void tm_times_Hopping_Matrix(const int ieo, spinor * const l, spinor * const k, complex double const cfactor) {
# ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
# endif
# ifdef OMP
# pragma omp parallel
{
su3 * restrict u0 ALIGN;
# endif
# define _MUL_G5_CMPLX
# if (defined BGQ && defined XLC)
complex double ALIGN bla = cfactor;
vector4double ALIGN cf = vec_ld2(0, (double*) &bla);
# elif (defined SSE2 || defined SSE3)
_Complex double ALIGN cf = cfactor;
# endif
# include "operator/halfspinor_body.c"
# undef _MUL_G5_CMPLX
# ifdef OMP
} /* OpenMP closing brace */
# endif
return;
}
void tm_sub_Hopping_Matrix(const int ieo, spinor * const l, spinor * p, spinor * const k,
complex double const cfactor) {
# ifdef XLC
# pragma disjoint(*l, *k)
# endif
# ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
# endif
# if (defined TM_USE_MPI)
xchange_field(k, ieo);
# endif
# ifdef TM_USE_OMP
# pragma omp parallel
{
# endif
# define _TM_SUB_HOP
spinor * pn;
# if (defined BGQ && defined XLC)
complex double ALIGN bla = cfactor;
vector4double ALIGN cf = vec_ld2(0, (double*) &bla);
# elif (defined SSE2 || defined SSE3)
_Complex double ALIGN cf = cfactor;
su3_vector ALIGN psi, psi2;
# endif
# include "operator/hopping_body_dbl.c"
# undef _TM_SUB_HOP
# ifdef TM_USE_OMP
} /* OpenMP closing brace */
# endif
return;
}
void tm_times_Hopping_Matrix(const int ieo, spinor * const l, spinor * const k, double complex const cfactor) {
# ifdef XLC
# pragma disjoint(*l, *k)
# endif
# ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
# endif
# if (defined MPI)
xchange_field(k, ieo);
# endif
# ifdef OMP
# pragma omp parallel
{
# endif
# define _MUL_G5_CMPLX
# if (defined BGQ && defined XLC)
complex double ALIGN bla = cfactor;
vector4double ALIGN cf = vec_ld2(0, (double*) &bla);
# elif (defined SSE2 || defined SSE3)
_Complex double ALIGN cf = cfactor;
# endif
# include "operator/hopping_body_dbl.c"
# undef _MUL_G5_CMPLX
# ifdef OMP
} /* OpenMP closing brace */
# endif
return;
}
void Hopping_Matrix(const int ieo, spinor * const l, spinor * const k) {
# ifdef XLC
# pragma disjoint(*l, *k)
# endif
# ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
# endif
# if (defined TM_USE_MPI && !(defined _NO_COMM))
xchange_field(k, ieo);
# endif
# ifdef TM_USE_OMP
# pragma omp parallel
{
# endif
# include "operator/hopping_body_dbl.c"
# ifdef TM_USE_OMP
} /* OpenMP closing brace */
# endif
return;
}
void Hopping_Matrix(const int ieo, spinor * const l, spinor * const k) {
#ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
#endif
#ifdef TM_USE_OMP
#pragma omp parallel
{
su3 * restrict u0 ALIGN;
#endif
# include "operator/halfspinor_body.c"
# ifdef TM_USE_OMP
} /* OpenMP closing brace */
# endif
return;
}
int main(int argc,char *argv[]) {
FILE *parameterfile=NULL,*rlxdfile=NULL, *countfile=NULL;
char * filename = NULL;
char datafilename[50];
char parameterfilename[50];
char gauge_filename[50];
char * nstore_filename = ".nstore_counter";
char * input_filename = NULL;
int rlxd_state[105];
int j,ix,mu;
int k;
struct timeval t1;
int g_nev, max_iter_ev;
double stop_prec_ev;
/* Energy corresponding to the Gauge part */
double eneg = 0., plaquette_energy = 0., rectangle_energy = 0.;
/* Acceptance rate */
int Rate=0;
/* Do we want to perform reversibility checks */
/* See also return_check_flag in read_input.h */
int return_check = 0;
/* For getopt */
int c;
/* For the Polyakov loop: */
int dir = 2;
_Complex double pl, pl4;
verbose = 0;
g_use_clover_flag = 0;
g_nr_of_psf = 1;
#ifndef XLC
signal(SIGUSR1,&catch_del_sig);
signal(SIGUSR2,&catch_del_sig);
signal(SIGTERM,&catch_del_sig);
signal(SIGXCPU,&catch_del_sig);
#endif
while ((c = getopt(argc, argv, "h?f:o:")) != -1) {
switch (c) {
case 'f':
input_filename = calloc(200, sizeof(char));
strcpy(input_filename,optarg);
break;
case 'o':
filename = calloc(200, sizeof(char));
strcpy(filename,optarg);
break;
case 'h':
case '?':
default:
usage();
break;
}
}
if(input_filename == NULL){
input_filename = "hmc.input";
}
if(filename == NULL){
filename = "output";
}
/* Read the input file */
read_input(input_filename);
mpi_init(argc, argv);
if(Nsave == 0){
Nsave = 1;
}
if(nstore == -1) {
countfile = fopen(nstore_filename, "r");
if(countfile != NULL) {
fscanf(countfile, "%d\n", &nstore);
fclose(countfile);
}
else {
nstore = 0;
}
}
if(g_rgi_C1 == 0.) {
g_dbw2rand = 0;
}
#ifndef TM_USE_MPI
g_dbw2rand = 0;
#endif
/* Reorder the mu parameter and the number of iterations */
if(g_mu3 > 0.) {
g_mu = g_mu1;
g_mu1 = g_mu3;
g_mu3 = g_mu;
j = int_n[1];
int_n[1] = int_n[3];
int_n[3] = j;
j = g_csg_N[0];
g_csg_N[0] = g_csg_N[4];
g_csg_N[4] = j;
g_csg_N[6] = j;
if(fabs(g_mu3) > 0) {
g_csg_N[6] = 0;
}
g_nr_of_psf = 3;
}
else if(g_mu2 > 0.) {
g_mu = g_mu1;
g_mu1 = g_mu2;
g_mu2 = g_mu;
int_n[3] = int_n[1];
int_n[1] = int_n[2];
int_n[2] = int_n[3];
/* For chronological inverter */
g_csg_N[4] = g_csg_N[0];
g_csg_N[0] = g_csg_N[2];
g_csg_N[2] = g_csg_N[4];
if(fabs(g_mu2) > 0) {
g_csg_N[4] = 0;
}
g_csg_N[6] = 0;
g_nr_of_psf = 2;
}
else {
g_csg_N[2] = g_csg_N[0];
if(fabs(g_mu2) > 0) {
g_csg_N[2] = 0;
}
g_csg_N[4] = 0;
g_csg_N[6] = 0;
}
for(j = 0; j < g_nr_of_psf+1; j++) {
if(int_n[j] == 0) int_n[j] = 1;
}
if(g_nr_of_psf == 3) {
g_eps_sq_force = g_eps_sq_force1;
g_eps_sq_force1 = g_eps_sq_force3;
g_eps_sq_force3 = g_eps_sq_force;
g_eps_sq_acc = g_eps_sq_acc1;
g_eps_sq_acc1 = g_eps_sq_acc3;
g_eps_sq_acc3 = g_eps_sq_acc;
}
if(g_nr_of_psf == 2) {
g_eps_sq_force = g_eps_sq_force1;
g_eps_sq_force1 = g_eps_sq_force2;
g_eps_sq_force2 = g_eps_sq_force;
g_eps_sq_acc = g_eps_sq_acc1;
g_eps_sq_acc1 = g_eps_sq_acc2;
g_eps_sq_acc2 = g_eps_sq_acc;
}
g_mu = g_mu1;
g_eps_sq_acc = g_eps_sq_acc1;
g_eps_sq_force = g_eps_sq_force1;
#ifdef _GAUGE_COPY
j = init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
j = init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
if ( j!= 0) {
fprintf(stderr, "Not enough memory for gauge_fields! Aborting...\n");
exit(0);
}
j = init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for geometry_indices! Aborting...\n");
exit(0);
}
j = init_spinor_field(VOLUMEPLUSRAND/2, NO_OF_SPINORFIELDS);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
j = init_bispinor_field(VOLUME/2, NO_OF_SPINORFIELDS);
j = init_csg_field(VOLUMEPLUSRAND/2, g_csg_N);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for csg fields! Aborting...\n");
exit(0);
}
j = init_moment_field(VOLUME, VOLUMEPLUSRAND);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for moment fields! Aborting...\n");
exit(0);
}
zero_spinor_field(g_spinor_field[DUM_DERI+4],VOLUME/2);
zero_spinor_field(g_spinor_field[DUM_DERI+5],VOLUME/2);
zero_spinor_field(g_spinor_field[DUM_DERI+6],VOLUME/2);
if(g_proc_id == 0){
/* fscanf(fp6,"%s",filename); */
/*construct the filenames for the observables and the parameters*/
strcpy(datafilename,filename); strcat(datafilename,".data");
strcpy(parameterfilename,filename); strcat(parameterfilename,".para");
parameterfile=fopen(parameterfilename, "w");
printf("# This is the hmc code for twisted Mass Wilson QCD\n\nVersion %s\n", Version);
#ifdef SSE
printf("# The code was compiled with SSE instructions\n");
#endif
#ifdef SSE2
printf("# The code was compiled with SSE2 instructions\n");
#endif
#ifdef SSE3
printf("# The code was compiled with SSE3 instructions\n");
#endif
#ifdef P4
printf("# The code was compiled for Pentium4\n");
#endif
#ifdef OPTERON
printf("# The code was compiled for AMD Opteron\n");
#endif
#ifdef _NEW_GEOMETRY
printf("# The code was compiled with -D_NEW_GEOMETRY\n");
#endif
#ifdef _GAUGE_COPY
printf("# The code was compiled with -D_GAUGE_COPY\n");
#endif
printf("# The lattice size is %d x %d x %d x %d\n",
(int)(T*g_nproc_t), (int)(LX*g_nproc_x), (int)(LY), (int)(LZ));
printf("# The local lattice size is %d x %d x %d x %d\n",
(int)(T), (int)(LX), (int)(LY),(int) LZ);
printf("# beta = %f , kappa= %f\n", g_beta, g_kappa);
printf("# mus = %f, %f, %f\n", g_mu1, g_mu2, g_mu3);
printf("# int_n_gauge = %d, int_n_ferm1 = %d, int_n_ferm2 = %d, int_n_ferm3 = %d\n",
int_n[0], int_n[1], int_n[2], int_n[3]);
printf("# g_rgi_C0 = %f, g_rgi_C1 = %f\n", g_rgi_C0, g_rgi_C1);
printf("# Number of pseudo-fermion fields: %d\n", g_nr_of_psf);
printf("# g_eps_sq_force = %e, g_eps_sq_acc = %e\n", g_eps_sq_force, g_eps_sq_acc);
printf("# Integration scheme: ");
if(integtyp == 1) printf("leap-frog (single time scale)\n");
if(integtyp == 2) printf("Sexton-Weingarten (single time scale)\n");
if(integtyp == 3) printf("leap-frog (multiple time scales)\n");
if(integtyp == 4) printf("Sexton-Weingarten (multiple time scales)\n");
if(integtyp == 5) printf("higher order and leap-frog (multiple time scales)\n");
printf("# Using %s precision for the inversions!\n",
g_relative_precision_flag ? "relative" : "absolute");
printf("# Using in chronological inverter for spinor_field 1,2,3 a history of %d, %d, %d, respectively\n",
g_csg_N[0], g_csg_N[2], g_csg_N[4]);
fprintf(parameterfile, "The lattice size is %d x %d x %d x %d\n", (int)(g_nproc_t*T), (int)(g_nproc_x*LX), (int)(LY), (int)(LZ));
fprintf(parameterfile, "The local lattice size is %d x %d x %d x %d\n", (int)(T), (int)(LX), (int)(LY), (int)(LZ));
fprintf(parameterfile, "g_beta = %f , g_kappa= %f, c_sw = %f \n",g_beta,g_kappa,g_c_sw);
fprintf(parameterfile, "boundary of fermion fields (t,x,y,z): %f %f %f %f \n",X0,X1,X2,X3);
fprintf(parameterfile, "EPS_SQ0=%e, EPS_SQ1=%e EPS_SQ2=%e, EPS_SQ3=%e \n"
,EPS_SQ0,EPS_SQ1,EPS_SQ2,EPS_SQ3);
fprintf(parameterfile, "g_eps_sq_force = %e, g_eps_sq_acc = %e\n", g_eps_sq_force, g_eps_sq_acc);
fprintf(parameterfile, "dtau=%f, Nsteps=%d, Nmeas=%d, Nsave=%d, integtyp=%d, nsmall=%d \n",
dtau,Nsteps,Nmeas,Nsave,integtyp,nsmall);
fprintf(parameterfile, "mu = %f, mu2=%f, mu3=%f\n ", g_mu, g_mu2, g_mu3);
fprintf(parameterfile, "int_n_gauge = %d, int_n_ferm1 = %d, int_n_ferm2 = %d, int_n_ferm3 = %d\n ",
int_n[0], int_n[1], int_n[2], int_n[3]);
fprintf(parameterfile, "g_rgi_C0 = %f, g_rgi_C1 = %f\n", g_rgi_C0, g_rgi_C1);
fprintf(parameterfile, "# Number of pseudo-fermion fields: %d\n", g_nr_of_psf);
fprintf(parameterfile, "# Integration scheme: ");
if(integtyp == 1) fprintf(parameterfile, "leap-frog (single time scale)\n");
if(integtyp == 2) fprintf(parameterfile, "Sexton-Weingarten (single time scale)\n");
if(integtyp == 3) fprintf(parameterfile, "leap-frog (multiple time scales)\n");
if(integtyp == 4) fprintf(parameterfile, "Sexton-Weingarten (multiple time scales)\n");
if(integtyp == 5) fprintf(parameterfile, "higher order and leap-frog (multiple time scales)\n");
fprintf(parameterfile, "Using %s precision for the inversions!\n",
g_relative_precision_flag ? "relative" : "absolute");
fprintf(parameterfile, "Using in chronological inverter for spinor_field 1,2,3 a history of %d, %d, %d, respectively\n",
g_csg_N[0], g_csg_N[2], g_csg_N[4]);
fflush(stdout); fflush(parameterfile);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary();
check_geometry();
if(g_proc_id == 0) {
#if defined GEOMETRIC
if(g_proc_id==0) fprintf(parameterfile,"The geometric series is used as solver \n\n");
#else
if(g_proc_id==0) fprintf(parameterfile,"The BICG_stab is used as solver \n\n");
#endif
fflush(parameterfile);
}
/* Continue */
if(startoption == 3){
rlxdfile = fopen(rlxd_input_filename,"r");
if(rlxdfile != NULL) {
if(g_proc_id == 0) {
fread(rlxd_state,sizeof(rlxd_state),1,rlxdfile);
}
}
else {
if(g_proc_id == 0) {
printf("%s does not exist, switching to restart...\n", rlxd_input_filename);
}
startoption = 2;
}
fclose(rlxdfile);
if(startoption != 2) {
if(g_proc_id == 0) {
rlxd_reset(rlxd_state);
printf("Reading Gauge field from file %s\n", gauge_input_filename); fflush(stdout);
}
read_gauge_field_time_p(gauge_input_filename,g_gauge_field);
}
}
if(startoption != 3){
/* Initialize random number generator */
if(g_proc_id == 0) {
rlxd_init(1, random_seed);
/* hot */
