int mat_invert_uml(field_offset src, field_offset dest, field_offset temp,
double mass ){
int cgn;
double finalrsq;
register int i;
register site *s;
if( src==temp ){
printf("BOTCH\n"); exit(0);
}
/* multiply by U - even sites only */
dslash( src, F_OFFSET(ttt), EVEN);
scalar_mult_add_latvec( F_OFFSET(ttt), src,
-2.0*mass, temp, EVEN);
scalar_mult_latvec( temp, -1.0, temp, EVEN);
/* invert with M_adj M even */
cgn = ks_congrad( temp, dest, mass, niter, rsqprop,
EVEN, &finalrsq );
/* multiply by (1/2m)L, does nothing to even sites */
/* fix up odd sites , 1/2m (Dslash_oe*dest_e + phi_odd) */
dslash( dest, F_OFFSET(ttt), ODD );
FORODDSITES(i,s){
sub_su3_vector( (su3_vector *)F_PT(s,src), &(s->ttt),
(su3_vector *)F_PT(s,dest) );
scalar_mult_su3_vector( (su3_vector *)F_PT(s,dest), 1.0/(2.0*mass),
(su3_vector *)F_PT(s,dest) );
}
void Matrix_Vec_mult(su3_vector *src, su3_vector *res, int parity,
imp_ferm_links_t *fn )
{
register site *s;
register int i;
int otherparity = EVENANDODD;
if(temp == NULL ){
temp = (su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
}
switch(parity){
case EVEN:
otherparity = ODD ;
break ;
case ODD:
otherparity = EVEN ;
break ;
case EVENANDODD:
otherparity = EVENANDODD ;
break ;
default:
node0_printf("ERROR: wrong parity in eigen_stuff::Matrix_Vec_mult\n") ;
terminate(1) ;
}
dslash_field(src , temp, otherparity, fn) ;
dslash_field(temp, res , parity , fn) ;
FORSOMEPARITY(i,s,parity){
scalar_mult_su3_vector( &(res[i]), -1.0, &(res[i])) ;
}
static
void qop_to_milc_normalization_field(su3_vector *sol, MASSREAL mass,
int parity)
{
site *s;
int i;
Real x = 1/(NORMFACT(mass));
FORSOMEPARITY(i,s,parity){
scalar_mult_su3_vector(sol+i, x, sol+i);
}
void scalar_mult_wvec( wilson_vector *src, Real s, wilson_vector *dest){
#ifndef NATIVEDOUBLE
register int i;
for(i=0;i<4;i++)scalar_mult_su3_vector( &(src->d[i]), s, &(dest->d[i]));
#else /* RS6000 version */
register double ss;
ss = s;
dest->d[0].c[0].real = ss*src->d[0].c[0].real;
dest->d[0].c[0].imag = ss*src->d[0].c[0].imag;
dest->d[0].c[1].real = ss*src->d[0].c[1].real;
dest->d[0].c[1].imag = ss*src->d[0].c[1].imag;
dest->d[0].c[2].real = ss*src->d[0].c[2].real;
dest->d[0].c[2].imag = ss*src->d[0].c[2].imag;
dest->d[1].c[0].real = ss*src->d[1].c[0].real;
dest->d[1].c[0].imag = ss*src->d[1].c[0].imag;
dest->d[1].c[1].real = ss*src->d[1].c[1].real;
dest->d[1].c[1].imag = ss*src->d[1].c[1].imag;
dest->d[1].c[2].real = ss*src->d[1].c[2].real;
dest->d[1].c[2].imag = ss*src->d[1].c[2].imag;
dest->d[2].c[0].real = ss*src->d[2].c[0].real;
dest->d[2].c[0].imag = ss*src->d[2].c[0].imag;
dest->d[2].c[1].real = ss*src->d[2].c[1].real;
dest->d[2].c[1].imag = ss*src->d[2].c[1].imag;
dest->d[2].c[2].real = ss*src->d[2].c[2].real;
dest->d[2].c[2].imag = ss*src->d[2].c[2].imag;
dest->d[3].c[0].real = ss*src->d[3].c[0].real;
dest->d[3].c[0].imag = ss*src->d[3].c[0].imag;
dest->d[3].c[1].real = ss*src->d[3].c[1].real;
dest->d[3].c[1].imag = ss*src->d[3].c[1].imag;
dest->d[3].c[2].real = ss*src->d[3].c[2].real;
dest->d[3].c[2].imag = ss*src->d[3].c[2].imag;
