void det_heatbath(const int id, hamiltonian_field_t * const hf) {
monomial * mnl = &monomial_list[id];
double atime, etime;
atime = gettime();
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_eo(mnl->w_fields[0], mnl->rngrepro, RN_GAUSS);
mnl->energy0 = square_norm(mnl->w_fields[0], VOLUME/2, 1);
mnl->Qp(mnl->pf, mnl->w_fields[0]);
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_lexic(mnl->w_fields[0], mnl->rngrepro,RN_GAUSS);
mnl->energy0 = square_norm(mnl->w_fields[0], VOLUME, 1);
Q_plus_psi(mnl->pf, mnl->w_fields[0]);
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);
etime = gettime();
if(g_proc_id == 0) {
if(g_debug_level > 1) {
printf("# Time for %s monomial heatbath: %e s\n", mnl->name, etime-atime);
}
if(g_debug_level > 3) {
printf("called det_heatbath for id %d energey %f\n", id, mnl->energy0);
}
}
return;
}
void cloverdet_heatbath(const int id, hamiltonian_field_t * const hf) {
monomial * mnl = &monomial_list[id];
double atime, etime;
atime = gettime();
int N = VOLUME/2;
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( (const su3**) hf->gaugefield, mnl->kappa, mnl->c_sw);
if(!mnl->even_odd_flag) {
N = VOLUME;
random_spinor_field_lexic(mnl->w_fields[0], mnl->rngrepro, RN_GAUSS);
}
else {
sw_invert(EE, mnl->mu);
random_spinor_field_eo(mnl->w_fields[0], mnl->rngrepro, RN_GAUSS);
}
mnl->energy0 = square_norm(mnl->w_fields[0], N, 1);
mnl->Qp(mnl->pf, mnl->w_fields[0]);
chrono_add_solution(mnl->pf, mnl->csg_field, mnl->csg_index_array,
mnl->csg_N, &mnl->csg_n, N);
g_mu = g_mu1;
g_mu3 = 0.;
boundary(g_kappa);
etime = gettime();
if(g_proc_id == 0) {
if(g_debug_level > 1) {
printf("# Time for %s monomial heatbath: %e s\n", mnl->name, etime-atime);
}
if(g_debug_level > 3) {
printf("called cloverdet_heatbath for id %d energy %f\n", id, mnl->energy0);
}
}
return;
}
int main(int argc, char *argv[])
{
FILE *parameterfile = NULL;
int c, j, i, ix = 0, isample = 0, op_id = 0;
char * filename = NULL;
char datafilename[50];
char parameterfilename[50];
char conf_filename[50];
char * input_filename = NULL;
double plaquette_energy;
struct stout_parameters params_smear;
spinor **s, *s_;
#ifdef _KOJAK_INST
#pragma pomp inst init
#pragma pomp inst begin(main)
#endif
#if (defined SSE || defined SSE2 || SSE3)
signal(SIGILL, &catch_ill_inst);
#endif
DUM_DERI = 8;
DUM_MATRIX = DUM_DERI + 5;
#if ((defined BGL && defined XLC) || defined _USE_TSPLITPAR)
NO_OF_SPINORFIELDS = DUM_MATRIX + 3;
#else
NO_OF_SPINORFIELDS = DUM_MATRIX + 3;
#endif
verbose = 0;
g_use_clover_flag = 0;
#ifdef MPI
# 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
while ((c = getopt(argc, argv, "h?vVf: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 'v':
verbose = 1;
break;
case 'V':
fprintf(stdout,"%s %s\n",PACKAGE_STRING,git_hash);
exit(0);
break;
case 'h':
case '?':
default:
usage();
break;
}
}
if (input_filename == NULL) {
input_filename = "invert.input";
}
if (filename == NULL) {
filename = "output";
}
/* Read the input file */
if( (j = read_input(input_filename)) != 0) {
fprintf(stderr, "Could not find input file: %s\nAborting...\n", input_filename);
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
/* this DBW2 stuff is not needed for the inversion ! */
if (g_dflgcr_flag == 1) {
even_odd_flag = 0;
}
g_rgi_C1 = 0;
if (Nsave == 0) {
Nsave = 1;
}
if (g_running_phmc) {
NO_OF_SPINORFIELDS = DUM_MATRIX + 8;
}
tmlqcd_mpi_init(argc, argv);
g_dbw2rand = 0;
/* starts the single and double precision random number */
