
QLA_ColorMatrix qcm;
QLA_Complex det, chk;
QLA_c_eq_r(chk, 0);
#endif
for(mu=0; mu<4; mu++){
QDP_M_eq_M_times_Ma(tmpmat, fblink[mu], tempmom_qdp[mu], QDP_all); // HERE?
QDP_M_eq_r_times_M_plus_M( tempmom_qdp[mu], &eb3, tmpmat, force->force[mu], QDP_all);// HERE?
QDP_M_eq_antiherm_M(force->force[mu], tempmom_qdp[mu], QDP_all);// HERE
#ifdef CHKSUM
QDP_m_eq_sum_M(&qcm, force->force[mu], QDP_all);
QLA_C_eq_det_M(&det, &qcm);
QLA_c_peq_c(chk, det);
#endif
}
#ifdef CHKSUM
QOP_printf0("chksum: %g %g\n", QLA_real(chk), QLA_imag(chk));
#endif
//DESTROY various fields
QDP_destroy_M(tmpmat);
QDP_destroy_M(tmpmat1);
QDP_destroy_M(tmpmat2);
QDP_destroy_M(tmpmat3);
QDP_destroy_M(staples);
QDP_destroy_M(tmpmat4);
for(mu=0; mu<4; mu++){
QDP_destroy_M(tempmom_qdp[mu]);
}
for(i=0; i<6; i++) {
QDP_destroy_M(Amu[i]);
}
for(i=4; i<8; i++) {
QDP_destroy_M(fblink[i]);
}
dtime += QOP_time();
double nflop = 96720;
info->final_sec = dtime;
info->final_flop = nflop*QDP_sites_on_node;
info->status = QOP_SUCCESS;
//QOP_printf0("Time in slow g_force: %e\n", info->final_sec);
}
/* Smearing level 0 */
static void
QOP_hisq_force_multi_smearing0_fnmat(QOP_info_t *info,
REAL *residues,
QDP_ColorVector *x[],
int nterms,
QDP_ColorMatrix *force_accum[4],
QDP_ColorMatrix *force_accum_naik[4])
{
int term;
int i,k;
int dir;
REAL coeff;
QDP_ColorMatrix *tmat;
QDP_ColorMatrix *oprod_along_path[MAX_PATH_LENGTH+1];
QDP_ColorMatrix *mat_tmp0;
QDP_ColorVector *tsrc[2], *vec_tmp[2];
size_t nflops = 0;
if( nterms==0 )return;
mat_tmp0 = QDP_create_M();
tmat = QDP_create_M();
tsrc[0] = QDP_create_V();
tsrc[1] = QDP_create_V();
vec_tmp[0] = QDP_create_V();
vec_tmp[1] = QDP_create_V();
for(i=0;i<=MAX_PATH_LENGTH;i++){
oprod_along_path[i] = QDP_create_M();
}
// clear force accumulators
for(dir=XUP;dir<=TUP;dir++)
QDP_M_eq_zero(force_accum[dir], QDP_all);
for(dir=XUP;dir<=TUP;dir++){ //AB loop on directions, path table is not needed
k=0; // which vec_tmp we are using (0 or 1)
QDP_V_eq_V(tsrc[k], x[0], QDP_all);
QDP_V_eq_sV(vec_tmp[k], tsrc[k],
fnshift(OPP_DIR(dir)), fndir(OPP_DIR(dir)), QDP_all);
QDP_M_eq_zero(oprod_along_path[0], QDP_all);
for(term=0;term<nterms;term++){
if(term<nterms-1) {
QDP_V_eq_V(tsrc[1-k], x[term+1], QDP_all);
QDP_V_eq_sV(vec_tmp[1-k], tsrc[1-k],
fnshift(OPP_DIR(dir)), fndir(OPP_DIR(dir)), QDP_all);
}
//QDP_M_eq_V_times_Va(tmat, x[term], vec_tmp[k], QDP_all);
QDP_M_eq_V_times_Va(tmat, tsrc[k], vec_tmp[k], QDP_all);
nflops += 54;
QDP_discard_V(vec_tmp[k]);
QDP_M_peq_r_times_M(oprod_along_path[0], &residues[term], tmat,
QDP_all);
nflops += 36;
k=1-k; // swap 0 and 1
} // end loop over terms in rational function expansion
link_gather_connection_qdp(oprod_along_path[1], oprod_along_path[0], tmat,
dir );
coeff = 1.;
QDP_M_peq_r_times_M(force_accum[dir],&coeff,oprod_along_path[1],QDP_all);
nflops += 36;
} // end of loop on directions //
// *** Naik part *** /
// clear force accumulators
for(dir=XUP;dir<=TUP;dir++)
QDP_M_eq_zero(force_accum_naik[dir], QDP_all);
for(dir=XUP;dir<=TUP;dir++){ //AB loop on directions, path table is not needed
