void cloverdet_heatbath(const int id, hamiltonian_field_t * const hf) {
monomial * mnl = &monomial_list[id];
g_mu = mnl->mu;
g_mu3 = mnl->rho;
g_c_sw = mnl->c_sw;
boundary(mnl->kappa);
mnl->csg_n = 0;
mnl->csg_n2 = 0;
mnl->iter0 = 0;
mnl->iter1 = 0;
init_sw_fields();
sw_term(hf->gaugefield, mnl->kappa, mnl->c_sw);
sw_invert(EE, mnl->mu);
random_spinor_field(g_spinor_field[2], VOLUME/2, mnl->rngrepro);
mnl->energy0 = square_norm(g_spinor_field[2], VOLUME/2, 1);
mnl->Qp(mnl->pf, g_spinor_field[2]);
chrono_add_solution(mnl->pf, mnl->csg_field, mnl->csg_index_array,
mnl->csg_N, &mnl->csg_n, VOLUME/2);
g_mu = g_mu1;
g_mu3 = 0.;
boundary(g_kappa);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("called cloverdet_heatbath for id %d %d\n", id, mnl->even_odd_flag);
}
return;
}
void det_heatbath(const int id, hamiltonian_field_t * const hf) {
monomial * mnl = &monomial_list[id];
g_mu = mnl->mu;
boundary(mnl->kappa);
mnl->csg_n = 0;
mnl->csg_n2 = 0;
mnl->iter0 = 0;
mnl->iter1 = 0;
if(mnl->even_odd_flag) {
random_spinor_field(g_spinor_field[2], VOLUME/2, mnl->rngrepro);
mnl->energy0 = square_norm(g_spinor_field[2], VOLUME/2, 1);
Qtm_plus_psi(mnl->pf, g_spinor_field[2]);
chrono_add_solution(mnl->pf, mnl->csg_field, mnl->csg_index_array,
mnl->csg_N, &mnl->csg_n, VOLUME/2);
if(mnl->solver != CG) {
chrono_add_solution(mnl->pf, mnl->csg_field2, mnl->csg_index_array2,
mnl->csg_N2, &mnl->csg_n2, VOLUME/2);
}
}
else {
random_spinor_field(g_spinor_field[2], VOLUME, mnl->rngrepro);
mnl->energy0 = square_norm(g_spinor_field[2], VOLUME, 1);
Q_plus_psi(mnl->pf, g_spinor_field[2]);
chrono_add_solution(mnl->pf, mnl->csg_field, mnl->csg_index_array,
mnl->csg_N, &mnl->csg_n, VOLUME/2);
if(mnl->solver != CG) {
chrono_add_solution(mnl->pf, mnl->csg_field2, mnl->csg_index_array2,
mnl->csg_N2, &mnl->csg_n2, VOLUME/2);
}
}
g_mu = g_mu1;
boundary(g_kappa);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("called det_heatbath for id %d %d\n", id, mnl->even_odd_flag);
}
return;
}
/*lambda: smallest eigenvalue, k eigenvector */
int evamax0(double *rz, int k, double q_off, double eps_sq) {
static double norm,norm0;
int j;
random_spinor_field(g_spinor_field[k], VOLUME/2);
norm0=square_norm(g_spinor_field[k], VOLUME/2, 1);
norm=1000.;
assign_mul_bra_add_mul_r( g_spinor_field[k], 1./sqrt(norm0),0., g_spinor_field[k], VOLUME/2);
for(j=1;j<ITER_MAX_BCG;j++)
{
Q_psi(k,k,q_off); Q_psi(k,k,q_off);
norm0=square_norm(g_spinor_field[k], VOLUME/2, 1);
norm0=sqrt(norm0);
assign_mul_bra_add_mul_r( g_spinor_field[k], 1./norm0,0., g_spinor_field[k], VOLUME/2);
if((norm-norm0)*(norm-norm0) <= eps_sq) break;
norm=norm0;
}
*rz=norm0;
return j;
}
void cloverdetratio_heatbath(const int id, hamiltonian_field_t * const hf) {
monomial * mnl = &monomial_list[id];
g_mu = mnl->mu;
g_c_sw = mnl->c_sw;
boundary(mnl->kappa);
mnl->csg_n = 0;
mnl->csg_n2 = 0;
mnl->iter0 = 0;
mnl->iter1 = 0;
init_sw_fields();
sw_term( (const su3**) hf->gaugefield, mnl->kappa, mnl->c_sw);
sw_invert(EE, mnl->mu);
random_spinor_field(g_spinor_field[4], VOLUME/2, mnl->rngrepro);
mnl->energy0 = square_norm(g_spinor_field[4], VOLUME/2, 1);
g_mu3 = mnl->rho;
mnl->Qp(g_spinor_field[3], g_spinor_field[4]);
g_mu3 = mnl->rho2;
zero_spinor_field(mnl->pf,VOLUME/2);
mnl->iter0 = cg_her(mnl->pf, g_spinor_field[3], mnl->maxiter, mnl->accprec,
g_relative_precision_flag, VOLUME/2, mnl->Qsq);
chrono_add_solution(mnl->pf, mnl->csg_field, mnl->csg_index_array,
mnl->csg_N, &mnl->csg_n, VOLUME/2);
mnl->Qm(mnl->pf, mnl->pf);
if(g_proc_id == 0 && g_debug_level > 3) {
printf("called cloverdetratio_heatbath for id %d \n", id);
}
g_mu3 = 0.;
g_mu = g_mu1;
boundary(g_kappa);
return;
}
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);
}
/*lambda: largest eigenvalue, k eigenvector */
int evamax(double *rz, int k, double q_off, double eps_sq) {
static double ritz,norm0,normg,normg0,beta_cg;
static double costh,sinth,cosd,sind,aaa,normp,xxx;
static double xs1,xs2,xs3;
int iteration;
/* Initialize k to be gaussian */