/* generator */
start_ranlux(rlxd_level, random_seed);
/* we need to make sure that we don't have even_odd_flag = 1 */
/* if any of the operators doesn't use it */
/* in this way even/odd can still be used by other operators */
for(j = 0; j < no_operators; j++) if(!operator_list[j].even_odd_flag) even_odd_flag = 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 (no_monomials > 0) {
if (even_odd_flag) {
j = init_monomials(VOLUMEPLUSRAND / 2, even_odd_flag);
}
else {
j = init_monomials(VOLUMEPLUSRAND, even_odd_flag);
}
if (j != 0) {
fprintf(stderr, "Not enough memory for monomial pseudo fermion fields! 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);
}
if (g_running_phmc) {
j = init_chi_spinor_field(VOLUMEPLUSRAND / 2, 20);
if (j != 0) {
fprintf(stderr, "Not enough memory for PHMC Chi fields! Aborting...\n");
exit(-1);
}
}
g_mu = g_mu1;
if (g_cart_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, "invert", git_hash);
fclose(parameterfile);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
phmc_invmaxev = 1.;
init_operators();
/* this could be maybe moved to init_operators */
#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)
if (even_odd_flag)
init_xchange_halffield();
# endif
#endif
for (j = 0; j < Nmeas; j++) {
sprintf(conf_filename, "%s.%.4d", gauge_input_filename, nstore);
if (g_cart_id == 0) {
printf("#\n# Trying to read gauge field from file %s in %s precision.\n",
conf_filename, (gauge_precision_read_flag == 32 ? "single" : "double"));
fflush(stdout);
}
if( (i = read_gauge_field(conf_filename)) !=0) {
fprintf(stderr, "Error %d while reading gauge field from %s\n Aborting...\n", i, conf_filename);
exit(-2);
}
if (g_cart_id == 0) {
printf("# Finished reading gauge field.\n");
fflush(stdout);
}
#ifdef MPI
xchange_gauge(g_gauge_field);
#endif
/*compute the energy of the gauge field*/
plaquette_energy = measure_gauge_action( (const su3**) g_gauge_field);
if (g_cart_id == 0) {
printf("# The computed 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( (const su3**) g_gauge_field);
if (g_cart_id == 0) {
printf("# The plaquette value after stouting is %e\n", plaquette_energy / (6.*VOLUME*g_nproc));
fflush(stdout);
}
}
if (reweighting_flag == 1) {
reweighting_factor(reweighting_samples, nstore);
}
/* Compute minimal eigenvalues, if wanted */
if (compute_evs != 0) {
eigenvalues(&no_eigenvalues, 5000, eigenvalue_precision,
0, compute_evs, nstore, even_odd_flag);
}
if (phmc_compute_evs != 0) {
#ifdef MPI
MPI_Finalize();
#endif
return(0);
}
/* Compute the mode number or topological susceptibility using spectral projectors, if wanted*/
if(compute_modenumber != 0 || compute_topsus !=0){
s_ = calloc(no_sources_z2*VOLUMEPLUSRAND+1, sizeof(spinor));
s = calloc(no_sources_z2, sizeof(spinor*));
if(s_ == NULL) {
printf("Not enough memory in %s: %d",__FILE__,__LINE__); exit(42);
}
if(s == NULL) {
printf("Not enough memory in %s: %d",__FILE__,__LINE__); exit(42);
}
for(i = 0; i < no_sources_z2; i++) {
#if (defined SSE3 || defined SSE2 || defined SSE)
s[i] = (spinor*)(((unsigned long int)(s_)+ALIGN_BASE)&~ALIGN_BASE)+i*VOLUMEPLUSRAND;
#else
s[i] = s_+i*VOLUMEPLUSRAND;
#endif
random_spinor_field_lexic(s[i], reproduce_randomnumber_flag,RN_Z2);
/* what is this here needed for?? */
/* spinor *aux_,*aux; */
/* #if ( defined SSE || defined SSE2 || defined SSE3 ) */
/* aux_=calloc(VOLUMEPLUSRAND+1, sizeof(spinor)); */
/* aux = (spinor *)(((unsigned long int)(aux_)+ALIGN_BASE)&~ALIGN_BASE); */