k=0; // which vec_tmp we are using (0 or 1)
QDP_V_eq_V(tsrc[k], x[0], QDP_all);
QDP_V_eq_sV(vec_tmp[k], tsrc[k], fnshift(OPP_3_DIR( DIR3(dir) )),
fndir(OPP_3_DIR( DIR3(dir) )), QDP_all);
QDP_M_eq_zero(oprod_along_path[0], QDP_all);
for(term=0;term<nterms;term++){
if(term<nterms-1) {
QDP_V_eq_V(tsrc[1-k], x[term+1], QDP_all);
QDP_V_eq_sV(vec_tmp[1-k], tsrc[1-k], fnshift(OPP_3_DIR( DIR3(dir) )),
fndir(OPP_3_DIR( DIR3(dir) )), QDP_all);
}
//QDP_M_eq_V_times_Va(tmat, x[term], vec_tmp[k], QDP_all);
QDP_M_eq_V_times_Va(tmat, tsrc[k], vec_tmp[k], QDP_all);
nflops += 54;
QDP_discard_V(vec_tmp[k]);
QDP_M_peq_r_times_M(oprod_along_path[0], &residues[term], tmat, QDP_all);
nflops += 36;
k=1-k; // swap 0 and 1
} // end loop over terms in rational function expansion
link_gather_connection_qdp(oprod_along_path[1], oprod_along_path[0], tmat,
DIR3(dir) );
coeff = 1; // fermion_eps is outside this routine in "wrapper" routine
QDP_M_peq_r_times_M(force_accum_naik[dir],&coeff,
oprod_along_path[1],QDP_all);
nflops += 36;
} // end of loop on directions
QDP_destroy_V( tsrc[0] );
QDP_destroy_V( tsrc[1] );
QDP_destroy_V( vec_tmp[0] );
QDP_destroy_V( vec_tmp[1] );
QDP_destroy_M( mat_tmp0 );
QDP_destroy_M( tmat );
for(i=0;i<=MAX_PATH_LENGTH;i++){
QDP_destroy_M( oprod_along_path[i] );
}
info->final_flop = ((double)nflops)*QDP_sites_on_node;
return;
} //hisq_force_multi_smearing0_fnmat
void
QOP_hisq_force_multi_wrapper_fnmat(QOP_info_t *info,
QOP_FermionLinksHisq *flh,
QOP_Force *Force,
QOP_hisq_coeffs_t *hisq_coeff,
REAL *residues,
QDP_ColorVector *x[],
int *n_orders_naik)
{
double dtime = QDP_time();
int i, ipath, dir;
REAL coeff_mult;
double *eps_naik = hisq_coeff->eps_naik;
int n_naiks = hisq_coeff->n_naiks;
QOP_hisq_unitarize_method_t umethod = hisq_coeff->umethod;
// Quark paths sorted by net displacement and last directions
static Q_path *q_paths_sorted_1 = NULL;
static Q_path *q_paths_sorted_2 = NULL;
static Q_path *q_paths_sorted_3 = NULL;
static int *netbackdir_table_1 = NULL;
static int *netbackdir_table_2 = NULL;
static int *netbackdir_table_3 = NULL;
static int first_force = 1;
if(first_force == 1)
QOP_make_paths_and_dirs_hisq(hisq_coeff, umethod);
int num_q_paths_1 = qop_get_num_q_paths_1();
int num_q_paths_2 = qop_get_num_q_paths_2();
int num_q_paths_3 = qop_get_num_q_paths_3();
Q_path *q_paths_1 = qop_get_q_paths_1();
Q_path *q_paths_2 = qop_get_q_paths_2();
Q_path *q_paths_3 = qop_get_q_paths_3();
Q_path *q_paths_sorted_current = NULL;
int *netbackdir_table_current = NULL;
int inaik;
int n_naik_shift;
double final_flop = 0.;
size_t nflops = 0;
QDP_ColorMatrix * force[4] = {Force->force[0], Force->force[1],
Force->force[2], Force->force[3]};
int num_q_paths_current,n_orders_naik_current;//==nterms
QDP_ColorMatrix *force_accum_0[4];
QDP_ColorMatrix *force_accum_0_naik[4];
QDP_ColorMatrix *force_accum_1[4];
QDP_ColorMatrix *force_accum_1u[4];
QDP_ColorMatrix *force_accum_2[4];
QDP_ColorMatrix *force_final[4];
QDP_ColorMatrix *Ugf[4], *Vgf[4], *Wgf[4];
int nterms = 0, n_order_naik_total;
