void update_u_cpu( Real eps ){
register int i,dir;
register site *s;
su3_matrix *link,temp1,temp2,htemp;
register Real t2,t3,t4,t5,t6,t7,t8;
/**TEMP**
Real gf_x,gf_av,gf_max;
int gf_i,gf_j;
**END TEMP **/
/**double dtime,dtime2,dclock();**/
/**dtime = -dclock();**/
/* Take divisions out of site loop (can't be done by compiler) */
t2 = eps/2.0;
t3 = eps/3.0;
t4 = eps/4.0;
t5 = eps/5.0;
t6 = eps/6.0;
t7 = eps/7.0;
t8 = eps/8.0;
/** TEMP **
gf_av=gf_max=0.0;
**END TEMP**/
#ifdef FN
invalidate_fermion_links(fn_links);
// free_fn_links(&fn_links);
// free_fn_links(&fn_links_dmdu0);
#endif
FORALLSITES(i,s){
for(dir=XUP; dir <=TUP; dir++){
uncompress_anti_hermitian( &(s->mom[dir]) , &htemp );
link = &(s->link[dir]);
mult_su3_nn(&htemp,link,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t8,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t7,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t6,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t5,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t4,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t3,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,t2,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,eps ,&temp2);
su3mat_copy(&temp2,link);
}
}
/**dtime += dclock();
node0_printf("LINK_UPDATE: time = %e mflops = %e\n",
dtime, (double)(5616.0*volume/(1.0e6*dtime*numnodes())) );**/
} /* update_u */
void update_u( double eps ){
register int i,dir;
register site *s;
su3_matrix *link,temp1,temp2,htemp;
register double t2,t3,t4,t5,t6;
/**TEMP**
double gf_x,gf_av,gf_max;
int gf_i,gf_j;
**END TEMP **/
/**double dtime,dtime2,dclock();**/
/**dtime = -dclock();**/
/* Temporary by-hand optimization until pgcc compiler bug is fixed */
t2 = eps/2.0;
t3 = eps/3.0;
t4 = eps/4.0;
t5 = eps/5.0;
t6 = eps/6.0;
/** TEMP **
gf_av=gf_max=0.0;
**END TEMP**/
FORALLSITES(i,s){
for(dir=XUP; dir <=TUP; dir++){
uncompress_anti_hermitian( &(s->mom[dir]) , &htemp );
link = &(s->link[dir]);
mult_su3_nn(&htemp,link,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/6.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t6,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/5.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t5,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/4.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t4,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/3.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t3,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/2.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t2,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,eps ,&temp2);
su3mat_copy(&temp2,link);
}
}
#ifdef FN
valid_longlinks=0;
valid_fatlinks=0;
#endif
/**dtime += dclock();
node0_printf("LINK_UPDATE: time = %e mflops = %e\n",
dtime, (double)(5616.0*volume/(1.0e6*dtime*numnodes())) );**/
} /* update_u */
// -----------------------------------------------------------------
// Calculate U = exp(A).U
// Goes to eighth order in the exponential:
// exp(A) * U = ( 1 + A + A^2/2 + A^3/3 ...) * U
// = U + A*(U + (A/2)*(U + (A/3)*( ... )))
void exp_mult(int dir, double eps, anti_hermitmat *A) {
register int i;
register site *s;
matrix *link, temp1, temp2, htemp;
register Real t2, t3, t4, t5, t6, t7, t8;
// Take divisions out of site loop (can't be done by compiler)
t2 = eps / 2.0;
t3 = eps / 3.0;
t4 = eps / 4.0;
t5 = eps / 5.0;
t6 = eps / 6.0;
t7 = eps / 7.0;
t8 = eps / 8.0;
FORALLSITES(i, s) {
