/* On a lattice with lattice constant len, along each axis, multiply
two adjacent smeared links in src to form a coarse link res on a
lattice s with lattice constant 2 * len */
void RG_create_gauge(QDP_ColorMatrix *res[RG_Nd],
QDP_ColorMatrix *src[RG_Nd],
QDP_Sub_Block s, int len)
{
int i,j,k;
int v[RG_Nd];
QDP_ColorMatrix *temp,*temp1;
QDP_Shift offset;
temp = QDP_create_M();
temp1 = QDP_create_M();
for(j=0; j<RG_Nd; ++j)
{
/* On axis displacement of length len */
for(k=0; k<RG_Nd; ++k) v[k] = 0;
v[j] = len;
offset = QDP_create_shift(v);
SQDP_M_eq_M_times_sM(res[j],src[j],src[j],offset,QDP_forward,s);
QDP_destroy_shift(offset);
// printf("Multp........node %d for %d\n",this_node,j); fflush(stdout);
}
// node0_printf(".......................done\n"); fflush(stdout);
QDP_destroy_M(temp);
QDP_destroy_M(temp1);
return;
}
/* Computes the staple :
mu
+-------+
nu | |
| |
X X
Where the mu link can be any su3_matrix. The result is saved in staple.
if staple==NULL then the result is not saved.
It also adds the computed staple to the fatlink[mu] with weight coef.
*/
static void
compute_gen_staple(QDP_ColorMatrix *staple, int mu, int nu,
QDP_ColorMatrix *link, double dcoef,
QDP_ColorMatrix *gauge[], QDP_ColorMatrix *fl[])
{
QLA_Real coef = dcoef;
QDP_ColorMatrix *ts0, *ts1;
QDP_ColorMatrix *tmat1, *tmat2;
QDP_ColorMatrix *tempmat;
ts0 = QDP_create_M();
ts1 = QDP_create_M();
tmat1 = QDP_create_M();
tmat2 = QDP_create_M();
tempmat = QDP_create_M();
/* Upper staple */
QDP_M_eq_sM(ts0, link, QDP_neighbor[nu], QDP_forward, QDP_all);
QDP_M_eq_sM(ts1, gauge[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
if(staple!=NULL) { /* Save the staple */
QDP_M_eq_M_times_Ma(tmat1, ts0, ts1, QDP_all);
QDP_M_eq_M_times_M(staple, gauge[nu], tmat1, QDP_all);
} else { /* No need to save the staple. Add it to the fatlinks */
QDP_M_eq_M_times_Ma(tmat1, ts0, ts1, QDP_all);
QDP_M_eq_M_times_M(tmat2, gauge[nu], tmat1, QDP_all);
QDP_M_peq_r_times_M(fl[mu], &coef, tmat2, QDP_all);
}
/* lower staple */
QDP_M_eq_sM(ts0, gauge[nu], QDP_neighbor[mu], QDP_forward, QDP_all);
QDP_M_eq_Ma_times_M(tmat1, gauge[nu], link, QDP_all);
QDP_M_eq_M_times_M(tempmat, tmat1, ts0, QDP_all);
QDP_M_eq_sM(ts0, tempmat, QDP_neighbor[nu], QDP_backward, QDP_all);
if(staple!=NULL) { /* Save the staple */
QDP_M_peq_M(staple, ts0, QDP_all);
QDP_M_peq_r_times_M(fl[mu], &coef, staple, QDP_all);
} else { /* No need to save the staple. Add it to the fatlinks */
QDP_M_peq_r_times_M(fl[mu], &coef, ts0, QDP_all);
}
QDP_destroy_M(ts0);
QDP_destroy_M(ts1);
QDP_destroy_M(tmat1);
QDP_destroy_M(tmat2);
QDP_destroy_M(tempmat);
} /* compute_gen_staple */
static int
qopqdp_force_random(lua_State *L)
{
#define NC QDP_get_nc(g->links[0])
BEGIN_ARGS;
GET_GAUGE(g);
END_ARGS;
QDP_RandomState *rs = g->lat->rs;
qassert(rs!=NULL);
if(QLA_Nc==3) { // use MILC conventions
QLA_RandomState *s = QDP_expose_S(rs);
for(int i=0; i<g->nd; i++) {
QDP_M_eq_funcia(g->links[i], randforce, s, QDP_all_L(g->qlat));
}
QDP_reset_S(rs);
} else {
QDP_ColorMatrix *m = QDP_create_M_L(g->qlat);
for(int i=0; i<g->nd; i++) {
QDP_M_eq_gaussian_S(m, rs, QDP_all_L(g->qlat));
QDP_M_eq_antiherm_M(g->links[i], m, QDP_all_L(g->qlat));
}
QDP_destroy_M(m);
}
return 0;
#undef NC
}
static void
qopqdp_gauge_free(lua_State *L, int idx)
{
gauge_t *g = qopqdp_gauge_check(L, idx);
for(int i=0; i<g->nd; i++) {
QDP_destroy_M(g->links[i]);
}
}
void RG_smearing(QDP_ColorMatrix *dest[RG_Nd], QDP_ColorMatrix *src[RG_Nd],QDP_Sub_Block s, int len)
{
int i;
QLA_Real staple_w,link0;
QLA_Int space_only;
QDP_ColorMatrix *temp[RG_Nd],*sm_link[RG_Nd],*pr_sm_link[RG_Nd];
for(i=0; i< RG_Nd; ++i)
{
sm_link[i] = QDP_create_M();
pr_sm_link[i] = QDP_create_M();
}
RG_value(&staple_w,&link0,&space_only);
/* Two smearing steps */
RG_smearing_qdp(sm_link, src, &staple_w, &link0, s, len);
#ifdef CHECK_SMEAR_QDP_MILC
project_qdp(sm_link, dest,&space_only);
#else
#ifdef CHECK_DEGRAND_W_SMEAR
project_qdp(sm_link, dest,&space_only);
#else
#ifdef CHECK_SMEAR_GAUGE_2
project_qdp(sm_link, dest,&space_only);
#else
project_qdp(sm_link, pr_sm_link,&space_only);
RG_smearing_qdp(sm_link, pr_sm_link,&staple_w,&link0,s,len);
project_qdp(sm_link, dest,&space_only);
#endif
#endif
#endif
for(i=0; i< RG_Nd; ++i)
{
QDP_destroy_M(sm_link[i]);
QDP_destroy_M(pr_sm_link[i]);
}
return;
}
/* Working from the finest lattice, smear the links at each level of
coarseness and multiply to form the links on the next higher
level. Results in rg_link. */
void RG_gauge(QDP_ColorMatrix *rg_link[NRG][RG_Nd],
QDP_ColorMatrix *link_qdp[RG_Nd],
QDP_Sub_Block s[NRG+1])
{
int i,j,len;
QDP_ColorMatrix *pr_sm_link[RG_Nd];
// node0_printf("Smearing links with Degrand trick........\n"); fflush(stdout);
for(i=0; i< RG_Nd; ++i)
pr_sm_link[i] = QDP_create_M();
#ifdef CHECK_DEGRAND_WO_SMEAR
for(i=0; i< RG_Nd; ++i)
SQDP_M_eq_M(rg_link[nrg-1][i],link_qdp[i],s[nrg]);
#else
RG_smearing(rg_link[nrg-1],link_qdp,s[nrg],1);
#ifdef CHECK_DEGRAND_W_SMEAR
SQDP_M_eq_M(rg_link[nrg-1][3],link_qdp[3],s[nrg]);
#endif
#endif
/* Work from the finest to the coarsest level */
for (i=1;i<nrg;i++)
{
len = intpow(2,i-1);
// printf("node %d: Smear links of length %d x a'\n",this_node,len); fflush(stdout);
// printf("node %d: rg_links of length %d x a'\n",this_node,2*len); fflush(stdout);
/* Smear the links */
#ifdef CHECK_DEGRAND_WO_SMEAR
for(j=0; j< RG_Nd; ++j)
SQDP_M_eq_M(pr_sm_link[j],rg_link[nrg-i][j],s[nrg-i+1]);
#else
RG_smearing(pr_sm_link,rg_link[nrg-i],s[nrg-i+1],len);