random_spinor_field(g_spinor_field[k], VOLUME/2);
norm0=square_norm(g_spinor_field[k], VOLUME/2, 1);
/*normalize k */
assign_mul_bra_add_mul_r( g_spinor_field[k], 1./sqrt(norm0),0., g_spinor_field[k], VOLUME/2);
Q_psi(DUM_SOLVER,k,q_off);
Q_psi(DUM_SOLVER,DUM_SOLVER,q_off);
/*compute the ritz functional */
/*put g on DUM_SOLVER+2 and p on DUM_SOLVER+1*/
ritz=scalar_prod_r(g_spinor_field[DUM_SOLVER], g_spinor_field[k], VOLUME/2, 1);
zero_spinor_field(g_spinor_field[DUM_SOLVER+2],VOLUME/2);
assign_add_mul_r_add_mul(g_spinor_field[DUM_SOLVER+2], g_spinor_field[DUM_SOLVER], g_spinor_field[k],
1., -ritz, VOLUME/2);
assign(g_spinor_field[DUM_SOLVER+1], g_spinor_field[DUM_SOLVER+2], VOLUME/2);
normg0=square_norm(g_spinor_field[DUM_SOLVER+2], VOLUME/2, 1);
/* main loop */
for(iteration=1;iteration<=ITER_MAX_BCG;iteration++) {
if(normg0 <= eps_sq) break;
Q_psi(DUM_SOLVER+2,DUM_SOLVER+1,q_off);
Q_psi(DUM_SOLVER+2,DUM_SOLVER+2,q_off);
/* compute costh and sinth */
normp=square_norm(g_spinor_field[DUM_SOLVER+1], VOLUME/2, 1);
xxx=scalar_prod_r(g_spinor_field[DUM_SOLVER+2], g_spinor_field[DUM_SOLVER+1], VOLUME/2, 1);
xs1=0.5*(ritz+xxx/normp);
xs2=0.5*(ritz-xxx/normp);
normp=sqrt(normp);
xs3=normg0/normp;
aaa=sqrt(xs2*xs2+xs3*xs3);
cosd=xs2/aaa;
sind=xs3/aaa;
if(cosd>=0.) {
costh=sqrt(0.5*(1.+cosd));
sinth=0.5*sind/costh;
}
else {
sinth=sqrt(0.5*(1.-cosd));
costh=0.5*sind/sinth;
}
ritz=xs1+aaa;
assign_add_mul_r_add_mul(g_spinor_field[k], g_spinor_field[k], g_spinor_field[DUM_SOLVER+1],
costh-1., sinth/normp, VOLUME/2);
assign_add_mul_r_add_mul(g_spinor_field[DUM_SOLVER], g_spinor_field[DUM_SOLVER], g_spinor_field[DUM_SOLVER+2],
costh-1., sinth/normp, VOLUME/2);
/* compute g */
zero_spinor_field(g_spinor_field[DUM_SOLVER+2],VOLUME/2);
assign_add_mul_r_add_mul(g_spinor_field[DUM_SOLVER+2], g_spinor_field[DUM_SOLVER], g_spinor_field[k],
1., -ritz, VOLUME/2);
/* calculate the norm of g' and beta_cg=costh g'^2/g^2 */
normg=square_norm(g_spinor_field[DUM_SOLVER+2], VOLUME/2, 1);
beta_cg=costh*normg/normg0;
if(beta_cg*costh*normp>20.*sqrt(normg)) beta_cg=0.;
normg0=normg;
/* compute the new value of p */
assign_add_mul_r(g_spinor_field[DUM_SOLVER+1], g_spinor_field[k], -scalar_prod_r(g_spinor_field[k], g_spinor_field[DUM_SOLVER+1], VOLUME/2), VOLUME/2, 1);
assign_mul_add_r(g_spinor_field[DUM_SOLVER+1],beta_cg, g_spinor_field[DUM_SOLVER+2], VOLUME/2);
/* restore the state of the iteration */
if(iteration%20==0) {
/* readjust x */
xxx=sqrt(square_norm(g_spinor_field[k], VOLUME/2), 1);
assign_mul_bra_add_mul_r( g_spinor_field[k], 1./xxx,0., g_spinor_field[k], VOLUME/2);
Q_psi(DUM_SOLVER,k,q_off);
Q_psi(DUM_SOLVER,DUM_SOLVER,q_off);
/*compute the ritz functional */
ritz=scalar_prod_r(g_spinor_field[DUM_SOLVER], g_spinor_field[k], VOLUME/2, 1);
/*put g on DUM_SOLVER+2 and p on DUM_SOLVER+1*/
zero_spinor_field(g_spinor_field[DUM_SOLVER+2],VOLUME/2);
assign_add_mul_r_add_mul(g_spinor_field[DUM_SOLVER+2], g_spinor_field[DUM_SOLVER], g_spinor_field[k],
1., -ritz, VOLUME/2);
normg0=square_norm(g_spinor_field[DUM_SOLVER+2], VOLUME/2, 1);
/*subtract a linear combination of x and g from p to
insure (x,p)=0 and (p,g)=(g,g) */
cosd=scalar_prod_r(g_spinor_field[k], g_spinor_field[DUM_SOLVER+1], VOLUME/2, 1);
assign_add_mul_r(g_spinor_field[DUM_SOLVER+1], g_spinor_field[k], -cosd, VOLUME/2);
cosd=scalar_prod_r(g_spinor_field[DUM_SOLVER+1], g_spinor_field[DUM_SOLVER+2], VOLUME/2, 1)-normg0;
assign_add_mul_r(g_spinor_field[DUM_SOLVER+1], g_spinor_field[DUM_SOLVER+2], -cosd/sqrt(normg0), VOLUME/2);
}
}
*rz=ritz;
return iteration;
}
void ndpoly_heatbath(const int id) {
int j;
double temp;
monomial * mnl = &monomial_list[id];
(*mnl).energy0 = 0.;
random_spinor_field(g_chi_up_spinor_field[0], VOLUME/2, (*mnl).rngrepro);
(*mnl).energy0 = square_norm(g_chi_up_spinor_field[0], VOLUME/2, 1);
if(g_epsbar!=0.0 || phmc_exact_poly == 0){
random_spinor_field(g_chi_dn_spinor_field[0], VOLUME/2, (*mnl).rngrepro);
