int main(int argc, char *argv[])
{
int prompt;
int i, j, iq0, iq1;
#ifdef CLOV_LEAN
int oldiq0, oldiq1, oldip0;
#endif
double starttime, endtime;
#ifdef PRTIME
double dtime;
#endif
wilson_prop_field *prop[MAX_PROP];
wilson_prop_field *quark[MAX_QK];
initialize_machine(&argc,&argv);
/* Remap standard I/O */
if(remap_stdio_from_args(argc, argv) == 1)terminate(1);
g_sync();
starttime=dclock();
/* set up */
STARTTIME;
prompt = setup();
ENDTIME("setup");
/* loop over input sets */
while( readin(prompt) == 0){
if(prompt == 2)continue;
total_iters=0;
#ifdef HISQ_SVD_COUNTER
hisq_svd_counter = 0;
#endif
/**************************************************************/
/* Set up gauge field */
if( param.fixflag == COULOMB_GAUGE_FIX)
{
if(this_node == 0)
printf("Fixing to Coulomb gauge\n");
STARTTIME;
gaugefix(TUP,(Real)1.5,500,GAUGE_FIX_TOL);
ENDTIME("gauge fix");
/* (Re)construct APE smeared links after gauge fixing.
No KS phases here! */
destroy_ape_links_3D(ape_links);
ape_links = ape_smear_3D( param.staple_weight, param.ape_iter );
invalidate_this_clov(gen_clov);
}
else
if(this_node == 0)printf("COULOMB GAUGE FIXING SKIPPED.\n");
/* save lattice if requested */
if( param.saveflag != FORGET ){
savelat_p = save_lattice( param.saveflag, param.savefile,
param.stringLFN );
} else {
savelat_p = NULL;
}
if(this_node==0)printf("END OF HEADER\n");
/**************************************************************/
/* Loop over the propagators */
STARTTIME;
for(i=0; i<param.num_prop; i++){
node0_printf("******* Creating propagator %d ********\n",i);fflush(stdout);
/**************************************************************/
/* Read and/or generate quark propagator */
if(param.prop_type[i] == CLOVER_TYPE)
{
int ncolor = convert_ksource_to_color(param.src_qs[i].nsource);
prop[i] = create_wp_field(ncolor);
node0_printf("Generate Dirac propagator\n");
node0_printf("Kappa= %g source %s residue= %g rel= %g\n",
(double)param.dcp[i].Kappa,
param.src_qs[i].descrp,
(double)param.qic[i].resid,
(double)param.qic[i].relresid);
/* For clover_info */
wqstmp = param.src_qs[i];
dcptmp = param.dcp[i];
total_iters += get_wprop_to_wp_field(param.prop_type[i],
param.startflag_w[i],
param.startfile_w[i],
param.saveflag_w[i],
param.savefile_w[i],
prop[i],
¶m.src_qs[i],
¶m.qic[i],
(void *)¶m.dcp[i],
param.bdry_phase[i],
param.coord_origin,
param.check[i]);
#ifdef CLOV_LEAN
/* Free clover prop memory if we have saved the prop to disk */
if(param.saveflag_w[i] != FORGET){
free_wp_field(prop[i]);
clear_qs(¶m.src_qs[i]);
node0_printf("destroy prop[%d]\n",i);
}
#endif
}
/* ------------------------------------------- */
else if(param.prop_type[i] == IFLA_TYPE)
{
int ncolor = convert_ksource_to_color(param.src_qs[i].nsource);
prop[i] = create_wp_field(ncolor);
node0_printf("Generate Dirac IFLA propagator\n");
if(this_node==0)printf("Kappa= %g source %s residue= %g rel= %g\n",
(double)param.nap[i].kapifla,
param.src_qs[i].descrp,
(double)param.qic[i].resid,
(double)param.qic[i].relresid);
/* For clover_info */
wqstmp = param.src_qs[i];
naptmp = param.nap[i];
total_iters += get_wprop_to_wp_field(param.prop_type[i],
param.startflag_w[i],
param.startfile_w[i],
param.saveflag_w[i],
param.savefile_w[i],
prop[i],
¶m.src_qs[i],
¶m.qic[i],
(void *)¶m.nap[i],
param.bdry_phase[i],
param.coord_origin,