#endif
}
int quark_renorm( void ) {
register int i, dir;
register site *s;
int j, cgn;
Real mass_x2, finalrsq;
Real pix, piy, piz, pit;
Real sin_pmu, q_mu, prop_a, prop_b, z_fac, m_func, ftmp = 0;
Real r1, r2, r3;
int pmu, px, py, pz, pt;
int pxn, pyn, pzn, ptn;
int currentnode;
int j1, jm2, k, dirs[4];
msg_tag *mtag[2];
int j_mass;
su3_vector **psim = NULL;
int xi, j2, j3, j4, parity;
int multiflag;
FILE *fp_mom_ks[MAX_NUM_MASS]; /* for writing mom propagator files */
char filename[50];
int prec = PRECISION; /* Make internal precision for CG the same as
the prevailing precision */
pix = 2.*PI / (Real)nx;
piy = 2.*PI / (Real)ny;
piz = 2.*PI / (Real)nz;
pit = 2.*PI / (Real)nt;
cgn = 0;
if( num_mass == 1){
multiflag = 0;
}
else{
multiflag = 1;
psim = (su3_vector **)malloc(num_mass*sizeof(su3_vector *));
for(j=0; j<num_mass; j++){
psim[j] = (su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
}
}
/* Open the momentum propagator files */
if(this_node == 0){
for(j=0; j<num_mass; j++){
sprintf(filename,"mom_pt_prop.m_%d",j);
fp_mom_ks[j] = fopen(filename, "ab");
if(fp_mom_ks[j] == NULL){
printf("quark_renorm: Node %d can't open file %s, error %d\n",
this_node,filename,errno);fflush(stdout);
terminate(1);
}
}
}
rephase( ON ); /* Turn staggered phases on */
/* Create fat and long links */
load_ferm_links(&fn_links, &ks_act_paths);
/* Loop over the 16 source points */
for(xi=0; xi<16; xi++){
/* Initialize color trace of the propagator */
FORALLSITES(i,s){
for(j=0; j<num_mass; j++){
s->trace_prop[j].real = 0.0;
s->trace_prop[j].imag = 0.0;
}
}
j1 = xi%2;
k = xi/2;
j2 = k%2;
k /= 2;
j3 = k%2;
k /= 2;
j4 = k%2;
parity = (j1+j2+j3+j4)%2;
/* dirs[XUP] = j1;
dirs[YUP] = j2;
dirs[ZUP] = j3;
dirs[TUP] = j4; */
/* Loop over colors of source vector */
for(j=0; j<3; j++){
/* initialize the source in phi */
FORALLSITES(i,s){
clearvec( &(s->phi));
}
/* Point source at site xi in the hypercube at origin */
if( node_number(j1,j2,j3,j4) == this_node ){
i=node_index(j1,j2,j3,j4);
lattice[i].phi.c[j].real = -1.0;
}
if( multiflag == 0){
for(j_mass=0; j_mass<num_mass; j_mass++){
FORALLSITES(i,s){
clearvec( &(s->xxx1));
}
if(parity == 0){
/* do a C.G. (source in phi, result in xxx1) */
cgn += ks_congrad( F_OFFSET(phi), F_OFFSET(xxx1),
mass[j_mass], niter, nrestart,
rsqprop, PRECISION,
EVEN, &finalrsq, &fn_links);
/* Multiply by -Madjoint */
dslash_site( F_OFFSET(xxx1), F_OFFSET(ttt), ODD,
&fn_links);
mass_x2 = 2.*mass[j_mass];
FOREVENSITES(i,s){
scalar_mult_su3_vector( &(s->xxx1), -mass_x2,
&(s->ttt));
}
}
else{
FORODDSITES(i,s){
scalar_mult_su3_vector( &(s->xxx1), -mass_x2,
&(s->ttt));
}
Until we have EO versions of dslash_field, we require FN
#endif
#include "generic_ks_includes.h" /* definitions files and prototypes */
#include "../include/dslash_ks_redefine.h" /* Actually not used, yet */
#include "../include/loopend.h"
/*#define CGTIME*/
int ks_congrad_two_src( /* Return value is number of iterations taken */
field_offset src1, /* source vector (type su3_vector) */
field_offset src2,
