double
bench_action(QOP_gauge_coeffs_t *coeffs, QOP_Force *out)
{
double sec=0, flop=0, mf=0;
QLA_Real acts, actt;
QOP_info_t info = QOP_INFO_ZERO;
for(int i=0; i<=nit; i++) {
QOP_symanzik_1loop_gauge_action(&info, gauge, &acts, &actt, coeffs);
if(i>0) {
sec += info.final_sec;
flop += info.final_flop;
mf += info.final_flop/(1e6*info.final_sec);
}
}
#if 1
printf0("action s: %g t: %g tot: %g\n", acts, actt, acts+actt);
coeffs->plaquette /= 4;
coeffs->rectangle /= 6;
coeffs->parallelogram /= 6;
coeffs->adjoint_plaquette /= 8;
QLA_Real eps=1;
QOP_verbose(QOP_VERB_DEBUG);
QOP_symanzik_1loop_gauge_force(&info, gauge, out, coeffs, eps);
QOP_verbose(QOP_VERB_OFF);
coeffs->plaquette *= 4;
coeffs->rectangle *= 6;
coeffs->parallelogram *= 6;
coeffs->adjoint_plaquette *= 8;
#endif
secs = sec/nit;
flops = flop/nit;
return mf/nit;
}
static int
ks_congrad_qop_generic( QOP_FermionLinksAsqtad* qop_links,
QOP_invert_arg_t *qop_invert_arg,
QOP_resid_arg_t ***qop_resid_arg,
MASSREAL *masses[], int nmass[],
QOP_ColorVector **qop_sol[],
QOP_ColorVector* qop_src[],
int nsrc, quark_invert_control *qic )
{
int isrc, imass;
int iters;
QOP_info_t info = QOP_INFO_ZERO;
char myname[] = "ks_congrad_qop_generic";
#ifdef AB_DEBUG_ENTRY_EXIT_ROUTINES
printf("Enter ks_congrad_qop_generic in d_congrad5_fn_qop_P.c\n");
#endif
#ifndef OLD_QOPQDP_NORM
/* Since version 0.9.0 the conventions for QOP_asqtad_invert* do not
give results consistent with other calls to ks_congrad and
ks_multicg... when the parity is EVENODD. You should change the
call and use mat_invert_uml instead. */
if(qop_invert_arg->evenodd == QOP_EVENODD){
node0_printf("%s: EVENANDODD is currently not supported.\n",myname);
terminate(1);
}
#endif
#ifdef CG_DEBUG
int save_qop_verb;
save_qop_verb = QOP_verbose(QOP_VERB_HI);
#endif
if(nsrc == 1 && nmass[0] == 1)
QOP_asqtad_invert( &info, qop_links, qop_invert_arg, qop_resid_arg[0][0],
masses[0][0], qop_sol[0][0], qop_src[0] );
else
QOP_asqtad_invert_multi( &info, qop_links, qop_invert_arg, qop_resid_arg,
masses, nmass, qop_sol, qop_src, nsrc );
#ifdef CG_DEBUG
QOP_verbose(save_qop_verb);
#endif
/* For now we return the largest value of the residual and iterations */
qic[0].final_rsq = 0;
qic[0].final_relrsq = 0;
qic[0].converged = 1;
iters = 0;
for(isrc = 0; isrc < nsrc; isrc++){
for(imass = 0; imass < nmass[isrc]; imass++){
if(this_node == 0){
if((qic[0].resid > 0 && qop_resid_arg[isrc][imass]->final_rsq <= qic[0].resid * qic[0].resid ) ||
(qic[0].relresid > 0 && qop_resid_arg[isrc][imass]->final_rel <= qic[0].relresid * qic[0].relresid )){
#ifdef CG_DEBUG
node0_printf(" OK converged (src %d, mass %d) ", isrc, imass);
node0_printf("final_rsq = %.2g rel = %.2g restarts = %d iters = %d\n",
qop_resid_arg[isrc][imass]->final_rsq,
qop_resid_arg[isrc][imass]->final_rel,
qop_resid_arg[isrc][imass]->final_restart,
qop_resid_arg[isrc][imass]->final_iter);
#endif
} else {
qic[0].converged = 0;
node0_printf(" NOT converged (src %d, mass %d) ", isrc, imass);
node0_printf("final_rsq = %.2g (cf %.2g) rel = %.2g (cf %.2g) restarts = %d iters = %d\n",
qop_resid_arg[isrc][imass]->final_rsq,
qic[0].resid * qic[0].resid,
qop_resid_arg[isrc][imass]->final_rel,
qic[0].relresid * qic[0].relresid,
qop_resid_arg[isrc][imass]->final_restart,
qop_resid_arg[isrc][imass]->final_iter);
}
}
if(qic[0].final_rsq < qop_resid_arg[isrc][imass]->final_rsq)
qic[0].final_rsq = qop_resid_arg[isrc][imass]->final_rsq;
if(qic[0].final_relrsq < qop_resid_arg[isrc][imass]->final_rel)
qic[0].final_relrsq = qop_resid_arg[isrc][imass]->final_rel;
#ifdef CG_DEBUG
if(nsrc > 1 || nmass[isrc] > 1)
node0_printf("CONGRAD5(src %d, mass %d): iters = %d resid = %.2g rel = %.2g\n",
isrc, imass,
qop_resid_arg[isrc][imass]->final_iter,
qop_resid_arg[isrc][imass]->final_rsq,
qop_resid_arg[isrc][imass]->final_rel);
#endif
}
if(qop_resid_arg[isrc][0]->final_iter > iters)
iters = qop_resid_arg[isrc][0]->final_iter;
}
// This structure isn't very friendly to multimass or multisource use
// So report restarts for the first mass in the list
qic[0].final_iters = iters;
qic[0].final_restart = qop_resid_arg[0][0]->final_restart;
#ifdef CGTIME
node0_printf("CONGRAD5: time = %e (fn_qop %s) ",
info.final_sec,qop_prec[QOP_PrecisionInt-1]);
for(isrc = 0; isrc < nsrc; isrc++)
node0_printf("nmass[%d] = %d iters = %d ",
isrc,nmass[isrc],qop_resid_arg[isrc][0]->final_iter);
if(info.final_sec == 0.)
node0_printf("mflops = 0.0\n");
else
node0_printf("mflops = %e\n", info.final_flop/(1.0e6*info.final_sec) );
fflush(stdout);
#endif
#ifdef AB_DEBUG_ENTRY_EXIT_ROUTINES
printf("Exit ks_congrad_qop_generic in d_congrad5_fn_qop_P.c\n");
#endif
fflush(stdout);
return iters;
}