param.check[i]);
#ifdef CLOV_LEAN
/* Free clover prop memory if we have saved the prop to disk */
if(param.saveflag_w[i] != FORGET){
free_wp_field(prop[i]);
clear_qs(¶m.src_qs[i]);
node0_printf("destroy prop[%d]\n",i);
}
#endif
}
else if(param.prop_type[i] == KS_TYPE || param.prop_type[i] == KS0_TYPE ) /* KS_TYPE */
{
prop[i] = create_wp_field(param.src_qs[i].ncolor);
if(this_node==0)printf("Mass= %g source %s residue= %g rel= %g\n",
(double)param.ksp[i].mass,
param.src_qs[i].descrp,
(double)param.qic[i].resid,
(double)param.qic[i].relresid);
total_iters += get_ksprop_to_wp_field(param.startflag_ks[i],
param.startfile_ks[i],
param.saveflag_ks[i],
param.savefile_ks[i],
prop[i],
¶m.src_qs[i],
¶m.qic[i],
¶m.ksp[i],
param.bdry_phase[i],
param.coord_origin,
param.check[i],
param.prop_type[i] == KS0_TYPE);
#ifdef CLOV_LEAN
/* (We don't free naive prop memory, since we don't save it
as a clover prop in get_ksprop_to_wp_field) */
#endif
}
else /* KS4_TYPE */
{
prop[i] = create_wp_field(param.src_qs[i].ncolor);
if(this_node==0)printf("Mass= %g source %s residue= %g rel= %g\n",
(double)param.ksp[i].mass,
param.src_qs[i].descrp,
(double)param.qic[i].resid,
(double)param.qic[i].relresid);
total_iters += get_ksprop4_to_wp_field(param.startflag_w[i],
param.startfile_w[i],
param.saveflag_w[i],
param.savefile_w[i],
prop[i],
¶m.src_qs[i],
¶m.qic[i],
¶m.ksp[i],
param.bdry_phase[i],
param.coord_origin,
param.check[i]);
}
} /* propagators */
ENDTIME("compute propagators");
/*****************************************************************/
/* Complete the quark propagators by applying the sink operators
to either the raw propagator or by building on an existing quark
propagator */
STARTTIME;
#ifdef CLOV_LEAN
oldip0 = -1;
oldiq0 = -1;
oldiq1 = -1;
#endif
for(j=0; j<param.num_qk; j++){
node0_printf("******* Creating quark %d ********\n",j); fflush(stdout);
i = param.prop_for_qk[j];
if(param.parent_type[j] == PROP_TYPE){
#ifdef CLOV_LEAN
/* Restore clover prop[i] from file. */
/* But first destroy the old one, unless we still need it */
if(oldip0 >= 0 && oldip0 != i)
if(param.prop_type[oldip0] == CLOVER_TYPE &&
param.saveflag_w[oldip0] != FORGET){
free_wp_field(prop[oldip0]);
node0_printf("destroy prop[%d]\n",oldip0);
}
/* In this case we won't need any old quarks */
if(oldiq0 >= 0)
if(param.saveflag_q[oldiq0] != FORGET){
free_wp_field(quark[oldiq0]);
node0_printf("destroy quark[%d]\n",oldiq0);
}
if(oldiq1 >= 0)
if(param.saveflag_q[oldiq1] != FORGET){
free_wp_field(quark[oldiq1]);
node0_printf("destroy quark[%d]\n",oldiq1);
}
if(prop[i]->swv[0] == NULL)
reread_wprop_to_wp_field(param.saveflag_w[i], param.savefile_w[i], prop[i]);
#endif
/* Before applying operator, apply momentum twist to
ape_links. Use the momentum twist of the parent
propagator */
momentum_twist_ape_links(i, +1);
/* Apply sink operator quark[j] <- Op[j] prop[i] */
quark[j] = create_wp_field_copy(prop[i]);
wp_sink_op(¶m.snk_qs_op[j], quark[j]);
/* Remove twist */
momentum_twist_ape_links(i, -1);
#ifdef CLOV_LEAN
oldip0 = i;
oldiq0 = -1;
#endif
}
else if(param.parent_type[j] == QUARK_TYPE) { /* QUARK_TYPE */
#ifdef CLOV_LEAN
/* Restore quark[i] from file */
/* But first destroy the old ones, unless we still need one of them */
/* In this case we won't need the old prop */
if(oldip0 >= 0)