(*mnl).energy0 += square_norm(g_chi_dn_spinor_field[0], VOLUME/2, 1);
}
else {
zero_spinor_field(g_chi_dn_spinor_field[0], VOLUME/2);
}
if((g_proc_id == g_stdio_proc) && (g_debug_level > 2)) {
printf("PHMC: Here comes the computation of H_old with \n \n");
printf("PHMC: First: random spinors and their norm \n ");
printf("PHMC: OLD Ennergy UP %e \n", (*mnl).energy0);
printf("PHMC: OLD Energy DN + UP %e \n\n", (*mnl).energy0);
}
if(phmc_exact_poly==0){
QNon_degenerate(g_chi_up_spinor_field[1], g_chi_dn_spinor_field[1],
g_chi_up_spinor_field[0], g_chi_dn_spinor_field[0]);
for(j = 1; j < (phmc_dop_n_cheby); j++){
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
assign(g_chi_dn_spinor_field[0], g_chi_dn_spinor_field[1], VOLUME/2);
Q_tau1_min_cconst_ND(g_chi_up_spinor_field[1], g_chi_dn_spinor_field[1],
g_chi_up_spinor_field[0], g_chi_dn_spinor_field[0],
phmc_root[phmc_dop_n_cheby-2+j]);
}
Poly_tilde_ND(g_chi_up_spinor_field[0], g_chi_dn_spinor_field[0], phmc_ptilde_cheby_coef,
phmc_ptilde_n_cheby, g_chi_up_spinor_field[1], g_chi_dn_spinor_field[1]);
}
else if( phmc_exact_poly==1 && g_epsbar!=0.0) {
/* Attention this is Q * tau1, up/dn are exchanged in the input spinor */
/* this is used as an preconditioner */
QNon_degenerate(g_chi_up_spinor_field[1],g_chi_dn_spinor_field[1],
g_chi_dn_spinor_field[0],g_chi_up_spinor_field[0]);
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
assign(g_chi_dn_spinor_field[0], g_chi_dn_spinor_field[1], VOLUME/2);
/* solve Q*tau1*P(Q^2) *x=y */
cg_her_nd(g_chi_up_spinor_field[1],g_chi_dn_spinor_field[1],
g_chi_up_spinor_field[0],g_chi_dn_spinor_field[0],
1000,1.e-16,0,VOLUME/2, Qtau1_P_ND);
/* phi= Bdagger phi */
for(j = 1; j < (phmc_dop_n_cheby); j++){
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
assign(g_chi_dn_spinor_field[0], g_chi_dn_spinor_field[1], VOLUME/2);
Q_tau1_min_cconst_ND(g_chi_up_spinor_field[1], g_chi_dn_spinor_field[1],
g_chi_up_spinor_field[0], g_chi_dn_spinor_field[0],
phmc_root[phmc_dop_n_cheby-2+j]);
}
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
assign(g_chi_dn_spinor_field[0], g_chi_dn_spinor_field[1], VOLUME/2);
}
else if(phmc_exact_poly==1 && g_epsbar==0.0) {
Qtm_pm_psi(g_chi_up_spinor_field[1], g_chi_up_spinor_field[0]);
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
/* solve (Q+)*(Q-)*P((Q+)*(Q-)) *x=y */
cg_her(g_chi_up_spinor_field[1], g_chi_up_spinor_field[0],
1000,1.e-16,0,VOLUME/2, Qtm_pm_Ptm_pm_psi);
/* phi= Bdagger phi */
for(j = 1; j < (phmc_dop_n_cheby); j++){
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
Qtm_pm_min_cconst_nrm(g_chi_up_spinor_field[1],
g_chi_up_spinor_field[0],
phmc_root[phmc_dop_n_cheby-2+j]);
}
assign(g_chi_up_spinor_field[0], g_chi_up_spinor_field[1], VOLUME/2);
}
assign(mnl->pf, g_chi_up_spinor_field[0], VOLUME/2);
assign(mnl->pf2, g_chi_dn_spinor_field[0], VOLUME/2);
temp = square_norm(g_chi_up_spinor_field[0], VOLUME/2, 1);
if((g_proc_id == g_stdio_proc) && (g_debug_level > 2)) {
printf("PHMC: Then: evaluate Norm of pseudofermion heatbath BHB \n ");
printf("PHMC: Norm of BHB up squared %e \n", temp);
}
if(g_epsbar!=0.0 || phmc_exact_poly==0)
temp += square_norm(g_chi_dn_spinor_field[0], VOLUME/2, 1);
if((g_proc_id == g_stdio_proc) && (g_debug_level > 2)){
printf("PHMC: Norm of BHB up + BHB dn squared %e \n\n", temp);
}
if(g_proc_id == 0 && g_debug_level > 3) {
printf("called ndpoly_heatbath for id %d with g_running_phmc = %d\n", id, g_running_phmc);
}
return;
}
int main(int argc,char *argv[])
{
int j,j_max,k,k_max = 1;
#ifdef HAVE_LIBLEMON
paramsXlfInfo *xlfInfo;
#endif
int status = 0;
static double t1,t2,dt,sdt,dts,qdt,sqdt;
double antioptaway=0.0;
#ifdef MPI
static double dt2;
DUM_DERI = 6;
DUM_SOLVER = DUM_DERI+2;
DUM_MATRIX = DUM_SOLVER+6;
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
# ifdef OMP
int mpi_thread_provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_thread_provided);
# else
MPI_Init(&argc, &argv);
# endif
MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id);
#else