#ifdef CHECK_DEGRAND_W_SMEAR
SQDP_M_eq_M(pr_sm_link[3],rg_link[nrg-i][3],s[nrg-i+1]);
#endif
#endif
/* Multiply the links */
RG_create_gauge(rg_link[nrg-i-1],pr_sm_link,s[nrg-i],len);
}
// node0_printf(".......................done\n"); fflush(stdout);
for(i=0; i< RG_Nd; ++i)
QDP_destroy_M(pr_sm_link[i]);
return;
}
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
unpackM(Layout *l, QDP_ColorMatrix *m, real *y)
{
QDP_Lattice *lat = QDP_get_lattice_M(m);
QDP_ColorMatrix *xx = QDP_create_M_L(lat);
real *x = QDP_expose_M(xx);
int nelem = 2*QDP_Nc*QDP_Nc;
toQDP_F(x, lat, y, l, nelem, 1);
QDP_reset_M(xx);
//QDP_M_eq_M(m, xx, QDP_all_L(lat));
QDP_M_eq_transpose_M(m, xx, QDP_all_L(lat));
QDP_destroy_M(xx);
}
void
packM(Layout *l, real *y, QDP_ColorMatrix *m)
{
QDP_Lattice *lat = QDP_get_lattice_M(m);
QDP_ColorMatrix *xx = QDP_create_M_L(lat);
//QDP_M_eq_M(xx, m, QDP_all_L(lat));
QDP_M_eq_transpose_M(xx, m, QDP_all_L(lat));
real *x = QDP_expose_M(xx);
int nelem = 2*QDP_Nc*QDP_Nc;
fromQDP_F(y, l, x, lat, nelem, 1);
QDP_reset_M(xx);
QDP_destroy_M(xx);
}
void RG_check_subset(QDP_Sub_Block QDP_block[NRG+1])
{
int i,j,len;
QDP_ColorMatrix *link_qdp[RG_Nd],*prova[RG_Nd];
QLA_Complex unit;
for(i=0; i< RG_Nd; ++i)
{
link_qdp[i] = QDP_create_M();
prova[i] = QDP_create_M();
}
QLA_c_eq_r(unit,1.0)
for(i=0; i<RG_Nd; ++i)
SQDP_M_eq_c(link_qdp[i],&unit,QDP_block[nrg]);
// printf("Created!!! this node %d\n",this_node); fflush(stdout);
for(i=0; i<RG_Nd; ++i)
SQDP_M_eq_sM(prova[i], link_qdp[i], QDP_neighbor[i], QDP_forward, QDP_block[nrg-1]);
// printf("I am out!!! this node %d\n",this_node); fflush(stdout);
// SQDP_M_eq_func(prova[0],print_gl,QDP_block[nrg-1]);
// printf("I have printed!!! this node %d\n",this_node); fflush(stdout);
for(i=0; i< RG_Nd; ++i)
{
QDP_destroy_M(link_qdp[i]);
QDP_destroy_M(prova[i]);
}
printf("I have destroyed every thing!!! this node %d\n",this_node); fflush(stdout);
return;
}
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
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);
}
void RG_create_path(QDP_ColorMatrix *pr_wlink[RG_Ncn], QDP_ColorMatrix *link_qdp[RG_Nd], QDP_Sub_Block s,int len)
{
int i,j,k,t,x[4];
int count,c2,space_only;
QDP_ColorMatrix *path_1[4];
QDP_ColorMatrix *path_2[12];
QDP_ColorMatrix *path_3[24];
QDP_ColorMatrix *path_4[24];
QDP_ColorMatrix *wlink[RG_Ncn];
QDP_Shift offset;
shift_v *d1,*d2,*d3,*d4;
QLA_Real c = 1.0;
QLA_Real fact2 = 1.0/2.0;
QLA_Real fact3 = 1.0/6.0;
QLA_Real fact4 = 1.0/24.0;
QLA_Complex unit;
d1 = (shift_v *) malloc(4*sizeof(shift_v));
d2 = (shift_v *) malloc(12*sizeof(shift_v));
d3 = (shift_v *) malloc(24*sizeof(shift_v));
d4 = (shift_v *) malloc(24*sizeof(shift_v));
for (i = 0; i < RG_Ncn; i++)
wlink[i] = QDP_create_M();
for (i = 0; i < 4; i++)
path_1[i] = QDP_create_M();
for (i = 0; i < 4; i++)
SQDP_M_eq_M(path_1[i],link_qdp[i],s);
for (i = 0; i < 12; i++)
path_2[i] = QDP_create_M();
for (i = 0; i < 24; i++)
{
path_3[i] = QDP_create_M();
path_4[i] = QDP_create_M();
}
// printf("Start building paths %d\n",this_node); fflush(stdout);
for (i = 0; i < RG_Nd; i++)
{
x[0] = i;
d1[i] = create_shift(x,1,len);
}
// printf("First shift %d\n",this_node);fflush(stdout);
count = 0;
for (i = 0; i < RG_Nd; i++)
{
x[0] = i;
c2 = find_count(d1,x,1);
offset = QDP_create_shift(d1[c2].s);
for (j = 0; j < RG_Nd ; j++) if ( j != i)
{
x[1] = j;
d2[count] = create_shift(x,2,len);
SQDP_M_eq_M_times_sM(path_2[count],path_1[c2],link_qdp[j],offset,QDP_forward,s);
count ++;
}
QDP_destroy_shift(offset);
}
// printf("Second shift %d\n",this_node);fflush(stdout);
count = 0;
for (i = 0; i < RG_Nd; i++)
for (j = 0; j < RG_Nd; j++) if (j != i)
{
x[0] = i; x[1] = j;
c2 = find_count(d2,x,2);
offset = QDP_create_shift(d2[c2].s);
for (k = 0; k < RG_Nd; k++) if (k != i) if (k != j)
{
x[2] = k;
d3[count] = create_shift(x,3,len);
SQDP_M_eq_M_times_sM(path_3[count],path_2[c2],link_qdp[k],offset,QDP_forward,s);
count++;
}
QDP_destroy_shift(offset);
}
// printf("Third shift %d\n",this_node);fflush(stdout);
count = 0;
for (i = 0; i < RG_Nd; i++)
for (j = 0; j < RG_Nd; j++) if (j != i)
for (k = 0; k < RG_Nd; k++) if (k != i) if (k != j)
{
x[0] = i; x[1] = j; x[2] = k;
c2 = find_count(d3,x,3);
offset = QDP_create_shift(d3[c2].s);
for (t = 0; t < RG_Nd; t++) if (t != i) if (t != j) if (t != k)
{
x[3] = t;
d4[count] = create_shift(x,4,len);
SQDP_M_eq_M_times_sM(path_4[count],path_3[c2],link_qdp[t],offset,QDP_forward,s);
count++;
}
QDP_destroy_shift(offset);
}
// printf("Fourth shift %d\n",this_node);fflush(stdout);
QLA_C_eq_R(&unit,&c);
SQDP_M_eq_c(wlink[0],&unit,s);
for (i=1;i<5;i++)
SQDP_M_eq_M(wlink[i],path_1[i-1],s);
for (i=5;i<RG_Ncn;i++)
{
SQDP_M_eq_zero(wlink[i],s);
for (j=0;j<12;j++) if(d2[j].rv == i)
SQDP_M_peq_r_times_M(wlink[i],&fact2,path_2[j],s);
for (j=0;j<24;j++) if(d3[j].rv == i)
SQDP_M_peq_r_times_M(wlink[i],&fact3,path_3[j],s);
for (j=0;j<24;j++) if(d4[j].rv == i)
SQDP_M_peq_r_times_M(wlink[i],&fact4,path_3[j],s);
}
space_only = RG_Ncn;
// printf("projection %d\n",this_node);fflush(stdout);
project_qdp(wlink, pr_wlink,&space_only);
for (i = 0; i < 4; i++)
QDP_destroy_M(path_1[i]);
for (i = 0; i < 12; i++)
QDP_destroy_M(path_2[i]);
for (i = 0; i < 24; i++)
{
QDP_destroy_M(path_3[i]);
QDP_destroy_M(path_4[i]);
}
for (i = 0; i < RG_Ncn; i++)
QDP_destroy_M(wlink[i]);
free(d1);
free(d2);
free(d3);
free(d4);
return;
}
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]);
}
}
int
congrad_w(int niter, Real rsqmin, Real *final_rsq_ptr)
{
int i;