if(param.prop_type[oldip0] == CLOVER_TYPE &&
param.saveflag_w[oldip0] != FORGET){
free_wp_field(prop[oldip0]);
node0_printf("destroy prop[%d]\n",oldip0);
}
if(oldiq0 >= 0 && oldiq0 != i)
if(param.saveflag_q[oldiq0] != FORGET){
free_wp_field(quark[oldiq0]);
node0_printf("destroy quark[%d]\n",oldiq0);
}
if(oldiq1 >= 0 && oldiq1 != i)
if(param.saveflag_q[oldiq1] != FORGET){
free_wp_field(quark[oldiq1]);
node0_printf("destroy quark[%d]\n",oldiq1);
}
if(quark[i]->swv[0] == NULL)
reread_wprop_to_wp_field(param.saveflag_q[i], param.savefile_q[i], quark[i]);
#endif
/* Apply sink operator quark[j] <- Op[j] quark[i] */
momentum_twist_ape_links(i, +1);
quark[j] = create_wp_field_copy(quark[i]);
wp_sink_op(¶m.snk_qs_op[j], quark[j]);
momentum_twist_ape_links(i, -1);
#ifdef CLOV_LEAN
oldip0 = -1;
oldiq0 = i;
#endif
} else { /* COMBO_TYPE */
int k;
int nc = quark[param.combo_qk_index[j][0]]->nc;
/* Create a zero field */
quark[j] = create_wp_field(nc);
/* Compute the requested linear combination */
for(k = 0; k < param.num_combo[j]; k++){
wilson_prop_field *q = quark[param.combo_qk_index[j][k]];
if(nc != q->nc){
printf("Error: Attempting to combine an inconsistent number of colors: %d != %d\n",nc, q->nc);
terminate(1);
}
scalar_mult_add_wprop_field(quark[j], q, param.combo_coeff[j][k], quark[j]);
}
}
/* Save the resulting quark[j] if requested */
dump_wprop_from_wp_field( param.saveflag_q[j], param.savetype_q[j],
param.savefile_q[j], quark[j]);
/* Can we delete any props and quarks now? */
/* If nothing later depends on a prop or quark, free it up. */
for(i = 0; i < param.num_prop; i++)
if( prop[i]->swv[0] != NULL && param.prop_dep_qkno[i] < j ){
free_wp_field(prop[i]);
node0_printf("free prop[%d]\n",i);
}
for(i = 0; i < j; i++)
if( quark[i]->swv[0] != NULL && param.quark_dep_qkno[i] < j ){
free_wp_field(quark[i]);
node0_printf("free quark[%d]\n",i);
}
#ifdef CLOV_LEAN
oldiq1 = j;
#endif
}
#ifdef CLOV_LEAN
/* Free remaining memory */
if(oldip0 >= 0)
if(param.prop_type[oldip0] == CLOVER_TYPE &&
param.saveflag_w[oldip0] != FORGET){
free_wp_field(prop[oldip0]);
node0_printf("destroy prop[%d]\n",oldip0);
}
if(oldiq0 >= 0)
if(param.saveflag_q[oldiq0] != FORGET){
free_wp_field(quark[oldiq0]);
node0_printf("destroy quark[%d]\n",oldiq0);
}
if(oldiq1 >= 0)
if(param.saveflag_q[oldiq1] != FORGET){
free_wp_field(quark[oldiq1]);
node0_printf("destroy quark[%d]\n",oldiq1);
}
#endif
/* Now destroy all remaining propagator fields */
for(i = 0; i < param.num_prop; i++){
if(prop[i] != NULL)node0_printf("destroy prop[%d]\n",i);
destroy_wp_field(prop[i]);
prop[i] = NULL;
}
ENDTIME("generate quarks");
/****************************************************************/
/* Compute the meson propagators */
STARTTIME;
for(i = 0; i < param.num_pair; i++){
/* Index for the quarks making up this meson */
iq0 = param.qkpair[i][0];
iq1 = param.qkpair[i][1];
node0_printf("Mesons for quarks %d and %d\n",iq0,iq1);
#ifdef CLOV_LEAN
/* Restore quarks from file and free old memory */
/* We try to reuse props that are already in memory, so we don't
destroy them immediately, but wait to see if we need
them again for the next pair. */
if(i > 0 && oldiq0 != iq0 && oldiq0 != iq1)
if(param.saveflag_q[oldiq0] != FORGET){
free_wp_field(quark[oldiq0]);
node0_printf("destroy quark[%d]\n",oldiq0);
}
