/* Initialize the C++ front end. This function is very sensitive to
the exact order that things are done here. It would be nice if the
initialization done by this routine were moved to its subroutines,
and the ordering dependencies clarified and reduced. */
bool
cxx_init (void)
{
location_t saved_loc;
unsigned int i;
static const enum tree_code stmt_codes[] = {
CTOR_INITIALIZER, TRY_BLOCK, HANDLER,
EH_SPEC_BLOCK, USING_STMT, TAG_DEFN,
IF_STMT, CLEANUP_STMT, FOR_STMT,
RANGE_FOR_STMT, WHILE_STMT, DO_STMT,
BREAK_STMT, CONTINUE_STMT, SWITCH_STMT,
EXPR_STMT
};
memset (&statement_code_p, 0, sizeof (statement_code_p));
for (i = 0; i < ARRAY_SIZE (stmt_codes); i++)
statement_code_p[stmt_codes[i]] = true;
saved_loc = input_location;
input_location = BUILTINS_LOCATION;
init_reswords ();
init_tree ();
init_cp_semantics ();
init_operators ();
init_method ();
init_error ();
current_function_decl = NULL;
class_type_node = ridpointers[(int) RID_CLASS];
cxx_init_decl_processing ();
if (c_common_init () == false)
{
input_location = saved_loc;
return false;
}
init_cp_pragma ();
init_repo ();
input_location = saved_loc;
return true;
}
int main(int argc, char *argv[])
{
stack_t stack; /* daten stack fuer die werte */
opmap_t ops; /* lookup table der functionen */
stack_init(&stack);
opmap_init(&ops);
init_operators(&ops);
while(1)
{
char *input = read_line(stdin); /* liess eine zeile ein (siehe io.c) */
if(!input) break; /* eine leere zeile signalisiert programmende */
evaluate(&ops, input, &stack);
free(input);
}
return EXIT_SUCCESS;
}
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, 1);
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
z2_random_spinor_field(s[i], VOLUME);
/* 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);
/* boundary(g_kappa); */
/* g_mu = g_mu1; */
/* Compute little Dirac operators */
/* alt_block_compute_little_D(); */
if (g_debug_level > 0) {
check_projectors();
check_local_D();
}
if (g_debug_level > 1) {
check_little_D_inversion();
}
}
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);
}
}
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
}
void online_measurement(const int traj, const int id, const int ieo) {
int i, j, t, tt, t0;
double *Cpp = NULL, *Cpa = NULL, *Cp4 = NULL;
double res = 0., respa = 0., resp4 = 0.;
double atime, etime;
float tmp;
operator * optr;
#ifdef MPI
double mpi_res = 0., mpi_respa = 0., mpi_resp4 = 0.;
// send buffer for MPI_Gather
double *sCpp = NULL, *sCpa = NULL, *sCp4 = NULL;
#endif
FILE *ofs;
char *filename;
char buf[100];
spinor phi;
filename=buf;
sprintf(filename,"%s%.6d", "onlinemeas." ,traj);
init_operators();
if(no_operators < 1 && g_proc_id == 0) {
if(g_proc_id == 0) {
fprintf(stderr, "Warning! no operators defined in input file, cannot perform online correlator mesurements!\n");
}
return;
}
if(no_operators > 1 && g_proc_id == 0) {
fprintf(stderr, "Warning! number of operators defined larger than 1, using only the first!\n");
}
optr = &operator_list[0];
// we don't want to do inversion twice for this purpose here