int iteration; /* counter for iterations */
double source_norm;
double rsqstop;
QLA_Real a, b;
double rsq,oldrsq,pkp; /* Sugar's a,b,resid**2,previous resid*2 */
/* pkp = cg_p.K.cg_p */
QLA_Real mkappa;
QLA_Real sum;
#ifdef CGTIME
double dtime;
#endif
#ifdef LU
mkappa = -kappa*kappa;
#else
mkappa = -kappa;
#endif
setup_cg();
for(i=0; i<4; i++) {
set_M_from_site(gaugelink[i], F_OFFSET(link[i]),EVENANDODD);
}
set_D_from_site(psi, F_OFFSET(psi),EVENANDODD);
set_D_from_site(chi, F_OFFSET(chi),EVENANDODD);
#ifdef PRESHIFT_LINKS
{
QDP_ColorMatrix *tcm;
tcm = QDP_create_M();
for(i=0; i<4; i++) {
QDP_M_eq_sM(tcm, gaugelink[i], QDP_neighbor[i], QDP_backward, QDP_all);
QDP_M_eq_Ma(gaugelink[i+4], tcm, QDP_all);
}
QDP_destroy_M(tcm);
}
#endif
#ifdef CGTIME
dtime = -dclock();
#endif
iteration=0;
start:
/* mp <- M_adjoint*M*psi
r,p <- chi - mp
rsq = |r|^2
source_norm = |chi|^2
*/
rsq = source_norm = 0.0;
#ifdef LU
QDP_D_eq_D(cgp, psi, QDP_even);
dslash_special_qdp(tt1, cgp, 1, QDP_odd, temp1);
dslash_special_qdp(ttt, tt1, 1, QDP_even, temp2);
QDP_D_eq_r_times_D_plus_D(ttt, &mkappa, ttt, cgp, QDP_even);
dslash_special_qdp(tt2, ttt, -1, QDP_odd, temp3);
dslash_special_qdp(mp, tt2, -1, QDP_even, temp4);
QDP_D_eq_r_times_D_plus_D(mp, &mkappa, mp, ttt, QDP_even);
QDP_D_eq_D_minus_D(cgr, chi, mp, QDP_even);
QDP_D_eq_D(cgp, cgr, QDP_even);
QDP_r_eq_norm2_D(&sum, chi, QDP_even);
source_norm = sum;
QDP_r_eq_norm2_D(&sum, cgr, QDP_even);
rsq = sum;
#else
QDP_D_eq_D(cgp, psi, QDP_even);
dslash_special_qdp(ttt, cgp, 1, QDP_all, temp1);
QDP_D_eq_r_times_D_plus_D(ttt, &mkappa, ttt, cgp, QDP_all);
dslash_special_qdp(mp, ttt, -1, QDP_all, temp1);
QDP_D_eq_r_times_D_plus_D(mp, &mkappa, mp, ttt, QDP_all);
QDP_D_eq_D_minus_D(cgr, chi, mp, QDP_all);
QDP_D_eq_D(cgp, cgr, QDP_all);
QDP_r_eq_norm2_D(&sum, chi, QDP_all);
source_norm = sum;
QDP_r_eq_norm2_D(&sum, cgr, QDP_all);
rsq = sum;
#endif
iteration++ ; /* iteration counts number of multiplications
by M_adjoint*M */
total_iters++;
/**if(this_node==0)printf("congrad2: source_norm = %e\n",source_norm);
if(this_node==0)printf("congrad2: iter %d, rsq %e, pkp %e, a %e\n",
iteration,(double)rsq,(double)pkp,(double)a );**/
rsqstop = rsqmin * source_norm;
if( rsq <= rsqstop ){
*final_rsq_ptr= (Real)rsq;
return (iteration);
}
/* main loop - do until convergence or time to restart */
/*
oldrsq <- rsq
mp <- M_adjoint*M*p
pkp <- p.M_adjoint*M.p
a <- rsq/pkp
psi <- psi + a*p
r <- r - a*mp
rsq <- |r|^2
b <- rsq/oldrsq
p <- r + b*p
*/
do {
oldrsq = rsq;
#ifdef LU
dslash_special_qdp(tt1, cgp, 1, QDP_odd, temp1);
dslash_special_qdp(ttt, tt1, 1, QDP_even, temp2);
QDP_D_eq_r_times_D_plus_D(ttt, &mkappa, ttt, cgp, QDP_even);
dslash_special_qdp(tt2, ttt, -1, QDP_odd, temp3);
dslash_special_qdp(mp, tt2, -1, QDP_even, temp4);
QDP_D_eq_r_times_D_plus_D(mp, &mkappa, mp, ttt, QDP_even);
QDP_r_eq_re_D_dot_D(&sum, cgp, mp, QDP_even);
pkp = sum;
#else
dslash_special_qdp(ttt, cgp, 1, QDP_all, temp1);
QDP_D_eq_r_times_D_plus_D(ttt, &mkappa, ttt, cgp, QDP_all);
dslash_special_qdp(mp, ttt, -1, QDP_all, temp1);
QDP_D_eq_r_times_D_plus_D(mp, &mkappa, mp, ttt, QDP_all);
QDP_r_eq_re_D_dot_D(&sum, cgp, mp, QDP_all);
pkp = sum;
#endif
iteration++;
total_iters++;
a = rsq / pkp;
QDP_D_peq_r_times_D(psi, &a, cgp, MYSUBSET);
QDP_D_meq_r_times_D(cgr, &a, mp, MYSUBSET);
QDP_r_eq_norm2_D(&sum, cgr, MYSUBSET);
rsq = sum;
/**if(this_node==0)printf("congrad2: iter %d, rsq %e, pkp %e, a %e\n",
iteration,(double)rsq,(double)pkp,(double)a );**/
if( rsq <= rsqstop ){
*final_rsq_ptr= (Real)rsq;
#ifdef CGTIME
dtime += dclock();
if(this_node==0)
printf("CONGRAD2: time = %.2e size_r= %.2e iters= %d MF = %.1f\n",
dtime,rsq,iteration,
(double)6480*iteration*even_sites_on_node/(dtime*1e6));
//(double)5616*iteration*even_sites_on_node/(dtime*1e6));
#endif
set_site_from_D(F_OFFSET(psi), psi,EVENANDODD);
return (iteration);
}
b = rsq / oldrsq;
QDP_D_eq_r_times_D_plus_D(cgp, &b, cgp, cgr, MYSUBSET);
} while( iteration%niter != 0);
set_site_from_D(F_OFFSET(psi), psi,EVENANDODD);
if( iteration < 3*niter ) goto start;
*final_rsq_ptr= (Real)rsq;
return(iteration);
}
static void
create_fn_links_qdp(QDP_ColorMatrix *fl[], QDP_ColorMatrix *ll[],
QDP_ColorMatrix *gf[], asqtad_path_coeff *coeffs)
{
int i, dir;
QDP_ColorMatrix *staple, *tempmat1;
int nu,rho,sig ;
QLA_Real one_link;
#ifdef LLTIME
double nflopfl = 61632;
double nflopll = 1804;
#endif
double dtimefl,dtimell;
for(i=0; i<4; i++) {
fl[i] = QDP_create_M();
ll[i] = QDP_create_M();
}
staple = QDP_create_M();
tempmat1 = QDP_create_M();
dtimefl = -dclock();
/* to fix up the Lepage term, included by a trick below */
one_link = coeffs->one_link - 6.0*coeffs->lepage;
for(dir=0; dir<4; dir++) {
QDP_M_eq_r_times_M(fl[dir], &one_link, gf[dir], QDP_all);
for(nu=0; nu<4; nu++) if(nu!=dir) {
compute_gen_staple(staple, dir, nu, gf[dir],
(double)coeffs->three_staple, gf, fl);
compute_gen_staple(NULL, dir, nu, staple, coeffs->lepage, gf, fl);
for(rho=0; rho<4; rho++) if((rho!=dir)&&(rho!=nu)) {
compute_gen_staple(tempmat1, dir, rho, staple,
(double)coeffs->five_staple, gf, fl);
for(sig=0; sig<4; sig++) {
if((sig!=dir)&&(sig!=nu)&&(sig!=rho)) {
compute_gen_staple(NULL, dir, sig, tempmat1,
(double)coeffs->seven_staple, gf, fl);
}
} /* sig */
} /* rho */
} /* nu */
} /* dir */
dtimell = -dclock();
dtimefl -= dtimell;
#ifdef LLTIME
node0_printf("LLTIME(Fat): time = %e (Asqtad opt) mflops = %e\n",dtimefl,
(Real)nflopfl*volume/(1e6*dtimefl*numnodes()) );
#endif
/* long links */