if(i > 0 && oldiq1 != iq0 && oldiq1 != iq1)
if(param.saveflag_q[oldiq1] != FORGET){
free_wp_field(quark[oldiq1]);
node0_printf("destroy quark[%d]\n",oldiq1);
}
if(quark[iq0]->swv[0] == NULL)
reread_wprop_to_wp_field(param.saveflag_q[iq0], param.savefile_q[iq0], quark[iq0]);
if(quark[iq1]->swv[0] == NULL){
reread_wprop_to_wp_field(param.saveflag_q[iq1], param.savefile_q[iq1], quark[iq1]);
}
#endif
/* Tie together to generate hadron spectrum */
spectrum_cl(quark[iq0], quark[iq1], i);
/* Remember, in case we need to free memory */
#ifdef CLOV_LEAN
oldiq0 = iq0;
oldiq1 = iq1;
#endif
}
#ifdef CLOV_LEAN
/* Free any remaining quark prop memory */
if(quark[oldiq0]->swv[0] != NULL)
if(param.saveflag_q[oldiq0] != FORGET){
free_wp_field(quark[oldiq0]);
node0_printf("destroy quark[%d]\n",oldiq0);
}
if(quark[oldiq1]->swv[0] != NULL)
if(param.saveflag_q[oldiq1] != FORGET){
free_wp_field(quark[oldiq1]);
node0_printf("destroy quark[%d]\n",oldiq1);
}
#endif
ENDTIME("tie hadron correlators");
node0_printf("RUNNING COMPLETED\n");
endtime=dclock();
node0_printf("Time = %e seconds\n",(double)(endtime-starttime));
node0_printf("total_iters = %d\n",total_iters);
#ifdef HISQ_SVD_COUNTER
printf("hisq_svd_counter = %d\n",hisq_svd_counter);
#endif
fflush(stdout);
for(i = 0; i < param.num_qk; i++){
if(quark[i] != NULL)node0_printf("destroy quark[%d]\n",i);
destroy_wp_field(quark[i]);
quark[i] = NULL;
}
destroy_ape_links_3D(ape_links);
/* Destroy fermion links (possibly created in make_prop()) */
#if FERM_ACTION == HISQ
destroy_fermion_links_hisq(fn_links);
#else
destroy_fermion_links(fn_links);
#endif
fn_links = NULL;
} /* readin(prompt) */
#ifdef HAVE_QUDA
qudaFinalize();
#endif
normal_exit(0);
return 0;
}
int main(int argc, char *argv[])
{
int meascount;
int prompt;
Real avm_iters,avs_iters;
double starttime,endtime;
#ifdef IOTIME
double dtime;
int iotime = 1;
#else
int iotime = 0;
#endif
int MaxCG;
Real RsdCG, RRsdCG;
int spin,color,k;
int flag;
int status;
int cl_cg = CL_CG;
w_prop_file *fp_in_w[MAX_KAP]; /* For propagator files */
w_prop_file *fp_out_w[MAX_KAP]; /* For propagator files */
wilson_vector *psi = NULL;
wilson_prop_field quark_propagator = NULL;
initialize_machine(&argc,&argv);
#ifdef HAVE_QDP
QDP_initialize(&argc, &argv);
#endif
/* Remap standard I/O */
if(remap_stdio_from_args(argc, argv) == 1)terminate(1);
g_sync();
/* set up */
prompt = setup_cl();
/* loop over input sets */
psi = create_wv_field();
quark_propagator = create_wp_field();
while( readin(prompt) == 0)
{
starttime=dclock();
MaxCG=niter;
wqstmp = wqs; /* For clover_info.c */
avm_iters=0.0;
meascount=0;
if( fixflag == COULOMB_GAUGE_FIX)
{
if(this_node == 0)
printf("Fixing to Coulomb gauge\n");
#ifdef IOTIME
dtime = -dclock();
#endif
gaugefix(TUP,(Real)1.5,500,GAUGE_FIX_TOL);
#ifdef IOTIME
dtime += dclock();
if(this_node==0)printf("Time to gauge fix = %e\n",dtime);
#endif
invalidate_this_clov(gen_clov);
}
else
if(this_node == 0)printf("COULOMB GAUGE FIXING SKIPPED.\n");
/* save lattice if requested */
if( saveflag != FORGET ){
savelat_p = save_lattice( saveflag, savefile, stringLFN );
}
if(this_node==0)printf("END OF HEADER\n");
/* if(this_node==0) printf("num_kap = %d\n", num_kap); */
/* Loop over kappas */
for(k=0;k<num_kap;k++){
kappa=kap[k];
RsdCG=resid[k];
RRsdCG=relresid[k];