optr->DownProp = 0;
if(optr->type != TMWILSON && optr->type != WILSON && optr->type != CLOVER) {
if(g_proc_id == 0) {
fprintf(stderr, "Warning! correlator online measurement currently only implemented for TMWILSON, WILSON and CLOVER\n");
fprintf(stderr, "Cannot perform online measurement!\n");
}
return;
}
/* generate random timeslice */
if(ranlxs_init == 0) {
rlxs_init(1, 123456);
}
ranlxs(&tmp, 1);
t0 = (int)(measurement_list[id].max_source_slice*tmp);
#ifdef MPI
MPI_Bcast(&t0, 1, MPI_INT, 0, MPI_COMM_WORLD);
#endif
if(g_debug_level > 1 && g_proc_id == 0) {
printf("# timeslice set to %d (T=%d) for online measurement\n", t0, g_nproc_t*T);
printf("# online measurements parameters: kappa = %g, mu = %g\n", g_kappa, g_mu/2./g_kappa);
}
atime = gettime();
#ifdef MPI
sCpp = (double*) calloc(T, sizeof(double));
sCpa = (double*) calloc(T, sizeof(double));
sCp4 = (double*) calloc(T, sizeof(double));
if(g_mpi_time_rank == 0) {
Cpp = (double*) calloc(g_nproc_t*T, sizeof(double));
Cpa = (double*) calloc(g_nproc_t*T, sizeof(double));
Cp4 = (double*) calloc(g_nproc_t*T, sizeof(double));
}
#else
Cpp = (double*) calloc(T, sizeof(double));
Cpa = (double*) calloc(T, sizeof(double));
Cp4 = (double*) calloc(T, sizeof(double));
#endif
source_generation_pion_only(g_spinor_field[0], g_spinor_field[1],
t0, 0, traj);
optr->sr0 = g_spinor_field[0];
optr->sr1 = g_spinor_field[1];
optr->prop0 = g_spinor_field[2];
optr->prop1 = g_spinor_field[3];
// op_id = 0, index_start = 0, write_prop = 0
optr->inverter(0, 0, 0);
/* now we bring it to normal format */
/* here we use implicitly DUM_MATRIX and DUM_MATRIX+1 */
convert_eo_to_lexic(g_spinor_field[DUM_MATRIX], g_spinor_field[2], g_spinor_field[3]);
/* now we sum only over local space for every t */
for(t = 0; t < T; t++) {
j = g_ipt[t][0][0][0];
res = 0.;
respa = 0.;
resp4 = 0.;
for(i = j; i < j+LX*LY*LZ; i++) {
res += _spinor_prod_re(g_spinor_field[DUM_MATRIX][j], g_spinor_field[DUM_MATRIX][j]);
_gamma0(phi, g_spinor_field[DUM_MATRIX][j]);
respa += _spinor_prod_re(g_spinor_field[DUM_MATRIX][j], phi);
_gamma5(phi, phi);
resp4 += _spinor_prod_im(g_spinor_field[DUM_MATRIX][j], phi);
}
#if defined MPI
MPI_Reduce(&res, &mpi_res, 1, MPI_DOUBLE, MPI_SUM, 0, g_mpi_time_slices);
res = mpi_res;
MPI_Reduce(&respa, &mpi_respa, 1, MPI_DOUBLE, MPI_SUM, 0, g_mpi_time_slices);
respa = mpi_respa;
MPI_Reduce(&resp4, &mpi_resp4, 1, MPI_DOUBLE, MPI_SUM, 0, g_mpi_time_slices);
resp4 = mpi_resp4;
sCpp[t] = +res/(g_nproc_x*LX)/(g_nproc_y*LY)/(g_nproc_z*LZ)*2.;
sCpa[t] = -respa/(g_nproc_x*LX)/(g_nproc_y*LY)/(g_nproc_z*LZ)*2.;
sCp4[t] = +resp4/(g_nproc_x*LX)/(g_nproc_y*LY)/(g_nproc_z*LZ)*2.;
#else
Cpp[t] = +res/(g_nproc_x*LX)/(g_nproc_y*LY)/(g_nproc_z*LZ)*2.;
Cpa[t] = -respa/(g_nproc_x*LX)/(g_nproc_y*LY)/(g_nproc_z*LZ)*2.;
Cp4[t] = +resp4/(g_nproc_x*LX)/(g_nproc_y*LY)/(g_nproc_z*LZ)*2.;
#endif
}
#ifdef MPI
/* some gymnastics needed in case of parallelisation */
if(g_mpi_time_rank == 0) {
MPI_Gather(sCpp, T, MPI_DOUBLE, Cpp, T, MPI_DOUBLE, 0, g_mpi_SV_slices);
