int
main(int argc, char **argv)
{
int status;
char* env = NULL;
env = getenv("GPAW_OFFLOAD");
if (env) {
errno = 0;
gpaw_offload_enabled = strtol(env, NULL, 10);
if (errno) {
fprintf(stderr,
"Wrong value for for GPAW_OFFLOAD.\nShould be either 0 or 1, but was %s\n",
env);
}
}
fprintf(stderr, "GPAW info: GPAW_OFFLOAD=%d\n", gpaw_offload_enabled);
#ifdef CRAYPAT
PAT_region_begin(1, "C-Initializations");
#endif
#ifndef GPAW_OMP
MPI_Init(&argc, &argv);
#else
int granted;
MPI_Init_thread(&argc, &argv, MPI_THREAD_MULTIPLE, &granted);
if(granted != MPI_THREAD_MULTIPLE) exit(1);
#endif // GPAW_OMP
// Get initial timing
double t0 = MPI_Wtime();
#ifdef GPAW_PERFORMANCE_REPORT
gpaw_perf_init();
#endif
#ifdef GPAW_MPI_MAP
int tag = 99;
int myid, numprocs, i, procnamesize;
char procname[MPI_MAX_PROCESSOR_NAME];
MPI_Comm_size(MPI_COMM_WORLD, &numprocs );
MPI_Comm_rank(MPI_COMM_WORLD, &myid );
MPI_Get_processor_name(procname, &procnamesize);
if (myid > 0) {
MPI_Send(&procnamesize, 1, MPI_INT, 0, tag, MPI_COMM_WORLD);
MPI_Send(procname, procnamesize, MPI_CHAR, 0, tag, MPI_COMM_WORLD);
}
else {
printf("MPI_COMM_SIZE is %d \n", numprocs);
printf("%s \n", procname);
for (i = 1; i < numprocs; ++i) {
MPI_Recv(&procnamesize, 1, MPI_INT, i, tag, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
MPI_Recv(procname, procnamesize, MPI_CHAR, i, tag, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
printf("%s \n", procname);
}
}
#endif // GPAW_MPI_MAP
#ifdef GPAW_MPI_DEBUG
// Default Errhandler is MPI_ERRORS_ARE_FATAL
MPI_Errhandler_set(MPI_COMM_WORLD, MPI_ERRORS_RETURN);
#endif
// Progname seems to be needed in some circumstances to resolve
// correct default sys.path
Py_SetProgramName(argv[0]);
Py_Initialize();
#pragma offload target(mic) if(gpaw_offload_enabled)
{
init_openmp();
}
if (PyType_Ready(&MPIType) < 0)
return -1;
if (PyType_Ready(&LFCType) < 0)
return -1;
if (PyType_Ready(&LocalizedFunctionsType) < 0)
return -1;
if (PyType_Ready(&OperatorType) < 0)
return -1;
if (PyType_Ready(&SplineType) < 0)
return -1;
if (PyType_Ready(&TransformerType) < 0)
return -1;
if (PyType_Ready(&XCFunctionalType) < 0)
return -1;
if (PyType_Ready(&lxcXCFunctionalType) < 0)
return -1;
PyObject* m = Py_InitModule3("_gpaw", functions,
"C-extension for GPAW\n\n...\n");
if (m == NULL)
return -1;
Py_INCREF(&MPIType);
PyModule_AddObject(m, "Communicator", (PyObject *)&MPIType);
// Add initial time to _gpaw object
PyModule_AddObject(m, "time0", PyFloat_FromDouble(t0));
Py_INCREF(&LFCType);
Py_INCREF(&LocalizedFunctionsType);
Py_INCREF(&OperatorType);
Py_INCREF(&SplineType);
Py_INCREF(&TransformerType);
Py_INCREF(&XCFunctionalType);
Py_INCREF(&lxcXCFunctionalType);
#ifdef GPAW_WITH_HDF5
init_gpaw_hdf5();
#endif
import_array1(-1);
MPI_Barrier(MPI_COMM_WORLD);
#ifdef CRAYPAT
PAT_region_end(1);
PAT_region_begin(2, "all other");
#endif
status = Py_Main(argc, argv);
#ifdef CRAYPAT
PAT_region_end(2);
#endif
#ifdef GPAW_PERFORMANCE_REPORT
gpaw_perf_finalize();
#endif
MPI_Finalize();
return status;
}
int main(int argc,char *argv[]) {
char spinorfilename[100];
char * filename = NULL;
int sample=0, ts=0, ss=1, typeflag = 1, t0=0, piononly = 0, ext_sourceflag = 0;
int is, ic, j, filenameflag = 0, appendflag = 0;
complex co;
int c;
int prec=32;
verbose = 0;
g_use_clover_flag = 0;
nstore = 0;
L=0;
T=0;
#ifdef MPI
MPI_Init(&argc, &argv);
#endif
#ifdef OMP
/* FIXME: in principle this should not be set like this as it could result
in thread oversubscription when more than one process is run locally
unfortunately, there does not seem to be a standard way to determine
the number of "local" MPI processes */
omp_num_threads = omp_get_max_threads();