for(dir=0; dir<4; dir++) {
QLA_Real naik = coeffs->naik;
QDP_M_eq_sM(staple, gf[dir], QDP_neighbor[dir], QDP_forward, QDP_all);
QDP_M_eq_M_times_M(tempmat1, gf[dir], staple, QDP_all);
QDP_discard_M(staple);
QDP_M_eq_sM(staple, tempmat1, QDP_neighbor[dir], QDP_forward, QDP_all);
QDP_M_eq_M_times_M(ll[dir], gf[dir], staple, QDP_all);
QDP_M_eq_r_times_M(ll[dir], &naik, ll[dir], QDP_all);
}
dtimell += dclock();
#ifdef LLTIME
node0_printf("LLTIME(long): time = %e (Asqtad opt) mflops = %e\n",dtimell,
(Real)nflopll*volume/(1e6*dtimell*numnodes()) );
#endif
QDP_destroy_M(staple);
QDP_destroy_M(tempmat1);
}
// topdir = 1..nd
// sidedir = -nd..nd
// toplinknum,sidelinknum = 0..nin-1
void
QOP_staples(QOP_info_t *info, int nout, int nin,
QDP_ColorMatrix *out[], QDP_ColorMatrix *in[],
int nstaples[], int *topdir[], int *sidedir[],
int *toplinknum[], int *sidelinknum[], QLA_Real *coef[])
{
#define NC QDP_get_nc(in[0])
double dtime = QOP_time();
double nflops = 0;
int nd = QDP_ndim();
QDP_ColorMatrix *ftmps[nin][nd], *t1, *t2, *bt2[nd];
for(int i=0; i<nin; i++)
for(int j=0; j<nd; j++)
ftmps[i][j] = NULL;
for(int i=0; i<nd; i++) bt2[i] = NULL;
t1 = QDP_create_M();
t2 = QDP_create_M();
for(int io=0; io<nout; io++) {
//QOP_printf0("%i: ns: %i\n", io, nstaples[io]);
for(int s=0; s<nstaples[io]; s++) {
QLA_Real c = coef[io][s];
int tn = toplinknum[io][s];
int sdir = sidedir[io][s];
//QOP_printf0(" %i: sdir: %i c: %g\n", s, sdir, c);
if(sdir==0) {
if(c==1) {
QDP_M_peq_M(out[io], in[tn], QDP_all);
nflops += PEQM;
} else {
QDP_M_peq_r_times_M(out[io], &c, in[tn], QDP_all);
nflops += 2*PEQM;
}
} else if(sdir>0) {
int nu = sdir-1;
int mu = topdir[io][s]-1;
int sn = sidelinknum[io][s];
QDP_ColorMatrix *Umunu = getU(tn, mu, nu);
QDP_ColorMatrix *Unumu = getU(sn, nu, mu);
QDP_M_eq_M_times_M(t1, in[sn], Umunu, QDP_all);
if(c==1) {
QDP_M_peq_M_times_Ma(out[io], t1, Unumu, QDP_all);
nflops += EQMTM+PEQMTM;
} else {
QDP_M_eq_M_times_Ma(t2, t1, Unumu, QDP_all);
QDP_M_peq_r_times_M(out[io], &c, t2, QDP_all);
nflops += 2*EQMTM+2*PEQM;
}
} else {
int nu = -sdir-1;
int mu = topdir[io][s]-1;
int sn = sidelinknum[io][s];
QDP_ColorMatrix *Unumu = getU(sn, nu, mu);
QDP_M_eq_M_times_M(t1, in[tn], Unumu, QDP_all);
QDP_M_eq_Ma_times_M(t2, in[sn], t1, QDP_all);
QDP_ColorMatrix *tb = shiftb(t2, nu);
if(c==1) {
QDP_M_peq_M(out[io], tb, QDP_all);
nflops += 2*EQMTM+PEQM;
} else {
QDP_M_peq_r_times_M(out[io], &c, tb, QDP_all);
nflops += 2*EQMTM+2*PEQM;
}
QDP_discard_M(tb);
}
}
}
for(int i=0; i<nin; i++)
for(int j=0; j<nd; j++)
if(ftmps[i][j]!=NULL) QDP_destroy_M(ftmps[i][j]);
for(int i=0; i<nd; i++) if(bt2[i]!=NULL) QDP_destroy_M(bt2[i]);
QDP_destroy_M(t1);
QDP_destroy_M(t2);
info->final_sec = QOP_time() - dtime;
info->final_flop = nflops*QDP_sites_on_node;
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 load_asqtad_links(int both, ferm_links_t *fn, ks_action_paths *ap) {
su3_matrix **t_fl = &fn->fat;
su3_matrix **t_ll = &fn->lng;
Real *act_path_coeff = ap->act_path_coeff;
QDP_ColorMatrix *fl[4];
QDP_ColorMatrix *ll[4];
QDP_ColorMatrix *gf[4];
int dir;
double remaptime = -dclock();
char myname[] = "load_asqtad_links";
asqtad_path_coeff c;
if( phases_in != 1){
node0_printf("%s: BOTCH: needs phases in\n",myname);
terminate(1);
}
/* Create QDP fields for fat links, long links, and temp for gauge field */
FORALLUPDIR(dir){
fl[dir] = QDP_create_M();
ll[dir] = QDP_create_M();
gf[dir] = QDP_create_M();
}
/* Map gauge links to QDP */
set4_M_from_site(gf, F_OFFSET(link), EVENANDODD);
/* Load Asqtad path coefficients from table */
c.one_link = act_path_coeff[0];
c.naik = act_path_coeff[1];
c.three_staple = act_path_coeff[2];
c.five_staple = act_path_coeff[3];
c.seven_staple = act_path_coeff[4];
c.lepage = act_path_coeff[5];
/* Compute fat and long links as QDP fields */
remaptime += dclock();
create_fn_links_qdp(fl, ll, gf, &c);
remaptime -= dclock();
/* Clean up */
FORALLUPDIR(dir){
QDP_destroy_M(gf[dir]);
}
/* Allocate space for t_fatlink if NULL */
if(*t_fl == NULL){
*t_fl = (su3_matrix *)special_alloc(sites_on_node*4*sizeof(su3_matrix));
if(*t_fl==NULL){
printf("%s(%d): no room for t_fatlink\n",myname,this_node);
terminate(1);
}
}
/* Allocate space for t_longlink if NULL and we are doing both fat
and long */
if(*t_ll == NULL && both){
*t_ll = (su3_matrix *)special_alloc(sites_on_node*4*sizeof(su3_matrix));
if(*t_ll==NULL){
printf("%s(%d): no room for t_longlink\n",myname,this_node);
terminate(1);
}
}
/* Map QDP fields to MILC order */
set4_field_from_M(*t_fl, fl, EVENANDODD);
if(both)set4_field_from_M(*t_ll, ll, EVENANDODD);
/* Clean up */
FORALLUPDIR(dir){
QDP_destroy_M(fl[dir]);
QDP_destroy_M(ll[dir]);
}
fn->valid = 1;
remaptime += dclock();
#ifdef LLTIME
#ifdef REMAP
node0_printf("LLREMAP: time = %e\n",remaptime);
#endif
#endif
}
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
}
/* 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_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
void RG_smear_dir (QDP_ColorMatrix *sm_link,
QDP_ColorMatrix *link[],
QLA_Real w_l, QLA_Real w_s,
QLA_Int dir, QDP_Sub_Block s, int len)
{
int i,v[RG_Nd],n;
QLA_Int nu,mu=dir;
QDP_Subset sub;
QDP_Shift offset[RG_Nd];
QLA_Complex unit;
QDP_ColorMatrix *temp1, *temp2, *temp3, *temp4, *temp5, *temp6;
temp1 = QDP_create_M();
temp2 = QDP_create_M();
temp3 = QDP_create_M();
temp4 = QDP_create_M();
temp5 = QDP_create_M();
temp6 = QDP_create_M();
for(nu=0; nu < RG_Nd ; nu++)
{
for(i=0; i<RG_Nd;i++) v[i] = 0;
v[nu] = len;
offset[nu] = QDP_create_shift(v);