for(inaik = 0; inaik < n_naiks; inaik++)
nterms += n_orders_naik[inaik];
n_order_naik_total = nterms;
for(i=0;i<4;i++) {
Ugf[i] = flh->U_links[i];
Vgf[i] = flh->V_links[i];
Wgf[i] = flh->W_unitlinks[i];
}
QDP_ColorMatrix *tmat;
QDP_ColorMatrix *mat_tmp0;
REAL treal;
if( first_force==1 ){
if( q_paths_sorted_1==NULL )
q_paths_sorted_1 = (Q_path *)malloc( num_q_paths_1*sizeof(Q_path) );
if(netbackdir_table_1==NULL )
netbackdir_table_1 = (int *)malloc( num_q_paths_1*sizeof(int) );
if( q_paths_sorted_2==NULL )
q_paths_sorted_2 = (Q_path *)malloc( num_q_paths_2*sizeof(Q_path) );
if(netbackdir_table_2==NULL )
netbackdir_table_2 = (int *)malloc( num_q_paths_2*sizeof(int) );
if( q_paths_sorted_3==NULL )
q_paths_sorted_3 = (Q_path *)malloc( num_q_paths_3*sizeof(Q_path) );
if(netbackdir_table_3==NULL )
netbackdir_table_3 = (int *)malloc( num_q_paths_3*sizeof(int) );
else{QOP_printf0("WARNING: remaking sorted path tables\n"); exit(0); }
// make sorted tables
sort_quark_paths_hisq( q_paths_1, q_paths_sorted_1, num_q_paths_1, 8 );
for( ipath=0; ipath<num_q_paths_1; ipath++ )
netbackdir_table_1[ipath] =
find_backwards_gather( &(q_paths_sorted_1[ipath]) );
sort_quark_paths_hisq( q_paths_2, q_paths_sorted_2, num_q_paths_2, 16 );
for( ipath=0; ipath<num_q_paths_2; ipath++ )
netbackdir_table_2[ipath] =
find_backwards_gather( &(q_paths_sorted_2[ipath]) );
sort_quark_paths_hisq( q_paths_3, q_paths_sorted_3, num_q_paths_3, 16 );
for( ipath=0; ipath<num_q_paths_3; ipath++ )
netbackdir_table_3[ipath] =
find_backwards_gather( &(q_paths_sorted_3[ipath]) );
first_force=0;
}
tmat = QDP_create_M();
mat_tmp0 = QDP_create_M();
for(i=XUP;i<=TUP;i++){
force_accum_0[i] = QDP_create_M();
force_accum_0_naik[i] = QDP_create_M();
force_accum_1[i] = QDP_create_M();
force_accum_1u[i] = QDP_create_M();
force_accum_2[i] = QDP_create_M();
force_final[i] = QDP_create_M();
}
for(dir=XUP;dir<=TUP;dir++)
QDP_M_eq_zero(force_accum_2[dir], QDP_all);
// loop on different naik masses
n_naik_shift = 0;
for( inaik=0; inaik<n_naiks; inaik++ ) {
// smearing level 0
if( 0==inaik ) {
n_orders_naik_current = n_order_naik_total;
}
else {
n_orders_naik_current = n_orders_naik[inaik];
}
QOP_hisq_force_multi_smearing0_fnmat(info,residues+n_naik_shift,
x+n_naik_shift, n_orders_naik_current,
force_accum_0, force_accum_0_naik);
final_flop += info->final_flop;
// smearing level 2
if( 0==inaik ) {
q_paths_sorted_current = q_paths_sorted_2;
num_q_paths_current = num_q_paths_2;
netbackdir_table_current = netbackdir_table_2;
}
else {
q_paths_sorted_current = q_paths_sorted_3;
num_q_paths_current = num_q_paths_3;
netbackdir_table_current = netbackdir_table_3;
}
QOP_hisq_force_multi_smearing_fnmat( info,Wgf,residues+n_naik_shift,
x+n_naik_shift,
n_orders_naik_current,
force_accum_1,
force_accum_0, force_accum_0_naik,
num_q_paths_current,
q_paths_sorted_current,
netbackdir_table_current );
//QOP_printf0("HISQ smear0 flops = %g\n", info->final_flop);
final_flop += info->final_flop;
if( 0==inaik ) {
coeff_mult = 1.0;
}
else {
coeff_mult = eps_naik[inaik];
}
for(dir=XUP;dir<=TUP;dir++) {