init_openmp();
#endif
while ((c = getopt(argc, argv, "h?NCpOEdao:L:T:n:t:s:S:P:")) != -1) {
switch (c) {
case 'L':
L = atoi(optarg);
LX = L;
LY = L;
LZ = L;
break;
case 'T':
T = atoi(optarg);
T_global = T;
break;
case 'N':
typeflag = 0;
break;
case 'd':
prec = 64;
break;
case 'O':
piononly = 1;
break;
case 'n':
nstore = atoi(optarg);
break;
case 's':
sample = atoi(optarg);
break;
case 't':
t0 = atoi(optarg);
break;
case 'S':
ss = atoi(optarg);
break;
case 'P':
ts = atoi(optarg);
break;
case 'o':
filename = calloc(200, sizeof(char));
strcpy(filename,optarg);
break;
case 'E':
ext_sourceflag = 1;
break;
case 'p':
filenameflag = 1;
break;
case 'a':
appendflag = 1;
break;
case 'h':
case '?':
default:
usage();
break;
}
}
if(ts == 0) {
ts = T;
}
if(filename == NULL){
filename = "source";
}
if(L==0 || T==0) {
if(g_proc_id == 0) {
fprintf(stderr, "L and T must be specified! Aborting...\n");
fflush( stderr );
}
exit(1);
}
tmlqcd_mpi_init(argc, argv);
j = init_geometry_indices(VOLUMEPLUSRAND);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for geometry_indices! Aborting...\n");
exit(0);
}
if(!ext_sourceflag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, 2);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND/2, 4);
}
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
/* define the geometry */
geometry();
if(!piononly) {
for(is = 0; is < 4; is ++) {
for(ic = 0; ic < 3; ic++) {
if(!filenameflag && !appendflag) {
sprintf(spinorfilename, "%s.%.4d.%.4d.%.2d.%.2d", filename, nstore, sample, t0, 3*is+ic);
}
else if(!filenameflag && appendflag) {
sprintf(spinorfilename, "%s.%.4d.%.4d.%.2d", filename, nstore, sample, t0);
}
else{
sprintf(spinorfilename, "%s.%.2d", filename, 3*is+ic);
}
if(!appendflag || (is == 0 && ic ==0)) {
printf("Generating source %s!\n", spinorfilename);
fflush(stdout);
}
source_generation_nucleon(g_spinor_field[0], g_spinor_field[1],
is, ic, t0, ts, ss, sample, nstore, typeflag);
co = scalar_prod(g_spinor_field[1], g_spinor_field[1], VOLUME/2, 1);
if((is == 0 && ic == 0) || appendflag == 0) {
write_source_type(0, spinorfilename);
}
write_source(g_spinor_field[0], g_spinor_field[1], spinorfilename, 1, prec);
}
}
}
else {
if(!ext_sourceflag) {
if(!filenameflag) {
sprintf(spinorfilename, "%s.%.4d.%.4d.%.2d", filename, nstore, sample, t0);
}
else {
sprintf(spinorfilename, "%s", filename);
}
printf("Generating source %s!\n", spinorfilename);
fflush(stdout);
source_generation_pion_only(g_spinor_field[0], g_spinor_field[1],
t0, sample, nstore);
co = scalar_prod(g_spinor_field[1], g_spinor_field[1], VOLUME/2, 1);
write_source_type(0, spinorfilename);
write_source(g_spinor_field[0], g_spinor_field[1], spinorfilename, 1, prec);
}
else {
if(!filenameflag) {
sprintf(spinorfilename, "%s.%.4d.%.4d.%.2d.inverted", filename, nstore, sample, t0);
}
else {
sprintf(spinorfilename, "%s.inverted", filename);
}
read_lime_spinor(g_spinor_field[0], g_spinor_field[1], spinorfilename, 0);
printf("Generating ext. pion source %s!\n", spinorfilename);
extended_pion_source(g_spinor_field[2], g_spinor_field[3],
g_spinor_field[0], g_spinor_field[1],
t0, 0., 0., 0.);
if(!filenameflag) {
sprintf(spinorfilename, "g%s.%.4d.%.4d.%.2d", filename, nstore, sample, t0);
}
else {
sprintf(spinorfilename, "g%s", filename);
}
write_source_type(0, spinorfilename);
write_source(g_spinor_field[2], g_spinor_field[3], spinorfilename, 1, prec);
}
}
#ifdef MPI
MPI_Finalize();
#endif
free_geometry_indices();
free_spinor_field();
return(0);
}
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);
}
int main(int argc,char *argv[]) {
FILE *parameterfile=NULL, *countfile=NULL;
char *filename = NULL;
char datafilename[206];