field_offset dest1, /* solution vectors */
field_offset dest2,
Real mass1,
Real mass2,
int niter, /* maximal number of CG interations */
int nrestart, /* maximal number of CG restarts */
Real rsqmin, /* desired residue squared */
int prec, /* internal precision for the inversion (ignored) */
int parity, /* parity to be worked on */
Real *final_rsq_ptr, /* final residue squared */
ferm_links_t *fn /* Storage for fermion links */
)
{
/* Site su3_vector's resid, cg_p and ttt are used as temporaries */
register int i;
register site *s;
int iteration; /* counter for iterations */
double c1, c2, rsq, oldrsq, pkp; /* pkp = cg_p.K.cg_p */
double source_norm,source_norm1; /* squared magnitude of source vector */
double rsqstop; /* stopping residual normalized by source norm */
int l_parity; /* parity we are currently doing */
int l_otherparity; /* the other parity */
msg_tag *tags1[16], *tags2[16]; /* tags for gathers to parity and opposite */
int special_started; /* 1 if dslash_special has been called */
int j, jud, jstrange;
double shift, msq_xm4;
double zeta_i[2], zeta_im1[2], zeta_ip1[2];
double beta_i[2], beta_im1[2], alpha[2];
int first;
su3_vector *temp;
su3_vector *destvec1;
su3_vector *destvec2;
su3_vector *pm_strange;
su3_vector *init_guess;
su3_vector *common_source;
su3_vector *ttt;
su3_vector *cg_p;
su3_vector *resid;
/* Timing */
#ifdef CGTIME
double dtimed,dtimec;
#endif
double nflop;
/* debug */
#ifdef CGTIME
dtimec = -dclock();
#endif
first = 0;
nflop = 1187; /* THIS LOOKS WRONG - DT */
if(parity==EVENANDODD)nflop *=2;
special_started = 0;
/* if we want both parities, we will do even first. */
switch(parity){
case(EVEN): l_parity=EVEN; l_otherparity=ODD; break;
case(ODD): l_parity=ODD; l_otherparity=EVEN; break;
case(EVENANDODD): l_parity=EVEN; l_otherparity=ODD; break;
}
jud = 0;
jstrange = 1;
temp=(su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
destvec1=(su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
destvec2=(su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
ttt=(su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
cg_p=(su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
resid=(su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
pm_strange = (su3_vector *)malloc(sites_on_node*sizeof(su3_vector));
init_guess = (su3_vector *)malloc(sites_on_node*sizeof(su3_vector) );
common_source = (su3_vector *)malloc(sites_on_node*sizeof(su3_vector) );
shift = 4.0*( mass2*mass2 - mass1*mass1);
msq_xm4 = -4.0*mass1*mass1;
iteration = 0;
#ifdef CGTIME
dtimec = -dclock();
#endif
/* initialization process */
start:
#ifdef FN
if(special_started==1) { /* clean up gathers */
cleanup_gathers(tags1, tags2);
special_started = 0;
}
#endif
/*This loop calculates init_guess = (phi2 - phi1)/shift = X0 , which
* is used to equalize the two sources*/
FORSOMEPARITY(i,s,l_parity){
sub_su3_vector( (su3_vector *)F_PT(s,src2),(su3_vector *)F_PT(s,src1),&temp[i] );
scalar_mult_su3_vector(&temp[i],1.0/shift,&init_guess[i] );
}END_LOOP
int main( int argc, char **argv ){