if(this_node==0)printf("Kappa= %g r0= %g residue= %g rel= %g\n",
(double)kappa,(double)wqs.r0,(double)RsdCG,
(double)RRsdCG);
/* open files for wilson propagators */
#ifdef IOTIME
dtime = -dclock();
#endif
fp_in_w[k] = r_open_wprop(startflag_w[k], startfile_w[k]);
#ifdef IOTIME
dtime += dclock();
if(startflag_w[k] != FRESH)
node0_printf("Time to open prop = %e\n",dtime);
#endif
fp_out_w[k] = w_open_wprop(saveflag_w[k], savefile_w[k],
wqs.type);
/* Loop over source spins */
for(spin=0;spin<4;spin++){
/* Loop over source colors */
for(color=0;color<3;color++){
meascount ++;
/*if(this_node==0)printf("color=%d spin=%d\n",color,spin); */
if(startflag_w[k] == CONTINUE)
{
node0_printf("Can not continue propagator here! Zeroing it instead\n");
startflag_w[k] = FRESH;
}
/* Saves one multiplication by zero in cgilu */
if(startflag_w[k] == FRESH)flag = 0;
else
flag = 1;
/* load psi if requested */
status = reload_wprop_sc_to_field( startflag_w[k], fp_in_w[k],
&wqs, spin, color, psi, iotime);
if(status != 0)
{
node0_printf("control_cl: Recovering from error by resetting initial guess to zero\n");
reload_wprop_sc_to_field( FRESH, fp_in_w[k], &wqs,
spin, color, psi, 0);
flag = 0;
}
/* Complete the source structure */
wqs.color = color;
wqs.spin = spin;
/* Load inversion control structure */
qic.prec = PRECISION;
qic.max = MaxCG;
qic.nrestart = nrestart;
qic.resid = RsdCG;
qic.relresid = RRsdCG;
qic.start_flag = flag;
/* Load Dirac matrix parameters */
dcp.Kappa = kappa;
dcp.Clov_c = clov_c;
dcp.U0 = u0;
/* compute the propagator. Result in psi. */
switch (cl_cg) {
case BICG:
avs_iters =
(Real)wilson_invert_field_wqs(&wqs, w_source_field, psi,
bicgilu_cl_field,
&qic,(void *)&dcp);
break;
case HOP:
avs_iters =
(Real)wilson_invert_field_wqs(&wqs, w_source_field, psi,
hopilu_cl_field,
&qic,(void *)&dcp);
break;
case MR:
avs_iters =
(Real)wilson_invert_field_wqs(&wqs, w_source_field, psi,
mrilu_cl_field,
&qic,(void *)&dcp);
break;
case CG:
avs_iters =
(Real)wilson_invert_field_wqs(&wqs, w_source_field, psi,
cgilu_cl_field,
&qic,(void *)&dcp);
break;
default:
node0_printf("main(%d): Inverter choice %d not supported\n",
this_node,cl_cg);
}
avm_iters += avs_iters;
copy_wp_from_wv(quark_propagator, psi, color, spin);
/* save psi if requested */
save_wprop_sc_from_field( saveflag_w[k],fp_out_w[k], &wqs,
spin,color,psi,"Fill in record info here",iotime);
} /* source spins */
} /* source colors */
/* close files for wilson propagators */
r_close_wprop(startflag_w[k],fp_in_w[k]);
#ifdef IOTIME
dtime = -dclock();
#endif
w_close_wprop(saveflag_w[k],fp_out_w[k]);
#ifdef IOTIME
dtime += dclock();
if(saveflag_w[k] != FORGET)
node0_printf("Time to close prop = %e\n",dtime);
#endif
/* spectrum, if requested */
if(strstr(spectrum_request,",spectrum,") != NULL)
spectrum_cl(quark_propagator, wqs.t0, k);
} /* kappas */
if(this_node==0)printf("RUNNING COMPLETED\n");
if(meascount>0){
if(this_node==0)printf("total cg iters for measurement= %e\n",
(double)avm_iters);
if(this_node==0)printf("cg iters for measurement= %e\n",
(double)avm_iters/(double)meascount);
}
endtime=dclock();
if(this_node==0){
printf("Time = %e seconds\n",(double)(endtime-starttime));
printf("total_iters = %d\n",total_iters);
}
fflush(stdout);
}
destroy_wv_field(psi);
destroy_wp_field(quark_propagator);
return 0;
}
