int main(int argc, char **argv) {
int c, i, mu, status;
int ispin, icol, isc;
int n_c = 3;
int n_s = 4;
int count = 0;
int filename_set = 0;
int dims[4] = {0,0,0,0};
int grid_size[4];
int l_LX_at, l_LXstart_at;
int x0, x1, x2, x3, ix, iix, iy, is, it, i3;
int sl0, sl1, sl2, sl3, have_source_flag=0;
int source_proc_coords[4], lsl0, lsl1, lsl2, lsl3;
int check_residuum = 0;
unsigned int VOL3, V5;
int do_gt = 0;
int full_orbit = 0;
int smear_source = 0;
char filename[200], source_filename[200], source_filename_write[200];
double ratime, retime;
double plaq_r=0., plaq_m=0., norm, norm2;
double spinor1[24];
double *gauge_qdp[4], *gauge_field_timeslice=NULL, *gauge_field_smeared=NULL;
double _1_2_kappa, _2_kappa, phase;
FILE *ofs;
int mu_trans[4] = {3, 0, 1, 2};
int threadid, nthreads;
int timeslice, source_timeslice;
char rng_file_in[100], rng_file_out[100];
int *source_momentum=NULL;
int source_momentum_class = -1;
int source_momentum_no = 0;
int source_momentum_runs = 1;
int imom;
int num_gpu_on_node=0, rank;
int source_location_5d_iseven;
int convert_sign=0;
#ifdef HAVE_QUDA
int rotate_gamma_basis = 1;
#else
int rotate_gamma_basis = 0;
#endif
omp_lock_t *lck = NULL, gen_lck[1];
int key = 0;
/****************************************************************************/
/* for smearing parallel to inversion */
double *smearing_spinor_field[] = {NULL,NULL};
int dummy_flag = 0;
/****************************************************************************/
/****************************************************************************/
#if (defined HAVE_QUDA) && (defined MULTI_GPU)
int x_face_size, y_face_size, z_face_size, t_face_size, pad_size;
#endif
/****************************************************************************/
/************************************************/
int qlatt_nclass;
int *qlatt_id=NULL, *qlatt_count=NULL, **qlatt_rep=NULL, **qlatt_map=NULL;
double **qlatt_list=NULL;
/************************************************/
/************************************************/
double boundary_condition_factor;
int boundary_condition_factor_set = 0;
/************************************************/
//#ifdef MPI
// kernelPackT = true;
//#endif
/***********************************************
* QUDA parameters
***********************************************/
#ifdef HAVE_QUDA
QudaPrecision cpu_prec = QUDA_DOUBLE_PRECISION;
QudaPrecision cuda_prec = QUDA_DOUBLE_PRECISION;
QudaPrecision cuda_prec_sloppy = QUDA_SINGLE_PRECISION;
QudaGaugeParam gauge_param = newQudaGaugeParam();
QudaInvertParam inv_param = newQudaInvertParam();
#endif
while ((c = getopt(argc, argv, "soch?vgf:p:b:S:R:")) != -1) {
switch (c) {
case 'v':
g_verbose = 1;
break;
case 'g':
do_gt = 1;
break;
case 'f':
strcpy(filename, optarg);
filename_set=1;
break;
case 'c':
check_residuum = 1;
fprintf(stdout, "# [invert_dw_quda] will check residuum again\n");
break;
case 'p':
n_c = atoi(optarg);
fprintf(stdout, "# [invert_dw_quda] will use number of colors = %d\n", n_c);
break;
case 'o':
full_orbit = 1;
fprintf(stdout, "# [invert_dw_quda] will invert for full orbit, if source momentum set\n");
case 's':
smear_source = 1;
fprintf(stdout, "# [invert_dw_quda] will smear the sources if they are read from file\n");
break;
case 'b':
boundary_condition_factor = atof(optarg);
boundary_condition_factor_set = 1;
fprintf(stdout, "# [invert_dw_quda] const. boundary condition factor set to %e\n", boundary_condition_factor);
break;
case 'S':
convert_sign = atoi(optarg);
fprintf(stdout, "# [invert_dw_quda] using convert sign %d\n", convert_sign);
break;
case 'R':
rotate_gamma_basis = atoi(optarg);
fprintf(stdout, "# [invert_dw_quda] rotate gamma basis %d\n", rotate_gamma_basis);
break;
case 'h':
case '?':
default:
usage();
break;
}
}
// get the time stamp
g_the_time = time(NULL);
/**************************************
* set the default values, read input
**************************************/
if(filename_set==0) strcpy(filename, "cvc.input");
if(g_proc_id==0) fprintf(stdout, "# Reading input from file %s\n", filename);
read_input_parser(filename);
#ifdef MPI
#ifdef HAVE_QUDA
grid_size[0] = g_nproc_x;
grid_size[1] = g_nproc_y;
grid_size[2] = g_nproc_z;
grid_size[3] = g_nproc_t;
fprintf(stdout, "# [] g_nproc = (%d,%d,%d,%d)\n", g_nproc_x, g_nproc_y, g_nproc_z, g_nproc_t);
initCommsQuda(argc, argv, grid_size, 4);
#else
MPI_Init(&argc, &argv);
#endif
#endif
#if (defined PARALLELTX) || (defined PARALLELTXY)
EXIT_WITH_MSG(1, "[] Error, 2-dim./3-dim. MPI-Version not yet implemented");
#endif
// some checks on the input data
if((T_global == 0) || (LX==0) || (LY==0) || (LZ==0)) {
if(g_proc_id==0) fprintf(stderr, "[invert_dw_quda] Error, T and L's must be set\n");
usage();
}
// set number of openmp threads
// initialize MPI parameters
mpi_init(argc, argv);
// the volume of a timeslice
VOL3 = LX*LY*LZ;
V5 = T*LX*LY*LZ*L5;
g_kappa5d = 0.5 / (5. + g_m5);
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] kappa5d = %e\n", g_kappa5d);
fprintf(stdout, "# [%2d] parameters:\n"\
"# [%2d] T = %3d\n"\
"# [%2d] Tstart = %3d\n"\
"# [%2d] L5 = %3d\n",\
g_cart_id, g_cart_id, T, g_cart_id, Tstart, g_cart_id, L5);
#ifdef MPI
if(T==0) {
fprintf(stderr, "[%2d] local T is zero; exit\n", g_cart_id);
MPI_Abort(MPI_COMM_WORLD, 1);
MPI_Finalize();
exit(2);
}
#endif
if(init_geometry() != 0) {
fprintf(stderr, "[invert_dw_quda] Error from init_geometry\n");
EXIT(1);
}
geometry();
if( init_geometry_5d() != 0 ) {
fprintf(stderr, "[invert_dw_quda] Error from init_geometry_5d\n");
EXIT(2);
}
geometry_5d();
/**************************************
* initialize the QUDA library
**************************************/
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] initializing quda\n");
#ifdef HAVE_QUDA
// cudaGetDeviceCount(&num_gpu_on_node);
if(g_gpu_per_node<0) {
if(g_cart_id==0) fprintf(stderr, "[] Error, number of GPUs per node not set\n");
EXIT(106);
} else {
num_gpu_on_node = g_gpu_per_node;
}
#ifdef MPI
rank = comm_rank();
#else
rank = 0;
#endif
g_gpu_device_number = rank % num_gpu_on_node;
fprintf(stdout, "# [] process %d/%d uses device %d\n", rank, g_cart_id, g_gpu_device_number);
initQuda(g_gpu_device_number);
#endif
/**************************************
* prepare the gauge field
**************************************/
// read the gauge field from file
alloc_gauge_field(&g_gauge_field, VOLUMEPLUSRAND);
if(strcmp( gaugefilename_prefix, "identity")==0 ) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] Setting up unit gauge field\n");
for(ix=0;ix<VOLUME; ix++) {
for(mu=0;mu<4;mu++) {
_cm_eq_id(g_gauge_field+_GGI(ix,mu));
}
}
} else if(strcmp( gaugefilename_prefix, "random")==0 ) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] Setting up random gauge field with seed = %d\n", g_seed);
init_rng_state(g_seed, &g_rng_state);
random_gauge_field(g_gauge_field, 1.);
plaquette(&plaq_m);
sprintf(filename, "%s.%.4d", gaugefilename_prefix, Nconf);
check_error(write_lime_gauge_field(filename, plaq_m, Nconf, 64), "write_lime_gauge_field", NULL, 12);
} else {
if(g_gauge_file_format == 0) {
// ILDG
sprintf(filename, "%s.%.4d", gaugefilename_prefix, Nconf);
if(g_cart_id==0) fprintf(stdout, "# Reading gauge field from file %s\n", filename);
status = read_lime_gauge_field_doubleprec(filename);
} else if(g_gauge_file_format == 1) {
// NERSC
sprintf(filename, "%s.%.5d", gaugefilename_prefix, Nconf);
if(g_cart_id==0) fprintf(stdout, "# Reading gauge field from file %s\n", filename);
status = read_nersc_gauge_field(g_gauge_field, filename, &plaq_r);
//status = read_nersc_gauge_field_3x3(g_gauge_field, filename, &plaq_r);
}
if(status != 0) {
fprintf(stderr, "[invert_dw_quda] Error, could not read gauge field");
EXIT(12);
}
}
#ifdef MPI
xchange_gauge();
#endif
// measure the plaquette
plaquette(&plaq_m);
if(g_cart_id==0) fprintf(stdout, "# Measured plaquette value: %25.16e\n", plaq_m);
if(g_cart_id==0) fprintf(stdout, "# Read plaquette value : %25.16e\n", plaq_r);
#ifndef HAVE_QUDA
if(N_Jacobi>0) {
#endif
// allocate the smeared / qdp ordered gauge field
alloc_gauge_field(&gauge_field_smeared, VOLUMEPLUSRAND);
for(i=0;i<4;i++) {
gauge_qdp[i] = gauge_field_smeared + i*18*VOLUME;
}
#ifndef HAVE_QUDA
}
#endif
#ifdef HAVE_QUDA
// transcribe the gauge field
omp_set_num_threads(g_num_threads);
#pragma omp parallel for private(ix,iy,mu)
for(ix=0;ix<VOLUME;ix++) {
iy = g_lexic2eot[ix];
for(mu=0;mu<4;mu++) {
_cm_eq_cm(gauge_qdp[mu_trans[mu]]+18*iy, g_gauge_field+_GGI(ix,mu));
}
}
// multiply timeslice T-1 with factor of -1 (antiperiodic boundary condition)
if(g_proc_coords[0]==g_nproc_t-1) {
if(!boundary_condition_factor_set) boundary_condition_factor = -1.;
fprintf(stdout, "# [] process %d multiplies gauge-field timeslice T_global-1 with boundary condition factor %e\n", g_cart_id,
boundary_condition_factor);
omp_set_num_threads(g_num_threads);
#pragma omp parallel for private(ix,iy)
for(ix=0;ix<VOL3;ix++) {
iix = (T-1)*VOL3 + ix;
iy = g_lexic2eot[iix];
_cm_ti_eq_re(gauge_qdp[mu_trans[0]]+18*iy, -1.);
}
}
// QUDA precision parameters
switch(g_cpu_prec) {
case 0: cpu_prec = QUDA_HALF_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] CPU prec = half\n"); break;
case 1: cpu_prec = QUDA_SINGLE_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] CPU prec = single\n"); break;
case 2: cpu_prec = QUDA_DOUBLE_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] CPU prec = double\n"); break;
default: cpu_prec = QUDA_DOUBLE_PRECISION; break;
}
switch(g_gpu_prec) {
case 0: cuda_prec = QUDA_HALF_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] GPU prec = half\n"); break;
case 1: cuda_prec = QUDA_SINGLE_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] GPU prec = single\n"); break;
case 2: cuda_prec = QUDA_DOUBLE_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] GPU prec = double\n"); break;
default: cuda_prec = QUDA_DOUBLE_PRECISION; break;
}
switch(g_gpu_prec_sloppy) {
case 0: cuda_prec_sloppy = QUDA_HALF_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] GPU sloppy prec = half\n"); break;
case 1: cuda_prec_sloppy = QUDA_SINGLE_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] GPU sloppy prec = single\n"); break;
case 2: cuda_prec_sloppy = QUDA_DOUBLE_PRECISION; if(g_cart_id==0) fprintf(stdout, "# [] GPU sloppy prec = double\n"); break;
default: cuda_prec_sloppy = QUDA_SINGLE_PRECISION; break;
}
// QUDA gauge parameters
gauge_param.X[0] = LX;
gauge_param.X[1] = LY;
gauge_param.X[2] = LZ;
gauge_param.X[3] = T;
inv_param.Ls = L5;
gauge_param.anisotropy = 1.0;
gauge_param.type = QUDA_WILSON_LINKS;
gauge_param.gauge_order = QUDA_QDP_GAUGE_ORDER;
gauge_param.t_boundary = QUDA_ANTI_PERIODIC_T;
gauge_param.cpu_prec = cpu_prec;
gauge_param.cuda_prec = cuda_prec;
gauge_param.reconstruct = QUDA_RECONSTRUCT_12;
gauge_param.cuda_prec_sloppy = cuda_prec_sloppy;
gauge_param.reconstruct_sloppy = QUDA_RECONSTRUCT_12;
gauge_param.gauge_fix = QUDA_GAUGE_FIXED_NO;
gauge_param.ga_pad = 0;
inv_param.sp_pad = 0;
inv_param.cl_pad = 0;
// For multi-GPU, ga_pad must be large enough to store a time-slice
#ifdef MULTI_GPU
x_face_size = inv_param.Ls * gauge_param.X[1]*gauge_param.X[2]*gauge_param.X[3]/2;
y_face_size = inv_param.Ls * gauge_param.X[0]*gauge_param.X[2]*gauge_param.X[3]/2;
z_face_size = inv_param.Ls * gauge_param.X[0]*gauge_param.X[1]*gauge_param.X[3]/2;
t_face_size = inv_param.Ls * gauge_param.X[0]*gauge_param.X[1]*gauge_param.X[2]/2;
pad_size = _MAX(x_face_size, y_face_size);
pad_size = _MAX(pad_size, z_face_size);
pad_size = _MAX(pad_size, t_face_size);
gauge_param.ga_pad = pad_size;
if(g_cart_id==0) printf("# [invert_dw_quda] pad_size = %d\n", pad_size);
#endif
// load the gauge field
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] loading gauge field\n");
loadGaugeQuda((void*)gauge_qdp, &gauge_param);
gauge_qdp[0] = NULL;
gauge_qdp[1] = NULL;
gauge_qdp[2] = NULL;
gauge_qdp[3] = NULL;
#endif
/*********************************************
* APE smear the gauge field
*********************************************/
if(N_Jacobi>0) {
memcpy(gauge_field_smeared, g_gauge_field, 72*VOLUMEPLUSRAND*sizeof(double));
fprintf(stdout, "# [invert_dw_quda] APE smearing gauge field with paramters N_APE=%d, alpha_APE=%e\n", N_ape, alpha_ape);
APE_Smearing_Step_threads(gauge_field_smeared, N_ape, alpha_ape);