}
SQDP_M_eq_r_times_M(temp6,&w_l,link[mu],s);
/* Set temp4 to zero */
SQDP_M_eq_zero(temp4,s);
n = RG_Nd;
#ifdef CHECK_SMEAR_QDP_MILC
n = 3;
#endif
/* Sum on staples */
for(nu=0; nu < n ; nu++)if(nu != mu)
{
/* For forward staples */
SQDP_M_eq_sM(temp1, link[mu], offset[nu], QDP_forward, s);
SQDP_M_eq_sM(temp2, link[nu], offset[mu], QDP_forward, s);
SQDP_M_eq_M_times_Ma(temp3, temp1, temp2, s);
SQDP_M_peq_M_times_M(temp4, link[nu], temp3, s);
/* For backward staples */
SQDP_M_eq_M_times_M(temp3, link[mu], temp2, s);
SQDP_M_eq_Ma_times_M(temp1, link[nu], temp3, s);
SQDP_M_eq_sM(temp5, temp1, offset[nu], QDP_backward, s);
SQDP_M_peq_M(temp4, temp5, s);
}
/* U_smeared = w_l * U + w_s * U_staple */
SQDP_M_eq_r_times_M_plus_M(sm_link,&w_s,temp4,temp6,s);
QDP_destroy_M(temp1);
QDP_destroy_M(temp2);
QDP_destroy_M(temp3);
QDP_destroy_M(temp4);
QDP_destroy_M(temp5);
QDP_destroy_M(temp6);
for(nu=0; nu < RG_Nd ; nu++)
QDP_destroy_shift(offset[nu]);
return ;
}
static void
get_staple_plaq(QDP_ColorMatrix *staple, int mu, QDP_ColorMatrix *u[],
QOP_gauge_coeffs_t *coeffs,
QDP_Subset subset, QDP_Subset osubset)
{
#define NC QDP_get_nc(staple)
QDP_Lattice *lat = QDP_get_lattice_M(staple);
int nd = QDP_ndim_L(lat);
QDP_Shift *neighbor = QDP_neighbor_L(lat);
QLA_Real plaq = coeffs->plaquette;
QLA_Real adpl = coeffs->adjoint_plaquette;
#if 1
QDP_ColorMatrix *temp1, *temp2, *temp3, *temp4, *temp5, *temp6;
//temp1 = QDP_create_M();
temp2 = QDP_create_M_L(lat);
//temp3 = QDP_create_M();
temp4 = QDP_create_M_L(lat);
//temp5 = QDP_create_M();
temp6 = QDP_create_M_L(lat);
QDP_Complex *tc = NULL;
if(adpl!=0) tc = QDP_create_C_L(lat);
/* staple += u[nu](x) u[mu](x+nu) u*[nu](x+mu)
* + u*[nu](x-nu) u[mu](x-nu) u[nu](x-nu+mu) */
for(int nu=0; nu<nd; nu++) {
if (nu == mu) continue;
temp1 = QDP_create_M_L(lat);
temp3 = QDP_create_M_L(lat);
temp5 = QDP_create_M_L(lat);
QDP_M_eq_sM(temp1, u[nu], neighbor[mu], QDP_forward, QDP_all_L(lat));
QDP_M_eq_Ma_times_M(temp2, u[nu], u[mu], osubset);
QDP_M_eq_sM(temp3, u[mu], neighbor[nu], QDP_forward, subset);
QDP_M_eq_M_times_M(temp4, temp2, temp1, osubset);
QDP_M_eq_sM(temp5, temp4, neighbor[nu], QDP_backward, subset);
QDP_M_eq_M_times_M(temp6, u[nu], temp3, subset);
//QDP_M_peq_M_times_Ma(staple, temp6, temp1, subset);
//QDP_M_peq_M(staple, temp5, subset);
if(adpl==0) {
QDP_M_peq_M_times_Ma(temp5, temp6, temp1, subset);
QDP_M_peq_r_times_M(staple, &plaq, temp5, subset);
} else {
QLA_Complex z;
QLA_c_eq_r(z, plaq/adpl);
QDP_C_eq_c(tc, &z, subset);
QDP_M_eq_M_times_Ma(temp2, temp6, temp1, subset);
QDP_C_peq_M_dot_M(tc, temp2, u[mu], subset);
QDP_C_eq_r_times_C(tc, &adpl, tc, subset);
QDP_M_peq_C_times_M(staple, tc, temp2, subset);
QDP_C_eq_c(tc, &z, subset);
QDP_C_peq_M_dot_M(tc, temp5, u[mu], subset);
QDP_C_eq_r_times_C(tc, &adpl, tc, subset);
QDP_M_peq_C_times_M(staple, tc, temp5, subset);
}
//QDP_discard_M(temp1);
//QDP_discard_M(temp3);
//QDP_discard_M(temp5);
QDP_destroy_M(temp1);
QDP_destroy_M(temp3);
QDP_destroy_M(temp5);
} /* closes nu loop */
if(adpl!=0) QDP_destroy_C(tc);
//QDP_destroy_M(temp1);
QDP_destroy_M(temp2);
//QDP_destroy_M(temp3);
QDP_destroy_M(temp4);
//QDP_destroy_M(temp5);
QDP_destroy_M(temp6);
#else
QDP_ColorMatrix *t = QDP_create_M_L(lat);
int nu, path[3];
QDP_Subset subs[2];
subs[0] = subset;
subs[1] = osubset;
for(nu=0; nu<nd; nu++) {
if (nu == mu) continue;
path[0] = 1+nu;
path[1] = -(1+mu);
path[2] = -(1+nu);
path_prod(u, t, path, 3, 1, subs, neighsubeo);
QDP_M_peq_M(staple, t, subset);
path[0] = -(1+nu);
path[1] = -(1+mu);
path[2] = 1+nu;
path_prod(u, t, path, 3, 1, subs, neighsubeo);
QDP_M_peq_M(staple, t, subset);
}
QDP_destroy_M(t);
#endif
}
// topdir = 1..nd
// sidedir = -nd..nd
// toplinknum,sidelinknum = 0..nin-1
void
QOP_staples_deriv(QOP_info_t *info, int nout, int nin,
QDP_ColorMatrix *deriv[], QDP_ColorMatrix *chain[],
QDP_ColorMatrix *in[],
int nstaples[], int *topdir[], int *sidedir[],
int *toplinknum[], int *sidelinknum[], QLA_Real *coef[])
{
#define NC QDP_get_nc(in[0])
double dtime = QOP_time();
double nflops = 0;
int nd = QDP_ndim();
QDP_ColorMatrix *ftmps[nin][nd], *t1, *t2, *t3, *t4, *tc, *bt2[nd], *bt3[nd], *ctmps[nd];
int ctn[nd];
for(int i=0; i<nin; i++)
for(int j=0; j<nd; j++)
ftmps[i][j] = NULL;
for(int i=0; i<nd; i++) bt2[i] = bt3[i] = ctmps[i] = NULL;
t1 = QDP_create_M();
t2 = QDP_create_M();
t3 = QDP_create_M();
t4 = QDP_create_M();
tc = QDP_create_M();
// process in reverse in case calculated staples used as input for others
for(int io=nout-1; io>=0; io--) {
for(int i=0; i<nd; i++) {
if(ctmps[i]) QDP_discard_M(ctmps[i]);
ctn[i] = 0;
}
QDP_M_eq_M(tc, chain[io], QDP_all);
for(int s=0; s<nstaples[io]; s++) {
QLA_Real c = coef[io][s];
int tn = toplinknum[io][s];
int sdir = sidedir[io][s];
//QOP_printf0("io: %i s: %i sdir: %i tn: %i c: %g\n", io, s, sdir, tn, c);
if(sdir==0) {
if(c==1) {
QDP_M_peq_M(deriv[tn], tc, QDP_all);
nflops += PEQM;
} else {
QDP_M_peq_r_times_M(deriv[tn], &c, tc, QDP_all);
nflops += 2*PEQM;
}
} else if(sdir>0) {
int nu = sdir-1;
int mu = topdir[io][s]-1;
int sn = sidelinknum[io][s];
//QOP_printf0(" mu: %i nu: %i sn: %i\n", mu, nu, sn);
QDP_ColorMatrix *Umunu = getU(tn, mu, nu);
QDP_ColorMatrix *Unumu = getU(sn, nu, mu);
QDP_M_eq_M_times_M(t1, in[sn], Umunu, QDP_all);
QDP_M_eq_Ma_times_M(t2, tc, t1, QDP_all);
QDP_ColorMatrix *tb2 = shiftb(t2, mu);
QDP_M_eq_M_times_M(t1, tc, Unumu, QDP_all);
QDP_M_eq_Ma_times_M(t3, in[sn], t1, QDP_all);
QDP_ColorMatrix *tb3 = shiftb(t3, nu);