char parameterfilename[206];
char gauge_filename[50];
char nstore_filename[50];
char tmp_filename[50];
char *input_filename = NULL;
int status = 0, accept = 0;
int j,ix,mu, trajectory_counter=0;
unsigned int const io_max_attempts = 5; /* Make this configurable? */
unsigned int const io_timeout = 5; /* Make this configurable? */
struct timeval t1;
/* Energy corresponding to the Gauge part */
double plaquette_energy = 0., rectangle_energy = 0.;
/* Acceptance rate */
int Rate=0;
/* Do we want to perform reversibility checks */
/* See also return_check_flag in read_input.h */
int return_check = 0;
/* For getopt */
int c;
paramsXlfInfo *xlfInfo;
/* For online measurements */
measurement * meas;
int imeas;
#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
strcpy(gauge_filename,"conf.save");
strcpy(nstore_filename,".nstore_counter");
strcpy(tmp_filename, ".conf.tmp");
verbose = 1;
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
process_args(argc,argv,&input_filename,&filename);
set_default_filenames(&input_filename,&filename);
/* Read the input file */
if( (status = read_input(input_filename)) != 0) {
fprintf(stderr, "Could not find input file: %s\nAborting...\n", input_filename);
exit(-1);
}
#ifdef OMP
init_openmp();
#endif
DUM_DERI = 4;
DUM_SOLVER = DUM_DERI+1;
DUM_MATRIX = DUM_SOLVER+6;
if(g_running_phmc) {
NO_OF_SPINORFIELDS = DUM_MATRIX+8;
}
else {
NO_OF_SPINORFIELDS = DUM_MATRIX+6;
}
DUM_BI_DERI = 6;
DUM_BI_SOLVER = DUM_BI_DERI+7;
DUM_BI_MATRIX = DUM_BI_SOLVER+6;
NO_OF_BISPINORFIELDS = DUM_BI_MATRIX+6;
tmlqcd_mpi_init(argc, argv);
if(nstore == -1) {
countfile = fopen(nstore_filename, "r");
if(countfile != NULL) {
j = fscanf(countfile, "%d %d %s\n", &nstore, &trajectory_counter, gauge_input_filename);
if(j < 1) nstore = 0;
if(j < 2) trajectory_counter = 0;
fclose(countfile);
}
else {
nstore = 0;
trajectory_counter = 0;
}
}
#ifndef MPI
g_dbw2rand = 0;
#endif
g_mu = g_mu1;
#ifdef _GAUGE_COPY
status = init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
status = init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
if (status != 0) {
fprintf(stderr, "Not enough memory for gauge_fields! Aborting...\n");
exit(0);
}
j = init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if (j != 0) {
fprintf(stderr, "Not enough memory for geometry_indices! Aborting...\n");
exit(0);
}
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(0);
}
if(even_odd_flag) {
j = init_csg_field(VOLUMEPLUSRAND/2);
}
else {
j = init_csg_field(VOLUMEPLUSRAND);
}
if (j != 0) {
fprintf(stderr, "Not enough memory for csg 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_running_phmc) {
j = init_bispinor_field(VOLUME/2, NO_OF_BISPINORFIELDS);
if (j!= 0) {
fprintf(stderr, "Not enough memory for bi-spinor fields! Aborting...\n");
exit(0);
}
}
/* list and initialize measurements*/
if(g_proc_id == 0) {
printf("\n");
for(j = 0; j < no_measurements; j++) {
printf("# measurement id %d, type = %d: Frequency %d\n", j, measurement_list[j].type, measurement_list[j].freq);
}
}
init_measurements();
/*construct the filenames for the observables and the parameters*/
strncpy(datafilename,filename,200);
strcat(datafilename,".data");
strncpy(parameterfilename,filename,200);
strcat(parameterfilename,".para");
if(g_proc_id == 0){
parameterfile = fopen(parameterfilename, "a");
write_first_messages(parameterfile, "hmc", git_hash);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
status = check_geometry();
if (status != 0) {
fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n");
exit(1);
}
#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) {
init_dirac_halfspinor32();
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