xchange_gauge_field(gauge_field_smeared);
}
// allocate memory for the spinor fields
#ifdef HAVE_QUDA
no_fields = 3+2;
#else
no_fields = 6+2;
#endif
g_spinor_field = (double**)calloc(no_fields, sizeof(double*));
for(i=0; i<no_fields; i++) alloc_spinor_field(&g_spinor_field[i], VOLUMEPLUSRAND*L5);
smearing_spinor_field[0] = g_spinor_field[no_fields-2];
smearing_spinor_field[1] = g_spinor_field[no_fields-1];
switch(g_source_type) {
case 0:
case 5:
// the source locaton
sl0 = g_source_location / (LX_global*LY_global*LZ);
sl1 = ( g_source_location % (LX_global*LY_global*LZ) ) / ( LY_global*LZ);
sl2 = ( g_source_location % ( LY_global*LZ) ) / ( LZ);
sl3 = g_source_location % LZ;
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] global sl = (%d, %d, %d, %d)\n", sl0, sl1, sl2, sl3);
source_proc_coords[0] = sl0 / T;
source_proc_coords[1] = sl1 / LX;
source_proc_coords[2] = sl2 / LY;
source_proc_coords[3] = sl3 / LZ;
#ifdef MPI
MPI_Cart_rank(g_cart_grid, source_proc_coords, &g_source_proc_id);
#else
g_source_proc_id = 0;
#endif
have_source_flag = g_source_proc_id == g_cart_id;
lsl0 = sl0 % T;
lsl1 = sl1 % LX;
lsl2 = sl2 % LY;
lsl3 = sl3 % LZ;
if(have_source_flag) {
fprintf(stdout, "# [invert_dw_quda] process %d has the source at (%d, %d, %d, %d)\n", g_cart_id, lsl0, lsl1, lsl2, lsl3);
}
break;
case 2:
case 3:
case 4:
// the source timeslice
#ifdef MPI
source_proc_coords[0] = g_source_timeslice / T;
source_proc_coords[1] = 0;
source_proc_coords[2] = 0;
source_proc_coords[3] = 0;
MPI_Cart_rank(g_cart_grid, source_proc_coords, &g_source_proc_id);
have_source_flag = ( g_source_proc_id == g_cart_id );
source_timeslice = have_source_flag ? g_source_timeslice % T : -1;
#else
g_source_proc_id = 0;
have_source_flag = 1;
source_timeslice = g_source_timeslice;
#endif
break;
}
#ifdef HAVE_QUDA
/*************************************************************
* QUDA inverter parameters
*************************************************************/
inv_param.dslash_type = QUDA_DOMAIN_WALL_DSLASH;
if(strcmp(g_inverter_type_name, "cg") == 0) {
inv_param.inv_type = QUDA_CG_INVERTER;
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] using cg inverter\n");
} else if(strcmp(g_inverter_type_name, "bicgstab") == 0) {
inv_param.inv_type = QUDA_BICGSTAB_INVERTER;
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] using bicgstab inverter\n");
#ifdef MULTI_GPU
} else if(strcmp(g_inverter_type_name, "gcr") == 0) {
inv_param.inv_type = QUDA_GCR_INVERTER;
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] using gcr inverter\n");
#endif
} else {
if(g_cart_id==0) fprintf(stderr, "[invert_dw_quda] Error, unrecognized inverter type %s\n", g_inverter_type_name);
EXIT(123);
}
if(inv_param.inv_type == QUDA_CG_INVERTER) {
inv_param.solution_type = QUDA_MAT_SOLUTION;
inv_param.solve_type = QUDA_NORMEQ_PC_SOLVE;
} else if(inv_param.inv_type == QUDA_BICGSTAB_INVERTER) {
inv_param.solution_type = QUDA_MAT_SOLUTION;
inv_param.solve_type = QUDA_DIRECT_PC_SOLVE;
} else {
inv_param.solution_type = QUDA_MATPC_SOLUTION;
inv_param.solve_type = QUDA_DIRECT_PC_SOLVE;
}
inv_param.m5 = g_m5;
inv_param.kappa = 0.5 / (5. + inv_param.m5);
inv_param.mass = g_m0;
inv_param.tol = solver_precision;
inv_param.maxiter = niter_max;
inv_param.reliable_delta = reliable_delta;
#ifdef MPI
// domain decomposition preconditioner parameters
if(inv_param.inv_type == QUDA_GCR_INVERTER) {
if(g_cart_id == 0) printf("# [] settup DD parameters\n");
inv_param.gcrNkrylov = 15;
inv_param.inv_type_precondition = QUDA_MR_INVERTER;
inv_param.tol_precondition = 1e-6;
inv_param.maxiter_precondition = 200;
inv_param.verbosity_precondition = QUDA_VERBOSE;
inv_param.prec_precondition = cuda_prec_sloppy;
inv_param.omega = 0.7;
}
#endif
inv_param.matpc_type = QUDA_MATPC_EVEN_EVEN;
inv_param.dagger = QUDA_DAG_NO;
inv_param.mass_normalization = QUDA_KAPPA_NORMALIZATION; //;QUDA_MASS_NORMALIZATION;
inv_param.cpu_prec = cpu_prec;
inv_param.cuda_prec = cuda_prec;
inv_param.cuda_prec_sloppy = cuda_prec_sloppy;
inv_param.verbosity = QUDA_VERBOSE;
inv_param.preserve_source = QUDA_PRESERVE_SOURCE_NO;
inv_param.dirac_order = QUDA_DIRAC_ORDER;
#ifdef MPI
inv_param.preserve_dirac = QUDA_PRESERVE_DIRAC_YES;
inv_param.prec_precondition = cuda_prec_sloppy;
inv_param.gamma_basis = QUDA_DEGRAND_ROSSI_GAMMA_BASIS;
inv_param.dirac_tune = QUDA_TUNE_NO;
#endif
#endif
/*******************************************
* write initial rng state to file
*******************************************/
if( g_source_type==2 && g_coherent_source==2 ) {
sprintf(rng_file_out, "%s.0", g_rng_filename);
status = init_rng_stat_file (g_seed, rng_file_out);
if( status != 0 ) {
fprintf(stderr, "[invert_dw_quda] Error, could not write rng status\n");
EXIT(210);
}
} else if( (g_source_type==2 /*&& g_coherent_source==1*/) || g_source_type==3 || g_source_type==4) {
if( init_rng_state(g_seed, &g_rng_state) != 0 ) {
fprintf(stderr, "[invert_dw_quda] Error, could initialize rng state\n");
EXIT(211);
}
}
/*******************************************
* prepare locks for openmp
*******************************************/
nthreads = g_num_threads - 1;
lck = (omp_lock_t*)malloc(nthreads * sizeof(omp_lock_t));
if(lck == NULL) {
EXIT_WITH_MSG(97, "[invert_dw_quda] Error, could not allocate lck\n");
}
// init locks
for(i=0;i<nthreads;i++) {
omp_init_lock(lck+i);
}
omp_init_lock(gen_lck);
// check the source momenta
if(g_source_momentum_set) {
source_momentum = (int*)malloc(3*sizeof(int));
if(g_source_momentum[0]<0) g_source_momentum[0] += LX_global;
if(g_source_momentum[1]<0) g_source_momentum[1] += LY_global;
if(g_source_momentum[2]<0) g_source_momentum[2] += LZ_global;
fprintf(stdout, "# [invert_dw_quda] using final source momentum ( %d, %d, %d )\n", g_source_momentum[0], g_source_momentum[1], g_source_momentum[2]);
if(full_orbit) {
status = make_qcont_orbits_3d_parity_avg( &qlatt_id, &qlatt_count, &qlatt_list, &qlatt_nclass, &qlatt_rep, &qlatt_map);
if(status != 0) {
if(g_cart_id==0) fprintf(stderr, "\n[invert_dw_quda] Error while creating O_3-lists\n");
EXIT(4);
}
source_momentum_class = qlatt_id[g_ipt[0][g_source_momentum[0]][g_source_momentum[1]][g_source_momentum[2]]];
source_momentum_no = qlatt_count[source_momentum_class];
source_momentum_runs = source_momentum_class==0 ? 1 : source_momentum_no + 1;
if(g_cart_id==0) fprintf(stdout, "# [] source momentum belongs to class %d with %d members, which means %d runs\n",
source_momentum_class, source_momentum_no, source_momentum_runs);
}
}
if(g_source_type == 5) {
if(g_seq_source_momentum_set) {
if(g_seq_source_momentum[0]<0) g_seq_source_momentum[0] += LX_global;
if(g_seq_source_momentum[1]<0) g_seq_source_momentum[1] += LY_global;
if(g_seq_source_momentum[2]<0) g_seq_source_momentum[2] += LZ_global;
} else if(g_source_momentum_set) {
g_seq_source_momentum[0] = g_source_momentum[0];
g_seq_source_momentum[1] = g_source_momentum[1];
g_seq_source_momentum[2] = g_source_momentum[2];
}
fprintf(stdout, "# [invert_dw_quda] using final sequential source momentum ( %d, %d, %d )\n",
g_seq_source_momentum[0], g_seq_source_momentum[1], g_seq_source_momentum[2]);
}
/***********************************************
* loop on spin-color-index
***********************************************/
for(isc=g_source_index[0]; isc<=g_source_index[1]; isc++)
// for(isc=g_source_index[0]; isc<=g_source_index[0]; isc++)
{
ispin = isc / n_c;
icol = isc % n_c;
for(imom=0; imom<source_momentum_runs; imom++) {
/***********************************************
* set source momentum
***********************************************/
if(g_source_momentum_set) {
if(imom == 0) {
if(full_orbit) {
source_momentum[0] = 0;
source_momentum[1] = 0;
source_momentum[2] = 0;
} else {
source_momentum[0] = g_source_momentum[0];
source_momentum[1] = g_source_momentum[1];
source_momentum[2] = g_source_momentum[2];
}
} else {
source_momentum[0] = qlatt_map[source_momentum_class][imom-1] / (LY_global*LZ_global);
source_momentum[1] = ( qlatt_map[source_momentum_class][imom-1] % (LY_global*LZ_global) ) / LZ_global;
source_momentum[2] = qlatt_map[source_momentum_class][imom-1] % LZ_global;
}
if(g_cart_id==0) fprintf(stdout, "# [] run no. %d, source momentum (%d, %d, %d)\n",
imom, source_momentum[0], source_momentum[1], source_momentum[2]);
}
/***********************************************
* prepare the souce
***********************************************/
if(g_read_source == 0) { // create source
switch(g_source_type) {
case 0:
// point source
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] Creating point source\n");
for(ix=0;ix<L5*VOLUME;ix++) { _fv_eq_zero(g_spinor_field[0]+ix); }
if(have_source_flag) {
if(g_source_momentum_set) {
phase = 2*M_PI*( source_momentum[0]*sl1/(double)LX_global + source_momentum[1]*sl2/(double)LY_global + source_momentum[2]*sl3/(double)LZ_global );
g_spinor_field[0][_GSI(g_ipt[lsl0][lsl1][lsl2][lsl3]) + 2*(n_c*ispin+icol) ] = cos(phase);
g_spinor_field[0][_GSI(g_ipt[lsl0][lsl1][lsl2][lsl3]) + 2*(n_c*ispin+icol)+1] = sin(phase);
} else {
g_spinor_field[0][_GSI(g_ipt[lsl0][lsl1][lsl2][lsl3]) + 2*(n_c*ispin+icol) ] = 1.;
}
}
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d.qx%.2dqy%.2dqz%.2d",
filename_prefix, Nconf, sl0, sl1, sl2, sl3, n_c*ispin+icol, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d", filename_prefix, Nconf, sl0, sl1, sl2, sl3, n_c*ispin+icol);
}
#ifdef HAVE_QUDA
// set matpc_tpye
source_location_5d_iseven = ( (g_iseven[g_ipt[lsl0][lsl1][lsl2][lsl3]] && ispin<n_s/2) || (!g_iseven[g_ipt[lsl0][lsl1][lsl2][lsl3]] && ispin>=n_s/2) ) ? 1 : 0;
if(source_location_5d_iseven) {
inv_param.matpc_type = QUDA_MATPC_EVEN_EVEN;
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] matpc type is MATPC_EVEN_EVEN\n");
} else {
inv_param.matpc_type = QUDA_MATPC_ODD_ODD;
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] matpc type is MATPC_ODD_ODD\n");
}
#endif
break;
case 2:
// timeslice source
if(g_coherent_source==1) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] Creating coherent timeslice source\n");
status = prepare_coherent_timeslice_source(g_spinor_field[0], gauge_field_smeared, g_coherent_source_base, g_coherent_source_delta, VOLUME, g_rng_state, 1);
if(status != 0) {
fprintf(stderr, "[invert_dw_quda] Error from prepare source, status was %d\n", status);
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 123);
MPI_Finalize();
#endif
exit(123);
}
check_error(prepare_coherent_timeslice_source(g_spinor_field[0], gauge_field_smeared, g_coherent_source_base, g_coherent_source_delta, VOLUME, g_rng_state, 1),
"prepare_coherent_timeslice_source", NULL, 123);
timeslice = g_coherent_source_base;
} else {
if(g_coherent_source==2) {
timeslice = (g_coherent_source_base+isc*g_coherent_source_delta)%T_global;
fprintf(stdout, "# [invert_dw_quda] Creating timeslice source\n");
check_error(prepare_timeslice_source(g_spinor_field[0], gauge_field_smeared, timeslice, VOLUME, g_rng_state, 1),
"prepare_timeslice_source", NULL, 123);
} else {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] Creating timeslice source\n");
check_error(prepare_timeslice_source(g_spinor_field[0], gauge_field_smeared, g_source_timeslice, VOLUME, g_rng_state, 1),
"prepare_timeslice_source", NULL, 124);
timeslice = g_source_timeslice;
}
}
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d.qx%.2dqy%.2dqz%.2d", filename_prefix, Nconf,
timeslice, isc, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d", filename_prefix, Nconf, timeslice, isc);
}
break;
case 3:
// timeslice sources for one-end trick (spin dilution)
fprintf(stdout, "# [invert_dw_quda] Creating timeslice source for one-end-trick\n");
check_error( prepare_timeslice_source_one_end(g_spinor_field[0], gauge_field_smeared, source_timeslice, source_momentum, isc%n_s, g_rng_state, \
( isc%n_s==(n_s-1) && imom==source_momentum_runs-1 )), "prepare_timeslice_source_one_end", NULL, 125 );
c = N_Jacobi > 0 ? isc%n_s + n_s : isc%n_s;
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d.qx%.2dqy%.2dqz%.2d", filename_prefix, Nconf,
g_source_timeslice, c, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d", filename_prefix, Nconf, g_source_timeslice, c);
}
break;
case 4:
// timeslice sources for one-end trick (spin and color dilution )
fprintf(stdout, "# [invert_dw_quda] Creating timeslice source for one-end-trick\n");