/* #else */
/* aux_=calloc(VOLUMEPLUSRAND, sizeof(spinor)); */
/* aux = aux_; */
/* #endif */
if(g_proc_id == 0) {
printf("source %d \n", i);
}
if(compute_modenumber != 0){
mode_number(s[i], mstarsq);
}
if(compute_topsus !=0) {
top_sus(s[i], mstarsq);
}
}
free(s);
free(s_);
}
/* move to operators as well */
if (g_dflgcr_flag == 1) {
/* set up deflation blocks */
init_blocks(nblocks_t, nblocks_x, nblocks_y, nblocks_z);
/* the can stay here for now, but later we probably need */
/* something like init_dfl_solver called somewhere else */
/* create set of approximate lowest eigenvectors ("global deflation subspace") */
/* g_mu = 0.; */
/* boundary(0.125); */
generate_dfl_subspace(g_N_s, VOLUME, reproduce_randomnumber_flag);
/* boundary(g_kappa); */
/* g_mu = g_mu1; */
/* Compute little Dirac operators */
/* alt_block_compute_little_D(); */
if (g_debug_level > 0) {
check_projectors(reproduce_randomnumber_flag);
check_local_D(reproduce_randomnumber_flag);
}
if (g_debug_level > 1) {
check_little_D_inversion(reproduce_randomnumber_flag);
}
}
if(SourceInfo.type == 1) {
index_start = 0;
index_end = 1;
}
g_precWS=NULL;
if(use_preconditioning == 1){
/* todo load fftw wisdom */
#if (defined HAVE_FFTW ) && !( defined MPI)
loadFFTWWisdom(g_spinor_field[0],g_spinor_field[1],T,LX);
#else
use_preconditioning=0;
#endif
}
if (g_cart_id == 0) {
fprintf(stdout, "#\n"); /*Indicate starting of the operator part*/
}
for(op_id = 0; op_id < no_operators; op_id++) {
boundary(operator_list[op_id].kappa);
g_kappa = operator_list[op_id].kappa;
g_mu = 0.;
if(use_preconditioning==1 && PRECWSOPERATORSELECT[operator_list[op_id].solver]!=PRECWS_NO ){
printf("# Using preconditioning with treelevel preconditioning operator: %s \n",
precWSOpToString(PRECWSOPERATORSELECT[operator_list[op_id].solver]));
/* initial preconditioning workspace */
operator_list[op_id].precWS=(spinorPrecWS*)malloc(sizeof(spinorPrecWS));
spinorPrecWS_Init(operator_list[op_id].precWS,
operator_list[op_id].kappa,
operator_list[op_id].mu/2./operator_list[op_id].kappa,
-(0.5/operator_list[op_id].kappa-4.),
PRECWSOPERATORSELECT[operator_list[op_id].solver]);
g_precWS = operator_list[op_id].precWS;
if(PRECWSOPERATORSELECT[operator_list[op_id].solver] == PRECWS_D_DAGGER_D) {
fitPrecParams(op_id);
}
}
for(isample = 0; isample < no_samples; isample++) {
for (ix = index_start; ix < index_end; ix++) {
if (g_cart_id == 0) {
fprintf(stdout, "#\n"); /*Indicate starting of new index*/
}
/* we use g_spinor_field[0-7] for sources and props for the moment */
/* 0-3 in case of 1 flavour */
/* 0-7 in case of 2 flavours */
prepare_source(nstore, isample, ix, op_id, read_source_flag, source_location);
operator_list[op_id].inverter(op_id, index_start, 1);
}
}
if(use_preconditioning==1 && operator_list[op_id].precWS!=NULL ){
/* free preconditioning workspace */
spinorPrecWS_Free(operator_list[op_id].precWS);
free(operator_list[op_id].precWS);
}
if(operator_list[op_id].type == OVERLAP){
free_Dov_WS();
}
}
nstore += Nsave;
}
#ifdef MPI
MPI_Finalize();
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_blocks();
free_dfl_subspace();
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
free_chi_spinor_field();
return(0);
#ifdef _KOJAK_INST
#pragma pomp inst end(main)
#endif
}
int main(int argc,char *argv[])
{
#ifdef _USE_HALFSPINOR
#undef _USE_HALFSPINOR
printf("# WARNING: USE_HALFSPINOR will be ignored (not supported here).\n");
#endif
if(even_odd_flag)
