// create a field with "n" entities on a site // (n=4 is a usual link field) static su3_matrix *create_mn_special(int n){ char myname[] = "create_mn_special"; su3_matrix *m; m = (su3_matrix *)special_alloc(sites_on_node*n*sizeof(su3_matrix)); if(m==NULL){ printf("%s: no room\n",myname); terminate(1); } return m; }
double imp_gauge_action() { register int i; int rep; register site *s; complex trace; double g_action; double action,act2,total_action; su3_matrix *tempmat1; int length; /* these are for loop_table */ int ln,iloop; g_action=0.0; tempmat1 = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix)); if(tempmat1 == NULL){ printf("imp_gauge_action: Can't malloc temporary\n"); terminate(1); } /* gauge action */ for(iloop=0;iloop<NLOOP;iloop++){ length=loop_length[iloop]; /* loop over rotations and reflections */ for(ln=0;ln<loop_num[iloop];ln++){ path_product( loop_table[iloop][ln] , length, tempmat1 ); FORALLSITES(i,s){ trace=trace_su3( &tempmat1[i] ); action = 3.0 - (double)trace.real; /* need the "3 -" for higher characters */ total_action= (double)loop_coeff[iloop][0]*action; act2=action; for(rep=1;rep<NREPS;rep++){ act2 *= action; total_action += (double)loop_coeff[iloop][rep]*act2; } g_action += total_action; } END_LOOP /* sites */ } /* ln */ } /* iloop */
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 }
static su3_matrix *create_longlinks_qop_milc(QOP_info_t *info, QOP_asqtad_coeffs_t *coeffs, QOP_GaugeField *gauge) { register int i; register site *s; int ipath,dir; int disp[4]; int num_q_paths = ks_act_paths.num_q_paths; Q_path *q_paths = ks_act_paths.q_paths; register su3_matrix *long1; su3_matrix *staple, *tempmat1; int nflop = 1804; double dtime; su3_matrix *t_ll; char myname[] = "create_longlinks_qop_milc"; dtime=-dclock(); if( phases_in != 1){ node0_printf("BOTCH: %s needs phases in\n",myname); terminate(0); } /* Allocate space for t_longlink if NULL */ t_ll = (su3_matrix *)special_alloc(sites_on_node*4*sizeof(su3_matrix)); if(t_ll==NULL){ printf("NODE %d: no room for t_ll\n",this_node); terminate(1); } staple = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix)); if(staple == NULL){ printf("%s: Can't malloc temporary\n",myname); terminate(1); } tempmat1 = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix)); if(tempmat1 == NULL){ printf("%s: Can't malloc temporary\n",myname); terminate(1); } for (dir=XUP; dir<=TUP; dir++){ /* loop over longlink directions */ /* set longlink to zero */ FORALLSITES(i,s){ long1 = &(t_ll[4*i+dir]); clear_su3mat( long1 ); } /* loop over paths, checking for ones with total displacement 3*dir */ for( ipath=0; ipath<num_q_paths; ipath++ ){ /* loop over paths */ /* compute total displacement of path */ for(i=XUP;i<=TUP;i++)disp[i]=0; for( i=0; i<q_paths[ipath].length; i++){ if( GOES_FORWARDS(q_paths[ipath].dir[i]) ) disp[ q_paths[ipath].dir[i] ]++; else disp[OPP_DIR(q_paths[ipath].dir[i]) ]--; } for( disp[dir]+=3,i=XUP; i<=TUP; i++)if(disp[i]!=0)break; if( i<=TUP )continue; /* skip if path doesn't go to right place */ /**printf("ipath = %d, found a path: ",ipath); for(j=0;j<q_paths[ipath].length;j++)printf("\t%d", q_paths[ipath].dir[j]); printf("\n");**/ path_product_qop_milc( q_paths[ipath].dir, q_paths[ipath].length, tempmat1, gauge ); FORALLSITES(i,s){ su3_adjoint( &tempmat1[i], &staple[i] ); long1 = &(t_ll[4*i+dir]); scalar_mult_add_su3_matrix( long1, &staple[i], -q_paths[ipath].coeff, long1 ); /* minus sign in coeff. because we used backward path*/ } } /* ipath */