MPI_Gather(sCpa, T, MPI_DOUBLE, Cpa, T, MPI_DOUBLE, 0, g_mpi_SV_slices);
MPI_Gather(sCp4, T, MPI_DOUBLE, Cp4, T, MPI_DOUBLE, 0, g_mpi_SV_slices);
}
#endif
/* and write everything into a file */
if(g_mpi_time_rank == 0 && g_proc_coords[0] == 0) {
ofs = fopen(filename, "w");
fprintf( ofs, "1 1 0 %e %e\n", Cpp[t0], 0.);
for(t = 1; t < g_nproc_t*T/2; t++) {
tt = (t0+t)%(g_nproc_t*T);
fprintf( ofs, "1 1 %d %e ", t, Cpp[tt]);
tt = (t0+g_nproc_t*T-t)%(g_nproc_t*T);
fprintf( ofs, "%e\n", Cpp[tt]);
}
tt = (t0+g_nproc_t*T/2)%(g_nproc_t*T);
fprintf( ofs, "1 1 %d %e %e\n", t, Cpp[tt], 0.);
fprintf( ofs, "2 1 0 %e %e\n", Cpa[t0], 0.);
for(t = 1; t < g_nproc_t*T/2; t++) {
tt = (t0+t)%(g_nproc_t*T);
fprintf( ofs, "2 1 %d %e ", t, Cpa[tt]);
tt = (t0+g_nproc_t*T-t)%(g_nproc_t*T);
fprintf( ofs, "%e\n", Cpa[tt]);
}
tt = (t0+g_nproc_t*T/2)%(g_nproc_t*T);
fprintf( ofs, "2 1 %d %e %e\n", t, Cpa[tt], 0.);
fprintf( ofs, "6 1 0 %e %e\n", Cp4[t0], 0.);
for(t = 1; t < g_nproc_t*T/2; t++) {
tt = (t0+t)%(g_nproc_t*T);
fprintf( ofs, "6 1 %d %e ", t, Cp4[tt]);
tt = (t0+g_nproc_t*T-t)%(g_nproc_t*T);
fprintf( ofs, "%e\n", Cp4[tt]);
}
tt = (t0+g_nproc_t*T/2)%(g_nproc_t*T);
fprintf( ofs, "6 1 %d %e %e\n", t, Cp4[tt], 0.);
fclose(ofs);
}
#ifdef MPI
if(g_mpi_time_rank == 0) {
free(Cpp);
free(Cpa);
free(Cp4);
}
free(sCpp);
free(sCpa);
free(sCp4);
#else
free(Cpp);
free(Cpa);
free(Cp4);
#endif
etime = gettime();
if(g_proc_id == 0 && g_debug_level > 0) {
printf("ONLINE: measurement done int t/s = %1.4e\n", etime - atime);
}
return;
}
int tmLQCD_invert_init(int argc, char *argv[], const int _verbose) {
DUM_DERI = 8;
DUM_MATRIX = DUM_DERI + 5;
NO_OF_SPINORFIELDS = DUM_MATRIX + 3;
// in read_input.h
verbose = _verbose;
g_use_clover_flag = 0;
#ifdef MPI
MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id);
#else
g_proc_id = 0;
#endif
/* Read the input file */
if( (read_input("invert.input")) != 0) {
fprintf(stderr, "tmLQCD_init_invert: Could not find input file: invert.input\nAborting...");
}
#ifdef OMP
init_openmp();
#endif
tmlqcd_mpi_init(argc, argv);
g_dbw2rand = 0;
for(int j = 0; j < no_operators; j++) if(!operator_list[j].even_odd_flag) even_odd_flag = 0;
#ifdef _GAUGE_COPY
int j = init_gauge_field(VOLUMEPLUSRAND, 1);
#else
int j = init_gauge_field(VOLUMEPLUSRAND, 0);
#endif
if (j != 0) {
fprintf(stderr, "tmLQCD_init_invert: Not enough memory for gauge_fields! Aborting...\n");
return(-1);
}
j = init_geometry_indices(VOLUMEPLUSRAND);
if (j != 0) {
fprintf(stderr, "tmLQCD_init_invert: Not enough memory for geometry indices! Aborting...\n");
return(-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, "tmLQCD_init_invert: Not enough memory for spinor fields! Aborting...\n");
return(-1);
}
// define the geometry
geometry();
// initialise the operators
init_operators();
#ifdef HAVE_GPU
if(usegpu_flag){
if(even_odd_flag){
init_mixedsolve_eo(g_gauge_field);
}
else{
init_mixedsolve(g_gauge_field);
}
# ifdef GPU_DOUBLE
/*init double fields w/o momenta*/
init_gpu_fields(0);