{
even_odd_flag=0;
printf("# WARNING: even_odd_flag will be ignored (not supported here).\n");
}
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("test_Dslash.input")) != 0) {
fprintf(stderr, "Could not find input file: test_Dslash.input\nAborting...\n");
exit(-1);
}
#ifdef OMP
init_openmp();
#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, g_gauge_field);
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge(g_gauge_field);
#endif
/* 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_lexic(g_spinor_field[k], reproduce_randomnumber_flag, RN_GAUSS);
}
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
/* here the actual Dslash */
D_psi(g_spinor_field[0], g_spinor_field[1]);
t2 = gettime();
dt=t2-t1;
#ifdef MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
if(g_proc_id==0) {
printf("# Time for Dslash %e sec.\n", sdt);
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 OMP
free_omp_accumulators();
#endif
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
return(0);
}
int main(int argc,char *argv[])
{
FILE *parameterfile = NULL;
char datafilename[206];
char parameterfilename[206];
char conf_filename[50];
char scalar_filename[50];
char * input_filename = NULL;
char * filename = NULL;
double plaquette_energy;
#ifdef _USE_HALFSPINOR
#undef _USE_HALFSPINOR
printf("# WARNING: USE_HALFSPINOR will be ignored (not supported here).\n");
#endif
if(even_odd_flag)
{
even_odd_flag=0;
printf("# WARNING: even_odd_flag will be ignored (not supported here).\n");
}
int j,j_max,k,k_max = 2;
_Complex double * drvsc;
#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.;
process_args(argc,argv,&input_filename,&filename);
set_default_filenames(&input_filename, &filename);
/* Read the input file */
if( (j = read_input(input_filename)) != 0) {
fprintf(stderr, "Could not find input file: %s\nAborting...\n", input_filename);
exit(-1);
}
if(g_proc_id==0) {
printf("parameter rho_BSM set to %f\n", rho_BSM);
printf("parameter eta_BSM set to %f\n", eta_BSM);
printf("parameter m0_BSM set to %f\n", m0_BSM);
}
#ifdef OMP
init_openmp();
#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);
j = init_bispinor_field(VOLUMEPLUSRAND, 12);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for bispinor fields! Aborting...\n");
exit(0);
}
j = init_spinor_field(VOLUMEPLUSRAND, 12);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
int numbScalarFields = 4;
j = init_scalar_field(VOLUMEPLUSRAND, numbScalarFields);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for scalar fields! Aborting...\n");
exit(0);
}
drvsc = malloc(18*VOLUMEPLUSRAND*sizeof(_Complex double));
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);
fflush(stdout);
}
/* define the geometry */
geometry();
j = init_bsm_2hop_lookup(VOLUME);
if ( j!= 0) {
// this should not be reached since the init function calls fatal_error anyway
fprintf(stderr, "Not enough memory for BSM2b 2hop lookup table! Aborting...\n");
exit(0);
}
/* define the boundary conditions for the fermion fields */
/* for the actual inversion, this is done in invert.c as the operators are iterated through */
//
// For the BSM operator we don't use kappa normalisation,
// as a result, when twisted boundary conditions are applied this needs to be unity.
// In addition, unlike in the Wilson case, the hopping term comes with a plus sign.
// However, in boundary(), the minus sign for the Wilson case is implicitly included.
// We therefore use -1.0 here.