/* update the momenta with the gauge force */ void QOP_symanzik_1loop_gauge_force(QOP_info_t *info, QOP_GaugeField *gauge, QOP_Force *force, QOP_gauge_coeffs_t *coeffs, Real eps) { register int i,dir; register site *st; su3_matrix tmat1; register Real eb3; /* Note: eps now includes eps*beta */ register su3_matrix* 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(); /* We make our own */ int max_length = get_max_length(); /* For Symanzik 1 loop! */ int nloop = get_nloop(); int nreps = get_nreps(); su3_matrix *forwardlink[4]; su3_matrix *tmpmom[4]; int nflop = 153004; /* For Symanzik1 action */ Real final_flop; double dtime; 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"; dtime=-dclock(); info->status = QOP_FAIL; /* Parity requirements */ if(gauge->evenodd != QOP_EVENODD || force->evenodd != QOP_EVENODD ) { printf("QOP_asqtad_force: Bad parity gauge %d force %d\n", gauge->evenodd, force->evenodd); return; } /* Map field pointers to local static pointers */ FORALLUPDIR(dir){ forwardlink[dir] = gauge->g + dir*sites_on_node; tmpmom[dir] = force->f + dir*sites_on_node; } /* Check loop coefficients */ if(coeffs->plaquette != loop_coeff[0][0] || coeffs->rectangle != loop_coeff[1][0] || coeffs->parallelogram != loop_coeff[2][0]) { printf("%s(%d): Path coeffs don't match\n",myname,this_node); return; } /* Allocate arrays according to action */ dirs = (int *)malloc(max_length*sizeof(int)); if(dirs == NULL){ printf("%s(%d): Can't malloc dirs\n",myname,this_node); return; } path_dir = (int *)malloc(max_length*sizeof(int)); if(path_dir == NULL){ printf("%s(%d): Can't malloc path_dir\n",myname,this_node); return; } 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); return; } 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); return; } eb3 = eps/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(forwardlink[dir]+i, &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( forwardlink[dir]+i, &(staple[i]), &tmat1 ); momentum = tmpmom[dir] + i; scalar_mult_sub_su3_matrix( momentum, &tmat1, eb3, momentum ); } END_LOOP
static void create_asqtad_links(int both, ferm_links_t *fn, ks_action_paths *ap) { Real *act_path_coeff = ap->act_path_coeff; su3_matrix **t_fl = &fn->fat; su3_matrix **t_ll = &fn->lng; char myname[] = "create_asqtad_links"; QOP_info_t info; QOP_asqtad_coeffs_t coeffs; #ifdef LLTIME double nflopfl = 61632; double nflopll = 1804; double nflop = nflopfl + nflopll; #endif double dtime; double remaptime = -dclock(); LOAD_QOP_ASQTAD_COEFFS(&coeffs, 0.5, act_path_coeff); if( phases_in != 1){ node0_printf("load_fermion_links_fn: BOTCH: needs phases in\n"); terminate(1); } /* Initialize QOP */ if(initialize_qop() != QOP_SUCCESS){ printf("%s(%d): Error initializing QOP\n",myname,this_node); terminate(1); } remaptime += dclock(); dtime = -dclock(); DESTROY_QOP_ASQTAD_FERMION_LINKS(fn); fn->QOP_L = CREATE_L_FROM_SITE_GAUGE( &info, &coeffs, F_OFFSET(link), EVENANDODD ); dtime += dclock(); remaptime -= dclock(); /* Allocate space for t_fl 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_fl\n",myname,this_node); terminate(1); } } /* Allocate space for t_ll 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_ll\n",myname,this_node); terminate(1); } } UNLOAD_L_TO_FIELDS( *t_fl, *t_ll, fn->QOP_L, EVENANDODD ); remaptime += dclock(); #ifdef LLTIME node0_printf("LLTIME(total): time = %e (Asqtad opt) mflops = %e\n",dtime, (Real)nflop*volume/(1e6*dtime*numnodes()) ); #endif #ifdef LLTIME #ifdef REMAP node0_printf("LLREMAP: time = %e\n",remaptime); #endif #endif }