check_error(prepare_timeslice_source_one_end_color(g_spinor_field[0], gauge_field_smeared, source_timeslice, source_momentum,\
isc%(n_s*n_c), g_rng_state, ( isc%(n_s*n_c)==(n_s*n_c-1) && imom==source_momentum_runs-1 )), "prepare_timeslice_source_one_end_color", NULL, 126);
c = N_Jacobi > 0 ? isc%(n_s*n_c) + (n_s*n_c) : isc%(n_s*n_c);
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d.qx%.2dqy%.2dqz%.2d", filename_prefix, Nconf,
g_source_timeslice, c, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d", filename_prefix, Nconf, g_source_timeslice, c);
}
break;
case 5:
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] preparing sequential point source\n");
check_error( prepare_sequential_point_source (g_spinor_field[0], isc, sl0, g_seq_source_momentum,
smear_source, g_spinor_field[1], gauge_field_smeared), "prepare_sequential_point_source", NULL, 33);
sprintf(source_filename, "%s.%.4d.t%.2dx%.2d.y%.2d.z%.2d.%.2d.qx%.2dqy%.2dqz%.2d", filename_prefix2, Nconf,
sl0, sl1, sl2, sl3, isc, g_source_momentum[0], g_source_momentum[1], g_source_momentum[2]);
break;
default:
fprintf(stderr, "\nError, unrecognized source type\n");
exit(32);
break;
}
} else { // read source
switch(g_source_type) {
case 0: // point source
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d.qx%.2dqy%.2dqz%.2d", \
filename_prefix2, Nconf, sl0, sl1, sl2, sl3, isc, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d", filename_prefix2, Nconf, sl0, sl1, sl2, sl3, isc);
}
fprintf(stdout, "# [invert_dw_quda] reading source from file %s\n", source_filename);
check_error(read_lime_spinor(g_spinor_field[0], source_filename, 0), "read_lime_spinor", NULL, 115);
break;
case 2: // timeslice source
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d.qx%.2dqy%.2dqz%.2d", filename_prefix2, Nconf, g_source_timeslice,
isc, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d", filename_prefix2, Nconf, g_source_timeslice, isc);
}
fprintf(stdout, "# [invert_dw_quda] reading source from file %s\n", source_filename);
check_error(read_lime_spinor(g_spinor_field[0], source_filename, 0), "read_lime_spinor", NULL, 115);
break;
default:
check_error(1, "source type", NULL, 104);
break;
case -1: // timeslice source
sprintf(source_filename, "%s", filename_prefix2);
fprintf(stdout, "# [invert_dw_quda] reading source from file %s\n", source_filename);
check_error(read_lime_spinor(g_spinor_field[0], source_filename, 0), "read_lime_spinor", NULL, 115);
break;
}
} // of if g_read_source
if(g_write_source) {
check_error(write_propagator(g_spinor_field[0], source_filename, 0, g_propagator_precision), "write_propagator", NULL, 27);
}
/***********************************************************************************************
* here threads split:
***********************************************************************************************/
if(dummy_flag==0) strcpy(source_filename_write, source_filename);
memcpy((void*)(smearing_spinor_field[0]), (void*)(g_spinor_field[0]), 24*VOLUME*sizeof(double));
if(dummy_flag>0) {
// copy only if smearing has been done; otherwise do not copy, do not invert
if(g_cart_id==0) fprintf(stdout, "# [] copy smearing field -> g field\n");
memcpy((void*)(g_spinor_field[0]), (void*)(smearing_spinor_field[1]), 24*VOLUME*sizeof(double));
}
omp_set_num_threads(g_num_threads);
#pragma omp parallel private(threadid, _2_kappa, is, ix, iy, iix, ratime, retime) shared(key,g_read_source, smear_source, N_Jacobi, kappa_Jacobi, smearing_spinor_field, g_spinor_field, nthreads, convert_sign, VOLUME, VOL3, T, L5, isc, rotate_gamma_basis, g_cart_id) firstprivate(inv_param, gauge_param, ofs)
{
threadid = omp_get_thread_num();
if(threadid < nthreads) {
fprintf(stdout, "# [] proc%.2d thread%.2d starting source preparation\n", g_cart_id, threadid);
// smearing
if( ( !g_read_source || (g_read_source && smear_source ) ) && N_Jacobi > 0 ) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] smearing source with N_Jacobi=%d, kappa_Jacobi=%e\n", N_Jacobi, kappa_Jacobi);
Jacobi_Smearing_threaded(gauge_field_smeared, smearing_spinor_field[0], smearing_spinor_field[1], kappa_Jacobi, N_Jacobi, threadid, nthreads);
}
/***********************************************
* create the 5-dim. source field
***********************************************/
if(convert_sign == 0) {
spinor_4d_to_5d_threaded(smearing_spinor_field[0], smearing_spinor_field[0], threadid, nthreads);
} else if(convert_sign == 1 || convert_sign == -1) {
spinor_4d_to_5d_sign_threaded(smearing_spinor_field[0], smearing_spinor_field[0], convert_sign, threadid, nthreads);
}
for(is=0; is<L5; is++) {
for(it=threadid; it<T; it+=nthreads) {
memcpy((void*)(g_spinor_field[0]+_GSI(g_ipt_5d[is][it][0][0][0])), (void*)(smearing_spinor_field[0]+_GSI(g_ipt_5d[is][it][0][0][0])), VOL3*24*sizeof(double));
}
}
// reorder, multiply with g2
for(is=0; is<L5; is++) {
for(it=threadid; it<T; it+=nthreads) {
for(i3=0; i3<VOL3; i3++) {
ix = (is*T+it)*VOL3 + i3;
_fv_eq_zero(smearing_spinor_field[1]+_GSI(ix));
}}}
if(rotate_gamma_basis) {
for(it=threadid; it<T; it+=nthreads) {
for(i3=0; i3<VOL3; i3++) {
ix = it * VOL3 + i3;
iy = lexic2eot_5d(0, ix);
_fv_eq_gamma_ti_fv(smearing_spinor_field[1]+_GSI(iy), 2, smearing_spinor_field[0]+_GSI(ix));
}}
for(it=threadid; it<T; it+=nthreads) {
for(i3=0; i3<VOL3; i3++) {
ix = it * VOL3 + i3;
iy = lexic2eot_5d(L5-1, ix);
_fv_eq_gamma_ti_fv(smearing_spinor_field[1]+_GSI(iy), 2, smearing_spinor_field[0]+_GSI(ix+(L5-1)*VOLUME));
}}
} else {
for(it=threadid; it<T; it+=nthreads) {
for(i3=0; i3<VOL3; i3++) {
ix = it * VOL3 + i3;
iy = lexic2eot_5d(0, ix);
_fv_eq_fv(smearing_spinor_field[1]+_GSI(iy), smearing_spinor_field[0]+_GSI(ix));
}}
for(it=threadid; it<T; it+=nthreads) {
for(i3=0; i3<VOL3; i3++) {
ix = it * VOL3 + i3;
iy = lexic2eot_5d(L5-1, ix);
_fv_eq_fv(smearing_spinor_field[1]+_GSI(iy), smearing_spinor_field[0]+_GSI(ix+(L5-1)*VOLUME));
}}
}
fprintf(stdout, "# [] proc%.2d thread%.2d finished source preparation\n", g_cart_id, threadid);
} else if(threadid == g_num_threads-1 && dummy_flag > 0) { // else branch on threadid
fprintf(stdout, "# [] proc%.2d thread%.2d starting inversion for dummy_flag = %d\n", g_cart_id, threadid, dummy_flag);
/***********************************************
* perform the inversion
***********************************************/
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] starting inversion\n");
xchange_field_5d(g_spinor_field[0]);
memset(g_spinor_field[1], 0, (VOLUME+RAND)*L5*24*sizeof(double));
ratime = CLOCK;
#ifdef MPI
if(inv_param.inv_type == QUDA_BICGSTAB_INVERTER || inv_param.inv_type == QUDA_GCR_INVERTER) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] calling invertQuda\n");
invertQuda(g_spinor_field[1], g_spinor_field[0], &inv_param);
} else if(inv_param.inv_type == QUDA_CG_INVERTER) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] calling testCG\n");
testCG(g_spinor_field[1], g_spinor_field[0], &inv_param);
} else {
if(g_cart_id==0) fprintf(stderr, "# [invert_dw_quda] unrecognized inverter\n");
}
#else
invertQuda(g_spinor_field[1], g_spinor_field[0], &inv_param);
#endif
retime = CLOCK;
if(g_cart_id==0) {
fprintf(stdout, "# [invert_dw_quda] QUDA time: %e seconds\n", inv_param.secs);
fprintf(stdout, "# [invert_dw_quda] QUDA Gflops: %e\n", inv_param.gflops/inv_param.secs);
fprintf(stdout, "# [invert_dw_quda] wall time: %e seconds\n", retime-ratime);
fprintf(stdout, "# [invert_dw_quda] Device memory used:\n\tSpinor: %f GiB\n\tGauge: %f GiB\n",
inv_param.spinorGiB, gauge_param.gaugeGiB);
}
} // of if threadid
// wait till all threads are here
#pragma omp barrier
if(inv_param.mass_normalization == QUDA_KAPPA_NORMALIZATION) {
_2_kappa = 2. * g_kappa5d;
for(ix=threadid; ix<VOLUME*L5;ix+=g_num_threads) {
_fv_ti_eq_re(g_spinor_field[1]+_GSI(ix), _2_kappa );
}
}
#pragma omp barrier
// reorder, multiply with g2
for(is=0;is<L5;is++) {
for(ix=threadid; ix<VOLUME; ix+=g_num_threads) {
iy = lexic2eot_5d(is, ix);
iix = is*VOLUME + ix;
_fv_eq_fv(g_spinor_field[0]+_GSI(iix), g_spinor_field[1]+_GSI(iy));
}}
#pragma omp barrier
if(rotate_gamma_basis) {
for(ix=threadid; ix<VOLUME*L5; ix+=g_num_threads) {
_fv_eq_gamma_ti_fv(g_spinor_field[1]+_GSI(ix), 2, g_spinor_field[0]+_GSI(ix));
}
} else {
for(ix=threadid; ix<VOLUME*L5;ix+=g_num_threads) {
_fv_eq_fv(g_spinor_field[1]+_GSI(ix), g_spinor_field[0]+_GSI(ix));
}
}
if(g_cart_id==0 && threadid==g_num_threads-1) fprintf(stdout, "# [invert_dw_quda] inversion done in %e seconds\n", retime-ratime);
#pragma omp single
{
#ifdef MPI
xchange_field_5d(g_spinor_field[1]);
#endif
/***********************************************
* check residuum
***********************************************/
if(check_residuum && dummy_flag>0) {
// apply the Wilson Dirac operator in the gamma-basis defined in cvc_linalg,
// which uses the tmLQCD conventions (same as in contractions)
// without explicit boundary conditions
#ifdef MPI
xchange_field_5d(g_spinor_field[2]);
xchange_field_5d(g_spinor_field[1]);
#endif
memset(g_spinor_field[0], 0, 24*(VOLUME+RAND)*L5*sizeof(double));
//sprintf(filename, "%s.inverted.ascii.%.2d", source_filename, g_cart_id);
//ofs = fopen(filename, "w");
//printf_spinor_field_5d(g_spinor_field[1], ofs);
//fclose(ofs);
Q_DW_Wilson_phi(g_spinor_field[0], g_spinor_field[1]);
for(ix=0;ix<VOLUME*L5;ix++) {
_fv_mi_eq_fv(g_spinor_field[0]+_GSI(ix), g_spinor_field[2]+_GSI(ix));
}
spinor_scalar_product_re(&norm2, g_spinor_field[2], g_spinor_field[2], VOLUME*L5);
spinor_scalar_product_re(&norm, g_spinor_field[0], g_spinor_field[0], VOLUME*L5);
if(g_cart_id==0) fprintf(stdout, "\n# [invert_dw_quda] absolut residuum squared: %e; relative residuum %e\n", norm, sqrt(norm/norm2) );
}
if(dummy_flag>0) {
/***********************************************
* create 4-dim. propagator
***********************************************/
if(convert_sign == 0) {
spinor_5d_to_4d(g_spinor_field[1], g_spinor_field[1]);
} else if(convert_sign == -1 || convert_sign == +1) {
spinor_5d_to_4d_sign(g_spinor_field[1], g_spinor_field[1], convert_sign);
}
/***********************************************
* write the solution
***********************************************/
sprintf(filename, "%s.inverted", source_filename_write);
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] writing propagator to file %s\n", filename);
check_error(write_propagator(g_spinor_field[1], filename, 0, g_propagator_precision), "write_propagator", NULL, 22);
//sprintf(filename, "prop.ascii.4d.%.2d.%.2d.%.2d", isc, g_nproc, g_cart_id);
//ofs = fopen(filename, "w");
//printf_spinor_field(g_spinor_field[1], ofs);
//fclose(ofs);
}
if(check_residuum) memcpy(g_spinor_field[2], smearing_spinor_field[0], 24*VOLUME*L5*sizeof(double));
} // of omp single
} // of omp parallel region
if(dummy_flag > 0) strcpy(source_filename_write, source_filename);
dummy_flag++;
} // of loop on momenta
} // of isc
#if 0
// last inversion
{
memcpy(g_spinor_field[0], smearing_spinor_field[1], 24*VOLUME*L5*sizeof(double));
if(g_cart_id==0) fprintf(stdout, "# [] proc%.2d starting last inversion\n", g_cart_id);
/***********************************************
* perform the inversion
***********************************************/
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] starting inversion\n");
xchange_field_5d(g_spinor_field[0]);
memset(g_spinor_field[1], 0, (VOLUME+RAND)*L5*24*sizeof(double));
ratime = CLOCK;
#ifdef MPI
if(inv_param.inv_type == QUDA_BICGSTAB_INVERTER || inv_param.inv_type == QUDA_GCR_INVERTER) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] calling invertQuda\n");
invertQuda(g_spinor_field[1], g_spinor_field[0], &inv_param);
} else if(inv_param.inv_type == QUDA_CG_INVERTER) {
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] calling testCG\n");
testCG(g_spinor_field[1], g_spinor_field[0], &inv_param);
} else {
if(g_cart_id==0) fprintf(stderr, "# [invert_dw_quda] unrecognized inverter\n");
}
#else
invertQuda(g_spinor_field[1], g_spinor_field[0], &inv_param);
#endif
retime = CLOCK;
if(g_cart_id==0) {
fprintf(stdout, "# [invert_dw_quda] QUDA time: %e seconds\n", inv_param.secs);
fprintf(stdout, "# [invert_dw_quda] QUDA Gflops: %e\n", inv_param.gflops/inv_param.secs);
fprintf(stdout, "# [invert_dw_quda] wall time: %e seconds\n", retime-ratime);