QDP_M_peq_r_times_M(force_accum_2[dir],&coeff_mult,
force_accum_1[dir],QDP_all);
nflops += 36;
}
n_naik_shift += n_orders_naik[inaik];
}
if ( umethod==QOP_UNITARIZE_NONE ){
// smearing level 1
QOP_hisq_force_multi_smearing_fnmat( info,Ugf,residues,
x,
nterms, force_accum_1,
force_accum_2, NULL,
num_q_paths_1,
q_paths_sorted_1,
netbackdir_table_1 );
final_flop += info->final_flop;
}
else if ( umethod==QOP_UNITARIZE_RATIONAL ){
// reunitarization
QOP_hisq_force_multi_reunit(info,Vgf,force_accum_1u,
force_accum_2);
//QOP_printf0("reunit flops = %g\n", info->final_flop);
final_flop += info->final_flop;
// smearing level 1
QOP_hisq_force_multi_smearing_fnmat( info,Ugf,residues,
x,
nterms, force_accum_1,
force_accum_1u, NULL,
num_q_paths_1,
q_paths_sorted_1,
netbackdir_table_1 );
//QOP_printf0("HISQ smear1 flops = %g\n", info->final_flop);
final_flop += info->final_flop;
}
else
{
QOP_printf0("Unknown or unsupported unitarization method\n");
exit(1);
}
// contraction with the link in question should be done here,
// after contributions from all levels of smearing are taken into account
for(dir=XUP;dir<=TUP;dir++){
QDP_M_eq_M_times_M(force_final[dir],Ugf[dir],force_accum_1[dir],QDP_all);
nflops += 198;
}
// take into account even/odd parity (it is NOT done in "smearing" routine)
//eps multiplication done outside QOP
for(dir=XUP;dir<=TUP;dir++){
QDP_M_eq_M(tmat,force_final[dir],QDP_all);
treal = 2.0;
QDP_M_eq_r_times_M(force_final[dir],&treal,tmat,QDP_even);
treal = -2.0;
QDP_M_eq_r_times_M(force_final[dir],&treal,tmat,QDP_odd);
nflops += 18;
}
// Put antihermitian traceless part into momentum
// add force to momentum
for(dir=XUP; dir<=TUP; dir++){
QDP_M_eq_antiherm_M(mat_tmp0, force_final[dir], QDP_all);
QDP_M_peq_M(force[dir], mat_tmp0, QDP_all);
nflops += 24+18;
//QDP_M_peq_M(force_final[dir], force[dir], QDP_all);
//QDP_M_eq_antiherm_M(force[dir], force_final[dir], QDP_all);
}
for(i=XUP;i<=TUP;i++){
QDP_destroy_M( force_accum_0[i] );
QDP_destroy_M( force_accum_0_naik[i] );
QDP_destroy_M( force_accum_1[i] );
QDP_destroy_M( force_accum_1u[i] );
QDP_destroy_M( force_accum_2[i] );
QDP_destroy_M( force_final[i] );
}
QDP_destroy_M( tmat );
QDP_destroy_M( mat_tmp0 );
final_flop += ((double)nflops)*QDP_sites_on_node;
info->final_sec = QDP_time() - dtime;
info->final_flop = final_flop;
info->status = QOP_SUCCESS;
//QOP_printf0("HISQ force flops = %g\n", info->final_flop);
} //hisq_force_multi_wrapper_fnmat
/* Smearing level i*/
static void
QOP_hisq_force_multi_smearing_fnmat(QOP_info_t *info,
QDP_ColorMatrix * gf[4],
REAL *residues,
QDP_ColorVector *x[],
int nterms,
QDP_ColorMatrix *force_accum[4],
QDP_ColorMatrix *force_accum_old[4],
QDP_ColorMatrix *force_accum_naik_old[4],
int internal_num_q_paths,
Q_path *internal_q_paths_sorted,
int *internal_netbackdir_table)
{
int i,j,k,lastdir=-99,ipath,ilink;
int length,dir,odir;
REAL coeff;
QDP_ColorMatrix *tmat;
QDP_ColorMatrix *oprod_along_path[MAX_PATH_LENGTH+1];
QDP_ColorMatrix *mats_along_path[MAX_PATH_LENGTH+1];
QDP_ColorMatrix *mat_tmp0,*mat_tmp1, *stmp[8];;