# endif
# ifdef TEMPORALGAUGE
int retval;
if((retval=init_temporalgauge(VOLUME, g_gauge_field)) !=0){
if(g_proc_id == 0) printf("tmLQCD_init_invert: Error while initializing temporal gauge. Aborting...\n");
exit(200);
}
# endif
}//usegpu_flag
#endif
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if (j != 0) {
fprintf(stderr, "tmLQCD_init_invert: Not enough memory for halffield! Aborting...\n");
return(-1);
}
if (g_sloppy_precision_flag == 1) {
j = init_dirac_halfspinor32();
if (j != 0) {
fprintf(stderr, "tmLQCD_init_invert: Not enough memory for 32-bit halffield! Aborting...\n");
return(-1);
}
}
# if (defined _PERSISTENT)
if (even_odd_flag)
init_xchange_halffield();
# endif
#endif
tmLQCD_invert_initialised = 1;
return(0);
}
/* Initialize the C++ front end. This function is very sensitive to
the exact order that things are done here. It would be nice if the
initialization done by this routine were moved to its subroutines,
and the ordering dependencies clarified and reduced. */
bool
cxx_init (void)
{
unsigned int i;
static const enum tree_code stmt_codes[] = {
CTOR_INITIALIZER, TRY_BLOCK, HANDLER,
EH_SPEC_BLOCK, USING_STMT, TAG_DEFN,
IF_STMT, CLEANUP_STMT, FOR_STMT,
WHILE_STMT, DO_STMT, BREAK_STMT,
CONTINUE_STMT, SWITCH_STMT, EXPR_STMT
};
memset (&statement_code_p, 0, sizeof (statement_code_p));
for (i = 0; i < ARRAY_SIZE (stmt_codes); i++)
statement_code_p[stmt_codes[i]] = true;
/* We cannot just assign to input_filename because it has already
been initialized and will be used later as an N_BINCL for stabs+
debugging. */
#ifdef USE_MAPPED_LOCATION
push_srcloc (BUILTINS_LOCATION);
#else
push_srcloc ("<built-in>", 0);
#endif
init_reswords ();
init_tree ();
init_cp_semantics ();
init_operators ();
init_method ();
init_error ();
current_function_decl = NULL;
class_type_node = ridpointers[(int) RID_CLASS];
cxx_init_decl_processing ();
/* The fact that G++ uses COMDAT for many entities (inline
functions, template instantiations, virtual tables, etc.) mean
that it is fundamentally unreliable to try to make decisions
about whether or not to output a particular entity until the end
of the compilation. However, the inliner requires that functions
be provided to the back end if they are to be inlined.
Therefore, we always use unit-at-a-time mode; in that mode, we
can provide entities to the back end and it will decide what to
emit based on what is actually needed. */
flag_unit_at_a_time = 1;
if (c_common_init () == false)
{
pop_srcloc();
return false;
}
init_cp_pragma ();
init_repo ();
pop_srcloc();
return true;
}
int main(int argc, char *argv[])
{
FILE *parameterfile = NULL;
int j, i, ix = 0, isample = 0, op_id = 0;
char datafilename[206];
char parameterfilename[206];
char conf_filename[50];
char * input_filename = NULL;
char * filename = NULL;
double plaquette_energy;
struct stout_parameters params_smear;
#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;
NO_OF_SPINORFIELDS = DUM_MATRIX + 4;
//4 extra fields (corresponding to DUM_MATRIX+0..5) for deg. and ND matrix mult.