if(startoption == 1) {
random_gauge_field();
}
rlxd_get(rlxd_state);
#ifdef TM_USE_MPI
MPI_Send(&rlxd_state[0], 105, MPI_INT, 1, 99, MPI_COMM_WORLD);
MPI_Recv(&rlxd_state[0], 105, MPI_INT, g_nproc-1, 99, MPI_COMM_WORLD, &status);
rlxd_reset(rlxd_state);
#endif
}
#ifdef TM_USE_MPI
else {
MPI_Recv(&rlxd_state[0], 105, MPI_INT, g_proc_id-1, 99, MPI_COMM_WORLD, &status);
rlxd_reset(rlxd_state);
/* hot */
if(startoption == 1) {
random_gauge_field();
}
k=g_proc_id+1;
if(k==g_nproc){
k=0;
}
rlxd_get(rlxd_state);
MPI_Send(&rlxd_state[0], 105, MPI_INT, k, 99, MPI_COMM_WORLD);
}
#endif
/* Cold */
if(startoption == 0) {
unit_g_gauge_field();
}
/* Restart */
else if(startoption == 2) {
if (g_proc_id == 0){
printf("Reading Gauge field from file %s\n", gauge_input_filename); fflush(stdout);
}
read_gauge_field_time_p(gauge_input_filename,g_gauge_field);
}
}
/*For parallelization: exchange the gaugefield */
#ifdef TM_USE_MPI
xchange_gauge(g_gauge_field);
#endif
#ifdef _GAUGE_COPY
update_backward_gauge();
#endif
/*compute the energy of the gauge field*/
plaquette_energy=measure_gauge_action();
if(g_rgi_C1 > 0. || g_rgi_C1 < 0.) {
rectangle_energy = measure_rectangles();
if(g_proc_id==0){
fprintf(parameterfile,"#First rectangle value: %14.12f \n",rectangle_energy/(12.*VOLUME*g_nproc));
}
}
eneg = g_rgi_C0 * plaquette_energy + g_rgi_C1 * rectangle_energy;
/* Measure and print the Polyakov loop: */
polyakov_loop(&pl, dir);
if(g_proc_id==0){
fprintf(parameterfile,"#First plaquette value: %14.12f \n", plaquette_energy/(6.*VOLUME*g_nproc));
fprintf(parameterfile,"#First Polyakov loop value in %d-direction |L(%d)|= %14.12f \n",
dir, dir, cabs(pl));
}
dir=3;
polyakov_loop(&pl, dir);
if(g_proc_id==0){
fprintf(parameterfile,"#First Polyakov loop value in %d-direction |L(%d)|= %14.12f \n",
dir, dir, cabs(pl));
fclose(parameterfile);
}
/* set ddummy to zero */
for(ix = 0; ix < VOLUME+RAND; ix++){
for(mu=0; mu<4; mu++){
ddummy[ix][mu].d1=0.;
ddummy[ix][mu].d2=0.;
ddummy[ix][mu].d3=0.;
ddummy[ix][mu].d4=0.;
ddummy[ix][mu].d5=0.;
ddummy[ix][mu].d6=0.;
ddummy[ix][mu].d7=0.;
ddummy[ix][mu].d8=0.;
}
}
if(g_proc_id == 0) {
gettimeofday(&t1,NULL);
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "!!! Timestamp %ld, Nsave = %d, g_mu = %e, g_mu1 = %e, g_mu_2 = %e, g_mu3 = %e, beta = %f, kappa = %f, C1 = %f, int0 = %d, int1 = %d, int2 = %d, int3 = %d, g_eps_sq_force = %e, g_eps_sq_acc = %e, ",
t1.tv_sec, Nsave, g_mu, g_mu1, g_mu2, g_mu3, g_beta, g_kappa, g_rgi_C1,
int_n[0], int_n[1], int_n[2], int_n[3], g_eps_sq_force, g_eps_sq_acc);
fprintf(countfile, "Nsteps = %d, dtau = %e, tau = %e, integtyp = %d, rel. prec. = %d\n",
Nsteps, dtau, tau, integtyp, g_relative_precision_flag);
fclose(countfile);
}
/* HERE THE CALLS FOR SOME EIGENVALUES */
/* for lowest
g_nev = 10;
*/
/* for largest
*/
g_nev = 10;
max_iter_ev = 1000;
stop_prec_ev = 1.e-10;
if(g_proc_id==0) {
printf(" Values of mu = %e mubar = %e eps = %e precision = %e \n \n", g_mu, g_mubar, g_epsbar, stop_prec_ev);
}
eigenvalues(&g_nev, operator_flag, max_iter_ev, stop_prec_ev);
g_nev = 4;
max_iter_ev = 200;
stop_prec_ev = 1.e-03;
max_eigenvalues(&g_nev, operator_flag, max_iter_ev, stop_prec_ev);
if(g_proc_id==0) {
printf(" Values of mu = %e mubar = %e eps = %e precision = %e \n \n", g_mu, g_mubar, g_epsbar, stop_prec_ev);
/*
printf(" Values of mu = %e precision = %e \n \n", g_mu, stop_prec_ev);
*/
}
/* END OF EIGENVALUES CALLS */
if(g_proc_id==0) {
rlxd_get(rlxd_state);
rlxdfile=fopen("last_state","w");
fwrite(rlxd_state,sizeof(rlxd_state),1,rlxdfile);
fclose(rlxdfile);
printf("Acceptance Rate was: %e Prozent\n", 100.*(double)Rate/(double)Nmeas);
fflush(stdout);
parameterfile = fopen(parameterfilename, "a");
fprintf(parameterfile, "Acceptance Rate was: %e Prozent\n", 100.*(double)Rate/(double)Nmeas);
fclose(parameterfile);
}
#ifdef TM_USE_MPI
MPI_Finalize();
#endif
free_gauge_tmp();
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_bispinor_field();
free_moment_field();
return(0);
}
int main(int argc,char *argv[])
{
int j,j_max,k,k_max = 2;
paramsXlfInfo *xlfInfo;
int ix, n, *nn,*mm,i;
double delta, deltamax;
spinor rsp;
int status = 0;
#ifdef MPI
DUM_DERI = 6;
DUM_SOLVER = DUM_DERI+2;
DUM_MATRIX = DUM_SOLVER+6;
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
MPI_Init(&argc, &argv);
#endif
g_rgi_C1 = 1.;
/* Read the input file */
read_input("hopping_test.input");
tmlqcd_mpi_init(argc, argv);
if(g_proc_id==0) {
#ifdef SSE
printf("# The code was compiled with SSE instructions\n");
#endif
#ifdef SSE2
printf("# The code was compiled with SSE2 instructions\n");
#endif
#ifdef SSE3
printf("# The code was compiled with SSE3 instructions\n");
#endif
#ifdef P4
printf("# The code was compiled for Pentium4\n");
#endif
#ifdef OPTERON
printf("# The code was compiled for AMD Opteron\n");
#endif
#ifdef _GAUGE_COPY
printf("# The code was compiled with -D_GAUGE_COPY\n");
#endif
#ifdef BGL
printf("# The code was compiled for Blue Gene/L\n");
#endif
#ifdef BGP
printf("# The code was compiled for Blue Gene/P\n");
#endif
#ifdef _USE_HALFSPINOR
printf("# The code was compiled with -D_USE_HALFSPINOR\n");
#endif
#ifdef _USE_SHMEM
printf("# the code was compiled with -D_USE_SHMEM\n");
# ifdef _PERSISTENT
printf("# the code was compiled for persistent MPI calls (halfspinor only)\n");
# endif
#endif
#ifdef _INDEX_INDEP_GEOM
printf("# the code was compiled with index independent geometry\n");
#endif
#ifdef MPI
# ifdef _NON_BLOCKING
printf("# the code was compiled for non-blocking MPI calls (spinor and gauge)\n");
# endif
# ifdef _USE_TSPLITPAR
printf("# the code was compiled with tsplit parallelization\n");
# endif
#endif
printf("\n");
fflush(stdout);
}
#ifdef _GAUGE_COPY
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if(even_odd_flag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, 2*k_max+1);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, 2*k_max);
}
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
j = init_moment_field(VOLUME, VOLUMEPLUSRAND);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for moment fields! Aborting...\n");
exit(0);
}
if(g_proc_id == 0) {
fprintf(stdout,"The number of processes is %d \n",g_nproc);
printf("# The lattice size is %d x %d x %d x %d\n",
(int)(T*g_nproc_t), (int)(LX*g_nproc_x), (int)(LY*g_nproc_y), (int)(g_nproc_z*LZ));
printf("# The local lattice size is %d x %d x %d x %d\n",
(int)(T), (int)(LX), (int)(LY),(int) LZ);
if(even_odd_flag) {
printf("# testinging the even/odd preconditioned Dirac operator\n");
}
else {
printf("# testinging the standard Dirac operator\n");
}
fflush(stdout);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for halfspinor fields! Aborting...\n");
exit(0);
}
if(g_sloppy_precision_flag == 1) {
g_sloppy_precision = 1;
j = init_dirac_halfspinor32();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for 32-Bit halfspinor fields! Aborting...\n");
exit(0);
}
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
status = check_geometry();
if (status != 0) {
fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n");
exit(1);
}
#if (defined MPI && !(defined _USE_SHMEM))
check_xchange();
#endif
start_ranlux(1, 123456);
xlfInfo = construct_paramsXlfInfo(0.5, 0);
random_gauge_field(reproduce_randomnumber_flag);
if ( startoption == 2 ) { /* restart */
write_gauge_field(gauge_input_filename,gauge_precision_write_flag,xlfInfo);
} else if ( startoption == 0 ) { /* cold */
unit_g_gauge_field();
} else if (startoption == 3 ) { /* continue */
read_gauge_field(gauge_input_filename);
} else if ( startoption == 1 ) { /* hot */
}
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge();
#endif
#ifdef _GAUGE_COPY
update_backward_gauge();
#endif
if(even_odd_flag) {
/*initialize the pseudo-fermion fields*/
j_max=1;
for (k = 0; k < k_max; k++) {
random_spinor_field(g_spinor_field[k], VOLUME/2, 0);
}
if (read_source_flag == 2) { /* save */
/* even first, odd second */
write_spinorfield_cm_single(g_spinor_field[0],g_spinor_field[1],SourceInfo.basename);
} else if (read_source_flag == 1) { /* yes */
/* even first, odd second */
read_spinorfield_cm_single(g_spinor_field[0],g_spinor_field[1],SourceInfo.basename,-1,0);
# if (!defined MPI)
if (write_cp_flag == 1) {
strcat(SourceInfo.basename,".2");
read_spinorfield_cm_single(g_spinor_field[2],g_spinor_field[3],SourceInfo.basename,-1,0);
nn=(int*)calloc(VOLUME,sizeof(int));
if((void*)nn == NULL) return(100);
mm=(int*)calloc(VOLUME,sizeof(int));
if((void*)mm == NULL) return(100);
n=0;
deltamax=0.0;
for(ix=0;ix<VOLUME/2;ix++){
(rsp.s0).c0.re = (g_spinor_field[2][ix].s0).c0.re - (g_spinor_field[0][ix].s0).c0.re;
(rsp.s0).c0.im = (g_spinor_field[2][ix].s0).c0.im - (g_spinor_field[0][ix].s0).c0.im;
(rsp.s0).c1.re = (g_spinor_field[2][ix].s0).c1.re - (g_spinor_field[0][ix].s0).c1.re;
(rsp.s0).c1.im = (g_spinor_field[2][ix].s0).c1.im - (g_spinor_field[0][ix].s0).c1.im;
(rsp.s0).c2.re = (g_spinor_field[2][ix].s0).c2.re - (g_spinor_field[0][ix].s0).c2.re;
(rsp.s0).c2.im = (g_spinor_field[2][ix].s0).c2.im - (g_spinor_field[0][ix].s0).c2.im;
(rsp.s1).c0.re = (g_spinor_field[2][ix].s1).c0.re - (g_spinor_field[0][ix].s1).c0.re;
(rsp.s1).c0.im = (g_spinor_field[2][ix].s1).c0.im - (g_spinor_field[0][ix].s1).c0.im;
(rsp.s1).c1.re = (g_spinor_field[2][ix].s1).c1.re - (g_spinor_field[0][ix].s1).c1.re;
(rsp.s1).c1.im = (g_spinor_field[2][ix].s1).c1.im - (g_spinor_field[0][ix].s1).c1.im;
(rsp.s1).c2.re = (g_spinor_field[2][ix].s1).c2.re - (g_spinor_field[0][ix].s1).c2.re;
(rsp.s1).c2.im = (g_spinor_field[2][ix].s1).c2.im - (g_spinor_field[0][ix].s1).c2.im;
(rsp.s2).c0.re = (g_spinor_field[2][ix].s2).c0.re - (g_spinor_field[0][ix].s2).c0.re;
(rsp.s2).c0.im = (g_spinor_field[2][ix].s2).c0.im - (g_spinor_field[0][ix].s2).c0.im;
(rsp.s2).c1.re = (g_spinor_field[2][ix].s2).c1.re - (g_spinor_field[0][ix].s2).c1.re;
(rsp.s2).c1.im = (g_spinor_field[2][ix].s2).c1.im - (g_spinor_field[0][ix].s2).c1.im;
(rsp.s2).c2.re = (g_spinor_field[2][ix].s2).c2.re - (g_spinor_field[0][ix].s2).c2.re;
(rsp.s2).c2.im = (g_spinor_field[2][ix].s2).c2.im - (g_spinor_field[0][ix].s2).c2.im;
(rsp.s3).c0.re = (g_spinor_field[2][ix].s3).c0.re - (g_spinor_field[0][ix].s3).c0.re;
(rsp.s3).c0.im = (g_spinor_field[2][ix].s3).c0.im - (g_spinor_field[0][ix].s3).c0.im;
(rsp.s3).c1.re = (g_spinor_field[2][ix].s3).c1.re - (g_spinor_field[0][ix].s3).c1.re;
(rsp.s3).c1.im = (g_spinor_field[2][ix].s3).c1.im - (g_spinor_field[0][ix].s3).c1.im;
(rsp.s3).c2.re = (g_spinor_field[2][ix].s3).c2.re - (g_spinor_field[0][ix].s3).c2.re;
(rsp.s3).c2.im = (g_spinor_field[2][ix].s3).c2.im - (g_spinor_field[0][ix].s3).c2.im;
_spinor_norm_sq(delta,rsp);
if (delta > 1.0e-12) {
nn[n] = g_eo2lexic[ix];
mm[n]=ix;
n++;
}
if(delta>deltamax) deltamax=delta;
}
if (n>0){
printf("mismatch in even spincolorfield in %d points:\n",n);
for(i=0; i< MIN(n,1000); i++){
printf("%d,(%d,%d,%d,%d):%f vs. %f\n",nn[i],g_coord[nn[i]][0],g_coord[nn[i]][1],g_coord[nn[i]][2],g_coord[nn[i]][3],(g_spinor_field[2][mm[i]].s0).c0.re, (g_spinor_field[0][mm[i]].s0).c0.re);fflush(stdout);
}
}
n = 0;
for(ix=0;ix<VOLUME/2;ix++){
(rsp.s0).c0.re = (g_spinor_field[3][ix].s0).c0.re - (g_spinor_field[1][ix].s0).c0.re;
(rsp.s0).c0.im = (g_spinor_field[3][ix].s0).c0.im - (g_spinor_field[1][ix].s0).c0.im;
(rsp.s0).c1.re = (g_spinor_field[3][ix].s0).c1.re - (g_spinor_field[1][ix].s0).c1.re;
(rsp.s0).c1.im = (g_spinor_field[3][ix].s0).c1.im - (g_spinor_field[1][ix].s0).c1.im;
(rsp.s0).c2.re = (g_spinor_field[3][ix].s0).c2.re - (g_spinor_field[1][ix].s0).c2.re;
(rsp.s0).c2.im = (g_spinor_field[3][ix].s0).c2.im - (g_spinor_field[1][ix].s0).c2.im;
(rsp.s1).c0.re = (g_spinor_field[3][ix].s1).c0.re - (g_spinor_field[1][ix].s1).c0.re;
(rsp.s1).c0.im = (g_spinor_field[3][ix].s1).c0.im - (g_spinor_field[1][ix].s1).c0.im;
(rsp.s1).c1.re = (g_spinor_field[3][ix].s1).c1.re - (g_spinor_field[1][ix].s1).c1.re;