void load_fatlinks_cpu(info_t *info, su3_matrix *fat, ks_component_paths *p, su3_matrix *links){ register int i; int dir; register su3_matrix *fat1; su3_matrix *staple = NULL, *tempmat1 = NULL; char myname[] = "load_fatlinks_cpu"; #ifdef ASQ_OPTIMIZED_FATTENING int nu,rho,sig ; Real one_link; #else int ipath; int disp[4]; int num_q_paths = p->num_q_paths; Q_path *q_paths = p->q_paths; #endif double dtime = -dclock(); staple = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix)); if(staple == NULL){ printf("%s: Can't malloc temporary\n",myname); terminate(1); } tempmat1 = (su3_matrix *)special_alloc(sites_on_node*sizeof(su3_matrix)); if(tempmat1 == NULL){ printf("%s: Can't malloc temporary\n",myname); terminate(1); } #ifndef ASQ_OPTIMIZED_FATTENING /* general case code */ for (dir=XUP; dir<=TUP; dir++){ /* loop over fatlink directions */ /* set fatlink to zero */ FORALLFIELDSITES(i){ fat1 = fat + 4*i + dir; clear_su3mat( fat1 ); } /* loop over paths, checking for ones with total displacement 1*dir */ for( ipath=0; ipath<num_q_paths; ipath++ ){ /* loop over paths */ /* compute total displacement of path */ for(i=XUP;i<=TUP;i++)disp[i]=0; for( i=0; i<q_paths[ipath].length; i++){ if( GOES_FORWARDS(q_paths[ipath].dir[i]) ) disp[ q_paths[ipath].dir[i] ]++; else disp[OPP_DIR(q_paths[ipath].dir[i]) ]--; } for( disp[dir]+=1,i=XUP; i<=TUP; i++)if(disp[i]!=0)break; if( i<=TUP )continue; /* skip if path doesn't go to right place */ /**printf("dir = %d, found a path: ",dir); for(j=0;j<q_paths.[ipath].length;j++)printf("\t%d", q_paths[ipath].dir[j]); printf("\n");**/ // path_product( q_paths[ipath].dir, q_paths[ipath].length, tempmat1 ); // path_product_field( q_paths[ipath].dir, q_paths[ipath].length, // tempmat1, links ); path_product_fields( links, q_paths[ipath].dir, q_paths[ipath].length, tempmat1 ); FORALLFIELDSITES(i){ su3_adjoint( &tempmat1[i], &staple[i] ); fat1 = fat + 4*i + dir; scalar_mult_add_su3_matrix( fat1, &staple[i], -q_paths[ipath].coeff, fat1 ); /* minus sign in coeff. because we used backward path*/ } } /* ipath */ } /* loop over directions */ #else /* ASQ_OPTIMIZED_FATTENING, for Asq and Asqtad actions */ /* Optimized fattening code for the Asq and Asqtad actions. * * I assume that path 0 is the one link path 2 the 3-staple * * path 3 the 5-staple path 4 the 7-staple and path 5 the Lepage term. * * Path 1 is the Naik term. */ /* to fix up the Lepage term, included by a trick below */ one_link = (p->act_path_coeff.one_link - 6.0*p->act_path_coeff.lepage); for (dir=XUP; dir<=TUP; dir++){ FORALLFIELDSITES(i) /* Intialize fat links with c_1*U_\mu(x) */ { fat1 = fat + 4*i + dir; scalar_mult_su3_matrix(links + 4*i + dir, one_link, fat1 ); } /* Skip the rest of the calculation if the remaining coefficients vanish */ if( p->act_path_coeff.three_staple == 0.0 && p->act_path_coeff.lepage == 0.0 && p->act_path_coeff.five_staple == 0.0)continue; for(nu=XUP; nu<=TUP; nu++) if(nu!=dir) { // compute_gen_staple_site(staple,dir,nu,F_OFFSET(link[dir]), // *t_fl, act_path_coeff.three_staple); compute_gen_staple_field(staple, dir, nu, links + dir, 4, fat, p->act_path_coeff.three_staple, links); /* The Lepage term */ /* Note this also involves modifying c_1 (above) */ compute_gen_staple_field(NULL, dir, nu, staple, 1, fat, p->act_path_coeff.lepage, links); for(rho=XUP; rho<=TUP; rho++) if((rho!=dir)&&(rho!=nu)) { compute_gen_staple_field( tempmat1, dir, rho, staple, 1, fat, p->act_path_coeff.five_staple, links); for(sig=XUP; sig<=TUP; sig++) if((sig!=dir)&&(sig!=nu)&&(sig!=rho)) { compute_gen_staple_field(NULL,dir,sig,tempmat1, 1, fat, p->act_path_coeff.seven_staple, links); } /* sig */ } /* rho */ } /* nu */ }/* dir */ #endif special_free(staple); staple = NULL; special_free(tempmat1); tempmat1 = NULL; dtime += dclock(); info->final_sec += dtime; info->final_flop = 61632.*volume/numnodes(); if( p->act_path_coeff.three_staple == 0.0 && p->act_path_coeff.lepage == 0.0 && p->act_path_coeff.five_staple == 0.0) info->final_flop = 72.*volume/numnodes(); } /* load_fatlinks_cpu */
/* 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