g_proc_id = 0;
#endif
g_rgi_C1 = 1.;
/* Read the input file */
if((status = read_input("benchmark.input")) != 0) {
fprintf(stderr, "Could not find input file: benchmark.input\nAborting...\n");
exit(-1);
}
#ifdef OMP
if(omp_num_threads > 0)
{
omp_set_num_threads(omp_num_threads);
}
else {
if( g_proc_id == 0 )
printf("# No value provided for OmpNumThreads, running in single-threaded mode!\n");
omp_num_threads = 1;
omp_set_num_threads(omp_num_threads);
}
init_omp_accumulators(omp_num_threads);
#endif
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 MPI
# ifdef _NON_BLOCKING
printf("# The code was compiled for non-blocking MPI calls (spinor and gauge)\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 + g_dbw2rand);
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("# benchmarking the even/odd preconditioned Dirac operator\n");
}
else {
printf("# benchmarking 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);
random_gauge_field(reproduce_randomnumber_flag);
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge(g_gauge_field);
#endif
if(even_odd_flag) {
/*initialize the pseudo-fermion fields*/
j_max=2048;
sdt=0.;
for (k = 0; k < k_max; k++) {
random_spinor_field(g_spinor_field[k], VOLUME/2, 0);
}
while(sdt < 30.) {
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
antioptaway=0.0;
for (j=0;j<j_max;j++) {
for (k=0;k<k_max;k++) {
Hopping_Matrix(0, g_spinor_field[k+k_max], g_spinor_field[k]);
Hopping_Matrix(1, g_spinor_field[2*k_max], g_spinor_field[k+k_max]);
antioptaway+=creal(g_spinor_field[2*k_max][0].s0.c0);
}
}
t2 = gettime();
dt = t2-t1;
#ifdef MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
qdt=dt*dt;
#ifdef MPI
MPI_Allreduce (&qdt, &sqdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sqdt = qdt;
#endif
sdt=sdt/((double)g_nproc);
sqdt=sqrt(sqdt/g_nproc-sdt*sdt);
j_max*=2;
}
j_max=j_max/2;
dts=dt;
sdt=1.0e6f*sdt/((double)(k_max*j_max*(VOLUME)));
sqdt=1.0e6f*sqdt/((double)(k_max*j_max*(VOLUME)));
if(g_proc_id==0) {
printf("# The following result is just to make sure that the calculation is not optimized away: %e\n", antioptaway);
printf("# Total compute time %e sec, variance of the time %e sec. (%d iterations).\n", sdt, sqdt, j_max);
printf("# Communication switched on:\n# (%d Mflops [%d bit arithmetic])\n", (int)(1608.0f/sdt),(int)sizeof(spinor)/3);
#ifdef OMP
printf("# Mflops per OpenMP thread ~ %d\n",(int)(1608.0f/(omp_num_threads*sdt)));
#endif
printf("\n");
fflush(stdout);
}
#ifdef MPI
/* isolated computation */
t1 = gettime();
antioptaway=0.0;
for (j=0;j<j_max;j++) {
for (k=0;k<k_max;k++) {
Hopping_Matrix_nocom(0, g_spinor_field[k+k_max], g_spinor_field[k]);
Hopping_Matrix_nocom(1, g_spinor_field[2*k_max], g_spinor_field[k+k_max]);
antioptaway += creal(g_spinor_field[2*k_max][0].s0.c0);
}
}
t2 = gettime();
dt2 = t2-t1;
/* compute the bandwidth */
dt=dts-dt2;
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
sdt=sdt/((double)g_nproc);
MPI_Allreduce (&dt2, &dt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
dt=dt/((double)g_nproc);
dt=1.0e6f*dt/((double)(k_max*j_max*(VOLUME)));
if(g_proc_id==0) {
printf("# The following result is printed just to make sure that the calculation is not optimized away: %e\n",antioptaway);
printf("# Communication switched off: \n# (%d Mflops [%d bit arithmetic])\n", (int)(1608.0f/dt),(int)sizeof(spinor)/3);
#ifdef OMP
printf("# Mflops per OpenMP thread ~ %d\n",(int)(1608.0f/(omp_num_threads*dt)));
#endif
printf("\n");
fflush(stdout);
}
sdt=sdt/((double)k_max);
sdt=sdt/((double)j_max);
sdt=sdt/((double)(2*SLICE));
if(g_proc_id==0) {
printf("# The size of the package is %d bytes.\n",(SLICE)*192);
#ifdef _USE_HALFSPINOR
printf("# The bandwidth is %5.2f + %5.2f MB/sec\n", 192./sdt/1024/1024, 192./sdt/1024./1024);
#else
printf("# The bandwidth is %5.2f + %5.2f MB/sec\n", 2.*192./sdt/1024/1024, 2.*192./sdt/1024./1024);
#endif
}
#endif
fflush(stdout);
}
else {
/* the non even/odd case now */
/*initialize the pseudo-fermion fields*/