(rsp.s1).c1.im = (g_spinor_field[3][ix].s1).c1.im - (g_spinor_field[1][ix].s1).c1.im;
(rsp.s1).c2.re = (g_spinor_field[3][ix].s1).c2.re - (g_spinor_field[1][ix].s1).c2.re;
(rsp.s1).c2.im = (g_spinor_field[3][ix].s1).c2.im - (g_spinor_field[1][ix].s1).c2.im;
(rsp.s2).c0.re = (g_spinor_field[3][ix].s2).c0.re - (g_spinor_field[1][ix].s2).c0.re;
(rsp.s2).c0.im = (g_spinor_field[3][ix].s2).c0.im - (g_spinor_field[1][ix].s2).c0.im;
(rsp.s2).c1.re = (g_spinor_field[3][ix].s2).c1.re - (g_spinor_field[1][ix].s2).c1.re;
(rsp.s2).c1.im = (g_spinor_field[3][ix].s2).c1.im - (g_spinor_field[1][ix].s2).c1.im;
(rsp.s2).c2.re = (g_spinor_field[3][ix].s2).c2.re - (g_spinor_field[1][ix].s2).c2.re;
(rsp.s2).c2.im = (g_spinor_field[3][ix].s2).c2.im - (g_spinor_field[1][ix].s2).c2.im;
(rsp.s3).c0.re = (g_spinor_field[3][ix].s3).c0.re - (g_spinor_field[1][ix].s3).c0.re;
(rsp.s3).c0.im = (g_spinor_field[3][ix].s3).c0.im - (g_spinor_field[1][ix].s3).c0.im;
(rsp.s3).c1.re = (g_spinor_field[3][ix].s3).c1.re - (g_spinor_field[1][ix].s3).c1.re;
(rsp.s3).c1.im = (g_spinor_field[3][ix].s3).c1.im - (g_spinor_field[1][ix].s3).c1.im;
(rsp.s3).c2.re = (g_spinor_field[3][ix].s3).c2.re - (g_spinor_field[1][ix].s3).c2.re;
(rsp.s3).c2.im = (g_spinor_field[3][ix].s3).c2.im - (g_spinor_field[1][ix].s3).c2.im;
_spinor_norm_sq(delta,rsp);
if (delta > 1.0e-12) {
nn[n]=g_eo2lexic[ix+(VOLUME+RAND)/2];
mm[n]=ix;
n++;
}
if(delta>deltamax) deltamax=delta;
}
if (n>0){
printf("mismatch in odd spincolorfield in %d points:\n",n);
for(i=0; i< MIN(n,1000); i++){
printf("%d,(%d,%d,%d,%d):%f vs. %f\n",nn[i],g_coord[nn[i]][0],g_coord[nn[i]][1],g_coord[nn[i]][2],g_coord[nn[i]][3],(g_spinor_field[3][mm[i]].s0).c0.re, (g_spinor_field[1][mm[i]].s0).c0.re);fflush(stdout);
}
}
printf("max delta=%e",deltamax);fflush(stdout);
}
# endif
}
if (read_source_flag > 0 && write_cp_flag == 0) { /* read-source yes or nobutsave; checkpoint no */
/* first spinorial arg is output, the second is input */
Hopping_Matrix(1, g_spinor_field[1], g_spinor_field[0]); /*ieo=1 M_{eo}*/
Hopping_Matrix(0, g_spinor_field[0], g_spinor_field[1]); /*ieo=0 M_{oe}*/
strcat(SourceInfo.basename,".out");
write_spinorfield_cm_single(g_spinor_field[0],g_spinor_field[1],SourceInfo.basename);
printf("Check-field printed. Exiting...\n");
fflush(stdout);
}
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
}
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
return(0);
}
/* Serially Checked ! */
void Dtm_psi(spinor * const P, spinor * const Q){
if(P==Q){
printf("Error in Dtm_psi (D_psi.c):\n");
printf("Arguments must be differen spinor fields\n");
printf("Program aborted\n");
exit(1);
}
#ifdef _GAUGE_COPY2
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
#endif
# if defined TM_USE_MPI
xchange_lexicfield(Q);
# endif
#ifdef TM_USE_OMP
#pragma omp parallel
{
#endif
int ix,iy,iz;
su3 *up,*um;
spinor *s,*sp,*sm,*rn;
_Complex double fact1, fact2;
spinor rs __attribute__ ((aligned (16)));
fact1 = 1. + g_mu * I;
fact2 = conj(fact1);
#ifndef TM_USE_OMP
iy=g_iup[0][0];
sp=(spinor *) Q + iy;
up=&g_gauge_field[0][0];
#endif
/************************ loop over all lattice sites *************************/
#ifdef TM_USE_OMP
#pragma omp for
#endif
for (ix=0;ix<VOLUME;ix++){
#ifdef TM_USE_OMP
iy=g_iup[ix][0];
up=&g_gauge_field[ix][0];
sp=(spinor *) Q + iy;
#endif
s=(spinor *) Q + ix;
_prefetch_spinor(s);
/******************************* direction +0 *********************************/
iy=g_idn[ix][0];
sm = (spinor *) Q + iy;
_prefetch_spinor(sm);
_sse_load(sp->s0);
_sse_load_up(sp->s2);
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_0);
_sse_store_up(rs.s2);
// the diagonal bit
_sse_load_up(s->s0);
_sse_vector_cmplx_mul(fact1);
_sse_load(rs.s2);
_sse_vector_add();
_sse_store(rs.s0);
// g5 in the twisted term
_sse_load_up(s->s2);
_sse_vector_cmplx_mul(fact2);
_sse_load(rs.s2);
_sse_vector_add();
_sse_store(rs.s2);
um=&g_gauge_field[iy][0];
_prefetch_su3(um);
_sse_load(sp->s1);
_sse_load_up(sp->s3);
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_0);
_sse_store_up(rs.s3);
// the diagonal bit
_sse_load_up(s->s1);
_sse_vector_cmplx_mul(fact1);
_sse_load(rs.s3);
_sse_vector_add();
_sse_store(rs.s1);
// g5 in the twisted term
_sse_load_up(s->s3);
_sse_vector_cmplx_mul(fact2);
_sse_load(rs.s3);
_sse_vector_add();
_sse_store(rs.s3);
/******************************* direction -0 *********************************/
iy=g_iup[ix][1];
sp = (spinor *) Q + iy;
_prefetch_spinor(sp);
_sse_load(sm->s0);
_sse_load_up(sm->s2);
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_0);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s2);
_sse_vector_sub();
_sse_store(rs.s2);
up+=1;
_prefetch_su3(up);
_sse_load(sm->s1);
_sse_load_up(sm->s3);
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_0);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s3);
_sse_vector_sub();
_sse_store(rs.s3);
/******************************* direction +1 *********************************/
iy=g_idn[ix][1];
sm = (spinor *) Q + iy;
_prefetch_spinor(sm);
_sse_load(sp->s0);
_sse_load_up(sp->s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_1);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rs.s3);
um=&g_gauge_field[iy][1];
_prefetch_su3(um);
_sse_load(sp->s1);
_sse_load_up(sp->s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_1);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rs.s2);
/******************************* direction -1 *********************************/
iy=g_iup[ix][2];
sp = (spinor *) Q + iy;
_prefetch_spinor(sp);
_sse_load(sm->s0);
_sse_load_up(sm->s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_1);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rs.s3);
up+=1;
_prefetch_su3(up);
_sse_load(sm->s1);
_sse_load_up(sm->s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_1);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rs.s2);
/******************************* direction +2 *********************************/
iy=g_idn[ix][2];
sm = (spinor *) Q + iy;
_prefetch_spinor(sm);
_sse_load(sp->s0);
_sse_load_up(sp->s3);
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_2);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_add();
_sse_store(rs.s3);
um=&g_gauge_field[iy][2];
_prefetch_su3(um);
_sse_load(sp->s1);
_sse_load_up(sp->s2);
_sse_vector_sub();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_2);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_sub();
_sse_store(rs.s2);
/******************************* direction -2 *********************************/
iy=g_iup[ix][3];
sp = (spinor *) Q + iy;
_prefetch_spinor(sp);
_sse_load(sm->s0);
_sse_load_up(sm->s3);
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_2);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_sub();
_sse_store(rs.s3);
up+=1;
_prefetch_su3(up);
_sse_load(sm->s1);
_sse_load_up(sm->s2);
_sse_vector_add();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_2);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_add();
_sse_store(rs.s2);
/******************************* direction +3 *********************************/
iy=g_idn[ix][3];
sm = (spinor *) Q + iy;
_prefetch_spinor(sm);
_sse_load(sp->s0);
_sse_load_up(sp->s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_3);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rs.s2);
um=&g_gauge_field[iy][3];
_prefetch_su3(um);
_sse_load(sp->s1);
_sse_load_up(sp->s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(phase_3);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rs.s3);
/******************************* direction -3 *********************************/
iz=(ix+1+VOLUME)%VOLUME;
iy=g_iup[iz][0];
sp = (spinor *) Q + iy;
_prefetch_spinor(sp);
_sse_load(sm->s0);
_sse_load_up(sm->s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_3);
rn = (spinor *) P + ix;
_sse_load(rs.s0);
_sse_vector_add();
_sse_store_nt(rn->s0);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store_nt(rn->s2);
up=&g_gauge_field[iz][0];
_prefetch_su3(up);
_sse_load(sm->s1);
_sse_load_up(sm->s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(phase_3);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store_nt(rn->s1);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store_nt(rn->s3);
/******************************** end of loop *********************************/
}
#ifdef TM_USE_OMP
} /* OpenMP closing brace */
#endif
}
void Hopping_Matrix(const int ieo, spinor * const l, spinor * const k){
int ix;
su3 * restrict ALIGN U;
spinor * restrict ALIGN s;
halfspinor * restrict * phi ALIGN;
halfspinor32 * restrict * phi32 ALIGN;
/* We have 32 registers available */
_declare_hregs();
#ifdef _KOJAK_INST
#pragma pomp inst begin(hoppingmatrix)
#endif
#pragma disjoint(*s, *U)
#ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
#endif
__alignx(16, l);
__alignx(16, k);
if(g_sloppy_precision == 1 && g_sloppy_precision_flag == 1) {
__alignx(16, HalfSpinor32);
/* We will run through the source vector now */
/* instead of the solution vector */
s = k;
_prefetch_spinor(s);
/* s contains the source vector */
if(ieo == 0) {
U = g_gauge_field_copy[0][0];
}
else {
U = g_gauge_field_copy[1][0];
}
phi32 = NBPointer32[ieo];
_prefetch_su3(U);
/**************** loop over all lattice sites ******************/
ix=0;
for(int i = 0; i < (VOLUME)/2; i++){
/*********************** direction +0 ************************/
_hop_t_p_pre32();
s++;
U++;
ix++;
/*********************** direction -0 ************************/
_hop_t_m_pre32();
ix++;
/*********************** direction +1 ************************/
_hop_x_p_pre32();
ix++;
U++;
/*********************** direction -1 ************************/
_hop_x_m_pre32();
ix++;
/*********************** direction +2 ************************/
_hop_y_p_pre32();
ix++;
U++;
/*********************** direction -2 ************************/
_hop_y_m_pre32();
ix++;
/*********************** direction +3 ************************/
_hop_z_p_pre32();
ix++;
U++;
/*********************** direction -3 ************************/
_hop_z_m_pre32();
ix++;
/************************ end of loop ************************/
}
# if (defined TM_USE_MPI && !defined _NO_COMM)
xchange_halffield32();
# endif
s = l;
phi32 = NBPointer32[2 + ieo];
if(ieo == 0) {
U = g_gauge_field_copy[1][0];
}
else {
U = g_gauge_field_copy[0][0];
}
//_prefetch_halfspinor(phi32[0]);
_prefetch_su3(U);
/* Now we sum up and expand to a full spinor */
ix = 0;
/* _prefetch_spinor_for_store(s); */
for(int i = 0; i < (VOLUME)/2; i++){
/* This causes a lot of trouble, do we understand this? */
/* _prefetch_spinor_for_store(s); */
//_prefetch_halfspinor(phi32[ix+1]);
/*********************** direction +0 ************************/
_hop_t_p_post32();
ix++;
/*********************** direction -0 ************************/
_hop_t_m_post32();
U++;
ix++;
/*********************** direction +1 ************************/
_hop_x_p_post32();
ix++;
/*********************** direction -1 ************************/
_hop_x_m_post32();
U++;
ix++;
/*********************** direction +2 ************************/
_hop_y_p_post32();
ix++;
/*********************** direction -2 ************************/
_hop_y_m_post32();
U++;
ix++;
/*********************** direction +3 ************************/
_hop_z_p_post32();
ix++;
/*********************** direction -3 ************************/
_hop_z_m_post32();
U++;
ix++;
s++;
}
}
else {
__alignx(16, HalfSpinor);
/* We will run through the source vector now */
/* instead of the solution vector */
s = k;
_prefetch_spinor(s);
/* s contains the source vector */
if(ieo == 0) {
U = g_gauge_field_copy[0][0];
}
else {
U = g_gauge_field_copy[1][0];
}
phi = NBPointer[ieo];
_prefetch_su3(U);
/**************** loop over all lattice sites ******************/
ix=0;
for(int i = 0; i < (VOLUME)/2; i++){
/*********************** direction +0 ************************/
_hop_t_p_pre();
s++;
U++;
ix++;
/*********************** direction -0 ************************/