fprintf(stdout, "# [invert_dw_quda] Device memory used:\n\tSpinor: %f GiB\n\tGauge: %f GiB\n",
inv_param.spinorGiB, gauge_param.gaugeGiB);
}
omp_set_num_threads(g_num_threads);
#pragma omp parallel private(threadid,_2_kappa,is,ix,iy,iix) shared(VOLUME,L5,g_kappa,g_spinor_field,g_num_threads)
{
threadid = omp_get_thread_num();
if(inv_param.mass_normalization == QUDA_KAPPA_NORMALIZATION) {
_2_kappa = 2. * g_kappa5d;
for(ix=threadid; ix<VOLUME*L5;ix+=g_num_threads) {
_fv_ti_eq_re(g_spinor_field[1]+_GSI(ix), _2_kappa );
}
}
#pragma omp barrier
// reorder, multiply with g2
for(is=0;is<L5;is++) {
for(ix=threadid; ix<VOLUME; ix+=g_num_threads) {
iy = lexic2eot_5d(is, ix);
iix = is*VOLUME + ix;
_fv_eq_fv(g_spinor_field[0]+_GSI(iix), g_spinor_field[1]+_GSI(iy));
}}
#pragma omp barrier
if(rotate_gamma_basis) {
for(ix=threadid; ix<VOLUME*L5; ix+=g_num_threads) {
_fv_eq_gamma_ti_fv(g_spinor_field[1]+_GSI(ix), 2, g_spinor_field[0]+_GSI(ix));
}
} else {
for(ix=threadid; ix<VOLUME*L5;ix+=g_num_threads) {
_fv_eq_fv(g_spinor_field[1]+_GSI(ix), g_spinor_field[0]+_GSI(ix));
}
}
} // end of parallel region
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] inversion done in %e seconds\n", retime-ratime);
#ifdef MPI
xchange_field_5d(g_spinor_field[1]);
#endif
/***********************************************
* check residuum
***********************************************/
if(check_residuum && dummy_flag>0) {
// apply the Wilson Dirac operator in the gamma-basis defined in cvc_linalg,
// which uses the tmLQCD conventions (same as in contractions)
// without explicit boundary conditions
#ifdef MPI
xchange_field_5d(g_spinor_field[2]);
#endif
memset(g_spinor_field[0], 0, 24*(VOLUME+RAND)*L5*sizeof(double));
//sprintf(filename, "%s.inverted.ascii.%.2d", source_filename, g_cart_id);
//ofs = fopen(filename, "w");
//printf_spinor_field_5d(g_spinor_field[1], ofs);
//fclose(ofs);
Q_DW_Wilson_phi(g_spinor_field[0], g_spinor_field[1]);
for(ix=0;ix<VOLUME*L5;ix++) {
_fv_mi_eq_fv(g_spinor_field[0]+_GSI(ix), g_spinor_field[2]+_GSI(ix));
}
spinor_scalar_product_re(&norm, g_spinor_field[0], g_spinor_field[0], VOLUME*L5);
spinor_scalar_product_re(&norm2, g_spinor_field[2], g_spinor_field[2], VOLUME*L5);
if(g_cart_id==0) fprintf(stdout, "\n# [invert_dw_quda] absolut residuum squared: %e; relative residuum %e\n", norm, sqrt(norm/norm2) );
}
/***********************************************
* create 4-dim. propagator
***********************************************/
if(convert_sign == 0) {
spinor_5d_to_4d(g_spinor_field[1], g_spinor_field[1]);
} else if(convert_sign == -1 || convert_sign == +1) {
spinor_5d_to_4d_sign(g_spinor_field[1], g_spinor_field[1], convert_sign);
}
/***********************************************
* write the solution
***********************************************/
sprintf(filename, "%s.inverted", source_filename_write);
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] writing propagator to file %s\n", filename);
check_error(write_propagator(g_spinor_field[1], filename, 0, g_propagator_precision), "write_propagator", NULL, 22);
//sprintf(filename, "prop.ascii.4d.%.2d.%.2d.%.2d", isc, g_nproc, g_cart_id);
//ofs = fopen(filename, "w");
//printf_spinor_field(g_spinor_field[1], ofs);
//fclose(ofs);
} // of last inversion
#endif // of if 0
/***********************************************
* free the allocated memory, finalize
***********************************************/
#ifdef HAVE_QUDA
// finalize the QUDA library
if(g_cart_id==0) fprintf(stdout, "# [invert_dw_quda] finalizing quda\n");
#ifdef MPI
freeGaugeQuda();
#endif
endQuda();
#endif
if(g_gauge_field != NULL) free(g_gauge_field);
if(gauge_field_smeared != NULL) free(gauge_field_smeared);
if(no_fields>0) {
if(g_spinor_field!=NULL) {
for(i=0; i<no_fields; i++) if(g_spinor_field[i]!=NULL) free(g_spinor_field[i]);
free(g_spinor_field);
}
}
free_geometry();
if(g_source_momentum_set && full_orbit) {
finalize_q_orbits(&qlatt_id, &qlatt_count, &qlatt_list, &qlatt_rep);
if(qlatt_map != NULL) {
free(qlatt_map[0]);
free(qlatt_map);
}
}
if(source_momentum != NULL) free(source_momentum);
if(lck != NULL) free(lck);
#ifdef MPI
#ifdef HAVE_QUDA
endCommsQuda();
#else
MPI_Finalize();
#endif
#endif
if(g_cart_id==0) {
g_the_time = time(NULL);
fprintf(stdout, "\n# [invert_dw_quda] %s# [invert_dw_quda] end of run\n", ctime(&g_the_time));
fprintf(stderr, "\n# [invert_dw_quda] %s# [invert_dw_quda] end of run\n", ctime(&g_the_time));
}
return(0);
}
void source_generation_pion_only(spinor * const P, spinor * const Q,
const int t, const int sample,
const int nstore, const unsigned int _seed) {
int reset = 0, i, x, y, z, is, ic, lt, lx, ly, lz, id=0;
int coords[4], seed, r;
double rnumber, si=0., co=0.;
int rlxd_state[105];
const double sqr2 = 1./sqrt(2.);
_Complex double * p = NULL;
zero_spinor_field(P,VOLUME/2);
zero_spinor_field(Q,VOLUME/2);
/* save the ranlxd_state if neccessary */
if(ranlxd_init == 1) {
rlxd_get(rlxd_state);
reset = 1;
}
/* Compute the seed */
seed =(int) abs(_seed + sample + t*10*97 + nstore*100*53);
rlxd_init(2, seed);
lt = t - g_proc_coords[0]*T;
coords[0] = t / T;
for(x = 0; x < LX*g_nproc_x; x++) {
lx = x - g_proc_coords[1]*LX;
coords[1] = x / LX;
for(y = 0; y < LY*g_nproc_y; y++) {
ly = y - g_proc_coords[2]*LY;
coords[2] = y / LY;
for(z = 0; z < LZ*g_nproc_z; z++) {
lz = z - g_proc_coords[3]*LZ;
coords[3] = z / LZ;
#ifdef TM_USE_MPI
MPI_Cart_rank(g_cart_grid, coords, &id);
#endif
for(is = 0; is < 4; is++) {
for(ic = 0; ic < 3; ic++) {
ranlxd(&rnumber, 1);
if(g_cart_id == id) {
r = (int)floor(4.*rnumber);
if(r == 0)
{
si = sqr2;
co = sqr2;
}
else if(r == 1) {
si = -sqr2;
co = sqr2;
}
else if(r==2) {
si = sqr2;
co = -sqr2;
}
else {
si = -sqr2;
co = -sqr2;
}
i = g_lexic2eosub[ g_ipt[lt][lx][ly][lz] ];
if((lt+lx+ly+lz+g_proc_coords[3]*LZ+g_proc_coords[2]*LY
+ g_proc_coords[0]*T+g_proc_coords[1]*LX)%2 == 0) {
p = (_Complex double*)(P + i);
}
else {
p = (_Complex double*)(Q + i);
}
(*(p+3*is+ic)) = co + si * I;
}
}
}
}
}
}
/* reset the ranlxd if neccessary */
if(reset) {
rlxd_reset(rlxd_state);
}
return;
}
void mpi_manager_2D::determin_OtherRanks() {
// Find neighbouring ranks:
MPI_Cart_shift(comm2d, 0, 1, &left , &right);
MPI_Cart_shift(comm2d, 1, 1, &front, &back);
// Determine ranks of neighbour processes:
int shiftcoord[DIM];
int lbound[DIM],ubound[DIM];
for(int dim=0;dim<DIM;dim++){
lbound[dim]=-nproc[dim];
ubound[dim]= nproc[dim];
}
Neighbours.resize(lbound,ubound);
Neighbours.clear();
for(int dim0=-nproc[0]; dim0<=nproc[0]; dim0++){
shiftcoord[0] = (coords[0]+dim0);
if(shiftcoord[0] < 0) shiftcoord[0]+=nproc[0];
for(int dim1=-nproc[1]; dim1<=nproc[1]; dim1++){
shiftcoord[1] = (coords[1]+dim1);
if(shiftcoord[1] < 0) shiftcoord[1]+=nproc[1];
if(shiftcoord[0]>=0 && shiftcoord[0]<nproc[0] &&
shiftcoord[1]>=0 && shiftcoord[1]<nproc[1]) {
// Now determine rank at relative shifted position
// std::cout << " Cart ";
// std::cout << shiftcoord[0] << " ";
// std::cout << shiftcoord[1] << " ";
// std::cout << rank << " ";
// std::cout << nproc[0] << " ";
// std::cout << nproc[1] << " ";
// std::cout << std::endl;
MPI_Cart_rank(comm2d, shiftcoord, &Neighbours(dim0,dim1));
} else {
// If outside domain set to error value
Neighbours(dim0, dim1) = MPI_PROC_NULL;
}
}
}
NeighboursCyclic.resize(lbound,ubound);
NeighboursCyclic.clear();
for(int dim0=-nproc[0]; dim0<=nproc[0]; dim0++){
shiftcoord[0] = (coords[0]+dim0)%nproc[0];
if(shiftcoord[0] < 0) shiftcoord[0]+=nproc[0];
for(int dim1=-nproc[1]; dim1<=nproc[1]; dim1++){
shiftcoord[1] = (coords[1]+dim1)%nproc[1];
if(shiftcoord[1] < 0) shiftcoord[1]+=nproc[1];
// Now determine rank at relative shifted position
MPI_Cart_rank(comm2d, shiftcoord, &NeighboursCyclic(dim0,dim1));
}
}
// Now determine absolute position of ranks
AllRanks.resize(Index::set(0,0),
Index::set(nproc[0]-1,nproc[1]-1));
for(int dim1=0; dim1<nproc[1]; ++dim1) {
for(int dim0=0; dim0<nproc[0]; ++dim0) {
int coord[2] = {dim0, dim1};
MPI_Cart_rank(comm2d, coord, &AllRanks(dim0, dim1));
}
}
}
int main(int argc, char **argv) {
int c, i, mu, status;
int ispin, icol, isc;
int n_c = 3;
int n_s = 4;
int count = 0;
int filename_set = 0;
int dims[4] = {0,0,0,0};
int l_LX_at, l_LXstart_at;
int x0, x1, x2, x3, ix, iix, iy;
int sl0, sl1, sl2, sl3, have_source_flag=0;
int source_proc_coords[4], lsl0, lsl1, lsl2, lsl3, source_proc_id;
int check_residuum = 0;
unsigned int VOL3;
int do_gt = 0;
int full_orbit = 0;
char filename[200], source_filename[200];
double ratime, retime;
double plaq_r=0., plaq_m=0., norm, norm2;
// double spinor1[24], spinor2[24];
double *gauge_qdp[4], *gauge_field_timeslice=NULL, *gauge_field_smeared=NULL;
double _1_2_kappa, _2_kappa, phase;
FILE *ofs;
int mu_trans[4] = {3, 0, 1, 2};
int threadid, nthreads;
int timeslice;
char rng_file_in[100], rng_file_out[100];
int *source_momentum=NULL;
int source_momentum_class = -1;
int source_momentum_no = 0;
int source_momentum_runs = 1;
int imom;
/************************************************/
int qlatt_nclass;
int *qlatt_id=NULL, *qlatt_count=NULL, **qlatt_rep=NULL, **qlatt_map=NULL;
double **qlatt_list=NULL;
/************************************************/
/***********************************************
* QUDA parameters
***********************************************/
QudaPrecision cpu_prec = QUDA_DOUBLE_PRECISION;
QudaPrecision cuda_prec = QUDA_DOUBLE_PRECISION;
QudaPrecision cuda_prec_sloppy = QUDA_DOUBLE_PRECISION;
QudaGaugeParam gauge_param = newQudaGaugeParam();
QudaInvertParam inv_param = newQudaInvertParam();
#ifdef MPI
MPI_Init(&argc, &argv);
#endif
while ((c = getopt(argc, argv, "och?vgf:p:")) != -1) {
switch (c) {
case 'v':
g_verbose = 1;
break;
case 'g':
do_gt = 1;
break;
case 'f':
strcpy(filename, optarg);
filename_set=1;
break;
case 'c':
check_residuum = 1;
fprintf(stdout, "# [invert_quda] will check residuum again\n");
break;
case 'p':
n_c = atoi(optarg);
fprintf(stdout, "# [invert_quda] will use number of colors = %d\n", n_c);
break;
case 'o':
full_orbit = 1;
fprintf(stdout, "# [invert_quda] will invert for full orbit, if source momentum set\n");
break;
case 'h':
case '?':
default:
usage();
break;
}
}
// get the time stamp
g_the_time = time(NULL);
/**************************************
* set the default values, read input
**************************************/
if(filename_set==0) strcpy(filename, "cvc.input");
if(g_proc_id==0) fprintf(stdout, "# Reading input from file %s\n", filename);
read_input_parser(filename);
/* some checks on the input data */
if((T_global == 0) || (LX==0) || (LY==0) || (LZ==0)) {
if(g_proc_id==0) fprintf(stderr, "[invert_quda] Error, T and L's must be set\n");
usage();
}
if(g_kappa == 0.) {
if(g_proc_id==0) fprintf(stderr, "[invert_quda] Error, kappa should be > 0.n");
usage();
}
// set number of openmp threads
#ifdef OPENMP
omp_set_num_threads(g_num_threads);
#else
fprintf(stdout, "[invert_quda_cg] Warning, resetting global number of threads to 1\n");
g_num_threads = 1;
#endif
/* initialize MPI parameters */
mpi_init(argc, argv);
// the volume of a timeslice
VOL3 = LX*LY*LZ;
fprintf(stdout, "# [%2d] parameters:\n"\
"# [%2d] T = %3d\n"\
"# [%2d] Tstart = %3d\n",\
g_cart_id, g_cart_id, T, g_cart_id, Tstart);
#ifdef MPI
if(T==0) {
fprintf(stderr, "[%2d] local T is zero; exit\n", g_cart_id);
MPI_Abort(MPI_COMM_WORLD, 1);
MPI_Finalize();
exit(2);
}
#endif
if(init_geometry() != 0) {
fprintf(stderr, "ERROR from init_geometry\n");
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 1);
MPI_Finalize();
#endif
exit(1);
}
geometry();
/**************************************
* initialize the QUDA library
**************************************/
fprintf(stdout, "# [invert_quda] initializing quda\n");
initQuda(g_gpu_device_number);
/**************************************
* prepare the gauge field
**************************************/
// read the gauge field from file
alloc_gauge_field(&g_gauge_field, VOLUMEPLUSRAND);
if(strcmp( gaugefilename_prefix, "identity")==0 ) {
if(g_cart_id==0) fprintf(stdout, "# [invert_quda] Setting up unit gauge field\n");
for(ix=0;ix<VOLUME; ix++) {
for(mu=0;mu<4;mu++) {
_cm_eq_id(g_gauge_field+_GGI(ix,mu));
}
}
} else {
if(g_gauge_file_format == 0) {
// ILDG
sprintf(filename, "%s.%.4d", gaugefilename_prefix, Nconf);
if(g_cart_id==0) fprintf(stdout, "# Reading gauge field from file %s\n", filename);