QDP_ColorVector *vec_tmp[2];
int netbackdir;
size_t nflops = 0;
// table of net path displacements (backwards from usual convention)
Q_path *this_path; // pointer to current path
/* Allocate fields */
for(i=0;i<=MAX_PATH_LENGTH;i++){
oprod_along_path[i] = QDP_create_M();
}
for(i=1;i<=MAX_PATH_LENGTH;i++){
// 0 element is never used (it's unit matrix)
mats_along_path[i] = QDP_create_M();
}
mat_tmp0 = QDP_create_M();
mat_tmp1 = QDP_create_M();
for(i=0; i<8; i++) stmp[i] = QDP_create_M();
tmat = QDP_create_M();
vec_tmp[0] = QDP_create_V();
vec_tmp[1] = QDP_create_V();
// clear force accumulators
for(dir=XUP;dir<=TUP;dir++)
QDP_M_eq_zero(force_accum[dir], QDP_all);
// loop over paths, and loop over links in path
for( ipath=0; ipath<internal_num_q_paths; ipath++ ){
this_path = &(internal_q_paths_sorted[ipath]);
if(this_path->forwback== -1)continue; // skip backwards dslash
length = this_path->length;
netbackdir = internal_netbackdir_table[ipath];
// move f(i-1) force from current site in positive direction,
// this corresponds to outer product |X><Y| calculated at the endpoint of the path
if( netbackdir<8) { // Not a Naik path
link_gather_connection_qdp(oprod_along_path[0] ,
force_accum_old[OPP_DIR(netbackdir)],
tmat, netbackdir );
}
else { // Naik path
if( NULL==force_accum_naik_old ) {
QOP_printf0( "hisq_force_multi_smearing_fnmat: mismatch:\n" );
QOP_printf0( "force_accum_naik_old is NULL, but path table contains Naik paths(!)\n" );
exit(0);
}
// CONVERSION FROM 3-LINK DIRECTION TO 1-LINK DIRECTION
link_gather_connection_qdp(oprod_along_path[0] ,
force_accum_naik_old[OPP_DIR(netbackdir-8)],
tmat, netbackdir );
}
// figure out how much of the outer products along the path must be
// recomputed. j is last one needing recomputation. k is first one.
j=length-1; // default is recompute all
if( GOES_BACKWARDS(this_path->dir[0]) ) k=1; else k=0;
for(ilink=j;ilink>=k;ilink--){
link_transport_connection_qdp( oprod_along_path[length-ilink],
oprod_along_path[length-ilink-1], gf,
mat_tmp0, stmp, this_path->dir[ilink] );
nflops += 198;
}
// maintain an array of transports "to this point" along the path.
// Don't recompute beginning parts of path if same as last path
ilink=0; // first link where new transport is needed
// Sometimes we don't need the matrix for the last link
if( GOES_FORWARDS(this_path->dir[length-1]) ) k=length-1; else k=length;
for( ; ilink<k; ilink++ ){
if( ilink==0 ){
dir = this_path->dir[0];
if( GOES_FORWARDS(dir) ){
QDP_M_eq_sM(tmat, gf[dir], QDP_neighbor[dir],
QDP_backward, QDP_all);
QDP_M_eq_Ma(mats_along_path[1], tmat, QDP_all);
QDP_discard_M(tmat);
}
else{
QDP_M_eq_M(mats_along_path[1], gf[OPP_DIR(dir)], QDP_all);
}
}
else { // ilink != 0
dir = OPP_DIR(this_path->dir[ilink]);
link_transport_connection_qdp( mats_along_path[ilink+1],
mats_along_path[ilink], gf,
mat_tmp0, stmp, dir );
nflops += 198;
}
} // end loop over links
// A path has (length+1) points, counting the ends. At first
// point, no "down" direction links have their momenta "at this
// point". At last, no "up" ...