_hop_t_m_pre();
ix++;
/*********************** direction +1 ************************/
_hop_x_p_pre();
ix++;
U++;
/*********************** direction -1 ************************/
_hop_x_m_pre();
ix++;
/*********************** direction +2 ************************/
_hop_y_p_pre();
ix++;
U++;
/*********************** direction -2 ************************/
_hop_y_m_pre();
ix++;
/*********************** direction +3 ************************/
_hop_z_p_pre();
ix++;
U++;
/*********************** direction -3 ************************/
_hop_z_m_pre();
ix++;
/************************ end of loop ************************/
}
# if (defined TM_USE_MPI && !defined _NO_COMM)
xchange_halffield();
# endif
s = l;
phi = NBPointer[2 + ieo];
//_prefetch_halfspinor(phi[0]);
if(ieo == 0) {
U = g_gauge_field_copy[1][0];
}
else {
U = g_gauge_field_copy[0][0];
}
_prefetch_su3(U);
/* Now we sum up and expand to a full spinor */
ix = 0;
/* _prefetch_spinor_for_store(s); */
for(int i = 0; i < (VOLUME)/2; i++){
/* This causes a lot of trouble, do we understand this? */
/* _prefetch_spinor_for_store(s); */
//_prefetch_halfspinor(phi[ix+1]);
/*********************** direction +0 ************************/
_hop_t_p_post();
ix++;
/*********************** direction -0 ************************/
_hop_t_m_post();
U++;
ix++;
/*********************** direction +1 ************************/
_hop_x_p_post();
ix++;
/*********************** direction -1 ************************/
_hop_x_m_post();
U++;
ix++;
/*********************** direction +2 ************************/
_hop_y_p_post();
ix++;
/*********************** direction -2 ************************/
_hop_y_m_post();
U++;
ix++;
/*********************** direction +3 ************************/
_hop_z_p_post();
ix++;
/*********************** direction -3 ************************/
_hop_z_m_post();
U++;
ix++;
s++;
}
}
#ifdef _KOJAK_INST
#pragma pomp inst end(hoppingmatrix)
#endif
}
/* for ieo=0, k resides on odd sites and l on even sites */
void Hopping_Matrix(int ieo, spinor * const l, spinor * const k){
int ix,iy;
int ioff,ioff2,icx,icy;
su3 * restrict up, * restrict um;
spinor * restrict r, * restrict sp, * restrict sm;
spinor temp;
#ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge();
}
#endif
/* for parallelization */
# if (defined MPI && !(defined _NO_COMM))
xchange_field(k, ieo);
# endif
if(k == l){
printf("Error in H_psi (simple.c):\n");
printf("Arguments k and l must be different\n");
printf("Program aborted\n");
exit(1);
}
if(ieo == 0){
ioff = 0;
}
else{
ioff = (VOLUME+RAND)/2;
}
ioff2 = (VOLUME+RAND)/2-ioff;
/**************** loop over all lattice sites ****************/
for (icx = ioff; icx < (VOLUME/2 + ioff); icx++){
ix=g_eo2lexic[icx];
r=l+(icx-ioff);
/*********************** direction +0 ************************/
iy=g_iup[ix][0]; icy=g_lexic2eosub[iy];
sp=k+icy;
# if ((defined _GAUGE_COPY))
up=&g_gauge_field_copy[icx][0];
# else
up=&g_gauge_field[ix][0];
# endif
_vector_add(psi,(*sp).s0,(*sp).s2);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka0,chi);
_vector_assign(temp.s0,psi);
_vector_assign(temp.s2,psi);
_vector_add(psi,(*sp).s1,(*sp).s3);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka0,chi);
_vector_assign(temp.s1,psi);
_vector_assign(temp.s3,psi);
/*********************** direction -0 ************************/
iy=g_idn[ix][0]; icy=g_lexic2eosub[iy];
sm=k+icy;
# if ((defined _GAUGE_COPY))
um = up+1;
# else
um=&g_gauge_field[iy][0];
# endif
_vector_sub(psi,(*sm).s0,(*sm).s2);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka0,chi);
_vector_add_assign(temp.s0,psi);
_vector_sub_assign(temp.s2,psi);
_vector_sub(psi,(*sm).s1,(*sm).s3);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka0,chi);
_vector_add_assign(temp.s1,psi);
_vector_sub_assign(temp.s3,psi);
/*********************** direction +1 ************************/
iy=g_iup[ix][1]; icy=g_lexic2eosub[iy];
sp=k+icy;
# if ((defined _GAUGE_COPY))
up=um+1;
# else
up+=1;
# endif
_vector_i_add(psi,(*sp).s0,(*sp).s3);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka1,chi);
_vector_add_assign(temp.s0,psi);
_vector_i_sub_assign(temp.s3,psi);
_vector_i_add(psi,(*sp).s1,(*sp).s2);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka1,chi);
_vector_add_assign(temp.s1,psi);
_vector_i_sub_assign(temp.s2,psi);
/*********************** direction -1 ************************/
iy=g_idn[ix][1]; icy=g_lexic2eosub[iy];
sm=k+icy;
# ifndef _GAUGE_COPY
um=&g_gauge_field[iy][1];
# else
um=up+1;
# endif
_vector_i_sub(psi,(*sm).s0,(*sm).s3);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka1,chi);
_vector_add_assign(temp.s0,psi);
_vector_i_add_assign(temp.s3,psi);
_vector_i_sub(psi,(*sm).s1,(*sm).s2);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka1,chi);
_vector_add_assign(temp.s1,psi);
_vector_i_add_assign(temp.s2,psi);
/*********************** direction +2 ************************/
iy=g_iup[ix][2]; icy=g_lexic2eosub[iy];
sp=k+icy;
# if ((defined _GAUGE_COPY))
up=um+1;
# else
up+=1;
# endif
_vector_add(psi,(*sp).s0,(*sp).s3);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka2,chi);
_vector_add_assign(temp.s0,psi);
_vector_add_assign(temp.s3,psi);
_vector_sub(psi,(*sp).s1,(*sp).s2);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka2,chi);
_vector_add_assign(temp.s1,psi);
_vector_sub_assign(temp.s2,psi);
/*********************** direction -2 ************************/
iy=g_idn[ix][2]; icy=g_lexic2eosub[iy];
sm=k+icy;
# ifndef _GAUGE_COPY
um = &g_gauge_field[iy][2];
# else
um = up +1;
# endif
_vector_sub(psi,(*sm).s0,(*sm).s3);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka2,chi);
_vector_add_assign(temp.s0,psi);
_vector_sub_assign(temp.s3,psi);
_vector_add(psi,(*sm).s1,(*sm).s2);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka2,chi);
_vector_add_assign(temp.s1,psi);
_vector_add_assign(temp.s2,psi);
/*********************** direction +3 ************************/
iy=g_iup[ix][3]; icy=g_lexic2eosub[iy];
sp=k+icy;
# if ((defined _GAUGE_COPY))
up=um+1;
# else
up+=1;
# endif
_vector_i_add(psi,(*sp).s0,(*sp).s2);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka3,chi);
_vector_add_assign(temp.s0,psi);
_vector_i_sub_assign(temp.s2,psi);
_vector_i_sub(psi,(*sp).s1,(*sp).s3);
_su3_multiply(chi,(*up),psi);
_complex_times_vector(psi,ka3,chi);
_vector_add_assign(temp.s1,psi);
_vector_i_add_assign(temp.s3,psi);
/*********************** direction -3 ************************/
iy=g_idn[ix][3]; icy=g_lexic2eosub[iy];
sm=k+icy;
# ifndef _GAUGE_COPY
um = &g_gauge_field[iy][3];
# else
um = up+1;
# endif
_vector_i_sub(psi,(*sm).s0,(*sm).s2);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka3,chi);
_vector_add((*r).s0, temp.s0, psi);
_vector_i_add((*r).s2, temp.s2, psi);
_vector_i_add(psi,(*sm).s1,(*sm).s3);
_su3_inverse_multiply(chi,(*um),psi);
_complexcjg_times_vector(psi,ka3,chi);
_vector_add((*r).s1, temp.s1, psi);
_vector_i_sub((*r).s3, temp.s3, psi);
/************************ end of loop ************************/
}
}
int main(int argc,char *argv[]) {
FILE *parameterfile=NULL;
int c, j, is=0, ic=0;
int x, X, y, Y, z, Z, t, tt, i, sum;
char * filename = NULL;
char datafilename[50];
char parameterfilename[50];
char conf_filename[50];
char * input_filename = NULL;
double plaquette_energy, nrm;
double * norm;
struct stout_parameters params_smear;
#ifdef _GAUGE_COPY
int kb=0;
#endif
#ifdef MPI
double atime=0., etime=0.;
#endif
#ifdef _KOJAK_INST
#pragma pomp inst init
#pragma pomp inst begin(main)
#endif
DUM_DERI = 6;
/* DUM_DERI + 2 is enough (not 7) */
DUM_SOLVER = DUM_DERI+2;
DUM_MATRIX = DUM_SOLVER+6;
/* DUM_MATRIX + 2 is enough (not 6) */
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
verbose = 0;
g_use_clover_flag = 0;
g_nr_of_psf = 1;
#ifdef MPI
MPI_Init(&argc, &argv);
#endif
while ((c = getopt(argc, argv, "h?f:o:")) != -1) {
switch (c) {
case 'f':
input_filename = calloc(200, sizeof(char));
strcpy(input_filename,optarg);
break;
case 'o':
filename = calloc(200, sizeof(char));
strcpy(filename,optarg);
break;
case 'h':
case '?':
default:
usage();
break;
}
}
if(input_filename == NULL){
input_filename = "hmc.input";
}
if(filename == NULL){
filename = "output";
}
/* Read the input file */
read_input(input_filename);
/* here we want no even/odd preconditioning */
even_odd_flag = 0;
/* this DBW2 stuff is not needed for the inversion ! */
g_rgi_C1 = 0;
if(Nsave == 0){
Nsave = 1;
}
tmlqcd_mpi_init(argc, argv);
g_dbw2rand = 0;
#ifndef MPI
g_dbw2rand = 0;
#endif
#ifdef _GAUGE_COPY
j = init_gauge_field(VOLUMEPLUSRAND, 1);
#else
j = init_gauge_field(VOLUMEPLUSRAND, 0);
#endif
if ( j!= 0) {
fprintf(stderr, "Not enough memory for gauge_fields! Aborting...\n");
exit(-1);
}
j = init_geometry_indices(VOLUMEPLUSRAND);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for geometry indices! Aborting...\n");
exit(-1);
}
if(even_odd_flag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, NO_OF_SPINORFIELDS);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS);
}
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(-1);
}
g_mu = g_mu1;
if(g_proc_id == 0){
/*construct the filenames for the observables and the parameters*/
strcpy(datafilename,filename); strcat(datafilename,".data");
strcpy(parameterfilename,filename); strcat(parameterfilename,".para");
parameterfile=fopen(parameterfilename, "w");
write_first_messages(parameterfile, 0, 1);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary();
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for halffield! Aborting...\n");
exit(-1);
}
if(g_sloppy_precision_flag == 1) {
j = init_dirac_halfspinor32();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for 32-Bit halffield! Aborting...\n");
exit(-1);
}
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
norm = (double*)calloc(3.*LX/2.+T/2., sizeof(double));
for(j=0;j<Nmeas; j++) {
sprintf(conf_filename,"%s.%.4d", gauge_input_filename, nstore);
if (g_proc_id == 0){
printf("Reading Gauge field from file %s\n", conf_filename); fflush(stdout);
}
read_lime_gauge_field(conf_filename);
if (g_proc_id == 0){
printf("done!\n"); fflush(stdout);
}
#ifdef MPI
xchange_gauge();
#endif
#ifdef _GAUGE_COPY
update_backward_gauge();
#endif
/* Compute minimal eigenvalues, if wanted */
if(compute_evs != 0) {
eigenvalues(&no_eigenvalues, 1000, eigenvalue_precision, 0, compute_evs, nstore, even_odd_flag);
}
/*compute the energy of the gauge field*/
plaquette_energy = measure_gauge_action();
if(g_proc_id == 0) {
printf("The plaquette value is %e\n", plaquette_energy/(6.*VOLUME*g_nproc)); fflush(stdout);
}
if (use_stout_flag == 1){
params_smear.rho = stout_rho;
params_smear.iterations = stout_no_iter;
if (stout_smear((su3_tuple*)(g_gauge_field[0]), ¶ms_smear, (su3_tuple*)(g_gauge_field[0])) != 0)
exit(1) ;
g_update_gauge_copy = 1;
g_update_gauge_energy = 1;
g_update_rectangle_energy = 1;
plaquette_energy = measure_gauge_action();
if (g_proc_id == 0) {
printf("# The plaquette value after stouting is %e\n", plaquette_energy / (6.*VOLUME*g_nproc));
fflush(stdout);
}
}
source_spinor_field(g_spinor_field[0], g_spinor_field[1], 0, 0);
convert_eo_to_lexic(g_spinor_field[DUM_DERI], g_spinor_field[0], g_spinor_field[1]);
D_psi(g_spinor_field[DUM_DERI+1], g_spinor_field[DUM_DERI]);
if(even_odd_flag) {
i = invert_eo(g_spinor_field[2], g_spinor_field[3], g_spinor_field[0], g_spinor_field[1],
solver_precision, max_solver_iterations, solver_flag, g_relative_precision_flag,
sub_evs_cg_flag, even_odd_flag);
convert_eo_to_lexic(g_spinor_field[DUM_DERI+1], g_spinor_field[2], g_spinor_field[3]);
}
for(i = 0; i < 3*LX/2+T/2; i++){
norm[i] = 0.;
}
for(x = 0; x < LX; x++){
if(x > LX/2) X = LX-x;
else X = x;
for(y = 0; y < LY; y++){
if(y > LY/2) Y = LY-y;
else Y = y;
for(z = 0; z < LZ; z++){
if(z > LZ/2) Z = LZ-z;
else Z = z;
for(t = 0; t < T; t++){
if(t > T/2) tt = T - t;
else tt = t;
sum = X + Y + Z + tt;