uncompress_anti_hermitian(&(A[i]), &htemp);
link = &(s->link[dir]);
mult_nn(&htemp, link, &temp1);
scalar_mult_add_matrix(link, &temp1, t8, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, t7, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, t6, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, t5, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, t4, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, t3, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, t2, &temp2);
mult_nn(&htemp, &temp2, &temp1);
scalar_mult_add_matrix(link, &temp1, eps, &temp2);
mat_copy(&temp2, link); // This step updates the link U[dir]
}
void update_u(Real eps) {
register int i,dir;
register site *s;
su3_matrix *link,temp1,temp2,htemp;
register Real t2,t3,t4,t5,t6;
/* Temporary by-hand optimization until pgcc compiler bug is fixed */
t2 = eps/2.0;
t3 = eps/3.0;
t4 = eps/4.0;
t5 = eps/5.0;
t6 = eps/6.0;
invalidate_fermion_links(fn_links);
FORALLSITES(i,s){
for(dir=XUP; dir <=TUP; dir++) if(dir==TUP || s->t>0){
uncompress_anti_hermitian( &(s->mom[dir]) , &htemp );
link = &(s->link[dir]);
mult_su3_nn(&htemp,link,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/6.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t6,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/5.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t5,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/4.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t4,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/3.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t3,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
/**scalar_mult_add_su3_matrix(link,&temp1,eps/2.0,&temp2);**/
scalar_mult_add_su3_matrix(link,&temp1,t2,&temp2);
mult_su3_nn(&htemp,&temp2,&temp1);
scalar_mult_add_su3_matrix(link,&temp1,eps ,&temp2);
su3mat_copy(&temp2,link);
}
}
} /* update_u */
void gluon_prop( void ) {
register int i,dir;
register int pmu;
register site *s;
anti_hermitmat ahtmp;
Real pix, piy, piz, pit;
Real sin_pmu, sin_pmu2, prop_s, prop_l, ftmp1, ftmp2;
complex ctmp;
su3_matrix mat;
struct {
Real f1, f2;
} msg;
double trace, dmuAmu;
int px, py, pz, pt;
int currentnode,newnode;
pix = PI / (Real)nx;
piy = PI / (Real)ny;
piz = PI / (Real)nz;
pit = PI / (Real)nt;
trace = 0.0;
/* Make A_mu as anti-hermition traceless part of U_mu */
/* But store as SU(3) matrix for call to FFT */
for(dir=XUP; dir<=TUP; dir++)
{
FORALLSITES(i,s){
trace += (double)(trace_su3( &(s->link[dir]))).real;
make_anti_hermitian( &(s->link[dir]), &ahtmp);
uncompress_anti_hermitian( &ahtmp, &(s->a_mu[dir]));
}
g_sync();
/* Now Fourier transform */
restrict_fourier_site(F_OFFSET(a_mu[dir]),
sizeof(su3_matrix), FORWARDS);
}
static void
copy_milc_to_D_F(dsu3_matrix *dest, anti_hermitmat *src){
su3_matrix t;
uncompress_anti_hermitian( src, &t );
p2d_mat( dest, &t );
}
/* update the momenta with the gauge force */
void imp_gauge_force_cpu( Real eps, field_offset mom_off ){
register int i,dir;
register site *st;
su3_matrix tmat1,tmat2;
register Real eb3;
register anti_hermitmat* momentum;
su3_matrix *staple, *tempmat1;
/* lengths of various kinds of loops */
int *loop_length = get_loop_length();
/* number of rotations/reflections for each kind */
int *loop_num = get_loop_num();
/* table of directions, 1 for each kind of loop */
int ***loop_table = get_loop_table();
/* table of coefficients in action, for various "representations"
(actually, powers of the trace) */