/* Initialise random number generator */
start_ranlux(rlxd_level, random_seed^trajectory_counter);
/* Set up the gauge field */
/* continue and restart */
if(startoption==3 || startoption == 2) {
if(g_proc_id == 0) {
printf("# Trying to read gauge field from file %s in %s precision.\n",
gauge_input_filename, (gauge_precision_read_flag == 32 ? "single" : "double"));
fflush(stdout);
}
if( (status = read_gauge_field(gauge_input_filename)) != 0) {
fprintf(stderr, "Error %d while reading gauge field from %s\nAborting...\n", status, gauge_input_filename);
exit(-2);
}
if (g_proc_id == 0){
printf("# Finished reading gauge field.\n");
fflush(stdout);
}
}
else if (startoption == 1) {
/* hot */
random_gauge_field(reproduce_randomnumber_flag, g_gauge_field);
}
else if(startoption == 0) {
/* cold */
unit_g_gauge_field();
}
/*For parallelization: exchange the gaugefield */
#ifdef MPI
xchange_gauge(g_gauge_field);
#endif
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(0);
}
init_integrator();
if(g_proc_id == 0) {
for(j = 0; j < no_monomials; j++) {
printf("# monomial id %d type = %d timescale %d\n", j, monomial_list[j].type, monomial_list[j].timescale);
}
}
plaquette_energy = measure_gauge_action( (const su3**) g_gauge_field);
if(g_rgi_C1 > 0. || g_rgi_C1 < 0.) {
rectangle_energy = measure_rectangles( (const su3**) g_gauge_field);
if(g_proc_id == 0){
fprintf(parameterfile,"# Computed rectangle value: %14.12f.\n",rectangle_energy/(12.*VOLUME*g_nproc));
}
}
//eneg = g_rgi_C0 * plaquette_energy + g_rgi_C1 * rectangle_energy;
if(g_proc_id == 0) {
fprintf(parameterfile,"# Computed plaquette value: %14.12f.\n", plaquette_energy/(6.*VOLUME*g_nproc));
printf("# Computed plaquette value: %14.12f.\n", plaquette_energy/(6.*VOLUME*g_nproc));
fclose(parameterfile);
}
/* set ddummy to zero */
for(ix = 0; ix < VOLUMEPLUSRAND; ix++){
for(mu=0; mu<4; mu++){
ddummy[ix][mu].d1=0.;
ddummy[ix][mu].d2=0.;
ddummy[ix][mu].d3=0.;
ddummy[ix][mu].d4=0.;
ddummy[ix][mu].d5=0.;
ddummy[ix][mu].d6=0.;
ddummy[ix][mu].d7=0.;
ddummy[ix][mu].d8=0.;
}
}
if(g_proc_id == 0) {
gettimeofday(&t1,NULL);
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "!!! Timestamp %ld, Nsave = %d, g_mu = %e, g_mu1 = %e, g_mu_2 = %e, g_mu3 = %e, beta = %f, kappa = %f, C1 = %f, ",
t1.tv_sec, Nsave, g_mu, g_mu1, g_mu2, g_mu3, g_beta, g_kappa, g_rgi_C1);
for(j = 0; j < Integrator.no_timescales; j++) {
fprintf(countfile, "n_int[%d] = %d ", j, Integrator.no_mnls_per_ts[j]);
}
fprintf(countfile, "\n");
fclose(countfile);
}
/* Loop for measurements */
for(j = 0; j < Nmeas; j++) {
if(g_proc_id == 0) {
printf("#\n# Starting trajectory no %d\n", trajectory_counter);
}
return_check = return_check_flag && (trajectory_counter%return_check_interval == 0);
accept = update_tm(&plaquette_energy, &rectangle_energy, datafilename,
return_check, Ntherm<trajectory_counter, trajectory_counter);
Rate += accept;
/* Save gauge configuration all Nsave times */
if((Nsave !=0) && (trajectory_counter%Nsave == 0) && (trajectory_counter!=0)) {
sprintf(gauge_filename,"conf.%.4d", nstore);
if(g_proc_id == 0) {
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "%.4d, measurement %d of %d, Nsave = %d, Plaquette = %e, trajectory nr = %d\n",
nstore, j, Nmeas, Nsave, plaquette_energy/(6.*VOLUME*g_nproc),
trajectory_counter);
fclose(countfile);
}
nstore ++;
}
else {
sprintf(gauge_filename,"conf.save");
}
if(((Nsave !=0) && (trajectory_counter%Nsave == 0) && (trajectory_counter!=0)) || (write_cp_flag == 1) || (j >= (Nmeas - 1))) {
/* If a reversibility check was performed this trajectory, and the trajectory was accepted,
* then the configuration is currently stored in .conf.tmp, written out by update_tm.