if( GOES_FORWARDS(this_path->dir[length-1]) ) k=length-1; else k=length;
for( ilink=0; ilink<=k; ilink++ ){
if(ilink<length)dir = this_path->dir[ilink];
else dir=NODIR;
coeff = this_path->coeff;
if( (ilink%2)==1 )coeff = -coeff;
// add in contribution to the force
if( ilink<length && GOES_FORWARDS(dir) ){
link_gather_connection_qdp(mat_tmp1,
oprod_along_path[length-ilink-1], tmat, dir );
if(ilink==0)
{
QDP_M_eq_M(mat_tmp0,mat_tmp1,QDP_all);
}
else
{
QDP_M_eq_M_times_Ma(mat_tmp0, mats_along_path[ilink],
mat_tmp1, QDP_all);
nflops += 198;
QDP_M_eq_Ma(mat_tmp1,mat_tmp0,QDP_all);
}
QDP_M_peq_r_times_M(force_accum[dir],&coeff,mat_tmp1,QDP_all);
nflops += 36;
}
if( ilink>0 && GOES_BACKWARDS(lastdir) ){
odir = OPP_DIR(lastdir);
if( ilink==1 ){
QDP_M_eq_M(mat_tmp0,oprod_along_path[length-ilink],QDP_all);
QDP_M_eq_Ma(mat_tmp1,mat_tmp0,QDP_all);
}
else{
link_gather_connection_qdp(mat_tmp1, mats_along_path[ilink-1],
tmat, odir );
QDP_M_eq_M_times_Ma(mat_tmp0, oprod_along_path[length-ilink],
mat_tmp1, QDP_all);
nflops += 198;
QDP_M_eq_Ma(mat_tmp1, mat_tmp0, QDP_all);
}
QDP_M_peq_r_times_M(force_accum[odir],&coeff,mat_tmp1,QDP_all);
nflops += 36;
}
lastdir = dir;
} // end loop over links in path //
} // end loop over paths //
QDP_destroy_V( vec_tmp[0] );
QDP_destroy_V( vec_tmp[1] );
QDP_destroy_M( mat_tmp0 );
QDP_destroy_M( mat_tmp1 );
QDP_destroy_M( tmat );
for(i=0; i<8; i++) QDP_destroy_M(stmp[i]);
for(i=0;i<=MAX_PATH_LENGTH;i++){
QDP_destroy_M( oprod_along_path[i] );
}
for(i=1;i<=MAX_PATH_LENGTH;i++){
QDP_destroy_M( mats_along_path[i] );
}
info->final_flop = ((double)nflops)*QDP_sites_on_node;
return;
}//hisq_force_multi_smearing_fnmat
void
QOP_hisq_deriv_multi_fnmat2_qdp(QOP_info_t *info,
QOP_FermionLinksHisq *flh,
QDP_ColorMatrix *deriv[],
QOP_hisq_coeffs_t *hisq_coeff,
REAL *residues,
QDP_ColorVector *x[],
int *n_orders_naik)
{
#define NC QDP_get_nc(deriv[0])
if(!QOP_asqtad.inited) QOP_asqtad_invert_init();
double dtime = QDP_time();
double totalflops = 0;
int siteflops = 0;
QOP_info_t tinfo;
QDP_ColorMatrix *Ugf[4], *Vgf[4], *Wgf[4];
for(int i=0; i<4; i++) {
Ugf[i] = flh->U_links[i];
Vgf[i] = flh->V_links[i];
Wgf[i] = flh->W_unitlinks[i];
}
QDP_ColorMatrix *force_accum_0[4];
QDP_ColorMatrix *force_accum_0_naik[4];
QDP_ColorMatrix *force_accum_1[4];
QDP_ColorMatrix *force_accum_1u[4];
QDP_ColorMatrix *force_accum_2[4];
QDP_ColorMatrix *force_final[4];
QDP_ColorMatrix *tmat = QDP_create_M();
for(int i=0; i<4; i++) {
force_accum_0[i] = QDP_create_M();
force_accum_0_naik[i] = QDP_create_M();
force_accum_1[i] = QDP_create_M();
force_accum_1u[i] = QDP_create_M();
force_accum_2[i] = QDP_create_M();
force_final[i] = QDP_create_M();
QDP_M_eq_zero(force_accum_2[i], QDP_all);
}
int n_naiks = hisq_coeff->n_naiks;
int nterms = 0;
for(int inaik = 0; inaik < n_naiks; inaik++)
nterms += n_orders_naik[inaik];
// loop on different naik masses
int n_naik_shift = 0;
for(int inaik=0; inaik<n_naiks; inaik++) {
int n_orders_naik_current;
if( inaik==0 ) {
n_orders_naik_current = nterms;
} else {