_spinor_norm_sq(nrm, g_spinor_field[DUM_DERI+1][ g_ipt[t][x][y][z] ]);
/* _spinor_norm_sq(nrm, qprop[0][0][1][ g_ipt[t][x][y][z] ]); */
printf("%e %e\n", g_spinor_field[DUM_DERI+1][ g_ipt[t][x][y][z] ].s0.c0.re, g_spinor_field[DUM_DERI+1][ g_ipt[t][x][y][z] ].s0.c0.im);
nrm = sqrt( nrm );
printf("%1.12e\n", nrm);
if(nrm > norm[sum]) norm[sum] = nrm;
}
}
}
}
for(i = 0; i < 3*L/2+T/2; i++){
printf("%d %1.12e\n", i, norm[i]);
}
printf("\n");
nstore+=Nsave;
}
#ifdef MPI
MPI_Finalize();
#endif
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
return(0);
#ifdef _KOJAK_INST
#pragma pomp inst end(main)
#endif
}
/* input on k; output on l */
void Hopping_Matrix(const int ieo, spinor * const l, spinor * const k){
int icx,icz,ioff;
int ix,iz;
int x0,icx0,jj;
su3 *restrict up;
su3 * restrict um;
spinor * restrict sp;
spinor * restrict sm;
spinor * restrict rn;
# if (defined MPI)
# ifdef PARALLELX
# define REQC 4
# elif defined PARALLELXY
# define REQC 8
# elif defined PARALLELXYZ
# define REQC 12
# endif
MPI_Request requests[REQC];
MPI_Status status[REQC];
# endif
#ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge(g_gauge_field);
}
#endif
if(ieo == 0){ /* even out - odd in */
ioff = 0;
}
else{ /* odd out - even in */
ioff = (VOLUME+RAND)/2;
}
/* Loop over time direction. This is the outmost loop */
for(x0=0;x0<T;x0++){
/* start the communication of the timslice borders (non-blocking send and receive)*/
# if (defined MPI && !defined _NO_COMM)
xchange_field_open(k, ieo, x0, requests, status);
# endif
/* loop over timeslice. At: contribution of timelike links */
icx0=g_1st_eot[x0][ieo];
jj =0;
um=&g_gauge_field_copyt[icx0][0]-1; /* allowed? */
for(icx = icx0; icx < icx0+TEOSLICE; icx++){
rn=l+(icx-ioff);
/*********************** direction +0 ************************/
sp=k+g_iup_eo[icx][0]; /* all sp,sm,up,um could be moved up */
up=um+1;
#if (defined AVX)
_avx_load(sp->s0);
_avx_load_up(sp->s2);
_avx_vector_add();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka0);
_avx_store_up(rn->s0);
_avx_store_up(rn->s2);
_avx_load(sp->s1);
_avx_load_up(sp->s3);
_avx_vector_add();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka0);
_avx_store_up(rn->s1);
_avx_store_up(rn->s3);
/*********************** direction -0 ************************/
sm=k+g_idn_eo[icx][0];
um=up+1;
_avx_load(sm->s0);
_avx_load_up(sm->s2);
_avx_vector_sub();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka0);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s2);
_avx_vector_sub();
_avx_store(rn->s2);
_avx_load(sm->s1);
_avx_load_up(sm->s3);
_avx_vector_sub();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka0);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s3);
_avx_vector_sub();
_avx_store(rn->s3);
#elif (defined SSE2 || defined SSE3)
_sse_load(sp->s0);
_sse_load_up(sp->s2);
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka0);
_sse_store_up(rn->s0);
_sse_store_up(rn->s2);
_sse_load(sp->s1);
_sse_load_up(sp->s3);
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka0);
_sse_store_up(rn->s1);
_sse_store_up(rn->s3);
/*********************** direction -0 ************************/
sm=k+g_idn_eo[icx][0];
um=up+1;
_sse_load(sm->s0);
_sse_load_up(sm->s2);
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka0);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s2);
_sse_vector_sub();
_sse_store(rn->s2);
_sse_load(sm->s1);
_sse_load_up(sm->s3);
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka0);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s3);
_sse_vector_sub();
_sse_store(rn->s3);
#endif
jj++;
} /* end of loop over timeslice (At)*/
/* complete the communication of the timslice borders (and wait) */
#if (defined MPI && !defined _NO_COMM)
xchange_field_close(requests, status, REQC); /* MPI_Waitall */
#endif
/* loop over timeslice. Bt: contribution of spacelike links */
um=&g_gauge_field_copys[icx0][0]-1;
for(icx = icx0; icx < icx0+TEOSLICE; icx++){
ix=g_eo2lexic[icx];
rn=l+(icx-ioff);
/*********************** direction +1 ************************/
sp=k+g_iup_eo[icx][1];
up=um+1;
#if (defined AVX)
_avx_load(sp->s0);
_avx_load_up(sp->s3);
_avx_vector_i_mul();
_avx_vector_add();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka1);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s3);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_store(rn->s3);
_avx_load(sp->s1);
_avx_load_up(sp->s2);
_avx_vector_i_mul();
_avx_vector_add();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka1);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s2);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_store(rn->s2);
/*********************** direction -1 ************************/
sm=k+g_idn_eo[icx][1];
um=up+1;
_avx_load(sm->s0);
_avx_load_up(sm->s3);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka1);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s3);
_avx_vector_i_mul();
_avx_vector_add();
_avx_store(rn->s3);
_avx_load(sm->s1);
_avx_load_up(sm->s2);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka1);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s2);
_avx_vector_i_mul();
_avx_vector_add();
_avx_store(rn->s2);
/*********************** direction +2 ************************/
sp=k+g_iup_eo[icx][2];
up=um+1;
_avx_load(sp->s0);
_avx_load_up(sp->s3);
_avx_vector_add();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka2);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s3);
_avx_vector_add();
_avx_store(rn->s3);
_avx_load(sp->s1);
_avx_load_up(sp->s2);
_avx_vector_sub();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka2);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s2);
_avx_vector_sub();
_avx_store(rn->s2);
/*********************** direction -2 ************************/
sm=k+g_idn_eo[icx][2];
um=up+1;
_avx_load(sm->s0);
_avx_load_up(sm->s3);
_avx_vector_sub();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka2);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s3);
_avx_vector_sub();
_avx_store(rn->s3);
_avx_load(sm->s1);
_avx_load_up(sm->s2);
_avx_vector_add();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka2);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s2);
_avx_vector_add();
_avx_store(rn->s2);
/*********************** direction +3 ************************/
sp=k+g_iup_eo[icx][3];
up=um+1;
_avx_load(sp->s0);
_avx_load_up(sp->s2);
_avx_vector_i_mul();
_avx_vector_add();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka3);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s2);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_store(rn->s2);
_avx_load(sp->s1);
_avx_load_up(sp->s3);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_su3_multiply((*up));
_avx_vector_cmplx_mul(ka3);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s3);
_avx_vector_i_mul();
_avx_vector_add();
_avx_store(rn->s3);
/*********************** direction -3 ************************/
sm=k+g_idn_eo[icx][3];
um=up+1;
_avx_load(sm->s0);
_avx_load_up(sm->s2);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka3);
_avx_load(rn->s0);
_avx_vector_add();
_avx_store(rn->s0);
_avx_load(rn->s2);
_avx_vector_i_mul();
_avx_vector_add();
_avx_store(rn->s2);
_avx_load(sm->s1);
_avx_load_up(sm->s3);
_avx_vector_i_mul();
_avx_vector_add();
_avx_su3_inverse_multiply((*um));
_avx_vector_cmplxcg_mul(ka3);
_avx_load(rn->s1);
_avx_vector_add();
_avx_store(rn->s1);
_avx_load(rn->s3);
_avx_vector_i_mul();
_avx_vector_sub();
_avx_store(rn->s3);
#elif (defined SSE2 || defined SSE3)
_sse_load(sp->s0);
_sse_load_up(sp->s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka1);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rn->s3);
_sse_load(sp->s1);
_sse_load_up(sp->s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka1);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rn->s2);
/*********************** direction -1 ************************/
sm=k+g_idn_eo[icx][1];
um=up+1;
_sse_load(sm->s0);
_sse_load_up(sm->s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka1);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rn->s3);
_sse_load(sm->s1);
_sse_load_up(sm->s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka1);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rn->s2);
/*********************** direction +2 ************************/
sp=k+g_iup_eo[icx][2];
up=um+1;
_sse_load(sp->s0);
_sse_load_up(sp->s3);
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka2);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s3);
_sse_vector_add();
_sse_store(rn->s3);
_sse_load(sp->s1);
_sse_load_up(sp->s2);
_sse_vector_sub();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka2);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s2);
_sse_vector_sub();
_sse_store(rn->s2);
/*********************** direction -2 ************************/
sm=k+g_idn_eo[icx][2];
um=up+1;
_sse_load(sm->s0);
_sse_load_up(sm->s3);
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka2);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s3);
_sse_vector_sub();
_sse_store(rn->s3);
_sse_load(sm->s1);
_sse_load_up(sm->s2);
_sse_vector_add();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka2);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s2);
_sse_vector_add();
_sse_store(rn->s2);
/*********************** direction +3 ************************/
sp=k+g_iup_eo[icx][3];
up=um+1;
_sse_load(sp->s0);
_sse_load_up(sp->s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka3);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rn->s2);
_sse_load(sp->s1);
_sse_load_up(sp->s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_multiply((*up));
_sse_vector_cmplx_mul(ka3);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rn->s3);
/*********************** direction -3 ************************/
sm=k+g_idn_eo[icx][3];
um=up+1;
_sse_load(sm->s0);
_sse_load_up(sm->s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka3);
_sse_load(rn->s0);
_sse_vector_add();
_sse_store(rn->s0);
_sse_load(rn->s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rn->s2);
_sse_load(sm->s1);
_sse_load_up(sm->s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_inverse_multiply((*um));
_sse_vector_cmplxcg_mul(ka3);
_sse_load(rn->s1);
_sse_vector_add();
_sse_store(rn->s1);
_sse_load(rn->s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rn->s3);
#endif
} /* end of loop over timeslice (Bt)*/
} /* x0=0; x0<T */
}
void jdher_bi(int n, int lda, double tau, double tol,
int kmax, int jmax, int jmin, int itmax,
int blksize, int blkwise,
int V0dim, complex *V0,
int solver_flag,
int linitmax, double eps_tr, double toldecay,
int verbosity,
int *k_conv, complex *Q, double *lambda, int *it,
int maxmin, const int shift_mode,
matrix_mult_bi A_psi){
/****************************************************************************
* *
* Local variables *
* *
****************************************************************************/
/* constants */
/* allocatables:
* initialize with NULL, so we can free even unallocated ptrs */
double *s = NULL, *resnrm = NULL, *resnrm_old = NULL, *dtemp = NULL, *rwork = NULL;
complex *V_ = NULL, *V, *Vtmp = NULL, *U = NULL, *M = NULL, *Z = NULL,
*Res_ = NULL, *Res,
*eigwork = NULL, *temp1_ = NULL, *temp1;
int *idx1 = NULL, *idx2 = NULL,
*convind = NULL, *keepind = NULL, *solvestep = NULL,
*actcorrits = NULL;
/* non-allocated ptrs */
complex *q, *v, *u, *r = NULL;
/* complex *matdummy, *vecdummy; */
/* scalar vars */
double theta, alpha, it_tol;
int i, k, j, actblksize, eigworklen, found, conv, keep, n2, N = n*sizeof(complex)/sizeof(bispinor);
int act, cnt, idummy, info, CntCorrIts=0, endflag=0;
/* variables for random number generator */
int IDIST = 1;
int ISEED[4] = {2, 3, 5, 7};
ISEED[0] = g_proc_id;
/****************************************************************************
* *
* Of course on the CRAY everything is different :( !! *
* that's why we need something more.