status = read_lime_gauge_field_doubleprec(filename);
} else if(g_gauge_file_format == 1) {
// NERSC
sprintf(filename, "%s.%.5d", gaugefilename_prefix, Nconf);
if(g_cart_id==0) fprintf(stdout, "# Reading gauge field from file %s\n", filename);
status = read_nersc_gauge_field(g_gauge_field, filename, &plaq_r);
}
if(status != 0) {
fprintf(stderr, "[invert_quda] Error, could not read gauge field");
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 12);
MPI_Finalize();
#endif
exit(12);
}
}
#ifdef MPI
xchange_gauge();
#endif
// measure the plaquette
plaquette(&plaq_m);
if(g_cart_id==0) fprintf(stdout, "# Measured plaquette value: %25.16e\n", plaq_m);
if(g_cart_id==0) fprintf(stdout, "# Read plaquette value : %25.16e\n", plaq_r);
// allocate the smeared / qdp ordered gauge field
alloc_gauge_field(&gauge_field_smeared, VOLUME);
for(i=0;i<4;i++) {
gauge_qdp[i] = gauge_field_smeared + i*18*VOLUME;
}
// transcribe the gauge field
#ifdef OPENMP
omp_set_num_threads(g_num_threads);
#pragma omp parallel for private(ix,iy,mu)
#endif
for(ix=0;ix<VOLUME;ix++) {
iy = g_lexic2eot[ix];
for(mu=0;mu<4;mu++) {
_cm_eq_cm(gauge_qdp[mu_trans[mu]]+18*iy, g_gauge_field+_GGI(ix,mu));
}
}
// multiply timeslice T-1 with factor of -1 (antiperiodic boundary condition)
#ifdef OPENMP
omp_set_num_threads(g_num_threads);
#pragma omp parallel for private(ix,iy)
#endif
for(ix=0;ix<VOL3;ix++) {
iix = (T-1)*VOL3 + ix;
iy = g_lexic2eot[iix];
_cm_ti_eq_re(gauge_qdp[mu_trans[0]]+18*iy, -1.);
}
// QUDA gauge parameters
gauge_param.X[0] = LX_global;
gauge_param.X[1] = LY_global;
gauge_param.X[2] = LZ_global;
gauge_param.X[3] = T_global;
gauge_param.anisotropy = 1.0;
gauge_param.type = QUDA_WILSON_LINKS;
gauge_param.gauge_order = QUDA_QDP_GAUGE_ORDER;
gauge_param.t_boundary = QUDA_ANTI_PERIODIC_T;
gauge_param.cpu_prec = cpu_prec;
gauge_param.cuda_prec = cuda_prec;
gauge_param.reconstruct = QUDA_RECONSTRUCT_12;
gauge_param.cuda_prec_sloppy = cuda_prec_sloppy;
gauge_param.reconstruct_sloppy = QUDA_RECONSTRUCT_12;
gauge_param.gauge_fix = QUDA_GAUGE_FIXED_NO;
gauge_param.ga_pad = 0;
// load the gauge field
fprintf(stdout, "# [invert_quda] loading gauge field\n");
loadGaugeQuda((void*)gauge_qdp, &gauge_param);
gauge_qdp[0] = NULL;
gauge_qdp[1] = NULL;
gauge_qdp[2] = NULL;
gauge_qdp[3] = NULL;
/*********************************************
* APE smear the gauge field
*********************************************/
memcpy(gauge_field_smeared, g_gauge_field, 72*VOLUME*sizeof(double));
if(N_ape>0) {
fprintf(stdout, "# [invert_quda] APE smearing gauge field with paramters N_APE=%d, alpha_APE=%e\n", N_ape, alpha_ape);
#ifdef OPENMP
APE_Smearing_Step_threads(gauge_field_smeared, N_ape, alpha_ape);
#else
for(i=0; i<N_ape; i++) {
APE_Smearing_Step(gauge_field_smeared, alpha_ape);
}
#endif
}
/* allocate memory for the spinor fields */
no_fields = 3;
g_spinor_field = (double**)calloc(no_fields, sizeof(double*));
for(i=0; i<no_fields; i++) alloc_spinor_field(&g_spinor_field[i], VOLUMEPLUSRAND);
/* the source locaton */
sl0 = g_source_location / (LX_global*LY_global*LZ);
sl1 = ( g_source_location % (LX_global*LY_global*LZ) ) / ( LY_global*LZ);
sl2 = ( g_source_location % ( LY_global*LZ) ) / ( LZ);
sl3 = g_source_location % LZ;
if(g_cart_id==0) fprintf(stdout, "# [invert_quda] global sl = (%d, %d, %d, %d)\n", sl0, sl1, sl2, sl3);
source_proc_coords[0] = sl0 / T;
source_proc_coords[1] = sl1 / LX;
source_proc_coords[2] = sl2 / LY;
source_proc_coords[3] = sl3 / LZ;
#ifdef MPI
MPI_Cart_rank(g_cart_grid, source_proc_coords, &source_proc_id);
#else
source_proc_id = 0;
#endif
have_source_flag = source_proc_id == g_cart_id;
lsl0 = sl0 % T;
lsl1 = sl1 % LX;
lsl2 = sl2 % LY;
lsl3 = sl3 % LZ;
if(have_source_flag) {
fprintf(stdout, "# [invert_quda] process %d has the source at (%d, %d, %d, %d)\n", g_cart_id, lsl0, lsl1, lsl2, lsl3);
}
// QUDA inverter parameters
inv_param.dslash_type = QUDA_WILSON_DSLASH;
// inv_param.inv_type = QUDA_BICGSTAB_INVERTER;
inv_param.inv_type = QUDA_CG_INVERTER;
inv_param.kappa = g_kappa;
inv_param.tol = solver_precision;
inv_param.maxiter = niter_max;
inv_param.reliable_delta = reliable_delta;
inv_param.solution_type = QUDA_MAT_SOLUTION;
// inv_param.solve_type = QUDA_DIRECT_PC_SOLVE;
inv_param.solve_type = QUDA_NORMEQ_PC_SOLVE;
inv_param.matpc_type = QUDA_MATPC_EVEN_EVEN; // QUDA_MATPC_EVEN_EVEN;
inv_param.dagger = QUDA_DAG_NO;
inv_param.mass_normalization = QUDA_KAPPA_NORMALIZATION; //;QUDA_MASS_NORMALIZATION;
inv_param.cpu_prec = cpu_prec;
inv_param.cuda_prec = cuda_prec;
inv_param.cuda_prec_sloppy = cuda_prec_sloppy;
inv_param.preserve_source = QUDA_PRESERVE_SOURCE_NO;
inv_param.dirac_order = QUDA_DIRAC_ORDER;
inv_param.sp_pad = 0;
inv_param.cl_pad = 0;
inv_param.verbosity = QUDA_VERBOSE;
// write initial rng state to file
if(g_source_type==2 && g_coherent_source==2) {
sprintf(rng_file_out, "%s.0", g_rng_filename);
if( init_rng_stat_file (g_seed, rng_file_out) != 0 ) {
fprintf(stderr, "[invert_quda] Error, could not write rng status\n");
exit(210);
}
} else if(g_source_type==3 || g_source_type==4) {
if( init_rng_state(g_seed, &g_rng_state) != 0 ) {
fprintf(stderr, "[invert_quda] Error, could initialize rng state\n");
exit(211);
}
}
// check the source momenta
if(g_source_momentum_set) {
source_momentum = (int*)malloc(3*sizeof(int));
if(g_source_momentum[0]<0) g_source_momentum[0] += LX;
if(g_source_momentum[1]<0) g_source_momentum[1] += LY;
if(g_source_momentum[2]<0) g_source_momentum[2] += LZ;
fprintf(stdout, "# [invert_quda] using final source momentum ( %d, %d, %d )\n", g_source_momentum[0], g_source_momentum[1], g_source_momentum[2]);
if(full_orbit) {
status = make_qcont_orbits_3d_parity_avg( &qlatt_id, &qlatt_count, &qlatt_list, &qlatt_nclass, &qlatt_rep, &qlatt_map);
if(status != 0) {
fprintf(stderr, "\n[invert_quda] Error while creating O_3-lists\n");
exit(4);
}
source_momentum_class = qlatt_id[g_ipt[0][g_source_momentum[0]][g_source_momentum[1]][g_source_momentum[2]]];
source_momentum_no = qlatt_count[source_momentum_class];
source_momentum_runs = source_momentum_class==0 ? 1 : source_momentum_no + 1;
fprintf(stdout, "# [] source momentum belongs to class %d with %d members, which means %d runs\n",
source_momentum_class, source_momentum_no, source_momentum_runs);
}
}
/***********************************************
* loop on spin-color-index
***********************************************/
for(isc=g_source_index[0]; isc<=g_source_index[1]; isc++) {
ispin = isc / n_c;
icol = isc % n_c;
for(imom=0; imom<source_momentum_runs; imom++) {
/***********************************************
* set source momentum
***********************************************/
if(g_source_momentum_set) {
if(imom == 0) {
if(full_orbit) {
source_momentum[0] = 0;
source_momentum[1] = 0;
source_momentum[2] = 0;
} else {
source_momentum[0] = g_source_momentum[0];
source_momentum[1] = g_source_momentum[1];
source_momentum[2] = g_source_momentum[2];
}
} else {
source_momentum[0] = qlatt_map[source_momentum_class][imom-1] / (LY*LZ);
source_momentum[1] = ( qlatt_map[source_momentum_class][imom-1] % (LY*LZ) ) / LZ;
source_momentum[2] = qlatt_map[source_momentum_class][imom-1] % LZ;
}
fprintf(stdout, "# [] run no. %d, source momentum (%d, %d, %d)\n", imom, source_momentum[0], source_momentum[1], source_momentum[2]);
}
/***********************************************
* prepare the souce
***********************************************/
if(g_read_source == 0) { // create source
switch(g_source_type) {
case 0:
// point source
fprintf(stdout, "# [invert_quda] Creating point source\n");
for(ix=0;ix<24*VOLUME;ix++) g_spinor_field[0][ix] = 0.;
if(have_source_flag) {
if(g_source_momentum_set) {
phase = 2*M_PI*( source_momentum[0]*lsl1/(double)LX + source_momentum[1]*lsl2/(double)LY + source_momentum[2]*lsl3/(double)LZ );
g_spinor_field[0][_GSI(g_source_location) + 2*(n_c*ispin+icol) ] = cos(phase);
g_spinor_field[0][_GSI(g_source_location) + 2*(n_c*ispin+icol)+1] = sin(phase);
} else {
g_spinor_field[0][_GSI(g_source_location) + 2*(n_c*ispin+icol) ] = 1.;
}
}
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d.qx%.2dqy%.2dqz%.2d",
filename_prefix, Nconf, sl0, sl1, sl2, sl3, n_c*ispin+icol, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d", filename_prefix, Nconf, sl0, sl1, sl2, sl3, n_c*ispin+icol);
}
break;
case 2:
// timeslice source
if(g_coherent_source==1) {
fprintf(stdout, "# [invert_quda] Creating coherent timeslice source\n");
status = prepare_coherent_timeslice_source(g_spinor_field[0], gauge_field_smeared, g_coherent_source_base, g_coherent_source_delta, VOLUME, g_rng_filename, NULL);
if(status != 0) {
fprintf(stderr, "[invert_quda] Error from prepare source, status was %d\n", status);
exit(123);
}
timeslice = g_coherent_source_base;
} else {
if(g_coherent_source==2) {
strcpy(rng_file_in, rng_file_out);
if(isc == g_source_index[1]) { strcpy(rng_file_out, g_rng_filename); }
else { sprintf(rng_file_out, "%s.%d", g_rng_filename, isc+1); }
timeslice = (g_coherent_source_base+isc*g_coherent_source_delta)%T_global;
fprintf(stdout, "# [invert_quda] Creating timeslice source\n");
status = prepare_timeslice_source(g_spinor_field[0], gauge_field_smeared, timeslice, VOLUME, rng_file_in, rng_file_out);
if(status != 0) {
fprintf(stderr, "[invert_quda] Error from prepare source, status was %d\n", status);
exit(123);
}
} else {
fprintf(stdout, "# [invert_quda] Creating timeslice source\n");
status = prepare_timeslice_source(g_spinor_field[0], gauge_field_smeared, g_source_timeslice, VOLUME, g_rng_filename, g_rng_filename);
if(status != 0) {
fprintf(stderr, "[invert_quda] Error from prepare source, status was %d\n", status);
exit(124);
}
timeslice = g_source_timeslice;
}
}
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d.qx%.2dqy%.2dqz%.2d", filename_prefix, Nconf,
timeslice, isc, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d", filename_prefix, Nconf, timeslice, isc);
}
break;
case 3:
// timeslice sources for one-end trick (spin dilution)
fprintf(stdout, "# [invert_quda] Creating timeslice source for one-end-trick\n");
status = prepare_timeslice_source_one_end(g_spinor_field[0], gauge_field_smeared, g_source_timeslice, source_momentum, isc%n_s, g_rng_state, \
( isc%n_s==(n_s-1) && imom==source_momentum_runs-1 ) );
if(status != 0) {
fprintf(stderr, "[invert_quda] Error from prepare source, status was %d\n", status);
exit(125);
}
c = N_Jacobi > 0 ? isc%n_s + n_s : isc%n_s;
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d.qx%.2dqy%.2dqz%.2d", filename_prefix, Nconf,
g_source_timeslice, c, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d", filename_prefix, Nconf, g_source_timeslice, c);
}
break;
case 4:
// timeslice sources for one-end trick (spin and color dilution )
fprintf(stdout, "# [invert_quda] Creating timeslice source for one-end-trick\n");
status = prepare_timeslice_source_one_end_color(g_spinor_field[0], gauge_field_smeared, g_source_timeslice, source_momentum,\
isc%(n_s*n_c), g_rng_state, ( isc%(n_s*n_c)==(n_s*n_c-1) && imom==source_momentum_runs-1 ) );
if(status != 0) {
fprintf(stderr, "[invert_quda] Error from prepare source, status was %d\n", status);
exit(126);
}
c = N_Jacobi > 0 ? isc%(n_s*n_c) + (n_s*n_c) : isc%(n_s*n_c);
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d.qx%.2dqy%.2dqz%.2d", filename_prefix, Nconf,
g_source_timeslice, c, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.2d", filename_prefix, Nconf, g_source_timeslice, c);
}
break;
default:
fprintf(stderr, "\nError, unrecognized source type\n");
exit(32);
break;
}
} else { // read source
switch(g_source_type) {
case 0: // point source
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d.qx%.2dqy%.2dqz%.2d", \
filename_prefix2, Nconf, sl0, sl1, sl2, sl3, isc, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.t%.2dx%.2dy%.2dz%.2d.%.2d", filename_prefix2, Nconf, sl0, sl1, sl2, sl3, isc);
}
fprintf(stdout, "# [invert_quda] reading source from file %s\n", source_filename);
status = read_lime_spinor(g_spinor_field[0], source_filename, 0);
if(status != 0) {
fprintf(stderr, "# [invert_quda] Errro, could not read source from file %s\n", source_filename);
exit(115);
}
break;
case 2: // timeslice source
if(g_source_momentum_set) {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d.qx%.2dqy%.2dqz%.2d", filename_prefix2, Nconf, g_source_timeslice,