n_orders_naik_current = n_orders_naik[inaik];
}
QOP_get_mid(&tinfo, force_accum_0, QDP_neighbor, 4, residues+n_naik_shift,
1, x+n_naik_shift, n_orders_naik_current);
totalflops += tinfo.final_flop;
QOP_get_mid(&tinfo, force_accum_0_naik, QOP_common.neighbor3, 4,
residues+n_naik_shift, 1, x+n_naik_shift,
n_orders_naik_current);
totalflops += tinfo.final_flop;
// compensate for -1 on odd sites here instead of at end
for(int dir=0; dir<4; dir++) {
QDP_M_eqm_M(force_accum_0[dir], force_accum_0[dir], QDP_odd);
QDP_M_eqm_M(force_accum_0_naik[dir], force_accum_0_naik[dir], QDP_odd);
}
// smearing level 0
for(int i=0; i<4; i++) QDP_M_eq_zero(force_accum_1[i], QDP_all);
if(inaik==0) {
QOP_asqtad_coeffs_t acoef;
acoef.one_link = hisq_coeff->asqtad_one_link;
acoef.three_staple = hisq_coeff->asqtad_three_staple;
acoef.five_staple = hisq_coeff->asqtad_five_staple;
acoef.seven_staple = hisq_coeff->asqtad_seven_staple;
acoef.lepage = hisq_coeff->asqtad_lepage;
acoef.naik = hisq_coeff->asqtad_naik;
QOP_asqtad_deriv(&tinfo, Wgf, force_accum_1, &acoef,
force_accum_0, force_accum_0_naik);
//QOP_printf0("HISQ smear0 flops = %g\n", tinfo.final_flop);
totalflops += tinfo.final_flop;
} else {
QOP_asqtad_coeffs_t acoef;
acoef.one_link = hisq_coeff->difference_one_link;
acoef.three_staple = 0;
acoef.five_staple = 0;
acoef.seven_staple = 0;
acoef.lepage = 0;
acoef.naik = hisq_coeff->difference_naik;
QOP_asqtad_deriv(&tinfo, Wgf, force_accum_1, &acoef,
force_accum_0, force_accum_0_naik);
totalflops += tinfo.final_flop;
}
QLA_Real coeff_mult;
if( inaik==0 ) {
coeff_mult = 1.0;
} else {
coeff_mult = hisq_coeff->eps_naik[inaik];
}
for(int dir=0; dir<4; dir++) {
QDP_M_peq_r_times_M(force_accum_2[dir], &coeff_mult,
force_accum_1[dir], QDP_all);
}
siteflops += 4*36;
n_naik_shift += n_orders_naik[inaik];
}
// smearing level 1
QOP_asqtad_coeffs_t acoef;
acoef.one_link = hisq_coeff->fat7_one_link;
acoef.three_staple = hisq_coeff->fat7_three_staple;
acoef.five_staple = hisq_coeff->fat7_five_staple;
acoef.seven_staple = hisq_coeff->fat7_seven_staple;
acoef.lepage = 0;
acoef.naik = 0;
if(QOP_hisq_links.use_fat7_lepage) {
acoef.lepage = hisq_coeff->fat7_lepage;
}
QOP_hisq_unitarize_method_t umethod = hisq_coeff->umethod;
if ( umethod==QOP_UNITARIZE_NONE ){
for(int dir=0; dir<4; dir++)
QDP_M_eq_zero(force_accum_1[dir], QDP_all);
QOP_asqtad_deriv(&tinfo, Ugf, force_accum_1, &acoef,
force_accum_2, NULL);
totalflops += tinfo.final_flop;
} else if ( umethod==QOP_UNITARIZE_RATIONAL ) {
for(int mu=0; mu<4; mu++) QDP_M_eq_Ma(force_accum_1u[mu], force_accum_2[mu], QDP_all);
// reunitarization
#if QOP_Colors == 3
QOP_hisq_force_multi_reunit(&tinfo, Vgf, force_accum_2, force_accum_1u);
#else
for(int mu=0; mu<4; mu++) {
QOP_projectU_deriv_qdp(&tinfo, force_accum_2[mu], Wgf[mu], Vgf[mu], force_accum_1u[mu]);
}
#endif
//QOP_printf0("reunit flops = %g\n", tinfo.final_flop);
for(int mu=0; mu<4; mu++) QDP_M_eq_Ma(force_accum_1u[mu], force_accum_2[mu], QDP_all);
totalflops += tinfo.final_flop;