register site *s;
int i,si;
int prompt;
double dtime;
su3_vector **eigVec ;
su3_vector *tmp ;
double *eigVal ;
int total_R_iters ;
double chirality ;
initialize_machine(&argc,&argv);
#ifdef HAVE_QDP
QDP_initialize(&argc, &argv);
#ifndef QDP_PROFILE
QDP_profcontrol(0);
#endif
#endif
/* Remap standard I/O */
if(remap_stdio_from_args(argc, argv) == 1)terminate(1);
g_sync();
/* set up */
prompt = setup();
/* loop over input sets */
while( readin(prompt) == 0){
dtime = -dclock();
invalidate_all_ferm_links(&fn_links);
make_path_table(&ks_act_paths, &ks_act_paths_dmdu0);
/* Load fat and long links for fermion measurements if needed */
load_ferm_links(&fn_links, &ks_act_paths);
/* call fermion_variable measuring routines */
/* results are printed in output file */
f_meas_imp( F_OFFSET(phi), F_OFFSET(xxx), mass,
&fn_links, &fn_links_dmdu0);
eigVal = (double *)malloc(Nvecs*sizeof(double));
eigVec = (su3_vector **)malloc(Nvecs*sizeof(su3_vector*));
for(i=0;i<Nvecs;i++)
eigVec[i]=
(su3_vector*)malloc(sites_on_node*sizeof(su3_vector));
total_R_iters=Kalkreuter(eigVec, eigVal, eigenval_tol,
error_decr, Nvecs, MaxIter, Restart,
Kiters, EVEN, &fn_links) ;
tmp = (su3_vector*)malloc(sites_on_node*sizeof(su3_vector));
for(i=0;i<Nvecs;i++)
{
/* Construct to odd part of the vector. *
* Note that the true odd part of the eigenvector is *
* i/sqrt(eigVal) Dslash Psi. But since I only compute *
* the chirality the i factor is irrelevant (-i)*i=1!! */
dslash_fn_field(eigVec[i], tmp, ODD, &fn_links) ;
FORSOMEPARITY(si,s,ODD){
scalar_mult_su3_vector( &(tmp[si]),
1.0/sqrt(eigVal[i]),
&(eigVec[i][si]) ) ;
}
measure_chirality(eigVec[i], &chirality, EVENANDODD);
/* Here I divide by 2 since the EVEN vector is normalized to
* 1. The EVENANDODD vector is normalized to 2. I could have
* normalized the EVENANDODD vector to 1 and then not devide
* by to. The measure_chirality routine assumes vectors
* normalized to 1. */
node0_printf("Chirality(%i): %g\n",i,chirality/2) ;
}
free(tmp);
/**
for(i=0;i<Nvecs;i++)
{
sprintf(label,"DENSITY(%i)",i) ;
print_densities(eigVec[i], label, ny/2,nz/2,nt/2, EVEN) ;
}
**/
for(i=0;i<Nvecs;i++)
free(eigVec[i]) ;
free(eigVec) ;
free(eigVal) ;
#ifdef FN
invalidate_all_ferm_links(&fn_links);
#endif
fflush(stdout);
node0_printf("RUNNING COMPLETED\n"); fflush(stdout);
dtime += dclock();
if(this_node==0){
printf("Time = %e seconds\n",dtime);
printf("total_iters = %d\n",total_iters);
printf("total Rayleigh iters = %d\n",total_R_iters);
}
fflush(stdout);
}
int main( int argc, char **argv ){
register site *s;
int i,si;
int prompt;
double dtime;
su3_vector **eigVec ;
su3_vector *tmp ;
double *eigVal ;
int total_R_iters ;
double chirality, chir_ev, chir_od ;
imp_ferm_links_t **fn;
initialize_machine(&argc,&argv);
/* Remap standard I/O */
if(remap_stdio_from_args(argc, argv) == 1)terminate(1);
g_sync();
/* set up */
prompt = setup();
/* loop over input sets */
while( readin(prompt) == 0){
dtime = -dclock();
#ifdef HISQ_SVD_COUNTER
hisq_svd_counter = 0;
#endif
restore_fermion_links_from_site(fn_links, PRECISION);