if(c==1) {
QDP_M_peq_M_times_Ma(deriv[sn], t1, Umunu, QDP_all);
QDP_M_peq_M(deriv[sn], tb2, QDP_all);
QDP_M_peq_M(deriv[tn], tb3, QDP_all);
nflops += 4*EQMTM+PEQMTM+2*PEQM;
} else {
QDP_M_eq_M_times_Ma(t4, t1, Umunu, QDP_all);
QDP_M_peq_r_times_M(deriv[sn], &c, t4, QDP_all);
QDP_M_peq_r_times_M(deriv[sn], &c, tb2, QDP_all);
QDP_M_peq_r_times_M(deriv[tn], &c, tb3, QDP_all);
nflops += 5*EQMTM+6*PEQM;
}
QDP_discard_M(tb2);
QDP_discard_M(tb3);
} else {
int nu = -sdir-1;
int mu = topdir[io][s]-1;
int sn = sidelinknum[io][s];
QDP_ColorMatrix *Cmunu = getC(nu);
QDP_ColorMatrix *Unumu = getU(sn, nu, mu);
QDP_M_eq_M_times_M(t1, in[sn], Cmunu, QDP_all);
QDP_M_eq_Ma_times_M(t2, in[tn], t1, QDP_all);
QDP_ColorMatrix *tb2 = shiftb(t2, mu);
QDP_M_eq_M_times_M(t3, in[tn], Unumu, QDP_all);
if(c==1) {
QDP_M_peq_M_times_Ma(deriv[tn], t1, Unumu, QDP_all);
QDP_M_peq_M_times_Ma(deriv[sn], t3, Cmunu, QDP_all);
QDP_M_peq_M(deriv[sn], tb2, QDP_all);
nflops += 3*EQMTM+2*PEQMTM+PEQM;
} else {
QDP_M_eq_M_times_Ma(t4, t1, Unumu, QDP_all);
QDP_M_peq_r_times_M(deriv[tn], &c, t4, QDP_all);
QDP_M_eq_M_times_Ma(t4, t3, Cmunu, QDP_all);
QDP_M_peq_r_times_M(deriv[sn], &c, t4, QDP_all);
QDP_M_peq_r_times_M(deriv[sn], &c, tb2, QDP_all);
nflops += 5*EQMTM+6*PEQM;
}
QDP_discard_M(tb2);
}
}
}
for(int i=0; i<nin; i++)
for(int j=0; j<nd; j++)
if(ftmps[i][j]!=NULL) QDP_destroy_M(ftmps[i][j]);
for(int i=0; i<nd; i++) {
if(bt2[i]!=NULL) QDP_destroy_M(bt2[i]);
if(bt3[i]!=NULL) QDP_destroy_M(bt3[i]);
if(ctmps[i]!=NULL) QDP_destroy_M(ctmps[i]);
}
QDP_destroy_M(t1);
QDP_destroy_M(t2);
QDP_destroy_M(t3);
QDP_destroy_M(t4);
QDP_destroy_M(tc);
info->final_sec = QOP_time() - dtime;
info->final_flop = nflops*QDP_sites_on_node;
info->status = QOP_SUCCESS;
#undef NC
}
static void
get_staple_imp(QDP_ColorMatrix *staple, int mu, QDP_ColorMatrix **u,
QOP_gauge_coeffs_t *coeffs,
int subl, QDP_Subset subs[], int (*neighsub)(int subl, int dir))
{
#define NC QDP_get_nc(staple)
QDP_Lattice *lat = QDP_get_lattice_M(staple);
int nd = QDP_ndim_L(lat);
int nd2 = 2*nd;
QLA_Real plaq = coeffs->plaquette;
QLA_Real rect = coeffs->rectangle;
QLA_Real pgm = coeffs->parallelogram;
QLA_Real adpl = coeffs->adjoint_plaquette;
QDP_ColorMatrix *sm0[2][nd2];
QDP_ColorMatrix *t = QDP_create_M_L(lat);
for(int i=0; i<2; i++) {
tm[i] = QDP_create_M_L(lat);
sm[i] = sm0[i];
for(int nu=0; nu<nd2; nu++) {
sm[i][nu] = QDP_create_M_L(lat);
}
}
QDP_Complex *tc = NULL;
if(adpl!=0) tc = QDP_create_C_L(lat);
int mup = 1 + mu;
int bsubl = neighsub(subl, mup);
int path[5];
QDP_Subset subset = subs[subl];
if(plaq!=0 || adpl!=0) {
for(int nu=-nd; nu<=nd; nu++) {
if ( nu==-mup || nu==0 || nu==mup ) continue;
path[0] = nu;
path[1] = -mup;
path[2] = -nu;
path_prod(u, t, path, 3, bsubl, subs, neighsub);
if(adpl==0) {
QDP_M_peq_r_times_M(staple, &plaq, t, subset);
} else {
QLA_Complex z;
QLA_c_eq_r(z, plaq/adpl);
QDP_C_eq_c(tc, &z, subset);
QDP_C_peq_M_dot_M(tc, t, u[mu], subset);
QDP_C_eq_r_times_C(tc, &adpl, tc, subset);
QDP_M_peq_C_times_M(staple, tc, t, subset);
}
}
}
if(rect) {
for(int nu=-nd; nu<=nd; nu++) {
if ( nu==-mup || nu==0 || nu==mup ) continue;
//s = QDP_create_M();
path[0] = nu;
path[1] = nu;
path[2] = -mup;
path[3] = -nu;
path[4] = -nu;
path_prod(u, t, path, 5, bsubl, subs, neighsub);
QDP_M_peq_r_times_M(staple, &rect, t, subset);
//QDP_destroy_M(s);
//s = QDP_create_M();
path[0] = nu;
path[1] = -mup;
path[2] = -mup;
path[3] = -nu;
path[4] = mup;
path_prod(u, t, path, 5, bsubl, subs, neighsub);
QDP_M_peq_r_times_M(staple, &rect, t, subset);
//QDP_destroy_M(s);
//s = QDP_create_M();
path[0] = mup;
path[1] = nu;
path[2] = -mup;
path[3] = -mup;
path[4] = -nu;
path_prod(u, t, path, 5, bsubl, subs, neighsub);
QDP_M_peq_r_times_M(staple, &rect, t, subset);
//QDP_destroy_M(s);
}
}
if(pgm) {
for(int nu=-nd; nu<=nd; nu++) {
if ( nu==-mup || nu==0 || nu==mup ) continue;
for(int rho=-nd; rho<=nd; rho++) {
if ( rho==-mup || rho==0 || rho==mup || rho==-nu || rho==nu ) continue;
path[0] = nu;
path[1] = rho;
path[2] = -mup;
path[3] = -nu;
path[4] = -rho;
path_prod(u, t, path, 5, bsubl, subs, neighsub);
QDP_M_peq_r_times_M(staple, &pgm, t, subset);
}
}
}
if(adpl!=0) QDP_destroy_C(tc);
QDP_destroy_M(t);
for(int i=0; i<2; i++) {
for(int nu=0; nu<nd2; nu++) {
QDP_destroy_M(sm[i][nu]);
}
QDP_destroy_M(tm[i]);
}
}
void
QOP_asqtad_force_multi_asvec_qdp(QOP_info_t *info, QDP_ColorMatrix *links[],
QDP_ColorMatrix *force[], QOP_asqtad_coeffs_t *coef,
REAL eps[], QDP_ColorVector *xin[], int nsrc)
{
#define NC QDP_get_nc(xin[0])
REAL coeff[nsrc];
REAL OneLink[nsrc], Lepage[nsrc], Naik[nsrc], FiveSt[nsrc], ThreeSt[nsrc], SevenSt[nsrc];
REAL mNaik[nsrc], mLepage[nsrc], mFiveSt[nsrc], mThreeSt[nsrc], mSevenSt[nsrc];
QDP_ColorVector *P3[8][nsrc];
QDP_ColorVector *P5[8][nsrc];
QDP_ColorVector *P5tmp[8][8][nsrc];
QDP_ColorVector *P5s[4][nsrc];
QDP_ColorVector *P5tmps[4][8][nsrc];
//QDP_ColorVector *xin[nsrc];
QDP_ColorVector *xintmp[8][nsrc];
QDP_ColorVector *Pmu[nsrc];
QDP_ColorVector *Pmutmp[8][nsrc];
QDP_ColorVector *Pnumu[nsrc];
QDP_ColorVector *Pnumutmp[8][nsrc];
QDP_ColorVector *Prhonumu[nsrc];
QDP_ColorVector *Prhonumutmp[8][nsrc];
QDP_ColorVector *P7[nsrc];
QDP_ColorVector *P7tmp[8][nsrc];
QDP_ColorVector *P7rho[nsrc];
QDP_ColorVector *ttv[nsrc];
int i, dir;
int mu, nu, rho, sig;
double nflop1 = 253935;
double nflop2 = 433968;
double nflop = nflop1 + (nflop2-nflop1)*(nsrc-1);
double dtime;
dtime = -QOP_time();
ASQTAD_FORCE_BEGIN;
QOP_trace("test 1\n");
/* setup parallel transport */