j_max=1;
sdt=0.;
for (k=0;k<k_max;k++) {
random_spinor_field(g_spinor_field[k], VOLUME, 0);
}
while(sdt < 3.) {
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
for (j=0;j<j_max;j++) {
for (k=0;k<k_max;k++) {
D_psi(g_spinor_field[k+k_max], g_spinor_field[k]);
antioptaway+=creal(g_spinor_field[k+k_max][0].s0.c0);
}
}
t2 = gettime();
dt=t2-t1;
#ifdef MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
qdt=dt*dt;
#ifdef MPI
MPI_Allreduce (&qdt, &sqdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sqdt = qdt;
#endif
sdt=sdt/((double)g_nproc);
sqdt=sqrt(sqdt/g_nproc-sdt*sdt);
j_max*=2;
}
j_max=j_max/2;
dts=dt;
sdt=1.0e6f*sdt/((double)(k_max*j_max*(VOLUME)));
sqdt=1.0e6f*sqdt/((double)(k_max*j_max*(VOLUME)));
if(g_proc_id==0) {
printf("# The following result is just to make sure that the calculation is not optimized away: %e\n", antioptaway);
printf("# Total compute time %e sec, variance of the time %e sec. (%d iterations).\n", sdt, sqdt, j_max);
printf("\n# (%d Mflops [%d bit arithmetic])\n", (int)(1680.0f/sdt),(int)sizeof(spinor)/3);
#ifdef OMP
printf("# Mflops per OpenMP thread ~ %d\n",(int)(1680.0f/(omp_num_threads*sdt)));
#endif
printf("\n");
fflush(stdout);
}
}
#ifdef HAVE_LIBLEMON
if(g_proc_id==0) {
printf("# Performing parallel IO test ...\n");
}
xlfInfo = construct_paramsXlfInfo(0.5, 0);
write_gauge_field( "conf.test", 64, xlfInfo);
free(xlfInfo);
if(g_proc_id==0) {
printf("# done ...\n");
}
#endif
#ifdef MPI
MPI_Finalize();
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
return(0);
}
double reweighting_factor_nd(const int N) {
int i, n_iter;
double sq_norm, corr, sum=0., sq_sum = 0., temp1;
double mu1, mu2;
complex temp2;
mu1 = g_mu1;
mu2 = g_mu1;
/* Use spinor_field 2,3,5 */
/* in order not to conflict with anything else... */
for(i = 0; i < N; i++) {
random_spinor_field(g_chi_up_spinor_field[2],VOLUME/2, 1);
random_spinor_field(g_chi_dn_spinor_field[2],VOLUME/2, 1);
zero_spinor_field(g_chi_up_spinor_field[3],VOLUME/2);
zero_spinor_field(g_chi_dn_spinor_field[3],VOLUME/2);
temp1 = phmc_ptilde_cheby_coef[0];
phmc_ptilde_cheby_coef[0] = temp1 - 1;
Poly_tilde_ND(g_chi_up_spinor_field[3], g_chi_dn_spinor_field[3], phmc_ptilde_cheby_coef, phmc_ptilde_n_cheby, g_chi_up_spinor_field[2], g_chi_dn_spinor_field[2]);
phmc_ptilde_cheby_coef[0] = temp1;
temp2 = scalar_prod(g_chi_up_spinor_field[2], g_chi_up_spinor_field[3], VOLUME/2, 1);
if(temp2.im > 1.0e-8) {
printf("!!! WARNING Immaginary part of CORR-UP LARGER than 10^-8 !!! \n");
printf(" CORR-UP: Re=%12.10e Im=%12.10e \n", temp2.re, temp2.im);
}
corr = temp2.re;
printf(" CORR-UP: Re=%12.10e \n", corr);
temp2 = scalar_prod(g_chi_dn_spinor_field[2], g_chi_dn_spinor_field[3], VOLUME/2, 1);
if(temp2.im > 1.0e-8) {
printf("!!! WARNING Immaginary part of CORR_DN LARGER than 10^-8 !!! \n");
printf(" CORR-DN: Re=%12.10e Im=%12.10e \n", temp2.re, temp2.im);
}
corr += temp2.re;
printf(" CORR-DN: Re=%12.10e \n", temp2.im);
temp1 = -corr;
sum += temp1;
sq_sum += temp1*temp1;
printf("rew: n_iter = %d, sq_norm = %e, corr = %e\n", n_iter, sq_norm, corr);
/*
random_spinor_field(g_spinor_field[2],VOLUME/2, 1);
g_mu = mu2;
zero_spinor_field(g_spinor_field[3],VOLUME/2);
n_iter = solve_cg(3, 2, 0., 1.e-15, 1);
g_mu = mu1;
Qtm_pm_psi(g_spinor_field[5] , g_spinor_field[3]);
sq_norm = square_norm(g_spinor_field[2], VOLUME/2, 1);
corr = scalar_prod_r(g_spinor_field[2], g_spinor_field[5], VOLUME/2, 1);
sq_norm -= corr;
temp1 = sq_norm;
sum += temp1;
sq_sum += temp1*temp1;
printf("rew: n_iter = %d, sq_norm = %e, corr = %e\n", n_iter, sq_norm, corr);
*/
}
sum/=(double)N;
sq_sum/=(double)N;
printf("rew: factor = %e, err = %e\n", sum, sqrt(sum*sum-sq_sum)/((double)N-1));
return(sum);
}
void index_jd(int * nr_of_eigenvalues_ov,
const int max_iterations, const double precision_ov, char *conf_filename,
const int nstore, const int method){
complex *eval;
spinor *eigenvectors_ov, *eigenvectors_ov_;
spinor *lowvectors, *lowvectors_;