/* call fermion_variable measuring routines */
/* results are printed in output file */
f_meas_imp( 1, PRECISION, F_OFFSET(phi), F_OFFSET(xxx), mass,
0, fn_links);
eigVal = (double *)malloc(Nvecs*sizeof(double));
eigVec = (su3_vector **)malloc(Nvecs*sizeof(su3_vector*));
for(i=0;i<Nvecs;i++)
eigVec[i]=
(su3_vector*)malloc(sites_on_node*sizeof(su3_vector));
fn = get_fm_links(fn_links);
active_parity = EVEN;
total_R_iters=Kalkreuter(eigVec, eigVal, eigenval_tol,
error_decr, Nvecs, MaxIter, Restart,
Kiters);
node0_printf("The above where eigenvalues of -Dslash^2 in MILC normalization\n");
node0_printf("Here we also list eigenvalues of iDslash in continuum normalization\n");
for(i=0;i<Nvecs;i++)
{
if ( eigVal[i] > 0.0 ){ chirality = sqrt(eigVal[i]) / 2.0; }
else { chirality = 0.0; }
node0_printf("eigenval(%i): %10g\n",i,chirality) ;
}
tmp = (su3_vector*)malloc(sites_on_node*sizeof(su3_vector));
for(i=0;i<Nvecs;i++)
{
// if ( eigVal[i] > 10.0*eigenval_tol )
// {
/* Construct to odd part of the vector. *
* Note that the true odd part of the eigenvector is *
* i/sqrt(eigVal) Dslash Psi. But since I only compute *
* the chirality the i factor is irrelevant (-i)*i=1!! */
dslash_field(eigVec[i], tmp, ODD, fn[0]) ;
FORSOMEPARITY(si,s,ODD){
scalar_mult_su3_vector( &(tmp[si]),
1.0/sqrt(eigVal[i]),
&(eigVec[i][si]) ) ;
}
// measure_chirality(eigVec[i], &chirality, EVENANDODD);
/* Here I divide by 2 since the EVEN vector is normalized to
* 1. The EVENANDODD vector is normalized to 2. I could have
* normalized the EVENANDODD vector to 1 and then not devide
* by to. The measure_chirality routine assumes vectors
* normalized to 1. */
// node0_printf("Chirality(%i): %g\n",i,chirality/2) ;
measure_chirality(eigVec[i], &chir_ev, EVEN);
measure_chirality(eigVec[i], &chir_od, ODD);
chirality = (chir_ev + chir_od) / 2.0;
node0_printf("Chirality(%i) -- even, odd, total: %10g, %10g, %10g\n",
i,chir_ev,chir_od,chirality) ;
// }
// else
// {
/* This is considered a "zero mode", and treated as such */
// measure_chirality(eigVec[i], &chirality, EVEN);
// node0_printf("Chirality(%i): %g\n",i,chirality) ;
// }
}
#ifdef EO
cleanup_dslash_temps();
#endif
free(tmp);
/**
for(i=0;i<Nvecs;i++)
{
sprintf(label,"DENSITY(%i)",i) ;
print_densities(eigVec[i], label, ny/2,nz/2,nt/2, EVEN) ;
}
**/
for(i=0;i<Nvecs;i++)
free(eigVec[i]) ;
free(eigVec) ;
free(eigVal) ;
invalidate_fermion_links(fn_links);
fflush(stdout);
node0_printf("RUNNING COMPLETED\n"); fflush(stdout);
dtime += dclock();
if(this_node==0){
printf("Time = %e seconds\n",dtime);
printf("total_iters = %d\n",total_iters);
printf("total Rayleigh iters = %d\n",total_R_iters);
#ifdef HISQ_SVD_COUNTER
printf("hisq_svd_counter = %d\n",hisq_svd_counter);
#endif
}
fflush(stdout);
}
/* Now devide by 2 eq. (4.2b) of Golderman's Meson paper*/
FORALLSITES(i,s)
{
scalar_mult_su3_vector( (su3_vector *)F_PT(s,dest), .5,
(su3_vector *)F_PT(s,dest) );
}
/* Now divide by 2 eq. (4.2b) of Golterman's Meson paper*/
FORALLSITES(i,s)
{
scalar_mult_su3_vector( dest+i, .5, dest+i );
}