* *
****************************************************************************/
#ifdef CRAY
fupl_u = _cptofcd(cupl_u, strlen(cupl_u));
fupl_c = _cptofcd(cupl_c, strlen(cupl_c));
fupl_n = _cptofcd(cupl_n, strlen(cupl_n));
fupl_a = _cptofcd(cupl_a, strlen(cupl_a));
fupl_v = _cptofcd(cupl_v, strlen(cupl_v));
filaenv = _cptofcd(cilaenv, strlen(cilaenv));
fvu = _cptofcd(cvu, strlen(cvu));
#endif
/****************************************************************************
* *
* Execution starts here... *
* *
****************************************************************************/
/* NEW PART FOR GAUGE_COPY */
#ifdef _GAUGE_COPY
update_backward_gauge();
#endif
/* END NEW PART */
/* print info header */
if (verbosity > 1 && g_proc_id == 0) {
printf("Jacobi-Davidson method for hermitian Matrices\n");
printf("Solving A*x = lambda*x \n\n");
printf(" N= %10d ITMAX=%4d\n", n, itmax);
printf(" KMAX=%3d JMIN=%3d JMAX=%3d V0DIM=%3d\n",
kmax, jmin, jmax, V0dim);
printf(" BLKSIZE= %2d BLKWISE= %5s\n",
blksize, blkwise ? "TRUE" : "FALSE");
printf(" TOL= %11.4e TAU= %11.4e\n",
tol, tau);
printf(" LINITMAX= %5d EPS_TR= %10.3e TOLDECAY=%9.2e\n",
linitmax, eps_tr, toldecay);
printf("\n Computing %s eigenvalues\n",
maxmin ? "maximal" : "minimal");
printf("\n");
fflush( stdout );
}
/* validate input parameters */
if(tol <= 0) jderrorhandler(401,"");
if(kmax <= 0 || kmax > n) jderrorhandler(402,"");
if(jmax <= 0 || jmax > n) jderrorhandler(403,"");
if(jmin <= 0 || jmin > jmax) jderrorhandler(404,"");
if(itmax < 0) jderrorhandler(405,"");
if(blksize > jmin || blksize > (jmax - jmin)) jderrorhandler(406,"");
if(blksize <= 0 || blksize > kmax) jderrorhandler(406,"");
if(blkwise < 0 || blkwise > 1) jderrorhandler(407,"");
if(V0dim < 0 || V0dim >= jmax) jderrorhandler(408,"");
if(linitmax < 0) jderrorhandler(409,"");
if(eps_tr < 0.) jderrorhandler(500,"");
if(toldecay <= 1.0) jderrorhandler(501,"");
CONE.re=1.; CONE.im=0.;
CZERO.re=0.; CZERO.im=0.;
CMONE.re=-1.; CMONE.im=0.;
/* Get hardware-dependent values:
* Opt size of workspace for ZHEEV is (NB+1)*j, where NB is the opt.
* block size... */
eigworklen = (2 + _FT(ilaenv)(&ONE, filaenv, fvu, &jmax, &MONE, &MONE, &MONE, 6, 2)) * jmax;
/* Allocating memory for matrices & vectors */
if((void*)(V_ = (complex *)malloc((lda * jmax + 4) * sizeof(complex))) == NULL) {
errno = 0;
jderrorhandler(300,"V in jdher_bi");
}
#if (defined SSE || defined SSE2 || defined SSE3)
V = (complex*)(((unsigned long int)(V_)+ALIGN_BASE)&~ALIGN_BASE);
#else
V = V_;
#endif
if((void*)(U = (complex *)malloc(jmax * jmax * sizeof(complex))) == NULL) {
jderrorhandler(300,"U in jdher_bi");
}
if((void*)(s = (double *)malloc(jmax * sizeof(double))) == NULL) {
jderrorhandler(300,"s in jdher_bi");
}
if((void*)(Res_ = (complex *)malloc((lda * blksize+4) * sizeof(complex))) == NULL) {
jderrorhandler(300,"Res in jdher_bi");
}
#if (defined SSE || defined SSE2 || defined SSE3)
Res = (complex*)(((unsigned long int)(Res_)+ALIGN_BASE)&~ALIGN_BASE);
#else
Res = Res_;
#endif
if((void*)(resnrm = (double *)malloc(blksize * sizeof(double))) == NULL) {
jderrorhandler(300,"resnrm in jdher_bi");
}
if((void*)(resnrm_old = (double *)calloc(blksize,sizeof(double))) == NULL) {
jderrorhandler(300,"resnrm_old in jdher_bi");
}
if((void*)(M = (complex *)malloc(jmax * jmax * sizeof(complex))) == NULL) {
jderrorhandler(300,"M in jdher_bi");
}
if((void*)(Vtmp = (complex *)malloc(jmax * jmax * sizeof(complex))) == NULL) {
jderrorhandler(300,"Vtmp in jdher_bi");
}
if((void*)(p_work_bi = (complex *)malloc(lda * sizeof(complex))) == NULL) {
jderrorhandler(300,"p_work_bi in jdher_bi");
}
/* ... */
if((void*)(idx1 = (int *)malloc(jmax * sizeof(int))) == NULL) {
jderrorhandler(300,"idx1 in jdher_bi");
}
if((void*)(idx2 = (int *)malloc(jmax * sizeof(int))) == NULL) {
jderrorhandler(300,"idx2 in jdher_bi");
}
/* Indices for (non-)converged approximations */
if((void*)(convind = (int *)malloc(blksize * sizeof(int))) == NULL) {
jderrorhandler(300,"convind in jdher_bi");
}
if((void*)(keepind = (int *)malloc(blksize * sizeof(int))) == NULL) {
jderrorhandler(300,"keepind in jdher_bi");
}
if((void*)(solvestep = (int *)malloc(blksize * sizeof(int))) == NULL) {
jderrorhandler(300,"solvestep in jdher_bi");
}
if((void*)(actcorrits = (int *)malloc(blksize * sizeof(int))) == NULL) {
jderrorhandler(300,"actcorrits in jdher_bi");
}
if((void*)(eigwork = (complex *)malloc(eigworklen * sizeof(complex))) == NULL) {
jderrorhandler(300,"eigwork in jdher_bi");
}
if((void*)(rwork = (double *)malloc(3*jmax * sizeof(double))) == NULL) {
jderrorhandler(300,"rwork in jdher_bi");
}
if((void*)(temp1_ = (complex *)malloc((lda+4) * sizeof(complex))) == NULL) {
jderrorhandler(300,"temp1 in jdher_bi");
}
#if (defined SSE || defined SSE2 || defined SSE3)
temp1 = (complex*)(((unsigned long int)(temp1_)+ALIGN_BASE)&~ALIGN_BASE);
#else
temp1 = temp1_;
#endif
if((void*)(dtemp = (double *)malloc(lda * sizeof(complex))) == NULL) {
jderrorhandler(300,"dtemp in jdher_bi");
}
/* Set variables for Projection routines */
n2 = 2*n;
p_n = n;
p_n2 = n2;
p_Q_bi = Q;
p_A_psi_bi = A_psi;
p_lda = lda;
/**************************************************************************
* *
* Generate initial search subspace V. Vectors are taken from V0 and if *
* necessary randomly generated. *
* *
**************************************************************************/
/* copy V0 to V */
_FT(zlacpy)(fupl_a, &n, &V0dim, V0, &lda, V, &lda, 1);
j = V0dim;
/* if V0dim < blksize: generate additional random vectors */
if (V0dim < blksize) {
idummy = (blksize - V0dim)*n; /* nof random numbers */
_FT(zlarnv)(&IDIST, ISEED, &idummy, V + V0dim*lda);
j = blksize;
}
for (cnt = 0; cnt < j; cnt ++) {
ModifiedGS_bi(V + cnt*lda, n, cnt, V, lda);
alpha = sqrt(square_norm_bi((bispinor*)(V+cnt*lda), N));
alpha = 1.0 / alpha;
_FT(dscal)(&n2, &alpha, (double *)(V + cnt*lda), &ONE);
}
/* Generate interaction matrix M = V^dagger*A*V. Only the upper triangle
is computed. */
for (cnt = 0; cnt < j; cnt++){
A_psi((bispinor*) temp1, (bispinor*) (V+cnt*lda));
idummy = cnt+1;
for(i = 0; i < idummy; i++) {
M[cnt*jmax+i] = scalar_prod_bi((bispinor*)(V+i*lda), (bispinor*) temp1, N);
}
}
/* Other initializations */
k = 0; (*it) = 0;
if((*k_conv) > 0) {
k = (*k_conv);
}
actblksize = blksize;
for(act = 0; act < blksize; act ++){
solvestep[act] = 1;
}
/****************************************************************************
* *
* Main JD-iteration loop *
* *
****************************************************************************/
while((*it) < itmax) {
/****************************************************************************
* *
* Solving the projected eigenproblem *
* *
* M*u = V^dagger*A*V*u = s*u *
* M is hermitian, only the upper triangle is stored *
* *
****************************************************************************/
_FT(zlacpy)(fupl_u, &j, &j, M, &jmax, U, &jmax, 1);
_FT(zheev)(fupl_v, fupl_u, &j, U, &jmax, s, eigwork, &eigworklen, rwork, &info, 1, 1);
if (info != 0) {
printf("error solving the projected eigenproblem.");
printf(" zheev: info = %d\n", info);
}
if(info != 0) jderrorhandler(502,"problem in zheev for jdher_bi");
/* Reverse order of eigenvalues if maximal value is needed */
if(maxmin == 1){
sorteig(j, s, U, jmax, s[j-1], dtemp, idx1, idx2, 0);
}
else{
sorteig(j, s, U, jmax, 0., dtemp, idx1, idx2, 0);
}
/****************************************************************************
* *
* Convergence/Restart Check *
* *
* In case of convergence, strip off a whole block or just the converged *
* ones and put 'em into Q. Update the matrices Q, V, U, s *
* *
* In case of a restart update the V, U and M matrices and recompute the *
* Eigenvectors *
* *
****************************************************************************/
found = 1;
while(found) {
/* conv/keep = Number of converged/non-converged Approximations */
conv = 0; keep = 0;
for(act=0; act < actblksize; act++){
/* Setting pointers for single vectors */
q = Q + (act+k)*lda;
u = U + act*jmax;
r = Res + act*lda;
/* Compute Ritz-Vector Q[:,k+cnt1]=V*U[:,cnt1] */
theta = s[act];
_FT(zgemv)(fupl_n, &n, &j, &CONE, V, &lda, u, &ONE, &CZERO, q, &ONE, 1);
/* Compute the residual */
A_psi((bispinor*) r, (bispinor*) q);
theta = -theta;
_FT(daxpy)(&n2, &theta, (double*) q, &ONE, (double*) r, &ONE);
/* Compute norm of the residual and update arrays convind/keepind*/
resnrm_old[act] = resnrm[act];
resnrm[act] = sqrt(square_norm_bi((bispinor*) r, N));
if (resnrm[act] < tol){
convind[conv] = act;
conv = conv + 1;
}
else{
keepind[keep] = act;
keep = keep + 1;
}
} /* for(act = 0; act < actblksize; act ++) */
/* Check whether the blkwise-mode is chosen and ALL the
approximations converged, or whether the strip-off mode is
active and SOME of the approximations converged */
found = ((blkwise==1 && conv==actblksize) || (blkwise==0 && conv!=0))
&& (j > actblksize || k == kmax - actblksize);
/***************************************************************************
* *
* Convergence Case *
* *
* In case of convergence, strip off a whole block or just the converged *
* ones and put 'em into Q. Update the matrices Q, V, U, s *
* *
**************************************************************************/
if (found) {
/* Store Eigenvalues */
for(act = 0; act < conv; act++)
lambda[k+act] = s[convind[act]];
/* Re-use non approximated Ritz-Values */
for(act = 0; act < keep; act++)
s[act] = s[keepind[act]];
/* Shift the others in the right position */
for(act = 0; act < (j-actblksize); act ++)
s[act+keep] = s[act+actblksize];
/* Update V. Re-use the V-Vectors not looked at yet. */
idummy = j - actblksize;
for (act = 0; act < n; act = act + jmax) {
cnt = act + jmax > n ? n-act : jmax;
_FT(zlacpy)(fupl_a, &cnt, &j, V+act, &lda, Vtmp, &jmax, 1);
_FT(zgemm)(fupl_n, fupl_n, &cnt, &idummy, &j, &CONE, Vtmp,
&jmax, U+actblksize*jmax, &jmax, &CZERO, V+act+keep*lda, &lda, 1, 1);
}
/* Insert the not converged approximations as first columns in V */
for(act = 0; act < keep; act++){
_FT(zlacpy)(fupl_a,&n,&ONE,Q+(k+keepind[act])*lda,&lda,V+act*lda,&lda,1);
}
/* Store Eigenvectors */
for(act = 0; act < conv; act++){
_FT(zlacpy)(fupl_a,&n,&ONE,Q+(k+convind[act])*lda,&lda,Q+(k+act)*lda,&lda,1);
}
/* Update SearchSpaceSize j */
j = j - conv;
/* Let M become a diagonalmatrix with the Ritzvalues as entries ... */
_FT(zlaset)(fupl_u, &j, &j, &CZERO, &CZERO, M, &jmax, 1);
for (act = 0; act < j; act++){
M[act*jmax + act].re = s[act];
}
/* ... and U the Identity(jnew,jnew) */
_FT(zlaset)(fupl_a, &j, &j, &CZERO, &CONE, U, &jmax, 1);
if(shift_mode == 1){
if(maxmin == 0){
for(act = 0; act < conv; act ++){
if (lambda[k+act] > tau){
tau = lambda[k+act];
}
}
}
else{
for(act = 0; act < conv; act ++){
if (lambda[k+act] < tau){
tau = lambda[k+act];
}
}
}
}
/* Update Converged-Eigenpair-counter and Pro_k */
k = k + conv;
/* Update the new blocksize */
actblksize=min(blksize, kmax-k);
/* Exit main iteration loop when kmax eigenpairs have been
approximated */
if (k == kmax){
endflag = 1;
break;
}
/* Counter for the linear-solver-accuracy */
for(act = 0; act < keep; act++)
solvestep[act] = solvestep[keepind[act]];
/* Now we expect to have the next eigenvalues */
/* allready with some accuracy */
/* So we do not need to start from scratch... */
for(act = keep; act < blksize; act++)
solvestep[act] = 1;
} /* if(found) */