int i=0 , k=0, returncode=0, index = 0, determined = 0, signed_index = 0;
char filename[120];
FILE * ifs = NULL;
matrix_mult Operator[2];
double absdifference;
const int N2 = VOLUMEPLUSRAND;
#ifdef MPI
double atime, etime;
#endif
double lowestmodes[20];
int intsign, max_iter, first_blocksize = 1;
int * idx = NULL;
/**********************
* For Jacobi-Davidson
**********************/
int verbosity = 3, converged = 0, blocksize = 1, blockwise = 0;
int solver_it_max = 50, j_max, j_min, v0dim = 0;
double * eigenvalues_ov = NULL;
double decay_min = 1.7, threshold_min = 1.e-3, prec;
WRITER *writer=NULL;
spinor *s;
double sqnorm;
paramsPropagatorFormat *propagatorFormat = NULL;
double ap_eps_sq;
int switch_on_adaptive_precision = 0;
double ov_s = 0;
/**********************
* General variables
**********************/
eval= calloc((*nr_of_eigenvalues_ov),sizeof(complex));
shift = 0.0;
// ov_s = 0.5*(1./g_kappa - 8.) - 1.;
ap_eps_sq = precision_ov*precision_ov;
#if (defined SSE || defined SSE2 )
eigenvectors_ov_= calloc(VOLUMEPLUSRAND*(*nr_of_eigenvalues_ov)+1, sizeof(spinor));
eigenvectors_ov = (spinor *)(((unsigned long int)(eigenvectors_ov_)+ALIGN_BASE)&~ALIGN_BASE);
lowvectors_ = calloc(2*first_blocksize*VOLUMEPLUSRAND+1, sizeof(spinor));
lowvectors = (spinor *)(((unsigned long int)(lowvectors_)+ALIGN_BASE)&~ALIGN_BASE);
#else
// eigenvectors_ov_ = calloc(VOLUMEPLUSRAND*(*nr_of_eigenvalues_ov), sizeof(spinor));
eigenvectors_ov_ = calloc(VOLUMEPLUSRAND*(*nr_of_eigenvalues_ov), sizeof(spinor));
lowvectors_ = calloc(2*first_blocksize*VOLUMEPLUSRAND, sizeof(spinor));
eigenvectors_ov = eigenvectors_ov_;
lowvectors = lowvectors_;
#endif
// idx = malloc((*nr_of_eigenvalues_ov)*sizeof(int));
idx = malloc((*nr_of_eigenvalues_ov)*sizeof(int));
Operator[0]=&Dov_proj_plus;
Operator[1]=&Dov_proj_minus;
if(g_proc_id == g_stdio_proc){
printf("Computing first the two lowest modes in the positive and negative chirality sector, respectively\n");
if(switch_on_adaptive_precision == 1) {
printf("We have switched on adaptive precision with ap_eps_sq = %e!\n", ap_eps_sq);
}
printf("We have set the mass to zero within this computation!\n");
fflush(stdout);
}
prec = precision_ov;
j_min = 8; j_max = 16;
max_iter = 70;
#ifdef MPI
atime = MPI_Wtime();
#endif
v0dim = first_blocksize;
blocksize = v0dim;
for(intsign = 0; intsign < 2; intsign++){
converged = 0;
if(g_proc_id == g_stdio_proc){
printf("%s chirality sector: \n", intsign ? "negative" : "positive");
fflush(stdout);
}
if(max_iter == 70){
/********************************************************************
*
* We need random start spinor fields, but they must be half zero,
* that's why we apply the Projektor once
*
********************************************************************/
for(i = 0; i < first_blocksize; i++) {
random_spinor_field(&lowvectors[(first_blocksize*intsign+i)*VOLUMEPLUSRAND],N2,0);
Proj(&lowvectors[(first_blocksize*intsign+i)*VOLUMEPLUSRAND],
&lowvectors[(first_blocksize*intsign+i)*VOLUMEPLUSRAND],N2, intsign);
}
}
jdher(VOLUME*sizeof(spinor)/sizeof(complex),
VOLUMEPLUSRAND*sizeof(spinor)/sizeof(complex),
shift, prec, blocksize, j_max, j_min,
max_iter, blocksize, blockwise, v0dim, (complex*) &lowvectors[first_blocksize*intsign*VOLUMEPLUSRAND],
CG, solver_it_max,
threshold_min, decay_min, verbosity,
&converged, (complex*) &lowvectors[first_blocksize*intsign*VOLUMEPLUSRAND],
&lowestmodes[first_blocksize*intsign],
&returncode, JD_MINIMAL, 1,
Operator[intsign]);
if(converged != blocksize && max_iter == 70){
if(g_proc_id == g_stdio_proc){
printf("Restarting %s chirality sector with more iterations!\n", intsign ? "negative" : "positive");
fflush(stdout);
}
max_iter = 140;
intsign-=1;
}
else {
max_iter = 70;
/* Save the allready computed eigenvectors_ov */
for(i = 0; i< first_blocksize; i++) {
sprintf(filename, "eigenvector_of_D%s.%.2d.%s.%.4d",((intsign==0)?"plus":"minus"),i , conf_filename, nstore);
construct_writer(&writer, filename, 0);
/* todo write propagator format */