QDP_ColorMatrix *tmpmat = QDP_create_M();
for(i=0; i<QOP_common.ndim; i++) {
fbshift[i] = QDP_neighbor[i];
fbshiftdir[i] = QDP_forward;
fblink[i] = links[i];
fbshift[OPP_DIR(i)] = QDP_neighbor[i];
fbshiftdir[OPP_DIR(i)] = QDP_backward;
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);
}
tv = ttv;
for(i=0; i<nsrc; i++) {
tv[i] = QDP_create_V();
}
QOP_trace("test 2\n");
/* Allocate temporary vectors */
for(i=0; i<nsrc; i++) {
Pmu[i] = QDP_create_V();
Pnumu[i] = QDP_create_V();
Prhonumu[i] = QDP_create_V();
P7[i] = QDP_create_V();
P7rho[i] = QDP_create_V();
for(dir=0; dir<8; dir++) {
xintmp[dir][i] = QDP_create_V();
Pmutmp[dir][i] = QDP_create_V();
Pnumutmp[dir][i] = QDP_create_V();
Prhonumutmp[dir][i] = QDP_create_V();
P7tmp[dir][i] = QDP_create_V();
}
#if 1
for(mu=0; mu<4; mu++) {
P5s[mu][i] = QDP_create_V();
for(dir=0; dir<8; dir++) {
P5tmps[mu][dir][i] = QDP_create_V();
}
}
#else
for(mu=0; mu<8; mu++) {
P5[mu][i] = QDP_create_V();
for(dir=0; dir<8; dir++) {
P5tmp[mu][dir][i] = QDP_create_V();
//printf("%p %p\n", P5tmp[mu][dir][i], &(P5tmp[mu][dir][i])); fflush(stdout);
if(P5tmp[mu][dir][i]==NULL) {
fprintf(stderr, "error: can't create V\n");
QDP_abort();
}
}
}
#endif
}
//printf("%p\n", P5tmp[0][4][0]); fflush(stdout);
for(mu=0; mu<8; mu++) {
for(i=0; i<nsrc; i++) {
P3[mu][i] = QDP_create_V();
//P5[mu][i] = QDP_create_V();
}
}
for(mu=0; mu<4; mu++) {
tempmom_qdp[mu] = force[mu];
QDP_M_eqm_M(tempmom_qdp[mu], tempmom_qdp[mu], QDP_odd);
}
/* Path coefficients times fermion epsilon */
/* Load path coefficients from table */
for(i=0; i<nsrc; i++) {
OneLink[i] = coef->one_link * eps[i];
Naik[i] = coef->naik * eps[i]; mNaik[i] = -Naik[i];
ThreeSt[i] = coef->three_staple * eps[i]; mThreeSt[i] = -ThreeSt[i];
FiveSt[i] = coef->five_staple * eps[i]; mFiveSt[i] = -FiveSt[i];
SevenSt[i] = coef->seven_staple * eps[i]; mSevenSt[i] = -SevenSt[i];
Lepage[i] = coef->lepage * eps[i]; mLepage[i] = -Lepage[i];
}
#if 0
printf("nsrc = %i\n", nsrc);
printf("coeffs = %g %g %g %g %g %g\n", OneLink[0], ThreeSt[0], FiveSt[0],
SevenSt[0], Lepage[0], Naik[0]);
#endif
/* *************************************** */
QOP_trace("start force loop\n");
for(mu=0; mu<8; mu++) {
//u_shift_hw_fermion(temp_x_qdp, Pmu, OPP_DIR(mu), temp_hw[OPP_DIR(mu)]);
u_shift_color_vecs(xin, Pmu, OPP_DIR(mu), nsrc, xintmp[OPP_DIR(mu)]);
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) ) {
//u_shift_hw_fermion(Pmu, P3[sig], sig, temp_hw[sig]);
u_shift_color_vecs(Pmu, P3[sig], sig, nsrc, Pmutmp[sig]);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu]: x--+ *
* | | *
* o o *
* the 1 link in the path: - (numbering starts form 0) */
add_forces_to_mom(P3[sig], Pmu, sig, mThreeSt, nsrc);
}
}
for(nu=0; nu<8; nu++) if( (nu!=mu)&&(nu!=OPP_DIR(mu)) ) {
int nP5 = 0;
//Pnumu = hw_qdp[OPP_DIR(nu)];
//u_shift_hw_fermion(Pmu, Pnumu, OPP_DIR(nu), temp_hw[OPP_DIR(nu)]);
u_shift_color_vecs(Pmu, Pnumu, OPP_DIR(nu), nsrc, Pmutmp[OPP_DIR(nu)]);
//QDP_V_veq_V(Pnumu, P3[OPP_DIR(nu)], QDP_all, nsrc);
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) &&
(sig!=nu)&&(sig!=OPP_DIR(nu)) ) {
#if 1
for(i=0; i<nsrc; i++) {
P5[sig][i] = P5s[nP5][i];
for(dir=0; dir<8; dir++) P5tmp[sig][dir][i] = P5tmps[nP5][dir][i];
}
#endif
nP5++;
//u_shift_hw_fermion(Pnumu, P5[sig], sig, temp_hw[sig]);
u_shift_color_vecs(Pnumu, P5[sig], sig, nsrc, Pnumutmp[sig]);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu+nu]: x--+ *
* | | *
* o o *
* the 2 link in the path: + (numbering starts form 0) */
add_forces_to_mom(P5[sig], Pnumu, sig, FiveSt, nsrc);
}
}
QOP_trace("test 4\n");
for(rho=0; rho<8; rho++) if( (rho!=mu)&&(rho!=OPP_DIR(mu)) &&
(rho!=nu)&&(rho!=OPP_DIR(nu)) ) {
//Prhonumu = hw_qdp[OPP_DIR(rho)];
//u_shift_hw_fermion(Pnumu, Prhonumu, OPP_DIR(rho),
// temp_hw[OPP_DIR(rho)] );
u_shift_color_vecs(Pnumu, Prhonumu, OPP_DIR(rho), nsrc,
Pnumutmp[OPP_DIR(rho)]);
//QDP_V_veq_V(Prhonumu, P5[OPP_DIR(rho)], QDP_all, nsrc);
for(sig=0; sig<8; sig++) if( (sig!=mu )&&(sig!=OPP_DIR(mu )) &&
(sig!=nu )&&(sig!=OPP_DIR(nu )) &&
(sig!=rho)&&(sig!=OPP_DIR(rho)) ) {
/* Length 7 paths */
//P7 = hw_qdp[sig];
//u_shift_hw_fermion(Prhonumu, P7, sig, temp_hw[sig] );
QOP_trace("test 43\n");
u_shift_color_vecs(Prhonumu, P7, sig, nsrc, Prhonumutmp[sig]);
QOP_trace("test 44\n");
//QDP_V_eq_r_times_V(P7[0], &SevenSt[0], P7[0], QDP_all);
//QDP_V_eq_r_times_V(P7[1], &SevenSt[1], P7[1], QDP_all);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu+nu+rho]: x--+ *
* | | *
* o o *
* the 3 link in the path: - (numbering starts form 0) */
QOP_trace("test 45\n");
add_forces_to_mom(P7, Prhonumu, sig, mSevenSt, nsrc);
QOP_trace("test 46\n");
//mom_meq_force(P7, Prhonumu, sig);
}
/* Add the force F_rho the 2(4) link in the path: + */
//P7rho = hw_qdp[rho];
//u_shift_hw_fermion(P7, P7rho, rho, temp_hw[rho]);
QOP_trace("test 47\n");
u_shift_color_vecs(P7, P7rho, rho, nsrc, P7tmp[rho]);
QOP_trace("test 48\n");
side_link_forces(rho,sig,SevenSt,Pnumu,P7,Prhonumu,P7rho, nsrc);
QOP_trace("test 49\n");
//side_link_3f_force2(rho,sig,Pnumu,P7,Prhonumu,P7rho);
/* Add the P7rho vector to P5 */
for(i=0; i<nsrc; i++) {
if(FiveSt[i]!=0) coeff[i] = SevenSt[i]/FiveSt[i];
else coeff[i] = 0;
QOP_trace("test 410\n");
QDP_V_peq_r_times_V(P5[sig][i], &coeff[i], P7rho[i], QDP_all);
QOP_trace("test 411\n");
}
} /* sig */
} /* rho */
QOP_trace("test 5\n");
#define P5nu P7
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) &&
(sig!=nu)&&(sig!=OPP_DIR(nu)) ) {
/* Length 5 paths */