boundary(-1.0);
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))
// fails, we're not using spinor fields
// check_xchange();
#endif
start_ranlux(1, 123456);
// read gauge field
if( strcmp(gauge_input_filename, "create_random_gaugefield") == 0 ) {
random_gauge_field(reproduce_randomnumber_flag, g_gauge_field);
}
else {
sprintf(conf_filename, "%s.%.4d", gauge_input_filename, nstore);
if (g_cart_id == 0) {
printf("#\n# Trying to read gauge field from file %s in %s precision.\n",
conf_filename, (gauge_precision_read_flag == 32 ? "single" : "double"));
fflush(stdout);
}
int i;
if( (i = read_gauge_field(conf_filename,g_gauge_field)) !=0) {
fprintf(stderr, "Error %d while reading gauge field from %s\n Aborting...\n", i, conf_filename);
exit(-2);
}
if (g_cart_id == 0) {
printf("# Finished reading gauge field.\n");
fflush(stdout);
}
}
// read scalar field
if( strcmp(scalar_input_filename, "create_random_scalarfield") == 0 ) {
for( int s=0; s<numbScalarFields; s++ )
ranlxd(g_scalar_field[s], VOLUME);
}
else {
sprintf(scalar_filename, "%s.%d", scalar_input_filename, nscalar);
if (g_cart_id == 0) {
printf("#\n# Trying to read scalar field from file %s in %s precision.\n",
scalar_filename, (scalar_precision_read_flag == 32 ? "single" : "double"));
fflush(stdout);
}
int i;
if( (i = read_scalar_field(scalar_filename,g_scalar_field)) !=0) {
fprintf(stderr, "Error %d while reading scalar field from %s\n Aborting...\n", i, scalar_filename);
exit(-2);
}
if (g_cart_id == 0) {
printf("# Finished reading scalar field.\n");
fflush(stdout);
}
}
#ifdef MPI
xchange_gauge(g_gauge_field);
#endif
/*compute the energy of the gauge field*/
plaquette_energy = measure_plaquette( (const su3**) g_gauge_field);
if (g_cart_id == 0) {
printf("# The computed plaquette value is %e.\n", plaquette_energy / (6.*VOLUME*g_nproc));
fflush(stdout);
}
#ifdef MPI
for( int s=0; s<numbScalarFields; s++ )
generic_exchange(g_scalar_field[s], sizeof(scalar));
#endif
/*initialize the bispinor fields*/
j_max=1;
sdt=0.;
// w
random_spinor_field_lexic( (spinor*)(g_bispinor_field[4]), reproduce_randomnumber_flag, RN_GAUSS);
random_spinor_field_lexic( (spinor*)(g_bispinor_field[4])+VOLUME, reproduce_randomnumber_flag, RN_GAUSS);
// for the D^\dagger test:
// v
random_spinor_field_lexic( (spinor*)(g_bispinor_field[5]), reproduce_randomnumber_flag, RN_GAUSS);
random_spinor_field_lexic( (spinor*)(g_bispinor_field[5])+VOLUME, reproduce_randomnumber_flag, RN_GAUSS);
#if defined MPI
generic_exchange(g_bispinor_field[4], sizeof(bispinor));
#endif
// print L2-norm of source:
double squarenorm = square_norm((spinor*)g_bispinor_field[4], 2*VOLUME, 1);
if(g_proc_id==0) {
printf("\n# square norm of the source: ||w||^2 = %e\n\n", squarenorm);
fflush(stdout);
}
double t_MG, t_BK;
/* inversion needs to be done first because it uses loads of the g_bispinor_fields internally */
#if TEST_INVERSION
if(g_proc_id==1)
printf("Testing inversion\n");
// Bartek's operator
t1 = gettime();
cg_her_bi(g_bispinor_field[9], g_bispinor_field[4],
25000, 1.0e-14, 0, VOLUME, &Q2_psi_BSM2b);
t_BK = gettime() - t1;
// Marco's operator
t1 = gettime();
cg_her_bi(g_bispinor_field[8], g_bispinor_field[4],
25000, 1.0e-14, 0, VOLUME, &Q2_psi_BSM2m);
t_MG = gettime() - t1;
if(g_proc_id==0)
printf("Operator inversion time: t_MG = %f sec \t t_BK = %f sec\n\n", t_MG, t_BK);
#endif
/* now apply the operators to the same bispinor field and do various comparisons */
// Marco's operator
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t_MG = 0.0;
t1 = gettime();
D_psi_BSM2m(g_bispinor_field[0], g_bispinor_field[4]);
t1 = gettime() - t1;
#ifdef MPI