propagatorFormat = construct_paramsPropagatorFormat(64, 1);
write_propagator_format(writer, propagatorFormat);
free(propagatorFormat);
s=(spinor*)&lowvectors[first_blocksize*intsign*VOLUMEPLUSRAND];
write_spinor(writer, &s,NULL, 1, 64);
destruct_writer(writer);
writer=NULL;
sqnorm=square_norm(s,VOLUME,1);
printf(" wrote eigenvector of overlap operator !!! | |^2 = %e \n",sqnorm);
}
}
}
#ifdef MPI
etime = MPI_Wtime();
if(g_proc_id == g_stdio_proc){
printf("It took %f sec to determine the sector with zero modes, if any!\n", etime-atime);
}
#endif
/*Compare the two lowest modes */
absdifference = fabs(lowestmodes[0]-lowestmodes[first_blocksize]);
if(absdifference < 0.1*max(lowestmodes[0],lowestmodes[first_blocksize])){
/* They are equal within the errors */
if(g_proc_id == g_stdio_proc){
printf("Index is 0!\n");
fflush(stdout);
sprintf(filename, "eigenvalues_of_overlap_proj.%s.%.4d", conf_filename, nstore);
ifs = fopen(filename, "w");
printf("\nThe following lowest modes have been computed:\n");
fprintf(ifs, "Index is 0\n\n");
fprintf(ifs, "Sector with positive chirality:\n");
for(i = 0; i < first_blocksize; i++) {
lowestmodes[i] = 2.*(1.+ov_s)*lowestmodes[i];
fprintf(ifs, "%d %e positive\n", i, lowestmodes[i]);
printf("%d %e positive\n", i, lowestmodes[i]);
}
fprintf(ifs, "Sector with negative chirality:\n");
for(i = 0; i < first_blocksize; i++) {
lowestmodes[i+first_blocksize] = 2.*(1.+ov_s)*lowestmodes[i+first_blocksize];
fprintf(ifs, "%d %e negative\n", i, lowestmodes[i+first_blocksize]);
printf("%d %e negative\n", i, lowestmodes[i+first_blocksize]);
}
fclose(ifs);
for(k = 0; k < 2; k++) {
sprintf(filename, "eigenvalues_of_D%s.%s.%.4d",
k ? "minus" : "plus", conf_filename, nstore);
ifs = fopen(filename, "w");
fwrite(&first_blocksize, sizeof(int), 1, ifs);
index = 0;
fwrite(&index, sizeof(int), 1, ifs);
for(i = 0; i < first_blocksize; i++) {
fwrite(&lowestmodes[((intsign+1)%2)*first_blocksize+i], sizeof(double), 1, ifs);
}
fclose(ifs);
}
}
}
else{
/* they are not equal */
/* determine the sector with not trivial topology */
if(lowestmodes[0] < lowestmodes[first_blocksize]){
intsign = 0;
}
else{
intsign = 1;
}
if(g_proc_id == g_stdio_proc){
printf("Computing now up to %d modes in the sector with %s chirality\n",
(*nr_of_eigenvalues_ov), intsign ? "negative" : "positive");
fflush(stdout);
}
/* Here we set the (absolute) precision to be */
/* such that we can compare to the lowest mode */
/* in the other sector */
prec = (lowestmodes[first_blocksize*((intsign+1)%2)])*1.e-1;
eigenvalues_ov = (double*)malloc((*nr_of_eigenvalues_ov)*sizeof(double));
/* Copy the allready computed eigenvectors_ov */
for(i = 0; i < first_blocksize; i++) {
assign(&eigenvectors_ov[i], &lowvectors[(first_blocksize*intsign+i)*VOLUMEPLUSRAND],N2);
eigenvalues_ov[i] = lowestmodes[first_blocksize*intsign+i];
}
#ifdef MPI
atime = MPI_Wtime();
#endif
blocksize = 3;
j_min = 8; j_max = 16;
converged = first_blocksize;
for(i = first_blocksize; i < (*nr_of_eigenvalues_ov); i+=3) {
if((i + blocksize) > (*nr_of_eigenvalues_ov) ) {
blocksize = (*nr_of_eigenvalues_ov) - i;
}
/* Fill up the rest with random spinor fields */
/* and project it to the corresponding sector */
for(v0dim = i; v0dim < i+blocksize; v0dim++){
random_spinor_field(&eigenvectors_ov[v0dim*VOLUMEPLUSRAND],N2,0);
Proj(&eigenvectors_ov[v0dim*VOLUMEPLUSRAND], &eigenvectors_ov[v0dim*VOLUMEPLUSRAND],N2, intsign);
}
v0dim = blocksize;
returncode = 0;
/* compute minimal eigenvalues */
#ifdef MPI
/* pjdher(VOLUME*sizeof(spinor)/sizeof(complex), VOLUMEPLUSRAND*sizeof(spinor)/sizeof(complex),
shift, prec, omega, n_omega, ev_tr,
i+blocksize, j_max, j_min,
max_iterations, blocksize, blockwise, v0dim, (complex*)(&eigenvectors_ov[i*VOLUMEPLUSRAND]),
CG, solver_it_max,
threshold_min, decay_min, verbosity,
&converged, (complex*) eigenvectors_ov, eigenvalues_ov,
&returncode, JD_MINIMAL, 1, use_AV,
Operator[intsign]);*/
#else
jdher(VOLUME*sizeof(spinor)/sizeof(complex),
VOLUMEPLUSRAND*sizeof(spinor)/sizeof(complex),