isc, source_momentum[0], source_momentum[1], source_momentum[2]);
} else {
sprintf(source_filename, "%s.%.4d.%.2d.%.5d", filename_prefix2, Nconf, g_source_timeslice, isc);
}
fprintf(stdout, "# [invert_quda] reading source from file %s\n", source_filename);
status = read_lime_spinor(g_spinor_field[0], source_filename, 0);
if(status != 0) {
fprintf(stderr, "# [invert_quda] Errro, could not read source from file %s\n", source_filename);
exit(115);
}
break;
default:
fprintf(stderr, "[] Error, unrecognized source type for reading\n");
exit(104);
break;
}
} // of if g_read_source
//sprintf(filename, "%s.ascii", source_filename);
//ofs = fopen(filename, "w");
//printf_spinor_field(g_spinor_field[0], ofs);
//fclose(ofs);
if(g_write_source) {
status = write_propagator(g_spinor_field[0], source_filename, 0, g_propagator_precision);
if(status != 0) {
fprintf(stderr, "Error from write_propagator, status was %d\n", status);
exit(27);
}
}
// smearing
if(N_Jacobi > 0) {
#ifdef OPENMP
Jacobi_Smearing_Step_one_threads(gauge_field_smeared, g_spinor_field[0], g_spinor_field[1], N_Jacobi, kappa_Jacobi);
#else
for(c=0; c<N_Jacobi; c++) {
Jacobi_Smearing_Step_one(gauge_field_smeared, g_spinor_field[0], g_spinor_field[1], kappa_Jacobi);
}
#endif
}
// multiply with g2
for(ix=0;ix<VOLUME;ix++) {
_fv_eq_gamma_ti_fv(g_spinor_field[1]+_GSI(ix), 2, g_spinor_field[0]+_GSI(ix));
}
// transcribe the spinor field to even-odd ordering with coordinates (x,y,z,t)
for(ix=0;ix<VOLUME;ix++) {
iy = g_lexic2eot[ix];
_fv_eq_fv(g_spinor_field[2]+_GSI(iy), g_spinor_field[1]+_GSI(ix));
}
/***********************************************
* perform the inversion
***********************************************/
fprintf(stdout, "# [invert_quda] starting inversion\n");
ratime = (double)clock() / CLOCKS_PER_SEC;
for(ix=0;ix<VOLUME;ix++) {
_fv_eq_zero(g_spinor_field[1]+_GSI(ix) );
}
invertQuda(g_spinor_field[1], g_spinor_field[2], &inv_param);
retime = (double)clock() / CLOCKS_PER_SEC;
fprintf(stdout, "# [invert_quda] inversion done in %e seconds\n", retime-ratime);
fprintf(stdout, "# [invert_quda] Device memory used:\n\tSpinor: %f GiB\n\tGauge: %f GiB\n",
inv_param.spinorGiB, gauge_param.gaugeGiB);
if(inv_param.mass_normalization == QUDA_KAPPA_NORMALIZATION) {
_2_kappa = 2. * g_kappa;
for(ix=0;ix<VOLUME;ix++) {
_fv_ti_eq_re(g_spinor_field[1]+_GSI(ix), _2_kappa );
}
}
// transcribe the spinor field to lexicographical order with (t,x,y,z)
for(ix=0;ix<VOLUME;ix++) {
iy = g_lexic2eot[ix];
_fv_eq_fv(g_spinor_field[2]+_GSI(ix), g_spinor_field[1]+_GSI(iy));
}
// multiply with g2
for(ix=0;ix<VOLUME;ix++) {
_fv_eq_gamma_ti_fv(g_spinor_field[1]+_GSI(ix), 2, g_spinor_field[2]+_GSI(ix));
}
/***********************************************
* check residuum
***********************************************/
if(check_residuum) {
// apply the Wilson Dirac operator in the gamma-basis defined in cvc_linalg,
// which uses the tmLQCD conventions (same as in contractions)
// without explicit boundary conditions
Q_Wilson_phi(g_spinor_field[2], g_spinor_field[1]);
for(ix=0;ix<VOLUME;ix++) {
_fv_mi_eq_fv(g_spinor_field[2]+_GSI(ix), g_spinor_field[0]+_GSI(ix));
}
spinor_scalar_product_re(&norm, g_spinor_field[2], g_spinor_field[2], VOLUME);
spinor_scalar_product_re(&norm2, g_spinor_field[0], g_spinor_field[0], VOLUME);
fprintf(stdout, "\n# [invert_quda] absolut residuum squared: %e; relative residuum %e\n", norm, sqrt(norm/norm2) );
}
/***********************************************
* write the solution
***********************************************/
sprintf(filename, "%s.inverted", source_filename);
fprintf(stdout, "# [invert_quda] writing propagator to file %s\n", filename);
status = write_propagator(g_spinor_field[1], filename, 0, g_propagator_precision);
if(status != 0) {
fprintf(stderr, "Error from write_propagator, status was %d\n", status);
exit(22);
}
} // of loop on momenta
} // of isc
/***********************************************
* free the allocated memory, finalize
***********************************************/
// finalize the QUDA library
fprintf(stdout, "# [invert_quda] finalizing quda\n");
endQuda();
free(g_gauge_field);
free(gauge_field_smeared);
for(i=0; i<no_fields; i++) free(g_spinor_field[i]);
free(g_spinor_field);
free_geometry();
if(g_source_momentum_set && full_orbit) {
finalize_q_orbits(&qlatt_id, &qlatt_count, &qlatt_list, &qlatt_rep);
if(qlatt_map != NULL) {
free(qlatt_map[0]);
free(qlatt_map);
}
}
if(source_momentum != NULL) free(source_momentum);
#ifdef MPI
MPI_Finalize();
#endif
if(g_cart_id==0) {
g_the_time = time(NULL);
fprintf(stdout, "\n# [invert_quda] %s# [invert_quda] end of run\n", ctime(&g_the_time));
fprintf(stderr, "\n# [invert_quda] %s# [invert_quda] end of run\n", ctime(&g_the_time));
}
return(0);
}
void mpi_manager_3D::setup(NumArray<int> &nproc, NumArray<int> &mx) {
// Save number of processors in each dimension
for(int dir=0; dir<DIM; ++dir) {
this->nproc[dir] = nproc[dir];
}
// Determine the rank of the current task
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
// Get number of ranks from MPI
int ntasks;
MPI_Comm_size(MPI_COMM_WORLD, &ntasks);
this->ntasks = ntasks;
// Set the distribution of processes:
if(ntasks != nproc[0]*nproc[1]*nproc[2]){
std::cerr << " Wrong number of processes " << std::endl;
std::cout << ntasks << " " << nproc[0]*nproc[1]*nproc[2] << std::endl;
Finalise();
}
if(rank==0) {
std::cout << " Number of tasks: " << ntasks << std::endl;
}
// Check if grid can be subdevided as desired
for(int dir = 0; dir < DIM; ++dir) {
if(mx[dir] < nproc[dir] && nproc[dir] > 1) {
if(rank == 0) {
std::cerr << " Wrong grid topology for dimension ";
std::cerr << dir << std::endl;
std::cerr << " mx[" << dir << "]:" << mx[dir] << std::endl;
std::cerr << " nproc[" << dir << "]:" << nproc[dir] << std::endl;
}
Finalise();
}
}
// Check if grid is a power of 2:
double eps = 1.e-12;
for(int dir = 0; dir < DIM; ++dir) {
double exponent = log(mx[dir])/log(2.);
int i_exponent = static_cast<int>(exponent+eps);
if(exponent - i_exponent > 2.*eps) {
if(rank == 0) {
std::cerr << " Error: grid must be of the form mx = 2^n ";
std::cerr << std::endl;
std::cerr << " Exiting " << std::endl;
}
Finalise();
}
}
// Grid is not periodic
int periods[3] = {false, false, false};
int reorder = false;
// If all is okay: Create new communicator "comm3d"
MPI_Cart_create(MPI_COMM_WORLD, DIM, nproc, periods, reorder, &comm3d);
// Retrieve the cartesian topology
if (rank == 0) {
int TopoType;
std::cout << " Cart topology: ";
MPI_Topo_test(comm3d, &TopoType);
switch (TopoType) {
case MPI_UNDEFINED :
std::cout << " MPI_UNDEFINED " << std::endl;
break;
case MPI_GRAPH :
std::cout << "MPI_GRAPH" << std::endl;
break;
case MPI_CART :
std::cout << "MPI_CART" << std::endl;
break;
}
}
// Determine rank again for cartesian communicator -> overwrite rank
MPI_Comm_rank(comm3d, &rank);
// std::cout << " my rank: " << rank << std::endl;
// Translate rank to coordinates
MPI_Cart_coords(comm3d, rank, DIM, coords);
// // Backwards translation
// int TranslateRank;
// MPI_Cart_rank(comm3d, coords, &TranslateRank);
// Find neighbouring ranks
// Syntax: comm3d, shift direction, displacement, source, destination
MPI_Cart_shift(comm3d, 0, 1, &left , &right);
MPI_Cart_shift(comm3d, 1, 1, &front, &back);
MPI_Cart_shift(comm3d, 2, 1, &bottom, &top);
// std::cout << " My rank " << rank << " " << left << " " << right << " " << front << " " << back << " " << bottom << " " << top << std::endl;
if(rank==0) {
std::cout << " nearby " << right << " " << back << " " << top << std::endl;
}
// Determine ranks of neighbour processes:
int shiftcoord[DIM];
int lbound[DIM],ubound[DIM];
for(int dim=0;dim<DIM;dim++){
lbound[dim]=-1;
ubound[dim]= 1;
}
Neighbour.resize(lbound,ubound);
Neighbour.clear();
for(int dim0=-1; dim0<=1; dim0++){
shiftcoord[0] = (coords[0]+dim0)%nproc[0];
if(shiftcoord[0] < 0) shiftcoord[0]+=nproc[0];
for(int dim1=-1; dim1<=1; dim1++){
shiftcoord[1] = (coords[1]+dim1)%nproc[1];
if(shiftcoord[1] < 0) shiftcoord[1]+=nproc[1];
for(int dim2=-1; dim2<=1; dim2++){
shiftcoord[2] = (coords[2]+dim2)%nproc[2];
if(shiftcoord[2] < 0) shiftcoord[2]+=nproc[2];
MPI_Cart_rank(comm3d, shiftcoord,&Neighbour(dim0,dim1,dim2));
}
}
}
// if(rank==1) {
// for(int dim0=-1; dim0<=1; dim0++){
// for(int dim1=-1; dim1<=1; dim1++){
// for(int dim2=-1; dim2<=1; dim2++){
// std::cout << " neighbour " << dim0 << " " << dim1 << " ";
// std::cout << dim2 << " " << Neighbour(dim0, dim1, dim2);
// std::cout << std::endl;
// }
// }
// }
// }
// Determine absolute position of any rank:
AllRanks.resize(Index::set(0,0,0),
Index::set(nproc[0]-1,nproc[1]-1,nproc[2]-1));
for(int dim0=0; dim0<nproc[0]; ++dim0) {
for(int dim1=0; dim1<nproc[1]; ++dim1) {
for(int dim2=0; dim2<nproc[2]; ++dim2) {
int coord[3] = {dim0, dim1, dim2};
MPI_Cart_rank(comm3d, coord, &AllRanks(dim0, dim1, dim2));
}
}
}
// if(rank==2) {
// std::cout << " Neigh: " << rank << " "<<Neighbour(0,0,0) << " " << AllRanks(2,0,0) << std::endl;
// }
// Now make additional mpi groups relating to planes:
int count(0);
int num_xy = nproc[0]*nproc[1];
int num_xz = nproc[0]*nproc[2];
int num_yz = nproc[1]*nproc[2];
NumMatrix<int,1> x_ranks[nproc[0]];
NumMatrix<int,1> y_ranks[nproc[1]];
NumMatrix<int,1> z_ranks[nproc[2]];
// Walk trough z-axis -- xy plane
for(int irz=0; irz<nproc[2]; irz++) {
count = 0;
z_ranks[irz].resize(Index::set(0), Index::set(num_xy));
for(int irx=0; irx<nproc[0]; irx++) {
for(int iry=0; iry<nproc[1]; iry++) {
z_ranks[irz](count) = AllRanks(irx,iry,irz);
count++;
}
}
}
// Walk trough y-axis -- xz plane
for(int iry=0; iry<nproc[1]; iry++) {
count = 0;
y_ranks[iry].resize(Index::set(0), Index::set(num_xz));
for(int irx=0; irx<nproc[0]; irx++) {
for(int irz=0; irz<nproc[2]; irz++) {
y_ranks[iry](count) = AllRanks(irx,iry,irz);
count++;
}
}
}
// Walk trough x-axis -- yz plane
for(int irx=0; irx<nproc[0]; irx++) {
count = 0;
x_ranks[irx].resize(Index::set(0), Index::set(num_yz));
for(int iry=0; iry<nproc[1]; iry++) {
for(int irz=0; irz<nproc[2]; irz++) {
x_ranks[irx](count) = AllRanks(irx,iry,irz);
count++;
}
}
}
// Build local communicator:
MPI_Group group_all, group_constz, group_consty, group_constx;
// Get standard group handle:
MPI_Comm_group(comm3d, &group_all);
// Devide tasks into groups based on z-position
MPI_Group_incl(group_all, num_xy, z_ranks[coords[2]], &group_constz);
// Devide tasks into groups based on z-position
MPI_Group_incl(group_all, num_xz, y_ranks[coords[1]], &group_consty);
// Devide tasks into groups based on x-position
MPI_Group_incl(group_all, num_yz, x_ranks[coords[0]], &group_constx);
// // Make corresponding communicators:
// MPI_Comm_create(comm3d, group_constz, &comm_plane_xy); // const z
// MPI_Comm_create(comm3d, group_consty, &comm_plane_xz); // const x
// MPI_Comm_create(comm3d, group_constx, &comm_plane_yz); // const x
// // Get corresponding rank
// MPI_Group_rank (group_constz, &rank_plane_xy);
// MPI_Group_rank (group_consty, &rank_plane_xz);
// MPI_Group_rank (group_constx, &rank_plane_yz);
int remain_dims[3];
// x-y plane:
remain_dims[0] = 1;
remain_dims[1] = 1;
remain_dims[2] = 0;
MPI_Cart_sub(comm3d, remain_dims, &comm_plane_xy);
MPI_Comm_rank(comm_plane_xy, &rank_plane_xy);
// x-z plane
remain_dims[0] = 1;
remain_dims[1] = 0;
remain_dims[2] = 1;
MPI_Cart_sub(comm3d, remain_dims, &comm_plane_xz);
MPI_Comm_rank(comm_plane_xz, &rank_plane_xz);
// y-z plane
remain_dims[0] = 0;
remain_dims[1] = 1;
remain_dims[2] = 1;
MPI_Cart_sub(comm3d, remain_dims, &comm_plane_yz);
MPI_Comm_rank(comm_plane_yz, &rank_plane_yz);
}
/**
* accumulates pieces of the spinor field on nodes with index 0 in the dimensions given in which
* the collected data is returned
*/
void spinor_fft_reduce_2d(spinor *localSpinorField,int *collectionRank,spinor*** field_collection,spinor **membuff){
/* this implementation is intended for four dimensional parallelisation */
#if (defined PARALLELXYZT && defined MPI && defined HAVE_FFTW)
int sendRecvCoord[4];
int i;
int dims[]={g_nproc_t,g_nproc_x,g_nproc_y,g_nproc_z};
/* logfile variables */
char *logFilePrefix="Process";
char logFileName[512];
FILE *logFile;
const int MSG_LOCALDATA = 457;
MPI_Status ierr;
MPI_Datatype mpi_local_spinor;
const int which[]={0,1};
(*field_collection)=NULL;