if(endflag == 1){
break;
}
/**************************************************************************
* *
* Restart *
* *
* The Eigenvector-Aproximations corresponding to the first jmin *
* Petrov-Vectors are kept. if (j+actblksize > jmax) { *
* *
**************************************************************************/
if (j+actblksize > jmax) {
idummy = j; j = jmin;
for (act = 0; act < n; act = act + jmax) { /* V = V * U(:,1:j) */
cnt = act+jmax > n ? n-act : jmax;
_FT(zlacpy)(fupl_a, &cnt, &idummy, V+act, &lda, Vtmp, &jmax, 1);
_FT(zgemm)(fupl_n, fupl_n, &cnt, &j, &idummy, &CONE, Vtmp,
&jmax, U, &jmax, &CZERO, V+act, &lda, 1, 1);
}
_FT(zlaset)(fupl_a, &j, &j, &CZERO, &CONE, U, &jmax, 1);
_FT(zlaset)(fupl_u, &j, &j, &CZERO, &CZERO, M, &jmax, 1);
for (act = 0; act < j; act++)
M[act*jmax + act].re = s[act];
}
} /* while(found) */
if(endflag == 1){
break;
}
/****************************************************************************
* *
* Solving the correction equations *
* *
* *
****************************************************************************/
/* Solve actblksize times the correction equation ... */
for (act = 0; act < actblksize; act ++) {
/* Setting start-value for vector v as zeros(n,1). Guarantees
orthogonality */
v = V + j*lda;
for (cnt = 0; cnt < n; cnt ++){
v[cnt].re = 0.;
v[cnt].im = 0.;
}
/* Adaptive accuracy and shift for the lin.solver. In case the
residual is big, we don't need a too precise solution for the
correction equation, since even in exact arithmetic the
solution wouldn't be too usefull for the Eigenproblem. */
r = Res + act*lda;
if (resnrm[act] < eps_tr && resnrm[act] < s[act] && resnrm_old[act] > resnrm[act]){
p_theta = s[act];
}
else{
p_theta = tau;
}
p_k = k + actblksize;
/* if we are in blockwise mode, we do not want to */
/* iterate solutions much more, if they have */
/* allready the desired precision */
if(blkwise == 1 && resnrm[act] < tol) {
it_tol = pow(toldecay, (double)(-5));
}
else {
it_tol = pow(toldecay, (double)(-solvestep[act]));
}
solvestep[act] = solvestep[act] + 1;
/* equation and project if necessary */
ModifiedGS_bi(r, n, k + actblksize, Q, lda);
/* for(i=0;i<n;i++){ */
/* r[i].re*=-1.; */
/* r[i].im*=-1.; */
/* } */
g_sloppy_precision = 1;
/* Solve the correction equation ... */
if (solver_flag == BICGSTAB){
info = bicgstab_complex_bi((bispinor*) v, (bispinor*) r, linitmax,
it_tol*it_tol, g_relative_precision_flag, VOLUME/2, &Proj_A_psi_bi);
}
else if(solver_flag == CG){
info = cg_her_bi((bispinor*) v, (bispinor*) r, linitmax,
it_tol*it_tol, g_relative_precision_flag, VOLUME/2, &Proj_A_psi_bi);
}
else{
info = bicgstab_complex_bi((bispinor*) v, (bispinor*) r, linitmax,
it_tol*it_tol, g_relative_precision_flag, VOLUME/2, &Proj_A_psi_bi);
}
g_sloppy_precision = 0;
/* Actualizing profiling data */
if (info == -1){
CntCorrIts += linitmax;
}
else{
CntCorrIts += info;
}
actcorrits[act] = info;
/* orthonormalize v to Q, cause the implicit
orthogonalization in the solvers may be too inaccurate. Then
apply "IteratedCGS" to prevent numerical breakdown
in order to orthogonalize v to V */
ModifiedGS_bi(v, n, k+actblksize, Q, lda);
IteratedClassicalGS_bi(v, &alpha, n, j, V, temp1, lda);
alpha = 1.0 / alpha;
_FT(dscal)(&n2, &alpha, (double*) v, &ONE);
/* update interaction matrix M */
A_psi((bispinor*) temp1, (bispinor*) v);
idummy = j+1;
for(i = 0; i < idummy; i++){
M[j*jmax+i] = scalar_prod_bi((bispinor*) (V+i*lda), (bispinor*) temp1, N);
}
/* Increasing SearchSpaceSize j */
j ++;
} /* for (act = 0;act < actblksize; act ++) */
/* Print information line */
if(g_proc_id == 0) {
print_status(verbosity, *it, k, j - blksize, kmax, blksize, actblksize,
s, resnrm, actcorrits);
}
/* Increase iteration-counter for outer loop */
(*it) = (*it) + 1;
} /* Main iteration loop */
/******************************************************************
* *
* Eigensolutions converged or iteration limit reached *
* *
* Print statistics. Free memory. Return. *
* *
******************************************************************/
*k_conv = k;
if (verbosity >= 1) {
if(g_proc_id == 0) {
printf("\nJDHER execution statistics\n\n");
printf("IT_OUTER=%d IT_INNER_TOT=%d IT_INNER_AVG=%8.2f\n",
(*it), CntCorrIts, (double)CntCorrIts/(*it));
printf("\nConverged eigensolutions in order of convergence:\n");
printf("\n I LAMBDA(I) RES(I)\n");
printf("---------------------------------------\n");
}
for (act = 0; act < *k_conv; act ++) {
/* Compute the residual for solution act */
q = Q + act*lda;
theta = -lambda[act];
A_psi((bispinor*) r, (bispinor*) q);
_FT(daxpy)(&n2, &theta, (double*) q, &ONE, (double*) r, &ONE);
alpha = sqrt(square_norm_bi((bispinor*) r, N));
if(g_proc_id == 0) {
printf("%3d %22.15e %12.5e\n", act+1, lambda[act],
alpha);
}
}
if(g_proc_id == 0) {
printf("\n");
fflush( stdout );
}
}
free(V_); free(Vtmp); free(U);
free(s); free(Res_);
free(resnrm); free(resnrm_old);
free(M); free(Z);
free(eigwork); free(temp1_);
free(dtemp); free(rwork);
free(p_work_bi);
free(idx1); free(idx2);
free(convind); free(keepind); free(solvestep); free(actcorrits);
} /* jdher(.....) */
void Hopping_Matrix(const int ieo, spinor * const l, spinor * const k){
int ix, i;
su3 * restrict U ALIGN;
static spinor rs;
spinor * restrict s ALIGN;
halfspinor ** phi ALIGN;
#if defined OPTERON
const int predist=2;
#else
const int predist=1;
#endif
#ifdef _KOJAK_INST
#pragma pomp inst begin(hoppingmatrix)
#endif
#ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge();
}
#endif
/* We will run through the source vector now */
/* instead of the solution vector */
s = k;
_prefetch_spinor(s);
if(ieo == 0) {
U = g_gauge_field_copy[0][0];
}
else {
U = g_gauge_field_copy[1][0];
}
phi = NBPointer[ieo];
_prefetch_su3(U);
/**************** loop over all lattice sites ******************/
ix=0;
for(i = 0; i < (VOLUME)/2; i++){
/*********************** direction +0 ************************/
_prefetch_su3(U+predist);
_sse_load((*s).s0);
_sse_load_up((*s).s2);
_sse_vector_add();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka0);
_sse_store_nt_up((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s3);
_sse_vector_add();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka0);
_sse_store_nt_up((*phi[ix]).s1);
U++;
ix++;
/*********************** direction -0 ************************/
_sse_load((*s).s0);
_sse_load_up((*s).s2);
_sse_vector_sub();
_sse_store_nt((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s3);
_sse_vector_sub();
_sse_store_nt((*phi[ix]).s1);
ix++;
/*********************** direction +1 ************************/
_prefetch_su3(U+predist);
_sse_load((*s).s0);
/*next not needed?*/
_sse_load_up((*s).s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka1);
_sse_store_nt_up((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka1);
_sse_store_nt_up((*phi[ix]).s1);
ix++;
U++;
/*********************** direction -1 ************************/
_sse_load((*s).s0);
_sse_load_up((*s).s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store_nt((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store_nt((*phi[ix]).s1);
ix++;
/*********************** direction +2 ************************/
_prefetch_su3(U+predist);
_sse_load((*s).s0);
_sse_load_up((*s).s3);
_sse_vector_add();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka2);
_sse_store_nt_up((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s2);
_sse_vector_sub();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka2);
_sse_store_nt_up((*phi[ix]).s1);
ix++;
U++;
/*********************** direction -2 ************************/
_sse_load((*s).s0);
_sse_load_up((*s).s3);
_sse_vector_sub();
_sse_store_nt((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s2);
_sse_vector_add();
_sse_store_nt((*phi[ix]).s1);
ix++;
/*********************** direction +3 ************************/
_prefetch_su3(U+predist);
_prefetch_spinor(s+1);
_sse_load((*s).s0);
_sse_load_up((*s).s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka3);
_sse_store_nt_up((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_su3_multiply((*U));
_sse_vector_cmplx_mul(ka3);
_sse_store_nt_up((*phi[ix]).s1);
ix++;
U++;
/*********************** direction -3 ************************/
_sse_load((*s).s0);
_sse_load_up((*s).s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store_nt((*phi[ix]).s0);
_sse_load((*s).s1);
_sse_load_up((*s).s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store_nt((*phi[ix]).s1);
ix++;
s++;
}
# if (defined MPI && !defined _NO_COMM)
xchange_halffield();
# endif
s = l;
phi = NBPointer[2 + ieo];
if(ieo == 0) {
U = g_gauge_field_copy[1][0];
}
else {
U = g_gauge_field_copy[0][0];
}
_prefetch_su3(U);
/* Now we sum up and expand to a full spinor */
ix = 0;
for(i = 0; i < (VOLUME)/2; i++){
/*********************** direction +0 ************************/
_vector_assign(rs.s0, (*phi[ix]).s0);
_vector_assign(rs.s2, (*phi[ix]).s0);
_vector_assign(rs.s1, (*phi[ix]).s1);
_vector_assign(rs.s3, (*phi[ix]).s1);
ix++;
/*********************** direction -0 ************************/
_prefetch_su3(U+predist);
_sse_load((*phi[ix]).s0);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka0);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s2);
_sse_vector_sub();
_sse_store(rs.s2);
_sse_load((*phi[ix]).s1);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka0);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s3);
_sse_vector_sub();
_sse_store(rs.s3);
ix++;
U++;
/*********************** direction +1 ************************/
_sse_load_up((*phi[ix]).s0);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rs.s3);
_sse_load_up((*phi[ix]).s1);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rs.s2);
ix++;
/*********************** direction -1 ************************/
_prefetch_su3(U+predist);
_sse_load((*phi[ix]).s0);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka1);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rs.s3);
_sse_load((*phi[ix]).s1);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka1);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rs.s2);
ix++;
U++;
/*********************** direction +2 ************************/
_sse_load_up((*phi[ix]).s0);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_add();
_sse_store(rs.s3);
_sse_load_up((*phi[ix]).s1);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_sub();
_sse_store(rs.s2);
ix++;
/*********************** direction -2 ************************/
_prefetch_su3(U+predist);
_sse_load((*phi[ix]).s0);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka2);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s3);
_sse_vector_sub();
_sse_store(rs.s3);
_sse_load((*phi[ix]).s1);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka2);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s2);
_sse_vector_add();
_sse_store(rs.s2);
ix++;
U++;
/*********************** direction +3 ************************/
_sse_load_up((*phi[ix]).s0);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store(rs.s0);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store(rs.s2);
_sse_load_up((*phi[ix]).s1);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store(rs.s1);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store(rs.s3);
ix++;
/*********************** direction -3 ************************/
_prefetch_su3(U+predist);
_prefetch_spinor(s+1);
_sse_load((*phi[ix]).s0);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka3);