MPI_Allreduce (&t1, &t_MG, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
t_MG = t1;
#endif
// Bartek's operator
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t_BK = 0.0;
t1 = gettime();
D_psi_BSM2b(g_bispinor_field[1], g_bispinor_field[4]);
t1 = gettime() - t1;
#ifdef MPI
MPI_Allreduce (&t1, &t_BK, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
t_BK = t1;
#endif
if(g_proc_id==0)
printf("Operator application time: t_MG = %f sec \t t_BK = %f sec\n\n", t_MG, t_BK);
squarenorm = square_norm((spinor*)g_bispinor_field[0], 2*VOLUME, 1);
if(g_proc_id==0) {
printf("# || D_MG w ||^2 = %.16e\n", squarenorm);
fflush(stdout);
}
squarenorm = square_norm((spinor*)g_bispinor_field[1], 2*VOLUME, 1);
if(g_proc_id==0) {
printf("# || D_BK w ||^2 = %.16e\n\n\n", squarenorm);
fflush(stdout);
}
diff( (spinor*)g_bispinor_field[3], (spinor*)g_bispinor_field[0], (spinor*)g_bispinor_field[1], 2*VOLUME);
printf("element-wise difference between (D_BK w) and (D_MG w)\n");
printf("( D_MG w - M_BK w )->sp_up.s0.c0= %.16e + I*(%.16e)\n\n", creal(g_bispinor_field[3][0].sp_up.s0.c0), cimag(g_bispinor_field[3][0].sp_up.s0.c0) );
double diffnorm = square_norm( (spinor*) g_bispinor_field[3], 2*VOLUME, 1 );
if(g_proc_id==0){
printf("Square norm of the difference\n");
printf("|| D_MG w - D_BK w ||^2 = %.16e \n\n\n", diffnorm);
}
// < D w, v >
printf("Check consistency of D and D^dagger\n");
_Complex double prod1_MG = scalar_prod( (spinor*)g_bispinor_field[0], (spinor*)g_bispinor_field[5], 2*VOLUME, 1 );
if(g_proc_id==0)
printf("< D_MG w, v > = %.16e + I*(%.16e)\n", creal(prod1_MG), cimag(prod1_MG));
_Complex double prod1_BK = scalar_prod( (spinor*)g_bispinor_field[1], (spinor*)g_bispinor_field[5], 2*VOLUME, 1 );
if(g_proc_id==0)
printf("< D_BK w, v > = %.16e + I*(%.16e)\n\n", creal(prod1_BK), cimag(prod1_BK));
// < w, D^\dagger v >
t_MG = gettime();
D_psi_dagger_BSM2m(g_bispinor_field[6], g_bispinor_field[5]);
t_MG = gettime()-t_MG;
t_BK = gettime();
D_psi_dagger_BSM2b(g_bispinor_field[7], g_bispinor_field[5]);
t_BK = gettime() - t_BK;
if(g_proc_id==0)
printf("Operator dagger application time: t_MG = %f sec \t t_BK = %f sec\n\n", t_MG, t_BK);
_Complex double prod2_MG = scalar_prod((spinor*)g_bispinor_field[4], (spinor*)g_bispinor_field[6], 2*VOLUME, 1);
_Complex double prod2_BK = scalar_prod((spinor*)g_bispinor_field[4], (spinor*)g_bispinor_field[7], 2*VOLUME, 1);
if( g_proc_id == 0 ){
printf("< w, D_MG^dagger v > = %.16e + I*(%.16e)\n", creal(prod2_MG), cimag(prod2_MG));
printf("< w, D_BK^dagger v > = %.16e + I*(%.16e)\n", creal(prod2_BK), cimag(prod2_BK));
printf("\n| < D_MG w, v > - < w, D_MG^dagger v > | = %.16e\n",cabs(prod2_MG-prod1_MG));
printf("| < D_BK w, v > - < w, D_BK^dagger v > | = %.16e\n\n",cabs(prod2_BK-prod1_BK));
}
#if TEST_INVERSION
// check result of inversion
Q2_psi_BSM2m(g_bispinor_field[10], g_bispinor_field[8]);
Q2_psi_BSM2b(g_bispinor_field[11], g_bispinor_field[8]);
assign_diff_mul((spinor*)g_bispinor_field[10], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME);
assign_diff_mul((spinor*)g_bispinor_field[11], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME);
double squarenorm_MGMG = square_norm((spinor*)g_bispinor_field[10], 2*VOLUME, 1);
double squarenorm_BKMG = square_norm((spinor*)g_bispinor_field[11], 2*VOLUME, 1);
if(g_proc_id==0) {
printf("# ||Q2_MG*(Q2_MG)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_MGMG);
printf("# ||Q2_BK*(Q2_MG)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_BKMG);
fflush(stdout);
}
Q2_psi_BSM2b(g_bispinor_field[10], g_bispinor_field[9]);
Q2_psi_BSM2m(g_bispinor_field[11], g_bispinor_field[9]);