(*membuff)=NULL;
/* int result; */
sprintf(logFileName,"./%s_%02d.log",logFilePrefix,g_cart_id);
logFile=fopen(logFileName,"a");
MPI_Type_contiguous(VOLUME, field_point, &mpi_local_spinor);
MPI_Type_commit(&mpi_local_spinor);
for(i=0;i<4;i++)
sendRecvCoord[i]=g_proc_coords[i];
if( g_proc_coords[which[0]] == 0 && g_proc_coords[which[1]] == 0 ){
/* i am one of the nodes where data is accumulated */
spinor **accu_field;
spinor **fft_field;
spinor *memory_buffer_accu_field;
spinor *memory_buffer_fft_field;
int REDUCTIONVOLUME=1;
int recvRank;
MPI_Request *requests;
MPI_Status *status;
int request_count=0;
int num_requests;
fftw_plan local_2d_fft_forward;
*collectionRank=TRUE;
/* calculate the number of reduced 2d volume accumulated in this node */
/* number of spinor fields in local units */
REDUCTIONVOLUME*=dims[which[0]]*dims[which[1]];
/* number of receive messages */
num_requests=REDUCTIONVOLUME-1;
/* reserve space for receive messages */
requests=(MPI_Request*)malloc(sizeof(MPI_Request)*num_requests);
status=(MPI_Status*)malloc(sizeof(MPI_Status)*num_requests);
fprintf(logFile,"reduction volume = %d\n",REDUCTIONVOLUME);
/* allocate space for spinor field collection */
allocate_spinor_field_array(&accu_field,&memory_buffer_accu_field,VOLUME,REDUCTIONVOLUME);
allocate_spinor_field_array(&fft_field,&memory_buffer_fft_field,VOLUME,REDUCTIONVOLUME);
/* receive from certain nodes pieces of the spinor field */
for(sendRecvCoord[which[0]] = 0 ; sendRecvCoord[which[0]]< dims[which[0]] ; sendRecvCoord[which[0]]++){
for(sendRecvCoord[which[1]] = 0 ; sendRecvCoord[which[1]]< dims[which[1]] ; sendRecvCoord[which[1]]++){
if( sendRecvCoord[which[0]] != 0 || sendRecvCoord[which[1]] != 0){
MPI_Cart_rank(g_cart_grid,sendRecvCoord,&recvRank);
MPI_Irecv(accu_field[sendRecvCoord[which[0]]*dims[which[1]]+sendRecvCoord[which[1]] ] /* buffer */,
1, /* how may */
mpi_local_spinor, /* mpi data type */
recvRank, /* from whom i get it */
MSG_LOCALDATA, /* msg id */
g_cart_grid, /* communicator , status */
requests+request_count);
++request_count;
}
}
}
/* wait until all request finished */
MPI_Waitall(num_requests, requests, status);
assign(accu_field[0],localSpinorField,VOLUME);
/* transpose in xp-t space */
spinor_fft_transpose_xp_t(fft_field[0],accu_field[0],dims[0],dims[1],TRUE,1.);
/* create fftw plan */
local_2d_fft_forward=spinor_fftw_plan2d(fft_field[0],accu_field[0],T*dims[0],LX*dims[1],LY*LZ,1,FFTW_ESTIMATE);
fftw_execute(local_2d_fft_forward);
fftw_destroy_plan(local_2d_fft_forward);
/* assign(accu_field[0],fft_field[0],VOLUME*REDUCTIONVOLUME); */
free_spinor_field_array(&memory_buffer_fft_field); memory_buffer_fft_field=NULL;
/* free_spinor_field_array(&memory_buffer_accu_field); memory_buffer_accu_field=NULL; */
(*field_collection)=accu_field;
(*membuff)=memory_buffer_accu_field;
free(requests); requests = NULL;
free(status); status=NULL;
} else {
int sendRank;
MPI_Request request;
MPI_Status status;
*collectionRank=FALSE;
/* coordinates of the "root" */
sendRecvCoord[which[0]]=0;
sendRecvCoord[which[1]]=0;
MPI_Cart_rank(g_cart_grid,sendRecvCoord,&sendRank);
MPI_Isend(localSpinorField,1,mpi_local_spinor,sendRank,MSG_LOCALDATA,g_cart_grid,&request);
MPI_Wait(&request,&status);
}
MPI_Type_free(&mpi_local_spinor);
fclose(logFile);
#else
if(g_proc_id==0)
fprintf(stderr,"Error: Please choose FOUR dimensional parallelization!!!\n");
#endif
}
int main( int argc, char **argv )
{
int rank, size, i;
int errors=0;
int dims[NUM_DIMS];
int periods[NUM_DIMS];
int coords[NUM_DIMS];
int new_coords[NUM_DIMS];
int reorder = 0;
MPI_Comm comm_temp, comm_cart, new_comm;
int topo_status;
int ndims;
int new_rank;
int remain_dims[NUM_DIMS];
int newnewrank;
MPI_Init( &argc, &argv );
MPI_Comm_rank( MPI_COMM_WORLD, &rank );
MPI_Comm_size( MPI_COMM_WORLD, &size );
/* Clear dims array and get dims for topology */
for(i=0;i<NUM_DIMS;i++) { dims[i] = 0; periods[i] = 0; }
MPI_Dims_create ( size, NUM_DIMS, dims );
/* Make a new communicator with a topology */
MPI_Cart_create ( MPI_COMM_WORLD, 2, dims, periods, reorder, &comm_temp );
MPI_Comm_dup ( comm_temp, &comm_cart );
/* Determine the status of the new communicator */
MPI_Topo_test ( comm_cart, &topo_status );
if (topo_status != MPI_CART) errors++;
/* How many dims do we have? */
MPI_Cartdim_get( comm_cart, &ndims );
if ( ndims != NUM_DIMS ) errors++;
/* Get the topology, does it agree with what we put in? */
for(i=0;i<NUM_DIMS;i++) { dims[i] = 0; periods[i] = 0; }
MPI_Cart_get ( comm_cart, NUM_DIMS, dims, periods, coords );
/* Check that the coordinates are correct */
#if NUM_DIMS == 2
if (rank != coords[1] + coords[0] * dims[1]) {
errors++;
fprintf( stderr,
"Did not get expected coordinate (row major required by MPI standard 6.2)\n" );
}
#endif
/* Does the mapping from coords to rank work? */
MPI_Cart_rank ( comm_cart, coords, &new_rank );
if ( new_rank != rank ) errors++;
/* Does the mapping from rank to coords work */
MPI_Cart_coords ( comm_cart, rank, NUM_DIMS, new_coords );
for (i=0;i<NUM_DIMS;i++)
if ( coords[i] != new_coords[i] )
errors++;
/* Let's shift in each dimension and see how it works! */
/* Because it's late and I'm tired, I'm not making this */
/* automatically test itself. */
for (i=0;i<NUM_DIMS;i++) {
int source, dest;
MPI_Cart_shift(comm_cart, i, 1, &source, &dest);
#ifdef VERBOSE
printf ("[%d] Shifting %d in the %d dimension\n",rank,1,i);
printf ("[%d] source = %d dest = %d\n",rank,source,dest);
#endif
}
/* Subdivide */
remain_dims[0] = 0;
for (i=1; i<NUM_DIMS; i++) remain_dims[i] = 1;
MPI_Cart_sub ( comm_cart, remain_dims, &new_comm );
/* Determine the status of the new communicator */
MPI_Topo_test ( new_comm, &topo_status );
if (topo_status != MPI_CART) errors++;
/* How many dims do we have? */
MPI_Cartdim_get( new_comm, &ndims );
if ( ndims != NUM_DIMS-1 ) errors++;
/* Get the topology, does it agree with what we put in? */
for(i=0;i<NUM_DIMS-1;i++) { dims[i] = 0; periods[i] = 0; }
MPI_Cart_get ( new_comm, ndims, dims, periods, coords );
/* Does the mapping from coords to rank work? */
MPI_Comm_rank ( new_comm, &newnewrank );
MPI_Cart_rank ( new_comm, coords, &new_rank );
if ( new_rank != newnewrank ) errors++;
/* Does the mapping from rank to coords work */
MPI_Cart_coords ( new_comm, new_rank, NUM_DIMS -1, new_coords );
for (i=0;i<NUM_DIMS-1;i++)
if ( coords[i] != new_coords[i] )
errors++;
/* We're at the end */
MPI_Comm_free( &new_comm );
MPI_Comm_free( &comm_temp );
MPI_Comm_free( &comm_cart );
Test_Waitforall( );
if (errors) printf( "[%d] done with %d ERRORS!\n", rank,errors );
MPI_Finalize();
return 0;
}
/*
Check that the MPI implementation properly handles zero-dimensional
Cartesian communicators - the original standard implies that these
should be consistent with higher dimensional topologies and thus
these should work with any MPI implementation. MPI 2.1 made this
requirement explicit.
*/
int main(int argc, char *argv[])
{
int errs = 0;
int size, rank, ndims;
MPI_Comm comm, newcomm;
MTest_Init(&argc, &argv);
/* Create a new cartesian communicator in a subset of the processes */
MPI_Comm_size(MPI_COMM_WORLD, &size);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if (size < 2) {
fprintf(stderr, "This test needs at least 2 processes\n");
MPI_Abort(MPI_COMM_WORLD, 1);
}
MPI_Cart_create(MPI_COMM_WORLD, 0, NULL, NULL, 0, &comm);
if (comm != MPI_COMM_NULL) {
int csize;
MPI_Comm_size(comm, &csize);
if (csize != 1) {
errs++;
fprintf(stderr, "Sizes is wrong in cart communicator. Is %d, should be 1\n", csize);
}
/* This function is not meaningful, but should not fail */
MPI_Dims_create(1, 0, NULL);
ndims = -1;
MPI_Cartdim_get(comm, &ndims);
if (ndims != 0) {
errs++;
fprintf(stderr, "MPI_Cartdim_get: ndims is %d, should be 0\n", ndims);
}
/* this function should not fail */
MPI_Cart_get(comm, 0, NULL, NULL, NULL);
MPI_Cart_rank(comm, NULL, &rank);
if (rank != 0) {
errs++;
fprintf(stderr, "MPI_Cart_rank: rank is %d, should be 0\n", rank);
}
/* this function should not fail */
MPI_Cart_coords(comm, 0, 0, NULL);
MPI_Cart_sub(comm, NULL, &newcomm);
ndims = -1;
MPI_Cartdim_get(newcomm, &ndims);
if (ndims != 0) {
errs++;
fprintf(stderr, "MPI_Cart_sub did not return zero-dimensional communicator\n");
}
MPI_Barrier(comm);
MPI_Comm_free(&comm);
MPI_Comm_free(&newcomm);
} else if (rank == 0) {
errs++;
fprintf(stderr, "Communicator returned is null!");
}
MTest_Finalize(errs);
return MTestReturnValue(errs);
}
int main(int argc, char **argv) {
const int n_c = 3; // number of colors
int c, i, j, mu, nu, ir, is, ia, imunu;
int filename_set = 0;
int dims[4] = {0,0,0,0};
int l_LX_at, l_LXstart_at;
int source_location, have_source_flag = 0;
int x0, x1, x2, x3, ix;
int sx0, sx1, sx2, sx3;
int isimag[4];
int gperm[5][4], gperm2[4][4];
int check_position_space_WI=0;
int num_threads = 1, nthreads=-1, threadid=-1;
int exitstatus;
int write_ascii=0;
int mms = 0, mass_id = -1;
int outfile_prefix_set = 0;
int source_proc_coords[4], source_proc_id = -1;
int ud_single_file = 0;
double gperm_sign[5][4], gperm2_sign[4][4];
double *conn = NULL;
double *conn2 = NULL;
double contact_term[8];
double *work=NULL;
int verbose = 0;
int do_gt = 0, status;
char filename[100], contype[400], outfile_prefix[400];
double ratime, retime;
double plaq;
double spinor1[24], spinor2[24], U_[18];
double *gauge_trafo=(double*)NULL;
double *phi=NULL, *chi=NULL;
complex w;
double Usourcebuff[72], *Usource[4];
FILE *ofs;
#ifdef MPI
int *status;
#endif
#ifdef MPI
MPI_Init(&argc, &argv);
#endif
while ((c = getopt(argc, argv, "swah?vgf:t:m:o:")) != -1) {
switch (c) {
case 'v':
verbose = 1;
break;
case 'g':
do_gt = 1;
break;
case 'f':
strcpy(filename, optarg);
filename_set=1;
break;
case 'w':
check_position_space_WI = 1;
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] will check Ward identity in position space\n");
break;
case 't':
num_threads = atoi(optarg);
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] will use %d threads in spacetime loops\n", num_threads);
break;
case 'a':
write_ascii = 1;
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] will write data in ASCII format too\n");
break;
case 'm':
mms = 1;
mass_id = atoi(optarg);
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] will read propagators in MMS format with mass id %d\n", mass_id);
break;
case 'o':
strcpy(outfile_prefix, optarg);
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] will use prefix %s for output filenames\n", outfile_prefix);
outfile_prefix_set = 1;
break;
case 's':
ud_single_file = 1;
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] will read up and down propagator from same file\n");
break;
case 'h':
case '?':
default:
usage();
break;
}
}
if(g_cart_id==0) {
g_the_time = time(NULL);
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] using global time stamp %s", ctime(&g_the_time));
}
/*********************************
* set number of openmp threads
*********************************/
#ifdef OPENMP
omp_set_num_threads(num_threads);
#endif
/* set the default values */
if(filename_set==0) strcpy(filename, "cvc.input");
fprintf(stdout, "# Reading input from file %s\n", filename);
read_input_parser(filename);
/* some checks on the input data */
if((T_global == 0) || (LX==0) || (LY==0) || (LZ==0)) {
if(g_proc_id==0) fprintf(stderr, "\n[avc_exact2_lowmem_xspace] T and L's must be set\n");
usage();
}
if(g_kappa == 0.) {
if(g_proc_id==0) fprintf(stderr, "\n[avc_exact2_lowmem_xspace] kappa should be > 0.n");
usage();
}
/* initialize MPI parameters */
mpi_init(argc, argv);
#ifdef MPI
if((status = (int*)calloc(g_nproc, sizeof(int))) == (int*)NULL) {
MPI_Abort(MPI_COMM_WORLD, 1);
MPI_Finalize();
exit(7);
}
#endif
dims[0]=T_global; dims[1]=LX; dims[2]=LY; dims[3]=LZ;
#ifndef MPI
T = T_global;
Tstart = 0;
l_LX_at = LX;
l_LXstart_at = 0;
#endif
fprintf(stdout, "# [%2d] parameters:\n"\
"# [%2d] T = %3d\n"\
"# [%2d] Tstart = %3d\n"\
"# [%2d] l_LX_at = %3d\n"\
"# [%2d] l_LXstart_at = %3d\n",
g_cart_id, g_cart_id, T, g_cart_id, Tstart, g_cart_id, l_LX_at,
g_cart_id, l_LXstart_at);
#ifdef MPI
if(T==0) {
fprintf(stderr, "[%2d] local T is zero; exit\n", g_cart_id);
MPI_Abort(MPI_COMM_WORLD, 1);
MPI_Finalize();
exit(2);
}
#endif
if(init_geometry() != 0) {
fprintf(stderr, "ERROR from init_geometry\n");
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 1);