_sse_load(rs.s0);
_sse_vector_add();
_sse_store_nt((*s).s0);
_sse_load(rs.s2);
_sse_vector_i_mul();
_sse_vector_add();
_sse_store_nt((*s).s2);
_sse_load((*phi[ix]).s1);
_sse_su3_inverse_multiply((*U));
_sse_vector_cmplxcg_mul(ka3);
_sse_load(rs.s1);
_sse_vector_add();
_sse_store_nt((*s).s1);
_sse_load(rs.s3);
_sse_vector_i_mul();
_sse_vector_sub();
_sse_store_nt((*s).s3);
ix++;
U++;
s++;
}
#ifdef _KOJAK_INST
#pragma pomp inst end(hoppingmatrix)
#endif
}
/* for ieo=0, k resides on odd sites and l on even sites */
void Hopping_Matrix(const int ieo, spinor * const l, spinor * const k){
int i,ix;
su3 * restrict U ALIGN;
spinor * restrict s ALIGN;
spinor rs;
static su3_vector psi, chi, psi2, chi2;
halfspinor * restrict * phi ALIGN;
halfspinor32 * restrict * phi32 ALIGN;
#ifdef _KOJAK_INST
#pragma pomp inst begin(hoppingmatrix)
#endif
#ifdef XLC
#pragma disjoint(*l, *k, *U, *s)
#endif
#ifdef _GAUGE_COPY
if(g_update_gauge_copy) {
update_backward_gauge();
}
#endif
if(k == l){
printf("Error in H_psi (simple.c):\n");
printf("Arguments k and l must be different\n");
printf("Program aborted\n");
exit(1);
}
s = k;
if(ieo == 0) {
U = g_gauge_field_copy[0][0];
}
else {
U = g_gauge_field_copy[1][0];
}
if(g_sloppy_precision == 1 && g_sloppy_precision_flag == 1) {
phi32 = NBPointer32[ieo];
/**************** loop over all lattice sites ****************/
ix=0;
for(i = 0; i < (VOLUME)/2; i++){
_vector_assign(rs.s0, (*s).s0);
_vector_assign(rs.s1, (*s).s1);
_vector_assign(rs.s2, (*s).s2);
_vector_assign(rs.s3, (*s).s3);
s++;
/*********************** direction +0 ************************/
_vector_add(psi, rs.s0, rs.s2);
_su3_multiply(chi,(*U),psi);
_complex_times_vector((*phi32[ix]).s0, ka0, chi);
_vector_add(psi, rs.s1, rs.s3);
_su3_multiply(chi,(*U),psi);
_complex_times_vector((*phi32[ix]).s1, ka0, chi);
U++;
ix++;
/*********************** direction -0 ************************/
_vector_sub((*phi32[ix]).s0, rs.s0, rs.s2);
_vector_sub((*phi32[ix]).s1, rs.s1, rs.s3);
ix++;
/*********************** direction +1 ************************/
_vector_i_add(psi, rs.s0, rs.s3);
_su3_multiply(chi, (*U), psi);
_complex_times_vector((*phi32[ix]).s0, ka1, chi);
_vector_i_add(psi, rs.s1, rs.s2);
_su3_multiply(chi, (*U), psi);
_complex_times_vector((*phi32[ix]).s1, ka1, chi);
U++;
ix++;
/*********************** direction -1 ************************/
_vector_i_sub((*phi32[ix]).s0, rs.s0, rs.s3);
_vector_i_sub((*phi32[ix]).s1, rs.s1, rs.s2);
ix++;
/*********************** direction +2 ************************/
_vector_add(psi, rs.s0, rs.s3);
_su3_multiply(chi,(*U),psi);
_complex_times_vector((*phi32[ix]).s0, ka2, chi);
_vector_sub(psi, rs.s1, rs.s2);
_su3_multiply(chi,(*U),psi);
_complex_times_vector((*phi32[ix]).s1, ka2, chi);
U++;
ix++;
/*********************** direction -2 ************************/
_vector_sub((*phi32[ix]).s0, rs.s0, rs.s3);
_vector_add((*phi32[ix]).s1, rs.s1, rs.s2);
ix++;
/*********************** direction +3 ************************/
_vector_i_add(psi, rs.s0, rs.s2);
_su3_multiply(chi, (*U), psi);
_complex_times_vector((*phi32[ix]).s0, ka3, chi);
_vector_i_sub(psi, rs.s1, rs.s3);
_su3_multiply(chi,(*U),psi);
_complex_times_vector((*phi32[ix]).s1, ka3, chi);
U++;
ix++;
/*********************** direction -3 ************************/
_vector_i_sub((*phi32[ix]).s0, rs.s0, rs.s2);
_vector_i_add((*phi32[ix]).s1, rs.s1, rs.s3);
ix++;
/************************ end of loop ************************/
}
# if (defined MPI && !defined _NO_COMM)
xchange_halffield32();
# endif
s = l;
phi32 = NBPointer32[2 + ieo];
if(ieo == 0) {
U = g_gauge_field_copy[1][0];
}
else {
U = g_gauge_field_copy[0][0];
}
ix = 0;
for(i = 0; i < (VOLUME)/2; i++){
/*********************** direction +0 ************************/
_vector_assign(rs.s0, (*phi32[ix]).s0);
_vector_assign(rs.s2, (*phi32[ix]).s0);
_vector_assign(rs.s1, (*phi32[ix]).s1);
_vector_assign(rs.s3, (*phi32[ix]).s1);
ix++;
/*********************** direction -0 ************************/
_vector_assign(psi, (*phi32[ix]).s0);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka0,chi);
_vector_add_assign(rs.s0, psi);
_vector_sub_assign(rs.s2, psi);
_vector_assign(psi, (*phi32[ix]).s1);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka0,chi);
_vector_add_assign(rs.s1, psi);
_vector_sub_assign(rs.s3, psi);
ix++;
U++;
/*********************** direction +1 ************************/
_vector_add_assign(rs.s0, (*phi32[ix]).s0);
_vector_i_sub_assign(rs.s3, (*phi32[ix]).s0);
_vector_add_assign(rs.s1, (*phi32[ix]).s1);
_vector_i_sub_assign(rs.s2, (*phi32[ix]).s1);
ix++;
/*********************** direction -1 ************************/
_vector_assign(psi, (*phi32[ix]).s0);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka1,chi);
_vector_add_assign(rs.s0, psi);
_vector_i_add_assign(rs.s3, psi);
_vector_assign(psi, (*phi32[ix]).s1);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka1,chi);
_vector_add_assign(rs.s1, psi);
_vector_i_add_assign(rs.s2, psi);
U++;
ix++;
/*********************** direction +2 ************************/
_vector_add_assign(rs.s0, (*phi32[ix]).s0);
_vector_add_assign(rs.s3, (*phi32[ix]).s0);
_vector_add_assign(rs.s1, (*phi32[ix]).s1);
_vector_sub_assign(rs.s2, (*phi32[ix]).s1);
ix++;
/*********************** direction -2 ************************/
_vector_assign(psi, (*phi32[ix]).s0);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka2,chi);
_vector_add_assign(rs.s0, psi);
_vector_sub_assign(rs.s3, psi);
_vector_assign(psi, (*phi32[ix]).s1);
_su3_inverse_multiply(chi, (*U), psi);
_complexcjg_times_vector(psi,ka2,chi);
_vector_add_assign(rs.s1, psi);
_vector_add_assign(rs.s2, psi);
U++;
ix++;
/*********************** direction +3 ************************/
_vector_add_assign(rs.s0, (*phi32[ix]).s0);
_vector_i_sub_assign(rs.s2, (*phi32[ix]).s0);
_vector_add_assign(rs.s1, (*phi32[ix]).s1);
_vector_i_add_assign(rs.s3, (*phi32[ix]).s1);
ix++;
/*********************** direction -3 ************************/
_vector_assign(psi, (*phi32[ix]).s0);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka3,chi);
_vector_add((*s).s0, rs.s0, psi);
_vector_i_add((*s).s2, rs.s2, psi);
_vector_assign(psi, (*phi32[ix]).s1);
_su3_inverse_multiply(chi,(*U), psi);
_complexcjg_times_vector(psi,ka3,chi);
_vector_add((*s).s1, rs.s1, psi);
_vector_i_sub((*s).s3, rs.s3, psi);
U++;
ix++;
s++;
}
}
else {
phi = NBPointer[ieo];
/**************** loop over all lattice sites ****************/
ix=0;
/* #pragma ivdep*/
for(i = 0; i < (VOLUME)/2; i++){
_vector_assign(rs.s0, (*s).s0);
_vector_assign(rs.s1, (*s).s1);
_vector_assign(rs.s2, (*s).s2);
_vector_assign(rs.s3, (*s).s3);
s++;
/*********************** direction +0 ************************/
_vector_add(psi, rs.s0, rs.s2);
_vector_add(psi2, rs.s1, rs.s3);
_su3_multiply(chi,(*U),psi);
_su3_multiply(chi2,(*U),psi2);
_complex_times_vector((*phi[ix]).s0, ka0, chi);
_complex_times_vector((*phi[ix]).s1, ka0, chi2);
U++;
ix++;
/*********************** direction -0 ************************/
_vector_sub((*phi[ix]).s0, rs.s0, rs.s2);
_vector_sub((*phi[ix]).s1, rs.s1, rs.s3);
ix++;
/*********************** direction +1 ************************/
_vector_i_add(psi, rs.s0, rs.s3);
_vector_i_add(psi2, rs.s1, rs.s2);
_su3_multiply(chi, (*U), psi);
_su3_multiply(chi2, (*U), psi2);
_complex_times_vector((*phi[ix]).s0, ka1, chi);
_complex_times_vector((*phi[ix]).s1, ka1, chi2);
U++;
ix++;
/*********************** direction -1 ************************/
_vector_i_sub((*phi[ix]).s0, rs.s0, rs.s3);
_vector_i_sub((*phi[ix]).s1, rs.s1, rs.s2);
ix++;
/*********************** direction +2 ************************/
_vector_add(psi, rs.s0, rs.s3);
_vector_sub(psi2, rs.s1, rs.s2);
_su3_multiply(chi,(*U),psi);
_su3_multiply(chi2,(*U),psi2);
_complex_times_vector((*phi[ix]).s0, ka2, chi);
_complex_times_vector((*phi[ix]).s1, ka2, chi2);
U++;
ix++;
/*********************** direction -2 ************************/
_vector_sub((*phi[ix]).s0, rs.s0, rs.s3);
_vector_add((*phi[ix]).s1, rs.s1, rs.s2);
ix++;
/*********************** direction +3 ************************/
_vector_i_add(psi, rs.s0, rs.s2);
_vector_i_sub(psi2, rs.s1, rs.s3);
_su3_multiply(chi, (*U), psi);
_su3_multiply(chi2,(*U),psi2);
_complex_times_vector((*phi[ix]).s0, ka3, chi);
_complex_times_vector((*phi[ix]).s1, ka3, chi2);
U++;
ix++;
/*********************** direction -3 ************************/
_vector_i_sub((*phi[ix]).s0, rs.s0, rs.s2);
_vector_i_add((*phi[ix]).s1, rs.s1, rs.s3);
ix++;
/************************ end of loop ************************/
}
# if (defined MPI && !defined _NO_COMM)
xchange_halffield();
# endif
s = l;
phi = NBPointer[2 + ieo];
if(ieo == 0) {
U = g_gauge_field_copy[1][0];
}
else {
U = g_gauge_field_copy[0][0];
}
ix = 0;
/* #pragma ivdep */
for(i = 0; i < (VOLUME)/2; i++){
/*********************** direction +0 ************************/
_vector_assign(rs.s0, (*phi[ix]).s0);
_vector_assign(rs.s2, (*phi[ix]).s0);
_vector_assign(rs.s1, (*phi[ix]).s1);
_vector_assign(rs.s3, (*phi[ix]).s1);
ix++;
/*********************** direction -0 ************************/
_su3_inverse_multiply(chi,(*U),(*phi[ix]).s0);
_su3_inverse_multiply(chi2,(*U),(*phi[ix]).s1);
_complexcjg_times_vector(psi,ka0,chi);
_complexcjg_times_vector(psi2,ka0,chi2);
_vector_add_assign(rs.s0, psi);
_vector_sub_assign(rs.s2, psi);
_vector_add_assign(rs.s1, psi2);
_vector_sub_assign(rs.s3, psi2);
ix++;
U++;
/*********************** direction +1 ************************/
_vector_add_assign(rs.s0, (*phi[ix]).s0);
_vector_i_sub_assign(rs.s3, (*phi[ix]).s0);
_vector_add_assign(rs.s1, (*phi[ix]).s1);
_vector_i_sub_assign(rs.s2, (*phi[ix]).s1);
ix++;
/*********************** direction -1 ************************/
_su3_inverse_multiply(chi,(*U), (*phi[ix]).s0);
_su3_inverse_multiply(chi2, (*U), (*phi[ix]).s1);
_complexcjg_times_vector(psi,ka1,chi);
_complexcjg_times_vector(psi2,ka1,chi2);
_vector_add_assign(rs.s0, psi);
_vector_i_add_assign(rs.s3, psi);
_vector_add_assign(rs.s1, psi2);
_vector_i_add_assign(rs.s2, psi2);
U++;
ix++;
/*********************** direction +2 ************************/
_vector_add_assign(rs.s0, (*phi[ix]).s0);
_vector_add_assign(rs.s3, (*phi[ix]).s0);
_vector_add_assign(rs.s1, (*phi[ix]).s1);
_vector_sub_assign(rs.s2, (*phi[ix]).s1);
ix++;
/*********************** direction -2 ************************/
_su3_inverse_multiply(chi,(*U), (*phi[ix]).s0);
_su3_inverse_multiply(chi2, (*U), (*phi[ix]).s1);
_complexcjg_times_vector(psi,ka2,chi);
_complexcjg_times_vector(psi2,ka2,chi2);
_vector_add_assign(rs.s0, psi);
_vector_sub_assign(rs.s3, psi);
_vector_add_assign(rs.s1, psi2);
_vector_add_assign(rs.s2, psi2);
U++;
ix++;
/*********************** direction +3 ************************/
_vector_add_assign(rs.s0, (*phi[ix]).s0);
_vector_i_sub_assign(rs.s2, (*phi[ix]).s0);
_vector_add_assign(rs.s1, (*phi[ix]).s1);
_vector_i_add_assign(rs.s3, (*phi[ix]).s1);
ix++;
/*********************** direction -3 ************************/
_su3_inverse_multiply(chi,(*U), (*phi[ix]).s0);
_su3_inverse_multiply(chi2, (*U), (*phi[ix]).s1);
_complexcjg_times_vector(psi,ka3,chi);
_complexcjg_times_vector(psi2,ka3,chi2);
_vector_add((*s).s0, rs.s0, psi);
_vector_i_add((*s).s2, rs.s2, psi);
_vector_add((*s).s1, rs.s1, psi2);
_vector_i_sub((*s).s3, rs.s3, psi2);
U++;
ix++;
s++;
}
}
#ifdef _KOJAK_INST
#pragma pomp inst end(hoppingmatrix)
#endif
}