shift, prec, blocksize, j_max, j_min,
max_iter, blocksize, blockwise, v0dim, (complex*) &eigenvectors_ov[i*VOLUMEPLUSRAND],
CG, solver_it_max,
threshold_min, decay_min, verbosity,
&converged, (complex*) eigenvectors_ov,
eigenvalues_ov,
&returncode, JD_MINIMAL, 1,
Operator[intsign]);
#endif
/* Save eigenvectors_ov temporary */
/* in order to be able to restart */
for (k=i; k < converged; k++){
if(intsign == 0){
sprintf(filename, "eigenvector_of_Dplus.%.2d.%s.%.4d", k, conf_filename, nstore);
}
else{
sprintf(filename, "eigenvector_of_Dminus.%.2d.%s.%.4d", k, conf_filename, nstore);
}
/* write_spinorfield(&eigenvectors_ov[k*VOLUMEPLUSRAND], filename);*/
}
/* order the eigenvalues_ov and vectors */
for(k = 0; k < converged; k++) {
idx[k] = k;
}
/* quicksort(converged, eigenvalues_ov, idx);*/
/* Check whether the index is detemined */
index = 0;
for(k = 0; k < converged; k++) {
absdifference = fabs(lowestmodes[first_blocksize*((intsign+1)%2)] - eigenvalues_ov[k]);
if(absdifference < 0.1*lowestmodes[first_blocksize*((intsign+1)%2)]) {
/* We have found the first non zero */
if(k < converged-1) {
determined = 1;
break;
}
else {
blocksize = 1;
shift = eigenvalues_ov[converged-1];
}
}
else {
index++;
}
}
/* If we have determined the index or */
/* hit the maximal number of ev */
if(determined == 1 || converged == (*nr_of_eigenvalues_ov)) {
break;
}
else if(g_proc_id == g_stdio_proc) {
if(blocksize != 1) {
printf("Index %s (or equal) than %s%d, continuing!\n\n",
intsign ? "lower" : "bigger",
intsign ? "-" : "+", index);
fflush( stdout );
}
else {
printf("Index is %s%d, one non zero is missing, continuing!\n\n",
intsign ? "-" : "+", index);
fflush( stdout );
}
}
}
#ifdef MPI
etime = MPI_Wtime();
#endif
/* Save the eigenvectors_ov */
for(i = 0; i < converged; i++){
eval[i].re = 2.*(1.+ov_s)*eigenvalues_ov[i];
eval[i].im = 0.;
if(intsign == 0){
sprintf(filename, "eigenvector_of_Dplus.%.2d.%s.%.4d", i, conf_filename, nstore);
}
else{
sprintf(filename, "eigenvector_of_Dminus.%.2d.%s.%.4d", i, conf_filename, nstore);
}
/* write_spinorfield(&eigenvectors_ov[idx[i]*VOLUMEPLUSRAND], filename);*/
}
/* Some Output */
if(g_proc_id == g_stdio_proc) {
printf("Index is %s%d!\n", intsign ? "-" : "+", index);
#ifdef MPI
printf("Zero modes determined in %f sec!\n", etime-atime);
#endif
}
if(g_proc_id == 0) {
sprintf(filename, "eigenvalues_of_overlap_proj.%s.%.4d", conf_filename, nstore);
ifs = fopen(filename, "w");
printf("\nThe following lowest modes have been computed:\n");
fprintf(ifs, "Index is %s%d!\n\n", intsign ? "-" : "+", index);
for(k = 0; k < 2; k++) {
if(k == intsign) {
for (i=0; i < converged; i++) {
fprintf(ifs, "%d %e %s\n", i, eval[i].re, intsign ? "negative" : "positive");
printf("%d %e %s\n", i, eval[i].re, intsign ? "negative" : "positive");
}
}
else {
for(i = 0; i < first_blocksize; i++) {
lowestmodes[((intsign+1)%2)*first_blocksize+i] = 2.*(1.+ov_s)*lowestmodes[((intsign+1)%2)*first_blocksize+i];
fprintf(ifs, "%d %e %s\n", i, lowestmodes[((intsign+1)%2)*first_blocksize+i], intsign ? "positive" : "negative");
printf("%d %e %s\n", i, lowestmodes[((intsign+1)%2)*first_blocksize+i], intsign ? "positive" : "negative");
}
}
}
fclose(ifs);
if(intsign != 0) signed_index = -index;
else signed_index = index;
for(k = 0; k < 2; k++) {
sprintf(filename, "eigenvalues_of_D%s.%s.%.4d",
k ? "minus" : "plus", conf_filename, nstore);
ifs = fopen(filename, "w");
if(k == intsign) {
fwrite(&converged, sizeof(int), 1, ifs);
fwrite(&signed_index, sizeof(int), 1, ifs);
for (i=index; i < converged; i++) {
fwrite(&eval[i].re, sizeof(double), 1, ifs);
}
}
else {
fwrite(&first_blocksize, sizeof(int), 1, ifs);
fwrite(&signed_index, sizeof(int), 1, ifs);
for(i = 0; i < first_blocksize; i++) {
fwrite(&lowestmodes[((intsign+1)%2)*first_blocksize+i], sizeof(double), 1, ifs);
}
}
fclose(ifs);
}
}
}
switch_on_adaptive_precision = 0;
/* Free memory */
free(eigenvectors_ov_);
free(lowvectors_);
free(eval);
free(eigenvalues_ov);
free(idx);
}