* In that case also a readback was performed, so no need to test .conf.tmp
* In all other cases the gauge configuration still needs to be written out here. */
if (!(return_check && accept))
for (unsigned int attempt = 1; attempt <= io_max_attempts; ++attempt)
{
if (g_proc_id == 0)
fprintf(stdout, "# Writing gauge field to %s.\n", tmp_filename);
xlfInfo = construct_paramsXlfInfo(plaquette_energy/(6.*VOLUME*g_nproc), trajectory_counter);
status = write_gauge_field( tmp_filename, gauge_precision_write_flag, xlfInfo);
free(xlfInfo);
if (status) {
/* Writing the gauge field failed directly */
fprintf(stderr, "Error %d while writing gauge field to %s\nAborting...\n", status, tmp_filename);
exit(-2);
}
if (g_disable_IO_checks) {
if (g_proc_id == 0)
fprintf(stdout, "# Write completed successfully. Write not verified!\n");
break;
}
/* Read gauge field back to verify the writeout */
if (g_proc_id == 0)
fprintf(stdout, "# Write completed, verifying write...\n");
status = read_gauge_field(tmp_filename);
if (!status) {
if (g_proc_id == 0)
fprintf(stdout, "# Write successfully verified.\n");
break;
}
if (g_proc_id == 0) {
fprintf(stdout, "# Writeout of %s returned no error, but verification discovered errors.\n", tmp_filename);
fprintf(stdout, "# Potential disk or MPI I/O error.\n");
fprintf(stdout, "# This was attempt %d out of %d.\n", attempt, io_max_attempts);
}
if (attempt == io_max_attempts)
kill_with_error(NULL, g_proc_id, "Persistent I/O failures!\n");
if (g_proc_id == 0)
fprintf(stdout, "# Will attempt to write again in %d seconds.\n", io_timeout);
sleep(io_timeout);
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
}
/* Now move .conf.tmp into place */
if(g_proc_id == 0) {
fprintf(stdout, "# Renaming %s to %s.\n", tmp_filename, gauge_filename);
if (rename(tmp_filename, gauge_filename) != 0) {
/* Errno can be inspected here for more descriptive error reporting */
fprintf(stderr, "Error while trying to rename temporary file %s to %s. Unable to proceed.\n", tmp_filename, gauge_filename);
exit(-2);
}
countfile = fopen(nstore_filename, "w");
fprintf(countfile, "%d %d %s\n", nstore, trajectory_counter+1, gauge_filename);
fclose(countfile);
}
}
/* online measurements */
for(imeas = 0; imeas < no_measurements; imeas++){
meas = &measurement_list[imeas];
if(trajectory_counter%meas->freq == 0){
if (g_proc_id == 0) {
fprintf(stdout, "#\n# Beginning online measurement.\n");
}
meas->measurefunc(trajectory_counter, imeas, even_odd_flag);
}
}
if(g_proc_id == 0) {
verbose = 1;
}
ix = reread_input("hmc.reread");
if(g_proc_id == 0) {
verbose = 0;
}
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
if(ix == 0 && g_proc_id == 0) {
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "# Changed input parameters according to hmc.reread: measurement %d of %d\n", j, Nmeas);
fclose(countfile);
printf("# Changed input parameters according to hmc.reread (see stdout): measurement %d of %d\n", j, Nmeas);
remove("hmc.reread");
}
trajectory_counter++;
} /* end of loop over trajectories */
if(g_proc_id == 0 && Nmeas != 0) {
printf("# Acceptance rate was %3.2f percent, %d out of %d trajectories accepted.\n", 100.*(double)Rate/(double)Nmeas, Rate, Nmeas);
fflush(stdout);
parameterfile = fopen(parameterfilename, "a");
fprintf(parameterfile, "# Acceptance rate was %3.2f percent, %d out of %d trajectories accepted.\n", 100.*(double)Rate/(double)Nmeas, Rate, Nmeas);
fclose(parameterfile);
}
#ifdef MPI
MPI_Finalize();
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_gauge_tmp();
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
free_monomials();
if(g_running_phmc) {
free_bispinor_field();
free_chi_spinor_field();
}
free(input_filename);
free(filename);
return(0);
#ifdef _KOJAK_INST
#pragma pomp inst end(main)
#endif
}
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);
}
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);
}
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;
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
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);
}
#ifdef OMP
init_openmp();
#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*/
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);
}
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,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 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);
}
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 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);
//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 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();
free(filename);
free(input_filename);
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
return(0);
#ifdef _KOJAK_INST
#pragma pomp inst end(main)
#endif
}