assign_diff_mul((spinor*)g_bispinor_field[10], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME);
assign_diff_mul((spinor*)g_bispinor_field[11], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME);
double squarenorm_BKBK = square_norm((spinor*)g_bispinor_field[10], 2*VOLUME, 1);
double squarenorm_MGBK = square_norm((spinor*)g_bispinor_field[11], 2*VOLUME, 1);
if(g_proc_id==0) {
printf("# ||Q2_BK*(Q2_BK)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_BKBK);
printf("# ||Q2_MG*(Q2_BK)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_MGBK);
fflush(stdout);
}
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_gauge_field();
free_geometry_indices();
free_bispinor_field();
free_scalar_field();
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
return(0);
}
void reweighting_factor(const int N, const int nstore) {
int n = VOLUME;
monomial * mnl;
FILE * ofs;
hamiltonian_field_t hf;
hf.gaugefield = g_gauge_field;
hf.momenta = NULL;
hf.derivative = NULL;
hf.update_gauge_copy = g_update_gauge_copy;
double * data = (double*)calloc(no_monomials*N, sizeof(double));
double * trlog = (double*)calloc(no_monomials, sizeof(double));
// we compute the trlog part first, because they are independent of
// stochastic noise. This is only needed for even/odd monomials
for(int j = 0; j < no_monomials; j++) {
mnl = &monomial_list[j];
if(mnl->even_odd_flag) {
init_sw_fields();
double c_sw = mnl->c_sw;
if(c_sw < 0.) c_sw = 0.;
sw_term( (const su3**) hf.gaugefield, mnl->kappa, c_sw);
if(mnl->type != NDDETRATIO) {
trlog[j] = -sw_trace(0, mnl->mu);
}
else {
trlog[j] = -sw_trace_nd(0, mnl->mubar, mnl->epsbar);
}
sw_term( (const su3**) hf.gaugefield, mnl->kappa2, c_sw);
if(mnl->type != NDDETRATIO) {
trlog[j] -= -sw_trace(0, mnl->mu2);
}
else {
trlog[j] -= -sw_trace_nd(0, mnl->mubar2, mnl->epsbar2);
}
}
else {
trlog[j] = 0.;
}
if(g_proc_id == 0 && g_debug_level > 0) {
printf("# monomial[%d] %s, trlog = %e\n", j, mnl->name, trlog[j]);
}
}
for(int i = 0; i < N; i++) {
if(g_proc_id == 0 && g_debug_level > 0) {
printf("# computing reweighting factors for sample %d\n", i);
}
for(int j = 0; j < no_monomials; j++) {
mnl = &monomial_list[j];
if(mnl->type != GAUGE) {
if(mnl->even_odd_flag) {
random_spinor_field_eo(mnl->pf, mnl->rngrepro, RN_GAUSS);
mnl->energy0 = square_norm(mnl->pf, n/2, 1);
}
else {
random_spinor_field_lexic(mnl->pf, mnl->rngrepro, RN_GAUSS);
mnl->energy0 = square_norm(mnl->pf, n, 1);
}
if(g_proc_id == 0 && g_debug_level > 1) {
printf("# monomial[%d] %s, energy0 = %e\n", j, mnl->name, mnl->energy0);
}
if(mnl->type == NDDETRATIO) {
if(mnl->even_odd_flag) {
random_spinor_field_eo(mnl->pf2, mnl->rngrepro, RN_GAUSS);
mnl->energy0 += square_norm(mnl->pf, n/2, 1);
}
else {
random_spinor_field_lexic(mnl->pf, mnl->rngrepro, RN_GAUSS);
mnl->energy0 += square_norm(mnl->pf2, n, 1);
}
}
}
}
for(int j = 0; j < no_monomials; j++) {
mnl = &monomial_list[j];
if(mnl->type != GAUGE) {
double y = mnl->accfunction(j, &hf);
data[i*no_monomials + j] = y;
if(g_proc_id == 0 && g_debug_level > 0) {
printf("# monomial[%d] %s, stochastic part: w_%d=%e exp(w_%d)=%e\n", j, mnl->name, j, j, y, exp(y));
}
}
}
}
if(g_proc_id == 0) {
char filename[50];
sprintf(filename, "reweighting_factor.data.%.5d", nstore);
if((ofs = fopen(filename, "w")) == NULL) {
fatal_error("Could not open file for data output", "reweighting_factor");
}
else {
for(int j = 0; j < no_monomials; j++) {
mnl = &monomial_list[j];
for(int i = 0; i < N; i++) {
fprintf(ofs, "%.2d %.5d %e %e %e %e %.10e\n", j, i, mnl->kappa, mnl->kappa2, mnl->mu, mnl->mu2, data[i*no_monomials + j] + trlog[j]);
}
}
fclose(ofs);
}
}
free(data);
free(trlog);
}
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 TM_USE_MPI