/* Add the force F_nu the 1(3) link in the path: - */
//P5nu = hw_qdp[nu];
//u_shift_hw_fermion(P5[sig], P5nu, nu, temp_hw[nu]);
u_shift_color_vecs(P5[sig], P5nu, nu, nsrc, P5tmp[sig][nu]);
side_link_forces(nu, sig, mFiveSt, Pmu, P5[sig], Pnumu, P5nu, nsrc);
/* Add the P5nu vector to P3 */
for(i=0; i<nsrc; i++) {
if(ThreeSt[i]!=0) coeff[i] = FiveSt[i]/ThreeSt[i];
else coeff[i] = 0;
QDP_V_peq_r_times_V(P3[sig][i], &coeff[i], P5nu[i], QDP_all);
}
} /* sig */
} /* nu */
#define Pmumu Pnumu
#define Pmumutmp Pnumutmp
#define P5sig Prhonumu
#define P5sigtmp Prhonumutmp
#define P3mu P7
#define Popmu P7
#define Pmumumu P7
/* Now the Lepage term... It is the same as 5-link paths with
nu=mu and FiveSt=Lepage. */
//u_shift_hw_fermion(Pmu, Pmumu, OPP_DIR(mu), temp_hw[OPP_DIR(mu)] );
u_shift_color_vecs(Pmu, Pmumu, OPP_DIR(mu), nsrc, Pmutmp[OPP_DIR(mu)]);
for(sig=0; sig<8; sig++) if( (sig!=mu)&&(sig!=OPP_DIR(mu)) ) {
//P5sig = hw_qdp[sig];
//u_shift_hw_fermion(Pmumu, P5sig, sig, temp_hw[sig]);
u_shift_color_vecs(Pmumu, P5sig, sig, nsrc, Pmumutmp[sig]);
if(GOES_FORWARDS(sig)) {
/* Add the force F_sig[x+mu+nu]: x--+ *
* | | *
* o o *
* the 2 link in the path: + (numbering starts form 0) */
add_forces_to_mom(P5sig, Pmumu, sig, Lepage, nsrc);
}
/* Add the force F_nu the 1(3) link in the path: - */
//P5nu = hw_qdp[mu];
//u_shift_hw_fermion(P5sig, P5nu, mu, temp_hw[mu]);
u_shift_color_vecs(P5sig, P5nu, mu, nsrc, P5sigtmp[mu]);
side_link_forces(mu, sig, mLepage, Pmu, P5sig, Pmumu, P5nu, nsrc);
/* Add the P5nu vector to P3 */
for(i=0; i<nsrc; i++) {
if(ThreeSt[i]!=0) coeff[i] = Lepage[i]/ThreeSt[i];
else coeff[i] = 0;
QDP_V_peq_r_times_V(P3[sig][i], &coeff[i], P5nu[i], QDP_all);
}
/* Length 3 paths (Not the Naik term) */
/* Add the force F_mu the 0(2) link in the path: + */
if(GOES_FORWARDS(mu)) {
//P3mu = hw_qdp[mu]; /* OK to clobber P5nu */
//u_shift_hw_fermion(P3[sig], P3mu, mu, temp_hw[mu]);
//u_shift_color_vecs(P3[sig], P3mu, mu, 2, temp_hw[mu]);
for(i=0; i<nsrc; i++) {
QDP_V_eq_V(P5sig[i], P3[sig][i], QDP_all);
}
u_shift_color_vecs(P5sig, P3mu, mu, nsrc, P5sigtmp[mu]);
}
/* The above shift is not needed if mu is backwards */
side_link_forces(mu, sig, ThreeSt, xin, P3[sig], Pmu, P3mu, nsrc);
}
/* Finally the OneLink and the Naik term */
if(GOES_BACKWARDS(mu)) {
/* Do only the forward terms in the Dslash */
/* Because I have shifted with OPP_DIR(mu) Pmu is a forward *
* shift. */
/* The one link */
add_forces_to_mom(Pmu, xin, OPP_DIR(mu), OneLink, nsrc);
/* For the same reason Pmumu is the forward double link */
/* Popmu is a backward shift */
//Popmu = hw_qdp[mu]; /* OK to clobber P3mu */
//u_shift_hw_fermion(xin, Popmu, mu, temp_hw[mu]);
u_shift_color_vecs(xin, Popmu, mu, nsrc, xintmp[mu]);
/* The Naik */
/* link no 1: - */
add_forces_to_mom(Pmumu, Popmu, OPP_DIR(mu), mNaik, nsrc);
/* Pmumumu can overwrite Popmu which is no longer needed */
//Pmumumu = hw_qdp[OPP_DIR(mu)];
//u_shift_hw_fermion(Pmumu, Pmumumu, OPP_DIR(mu), temp_hw[OPP_DIR(mu)]);
u_shift_color_vecs(Pmumu, Pmumumu, OPP_DIR(mu), nsrc, Pmumutmp[OPP_DIR(mu)]);
/* link no 0: + */
add_forces_to_mom(Pmumumu, xin, OPP_DIR(mu), Naik, nsrc);
} else {
/* The rest of the Naik terms */
//Popmu = hw_qdp[mu]; /* OK to clobber P3mu */
//u_shift_hw_fermion(xin, Popmu, mu, temp_hw[mu]);
u_shift_color_vecs(xin, Popmu, mu, nsrc, xintmp[mu]);
/* link no 2: + */
/* Pmumu is double backward shift */
add_forces_to_mom(Popmu, Pmumu, mu, Naik, nsrc);
}
/* Here we have to do together the Naik term and the one link term */
}/* mu */
QOP_trace("test 6\n");
QOP_trace("test 7\n");
for(mu=0; mu<4; mu++) {
QDP_M_eq_M(tmpmat, tempmom_qdp[mu], QDP_even);
QDP_M_eqm_M(tmpmat, tempmom_qdp[mu], QDP_odd);
QDP_M_eq_antiherm_M(tempmom_qdp[mu], tmpmat, QDP_all);
}
QDP_destroy_M(tmpmat);
//printf("%p\n", P5tmp[0][4][0]); fflush(stdout);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
/* Free temporary vectors */
for(i=0; i<nsrc; i++) {
QDP_destroy_V(Pmu[i]);
QDP_destroy_V(Pnumu[i]);
QDP_destroy_V(Prhonumu[i]);
QDP_destroy_V(P7[i]);
QDP_destroy_V(P7rho[i]);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
for(dir=0; dir<8; dir++) {
QDP_destroy_V(xintmp[dir][i]);
QDP_destroy_V(Pmutmp[dir][i]);
QDP_destroy_V(Pnumutmp[dir][i]);
QDP_destroy_V(Prhonumutmp[dir][i]);
QDP_destroy_V(P7tmp[dir][i]);
}
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
for(mu=0; mu<4; mu++) {
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
QDP_destroy_V(P5s[mu][i]);
//QDP_destroy_V(P5[mu][i]);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
for(dir=0; dir<8; dir++) {
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
QDP_destroy_V(P5tmps[mu][dir][i]);
//printf("%p\n", P5tmp[mu][dir][i]); fflush(stdout);
//QDP_destroy_V(P5tmp[mu][dir][i]);
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
}
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
}
//if(QDP_this_node==0) { printf("line %i\n",__LINE__); fflush(stdout); }
}
//if(QDP_this_node==0) { printf("here3\n"); fflush(stdout); }
for(mu=0; mu<8; mu++) {
for(i=0; i<nsrc; i++) {
QDP_destroy_V(P3[mu][i]);
}
//QDP_destroy_V(P5[mu][0]);
//QDP_destroy_V(P5[mu][1]);
}
for(i=0; i<nsrc; i++) {
QDP_destroy_V(tv[i]);
}
//if(QDP_this_node==0) { printf("here4\n"); fflush(stdout); }
for(i=4; i<8; i++) {
QDP_destroy_M(fblink[i]);
}
dtime += QOP_time();
info->final_sec = dtime;
info->final_flop = nflop*QDP_sites_on_node;
info->status = QOP_SUCCESS;
ASQTAD_FORCE_END;
#undef NC
}
/* 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