Real **loop_coeff = get_loop_coeff();
int max_length = get_max_length();
int nloop = get_nloop();
int nreps = get_nreps();
#ifdef GFTIME
int nflop = 153004; /* For Symanzik1 action */
double dtime;
#endif
int j,k;
int *dirs,length;
int *path_dir,path_length;
int ln,iloop;
Real action,act2,new_term;
int ncount;
char myname[] = "imp_gauge_force";
#ifdef GFTIME
dtime=-dclock();
#endif
dirs = (int *)malloc(max_length*sizeof(int));
if(dirs == NULL){
printf("%s(%d): Can't malloc dirs\n",myname,this_node);
terminate(1);
}
path_dir = (int *)malloc(max_length*sizeof(int));
if(path_dir == NULL){
printf("%s(%d): Can't malloc path_dir\n",myname,this_node);
terminate(1);
}
staple = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix));
if(staple == NULL){
printf("%s(%d): Can't malloc temporary\n",myname,this_node);
terminate(1);
}
tempmat1 = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix));
if(tempmat1 == NULL){
printf("%s(%d): Can't malloc temporary\n",myname,this_node);
terminate(1);
}
eb3 = eps*beta/3.0;
/* Loop over directions, update mom[dir] */
for(dir=XUP; dir<=TUP; dir++){
FORALLSITES(i,st)for(j=0;j<3;j++)for(k=0;k<3;k++){
staple[i].e[j][k]=cmplx(0.0,0.0);
} END_LOOP
ncount=0;
for(iloop=0;iloop<nloop;iloop++){
length=loop_length[iloop];
for(ln=0;ln<loop_num[iloop];ln++){
/**printf("UPD: "); printpath( loop_table[iloop][ln], length );**/
/* set up dirs. we are looking at loop starting in "XUP"
direction, rotate so it starts in "dir" direction. */
for(k=0;k<length;k++){
if( GOES_FORWARDS(loop_table[iloop][ln][k]) ){
dirs[k]=(dir+loop_table[iloop][ln][k] )% 4;
}
else {
dirs[k]=OPP_DIR(
(dir+OPP_DIR(loop_table[iloop][ln][k]))%4 );
}
}
path_length= length-1; /* generalized "staple" */
/* check for links in direction of momentum to be
updated, each such link gives a contribution. Note
the direction of the path - opposite the link. */
for(k=0;k<length;k++)if( dirs[k]==dir||dirs[k]==OPP_DIR(dir)) {
if( GOES_FORWARDS(dirs[k]) ) for(j=0;j<path_length;j++) {
path_dir[j] = dirs[(k+j+1)%length];
}
if( GOES_BACKWARDS(dirs[k]) ) for(j=0;j<path_length;j++) {
path_dir[path_length-1-j] =
OPP_DIR(dirs[(k+j+1)%length]);
}
/**if(dir==XUP)printf("X_UPDATE PATH: "); printpath( path_dir, path_length );**/
path_product(path_dir,path_length, tempmat1);
/* We took the path in the other direction from our
old convention in order to get it to end up
"at our site", so now take adjoint */
/* then compute "single_action" contribution to
staple */
FORALLSITES(i,st){
su3_adjoint( &(tempmat1[i]), &tmat1 );
/* first we compute the fundamental term */
new_term = loop_coeff[iloop][0];
/* now we add in the higher representations */
if(nreps > 1){
node0_printf("WARNING: THIS CODE IS NOT TESTED\n"); exit(0);
act2=1.0;
action = 3.0 - realtrace_su3(&(st->link[dir]),
&tmat1 );
for(j=1;j<nreps;j++){
act2 *= action;
new_term +=
loop_coeff[iloop][j]*act2*(Real)(j+1);
}
} /* end if nreps > 1 */
scalar_mult_add_su3_matrix( &(staple[i]), &tmat1,
new_term, &(staple[i]) );
} END_LOOP
ncount++;
} /* k (location in path) */
} /* ln */
} /* iloop */
/* Now multiply the staple sum by the link, then update momentum */
FORALLSITES(i,st){
mult_su3_na( &(st->link[dir]), &(staple[i]), &tmat1 );
momentum = (anti_hermitmat *)F_PT(st,mom_off);
uncompress_anti_hermitian( &momentum[dir], &tmat2 );
scalar_mult_sub_su3_matrix( &tmat2, &tmat1,
eb3, &(staple[i]) );
make_anti_hermitian( &(staple[i]), &momentum[dir] );
} END_LOOP