static double dt2;
DUM_DERI = 6;
DUM_MATRIX = DUM_DERI+8;
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
# ifdef TM_USE_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 TM_USE_OMP
init_openmp();
#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 TM_USE_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 TM_USE_MPI && !(defined _USE_SHMEM))
check_xchange();
#endif
start_ranlux(1, 123456);
random_gauge_field(reproduce_randomnumber_flag, g_gauge_field);
#ifdef TM_USE_MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge(g_gauge_field);
#endif
if(even_odd_flag) {
sdt=0.; sqdt=0.0;
/*initialize the pseudo-fermion fields*/
for (k = 0; k < k_max; k++) {
random_spinor_field_eo(g_spinor_field[k], reproduce_randomnumber_flag, RN_GAUSS);
}
j_max=512;
antioptaway=0.0;
/* compute approximately how many applications we need to do to get a reliable measurement */
#ifdef TM_USE_MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
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);
}
}
dt = gettime()-t1;
// division by g_nproc because we will average over processes
j = (int)(ceil(j_max*31.0/dt/g_nproc));
#ifdef TM_USE_MPI
MPI_Allreduce(&j,&j_max, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
#else
j_max = j;
#endif
/* perform the actual benchmark */
#ifdef TM_USE_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);
}
}
dt = gettime()-t1;
#ifdef TM_USE_MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
qdt=dt*dt;
#ifdef TM_USE_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);
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);
#ifdef TM_USE_MPI
printf("# Communication switched on: \n");
#endif
printf("\n%12d Mflops(total) %8d Mflops(process)", (int)(g_nproc*1608.0f/sdt),(int)(1608.0f/sdt));
#ifdef TM_USE_OMP
printf(" %8d Mflops(thread)",(int)(1608.0f/(omp_num_threads*sdt)));
#endif
printf(" [ %d bit arithmetic ]\n\n",(int)(sizeof(spinor)/3));
fflush(stdout);
}
#ifdef TM_USE_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\n%12d Mflops(total) %8d Mflops(process)", (int)(g_nproc*1608.0f/dt),(int)(1608.0f/dt));
#ifdef TM_USE_OMP
printf(" %8d Mflops(thread)",(int)(1608.0f/(omp_num_threads*dt)));
#endif
printf(" [ %d bit arithmetic ]\n\n",(int)(sizeof(spinor)/3));
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=128;
sdt=0.;
for (k=0;k<k_max;k++) {
random_spinor_field_lexic(g_spinor_field[k], reproduce_randomnumber_flag, RN_GAUSS);
}
/* estimate a reasonable number of applications to get a reliable measurement */
#ifdef TM_USE_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;
// division by g_nproc because we will average over processes using MPI_SUM
j = (int)(ceil(j_max*31.0/dt/g_nproc));
#ifdef TM_USE_MPI
MPI_Allreduce(&j,&j_max, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
#else
j_max = j;
#endif
/* perform the actual measurement */
#ifdef TM_USE_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 TM_USE_MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
qdt=dt*dt;
#ifdef TM_USE_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);
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\n", sdt, sqdt, j_max);
printf(" %12d Mflops(total) %8d Mflops(process)", (int)(1680.0f*g_nproc/sdt),(int)(1680.0f/sdt));
#ifdef TM_USE_OMP
printf(" %8d Mflops(thread)",(int)(1680.0f/(omp_num_threads*sdt)));
#endif
printf(" [ %d bit arithmetic ]\n\n",(int)(sizeof(spinor)/3));
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 TM_USE_OMP
free_omp_accumulators();
#endif
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
#ifdef TM_USE_MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
return(0);
}