for(int dir=0; dir<4; dir++) QDP_M_eq_zero(force_accum_1[dir], QDP_all);
QOP_asqtad_deriv(&tinfo, Ugf, force_accum_1, &acoef,
force_accum_1u, NULL);
//QOP_printf0("HISQ smear1 flops = %g\n", tinfo.final_flop);
totalflops += tinfo.final_flop;
} else {
QOP_printf0("Unknown or unsupported unitarization method\n");
exit(1);
}
// take into account even/odd parity (it is NOT done in "smearing" routine)
// eps multiplication done outside QOP
// extra factor of 2
for(int dir=0; dir<4; dir++) {
QLA_Real treal = 2;
//QDP_M_peq_r_times_M(deriv[dir], &treal, force_accum_1[dir], QDP_even);
//QDP_M_meq_r_times_M(deriv[dir], &treal, force_accum_1[dir], QDP_odd);
QDP_M_peq_r_times_M(deriv[dir], &treal, force_accum_1[dir], QDP_all);
}
siteflops += 4*36;
for(int i=0; i<4; i++) {
QDP_destroy_M( force_accum_0[i] );
QDP_destroy_M( force_accum_0_naik[i] );
QDP_destroy_M( force_accum_1[i] );
QDP_destroy_M( force_accum_1u[i] );
QDP_destroy_M( force_accum_2[i] );
QDP_destroy_M( force_final[i] );
}
QDP_destroy_M( tmat );
totalflops += ((double)siteflops)*QDP_sites_on_node;
info->final_sec = QDP_time() - dtime;
info->final_flop = totalflops;
info->status = QOP_SUCCESS;
#undef NC
}
void
QOP_symanzik_1loop_gauge_heatbath_qdp(QOP_info_t *info,
QDP_ColorMatrix *links[],
QLA_Real beta,
QOP_gauge_coeffs_t *coeffs,
QDP_RandomState *rs0,
int nup, int nhb, int nover)
{
#define NC QDP_get_nc(links[0])
double dtime = QOP_time();
double nflops = 0;
if(coeffs->adjoint_plaquette) {
QOP_error("%s: adj plaq not supported\n", __func__);
}
fac = beta/QLA_Nc;
int imp = (coeffs->rectangle!=0)||(coeffs->parallelogram!=0);
QDP_Lattice *lat = QDP_get_lattice_M(links[0]);
int nd = QDP_ndim_L(lat);
QDP_Subset *cbs=QDP_even_and_odd_L(lat);
int ncb = 2;
if(imp) {
ncb = 32;
cbs = QOP_get_sub32(lat);
}
QDP_ColorMatrix *staple = QDP_create_M_L(lat);
QDP_ColorMatrix *v = QDP_create_M_L(lat);
QDP_ColorMatrix *tmp = QDP_create_M_L(lat);
rs = QDP_expose_S(rs0);
for(int up=0; up<nup; up++) {
for(int hb=0; hb<nhb; hb++) {
for(int cb=0; cb<ncb; cb++) {
QDP_Subset subset = cbs[cb];
for(int mu=0; mu<nd; mu++) {
QDP_M_eq_zero(staple, subset);
QOP_symanzik_1loop_gauge_staple_qdp(info, links, staple, mu, coeffs, cbs, cb);
QDP_M_eq_M_times_Ma(v, links[mu], staple, subset);
QDP_M_eq_funcit(v, hb_func, subset);
QDP_M_eq_M_times_M(tmp, v, links[mu], subset);
QDP_M_eq_M(links[mu], tmp, subset);
}
}
}
for(int over=0; over<nover; over++) {
for(int cb=0; cb<ncb; cb++) {
QDP_Subset subset = cbs[cb];
for(int mu=0; mu<nd; mu++) {
QDP_M_eq_zero(staple, subset);
QOP_symanzik_1loop_gauge_staple_qdp(info, links, staple, mu, coeffs, cbs, cb);
QDP_M_eq_M_times_Ma(v, links[mu], staple, subset);
QDP_M_eq_funcit(v, over_func, subset);
QDP_M_eq_M_times_M(tmp, v, links[mu], subset);
QDP_M_eq_M(links[mu], tmp, subset);
}
}
}
}
QDP_reset_S(rs0);
QDP_destroy_M(tmp);
QDP_destroy_M(v);
QDP_destroy_M(staple);
info->final_sec = QOP_time() - dtime;
info->final_flop = nflops*QDP_sites_on_node;
info->status = QOP_SUCCESS;
#undef NC
}
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();
}