NO_OF_SPINORFIELDS_32 = 6;
verbose = 0;
g_use_clover_flag = 0;
process_args(argc,argv,&input_filename,&filename);
set_default_filenames(&input_filename, &filename);
init_parallel_and_read_input(argc, argv, input_filename);
/* 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^nstore);
/* 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 TM_USE_MPI
g_dbw2rand = 0;
#endif
#ifdef _GAUGE_COPY
j = init_gauge_field(VOLUMEPLUSRAND, 1);
j += init_gauge_field_32(VOLUMEPLUSRAND, 1);
#else
j = init_gauge_field(VOLUMEPLUSRAND, 0);
j += init_gauge_field_32(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);
j += init_spinor_field_32(VOLUMEPLUSRAND / 2, NO_OF_SPINORFIELDS_32);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS);
j += init_spinor_field_32(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS_32);
}
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*/
strncpy(datafilename, filename, 200);
strcat(datafilename, ".data");
strncpy(parameterfilename, filename, 200);
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();
/* list and initialize measurements*/
if(g_proc_id == 0) {
printf("\n");
for(int j = 0; j < no_measurements; j++) {
printf("# measurement id %d, type = %d\n", j, measurement_list[j].type);
}
}
init_measurements();
/* 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);
}
/* for mixed precision solvers, the 32 bit halfspinor field must always be there */
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,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);
}
#ifdef TM_USE_MPI
xchange_gauge(g_gauge_field);
#endif
/*Convert to a 32 bit gauge field, after xchange*/
convert_32_gauge_field(g_gauge_field_32, g_gauge_field, VOLUMEPLUSRAND);
/*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);
}
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;
plaquette_energy = measure_plaquette( (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 any measurements are defined in the input file, do them here */
measurement * meas;
for(int imeas = 0; imeas < no_measurements; imeas++){
meas = &measurement_list[imeas];
if (g_proc_id == 0) {
fprintf(stdout, "#\n# Beginning online measurement.\n");
}
meas->measurefunc(nstore, imeas, even_odd_flag);
}
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 TM_USE_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){
invert_compute_modenumber();
}
// set up blocks if Deflation is used
if (g_dflgcr_flag)
init_blocks(nblocks_t, nblocks_x, nblocks_y, nblocks_z);
if(SourceInfo.type == SRC_TYPE_VOL || SourceInfo.type == SRC_TYPE_PION_TS || SourceInfo.type == SRC_TYPE_GEN_PION_TS) {
index_start = 0;
index_end = 1;
}
g_precWS=NULL;
if(use_preconditioning == 1){
/* todo load fftw wisdom */
#if (defined HAVE_FFTW ) && !( defined TM_USE_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 = operator_list[op_id].mu;
g_c_sw = operator_list[op_id].c_sw;
// DFLGCR and DFLFGMRES
if(operator_list[op_id].solver == DFLGCR || operator_list[op_id].solver == DFLFGMRES) {
generate_dfl_subspace(g_N_s, VOLUME, reproduce_randomnumber_flag);
}
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, random_seed);
//randmize initial guess for eigcg if needed-----experimental
if( (operator_list[op_id].solver == INCREIGCG) && (operator_list[op_id].solver_params.eigcg_rand_guess_opt) ){ //randomize the initial guess
gaussian_volume_source( operator_list[op_id].prop0, operator_list[op_id].prop1,isample,ix,0); //need to check this
}
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 TM_USE_OMP
free_omp_accumulators();
#endif
free_blocks();
free_dfl_subspace();
free_gauge_field();
free_gauge_field_32();
free_geometry_indices();
free_spinor_field();
free_spinor_field_32();
free_moment_field();
free_chi_spinor_field();
free(filename);
free(input_filename);
free(SourceInfo.basename);
free(PropInfo.basename);
#ifdef TM_USE_QUDA
_endQuda();
#endif
#ifdef TM_USE_MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
return(0);
#ifdef _KOJAK_INST
#pragma pomp inst end(main)
#endif
}