MPI_Finalize();
#endif
exit(1);
}
geometry();
alloc_gauge_field(&g_gauge_field, VOLUMEPLUSRAND);
if(!(strcmp(gaugefilename_prefix,"identity")==0)) {
/* read the gauge field */
sprintf(filename, "%s.%.4d", gaugefilename_prefix, Nconf);
if(g_cart_id==0) fprintf(stdout, "reading gauge field from file %s\n", filename);
read_lime_gauge_field_doubleprec(filename);
} else {
/* initialize unit matrices */
if(g_cart_id==0) fprintf(stdout, "\n# [avc_exact] initializing unit matrices\n");
for(ix=0;ix<VOLUME;ix++) {
_cm_eq_id( g_gauge_field + _GGI(ix, 0) );
_cm_eq_id( g_gauge_field + _GGI(ix, 1) );
_cm_eq_id( g_gauge_field + _GGI(ix, 2) );
_cm_eq_id( g_gauge_field + _GGI(ix, 3) );
}
}
#ifdef MPI
xchange_gauge();
#endif
/* measure the plaquette */
plaquette(&plaq);
if(g_cart_id==0) fprintf(stdout, "measured plaquette value: %25.16e\n", plaq);
/*
sprintf(filename, "gauge.%.2d", g_cart_id);
ofs = fopen(filename, "w");
for(x0=0;x0<T;x0++) {
for(x1=0;x1<LX;x1++) {
for(x2=0;x2<LY;x2++) {
for(x3=0;x3<LZ;x3++) {
ix = g_ipt[x0][x1][x2][x3];
for(mu=0;mu<4;mu++) {
for(i=0;i<9;i++) {
fprintf(ofs, "%8d%3d%3d%3d%3d%3d%3d%25.16e%25.16e\n", ix, x0+Tstart, x1+LXstart, x2+LYstart, x3, mu, i, g_gauge_field[_GGI(ix,mu)+2*i], g_gauge_field[_GGI(ix,mu)+2*i+1]);
}
}
}}}}
fclose(ofs);
if(g_cart_id==0) fprintf(stdout, "\nWarning: forced exit\n");
fflush(stdout);
fflush(stderr);
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 255);
MPI_Finalize();
#endif
exit(255);
*/
/* allocate memory for the spinor fields */
no_fields = 2;
if(mms) no_fields++;
g_spinor_field = (double**)calloc(no_fields, sizeof(double*));
for(i=0; i<no_fields; i++) alloc_spinor_field(&g_spinor_field[i], VOLUMEPLUSRAND);
if(mms) {
work = g_spinor_field[no_fields-1];
}
/* allocate memory for the contractions */
conn = (double*)calloc(2 * 16 * VOLUME, sizeof(double));
if( conn==(double*)NULL ) {
fprintf(stderr, "could not allocate memory for contr. fields\n");
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 3);
MPI_Finalize();
#endif
exit(3);
}
#ifdef OPENMP
#pragma omp parallel for
#endif
for(ix=0; ix<32*VOLUME; ix++) conn[ix] = 0.;
conn2 = (double*)calloc(2 * 16 * VOLUME, sizeof(double));
if( conn2 == NULL ) {
fprintf(stderr, "could not allocate memory for contr. fields\n");
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 3);
MPI_Finalize();
#endif
exit(3);
}
#ifdef OPENMP
#pragma omp parallel for
#endif
for(ix=0; ix<32*VOLUME; ix++) conn2[ix] = 0.;
/***********************************************************
* determine source coordinates, find out, if source_location is in this process
***********************************************************/
#if (defined PARALLELTX) || (defined PARALLELTXY)
sx0 = g_source_location / (LX_global*LY_global*LZ);
sx1 = (g_source_location%(LX_global*LY_global*LZ)) / (LY_global*LZ);
sx2 = (g_source_location%(LY_global*LZ)) / LZ;
sx3 = (g_source_location%LZ);
source_proc_coords[0] = sx0 / T;
source_proc_coords[1] = sx1 / LX;
source_proc_coords[2] = sx2 / LY;
source_proc_coords[3] = 0;
MPI_Cart_rank(g_cart_grid, source_proc_coords, &source_proc_id);
have_source_flag = (int)(g_cart_id == source_proc_id);
if(have_source_flag==1) {
fprintf(stdout, "\n# process %2d has source location\n", source_proc_id);
fprintf(stdout, "\n# global source coordinates: (%3d,%3d,%3d,%3d)\n", sx0, sx1, sx2, sx3);
fprintf(stdout, "\n# source proc coordinates: (%3d,%3d,%3d,%3d)\n", source_proc_coords[0],
source_proc_coords[1], source_proc_coords[2], source_proc_coords[3]);
}
sx0 = sx0 % T;
sx1 = sx1 % LX;
sx2 = sx2 % LY;
sx3 = sx3 % LZ;
# else
have_source_flag = (int)(g_source_location/(LX*LY*LZ)>=Tstart && g_source_location/(LX*LY*LZ)<(Tstart+T));
if(have_source_flag==1) fprintf(stdout, "process %2d has source location\n", g_cart_id);
sx0 = g_source_location/(LX*LY*LZ)-Tstart;
sx1 = (g_source_location%(LX*LY*LZ)) / (LY*LZ);
sx2 = (g_source_location%(LY*LZ)) / LZ;
sx3 = (g_source_location%LZ);
#endif
if(have_source_flag==1) {
fprintf(stdout, "local source coordinates: (%3d,%3d,%3d,%3d)\n", sx0, sx1, sx2, sx3);
source_location = g_ipt[sx0][sx1][sx2][sx3];
}
#ifdef MPI
# if (defined PARALLELTX) || (defined PARALLELTXY)
have_source_flag = source_proc_id;
MPI_Bcast(Usourcebuff, 72, MPI_DOUBLE, have_source_flag, g_cart_grid);
# else
MPI_Gather(&have_source_flag, 1, MPI_INT, status, 1, MPI_INT, 0, g_cart_grid);
if(g_cart_id==0) {
for(mu=0; mu<g_nproc; mu++) fprintf(stdout, "status[%1d]=%d\n", mu,status[mu]);
}
if(g_cart_id==0) {
for(have_source_flag=0; status[have_source_flag]!=1; have_source_flag++);
fprintf(stdout, "have_source_flag= %d\n", have_source_flag);
}
MPI_Bcast(&have_source_flag, 1, MPI_INT, 0, g_cart_grid);
# endif
fprintf(stdout, "[%2d] have_source_flag = %d\n", g_cart_id, have_source_flag);
#else
have_source_flag = 0;
#endif
/*
if(g_cart_id==0) fprintf(stdout, "\nWarning: forced exit\n");
fflush(stdout);
fflush(stderr);
#ifdef MPI
MPI_Abort(MPI_COMM_WORLD, 255);
MPI_Finalize();
#endif
exit(255);
*/
#ifdef MPI
ratime = MPI_Wtime();
#else
ratime = (double)clock() / CLOCKS_PER_SEC;
#endif
/***********************************************************
* initialize the Gamma matrices
***********************************************************/
// gamma_5:
gperm[4][0] = gamma_permutation[5][ 0] / 6;
gperm[4][1] = gamma_permutation[5][ 6] / 6;
gperm[4][2] = gamma_permutation[5][12] / 6;
gperm[4][3] = gamma_permutation[5][18] / 6;
gperm_sign[4][0] = gamma_sign[5][ 0];
gperm_sign[4][1] = gamma_sign[5][ 6];
gperm_sign[4][2] = gamma_sign[5][12];
gperm_sign[4][3] = gamma_sign[5][18];
// gamma_nu gamma_5
for(nu=0;nu<4;nu++) {
// permutation
gperm[nu][0] = gamma_permutation[6+nu][ 0] / 6;
gperm[nu][1] = gamma_permutation[6+nu][ 6] / 6;
gperm[nu][2] = gamma_permutation[6+nu][12] / 6;
gperm[nu][3] = gamma_permutation[6+nu][18] / 6;
// is imaginary ?
isimag[nu] = gamma_permutation[6+nu][0] % 2;
// (overall) sign
gperm_sign[nu][0] = gamma_sign[6+nu][ 0];
gperm_sign[nu][1] = gamma_sign[6+nu][ 6];
gperm_sign[nu][2] = gamma_sign[6+nu][12];
gperm_sign[nu][3] = gamma_sign[6+nu][18];
// write to stdout
if(g_cart_id == 0) {
fprintf(stdout, "# gamma_%d5 = (%f %d, %f %d, %f %d, %f %d)\n", nu,
gperm_sign[nu][0], gperm[nu][0], gperm_sign[nu][1], gperm[nu][1],
gperm_sign[nu][2], gperm[nu][2], gperm_sign[nu][3], gperm[nu][3]);
}
}
// gamma_nu
for(nu=0;nu<4;nu++) {
// permutation
gperm2[nu][0] = gamma_permutation[nu][ 0] / 6;
gperm2[nu][1] = gamma_permutation[nu][ 6] / 6;
gperm2[nu][2] = gamma_permutation[nu][12] / 6;
gperm2[nu][3] = gamma_permutation[nu][18] / 6;
// (overall) sign
gperm2_sign[nu][0] = gamma_sign[nu][ 0];
gperm2_sign[nu][1] = gamma_sign[nu][ 6];
gperm2_sign[nu][2] = gamma_sign[nu][12];
gperm2_sign[nu][3] = gamma_sign[nu][18];
// write to stdout
if(g_cart_id == 0) {
fprintf(stdout, "# gamma_%d = (%f %d, %f %d, %f %d, %f %d)\n", nu,
gperm2_sign[nu][0], gperm2[nu][0], gperm2_sign[nu][1], gperm2[nu][1],
gperm2_sign[nu][2], gperm2[nu][2], gperm2_sign[nu][3], gperm2[nu][3]);
}
}
/**********************************************************
**********************************************************
**
** first contribution
**
**********************************************************
**********************************************************/
/**********************************************
* loop on the Lorentz index nu at source
**********************************************/
for(ia=0; ia<n_c; ia++) {
for(nu=0; nu<4; nu++)
//for(nu=0; nu<4; nu++)
{
// fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] 1st part, processing nu = %d ...\n", nu);
for(ir=0; ir<4; ir++) {
// read 1 up-type propagator color components for spinor index ir
if(!mms) {
get_filename(filename, 0, 3*ir+ia, 1);
exitstatus = read_lime_spinor(g_spinor_field[0], filename, 0);
if(exitstatus != 0) {
fprintf(stderr, "\n# [avc_exact2_lowmem_xspace] Error from read_lime_spinor\n");
exit(111);
}
xchange_field(g_spinor_field[0]);
} else {
sprintf(filename, "%s.%.4d.00.%.2d.cgmms.%.2d.inverted", filename_prefix, Nconf, 3*ir+ia, mass_id);
exitstatus = read_lime_spinor(work, filename, 0);
if(exitstatus != 0) {
fprintf(stderr, "\n# [avc_exact2_lowmem_xspace] Error from read_lime_spinor\n");
exit(111);
}
xchange_field(work);
Qf5(g_spinor_field[0], work, -g_mu);
xchange_field(g_spinor_field[0]);
}
// read 1 dn-type propagator color components for spinor index gamma_perm ( ir )
if(!mms) {
if(ud_single_file) {
get_filename(filename, 0, 3*gperm[nu][ir]+ia, 1);
exitstatus = read_lime_spinor(g_spinor_field[1], filename, 1);
} else {
get_filename(filename, 0, 3*gperm[nu][ir]+ia, -1);
exitstatus = read_lime_spinor(g_spinor_field[1], filename, 0);
}
if(exitstatus != 0) {
fprintf(stderr, "\n# [avc_exact2_lowmem_xspace] Error from read_lime_spinor\n");
exit(111);
}
xchange_field(g_spinor_field[1]);
} else {
sprintf(filename, "%s.%.4d.%.2d.%.2d.cgmms.%.2d.inverted", filename_prefix, Nconf, 4, 3*gperm[nu][ir]+ia, mass_id);
exitstatus = read_lime_spinor(work, filename, 0);
if(exitstatus != 0) {
fprintf(stderr, "\n# [avc_exact2_lowmem_xspace] Error from read_lime_spinor\n");
exit(111);
}
xchange_field(work);
Qf5(g_spinor_field[1], work, g_mu);
xchange_field(g_spinor_field[1]);
}
phi = g_spinor_field[0];
chi = g_spinor_field[1];
//fprintf(stdout, "\n# [nu5] spin index pair (%d, %d); col index %d\n", ir, gperm[nu][ir], ia);
// 1) gamma_nu gamma_5 x U
for(mu=0; mu<4; mu++)
//for(mu=0; mu<1; mu++)
{
imunu = 4*mu+nu;
#ifdef OPENMP
#pragma omp parallel for private(ix, spinor1, spinor2, U_, w) shared(imunu, ia, nu, mu)
#endif
for(ix=0; ix<VOLUME; ix++) {
/*
threadid = omp_get_thread_num();
nthreads = omp_get_num_threads();
fprintf(stdout, "[thread%d] number of threads = %d\n", threadid, nthreads);
*/
_cm_eq_cm_ti_co(U_, &g_gauge_field[_GGI(ix,mu)], &co_phase_up[mu]);
_fv_eq_cm_ti_fv(spinor1, U_, phi+_GSI(g_iup[ix][mu]));
_fv_eq_gamma_ti_fv(spinor2, mu, spinor1);
_fv_mi_eq_fv(spinor2, spinor1);
_fv_eq_gamma_ti_fv(spinor1, 5, spinor2);
_co_eq_fv_dag_ti_fv(&w, chi+_GSI(ix), spinor1);
if(!isimag[nu]) {
conn[_GWI(imunu,ix,VOLUME) ] += gperm_sign[nu][ir] * w.re;
conn[_GWI(imunu,ix,VOLUME)+1] += gperm_sign[nu][ir] * w.im;
} else {
conn[_GWI(imunu,ix,VOLUME) ] += gperm_sign[nu][ir] * w.im;
conn[_GWI(imunu,ix,VOLUME)+1] -= gperm_sign[nu][ir] * w.re;
}
} // of ix
#ifdef OPENMP
#pragma omp parallel for private(ix, spinor1, spinor2, U_, w) shared(imunu, ia, nu, mu)
#endif
for(ix=0; ix<VOLUME; ix++) {
_cm_eq_cm_ti_co(U_, &g_gauge_field[_GGI(ix,mu)], &co_phase_up[mu]);
_fv_eq_cm_dag_ti_fv(spinor1, U_, phi+_GSI(ix));
_fv_eq_gamma_ti_fv(spinor2, mu, spinor1);
_fv_pl_eq_fv(spinor2, spinor1);
_fv_eq_gamma_ti_fv(spinor1, 5, spinor2);
_co_eq_fv_dag_ti_fv(&w, chi+_GSI(g_iup[ix][mu]), spinor1);
if(!isimag[nu]) {
conn[_GWI(imunu,ix,VOLUME) ] += gperm_sign[nu][ir] * w.re;
conn[_GWI(imunu,ix,VOLUME)+1] += gperm_sign[nu][ir] * w.im;
} else {
conn[_GWI(imunu,ix,VOLUME) ] += gperm_sign[nu][ir] * w.im;
conn[_GWI(imunu,ix,VOLUME)+1] -= gperm_sign[nu][ir] * w.re;
}
} // of ix
// contribution to local-local correlator
#ifdef OPENMP
#pragma omp parallel for private(ix, spinor1, spinor2, U_, w) shared(imunu, ia, nu, mu)
#endif
for(ix=0; ix<VOLUME; ix++) {
_fv_eq_gamma_ti_fv(spinor2, mu, phi+_GSI(ix) );
_fv_eq_gamma_ti_fv(spinor1, 5, spinor2);
_co_eq_fv_dag_ti_fv(&w, chi+_GSI(ix), spinor1);
if(!isimag[nu]) {
conn2[_GWI(imunu,ix,VOLUME) ] += gperm_sign[nu][ir] * w.re;
conn2[_GWI(imunu,ix,VOLUME)+1] += gperm_sign[nu][ir] * w.im;
} else {
conn2[_GWI(imunu,ix,VOLUME) ] += gperm_sign[nu][ir] * w.im;
conn2[_GWI(imunu,ix,VOLUME)+1] -= gperm_sign[nu][ir] * w.re;
}
} // of ix
} // of mu
} // of ir
} // of nu
} // of ia loop on colors
// normalisation of contractions
#ifdef OPENMP
#pragma omp parallel for
#endif
for(ix=0; ix<32*VOLUME; ix++) conn[ix] *= -0.5;
#ifdef OPENMP
#pragma omp parallel for
#endif
for(ix=0; ix<32*VOLUME; ix++) conn2[ix] *= -1.;
#ifdef MPI
retime = MPI_Wtime();
#else
retime = (double)clock() / CLOCKS_PER_SEC;
#endif
if(g_cart_id==0) fprintf(stdout, "contractions in %e seconds\n", retime-ratime);
// save results
#ifdef MPI
ratime = MPI_Wtime();
#else
ratime = (double)clock() / CLOCKS_PER_SEC;
#endif
if(outfile_prefix_set) {
sprintf(filename, "%s/cvc_lvc_x.%.4d.t%.2dx%.2dy%.2dz%.2d", outfile_prefix, Nconf, sx0, sx1, sx2, sx3);
} else {
sprintf(filename, "cvc_lvc_x.%.4d.t%.2dx%.2dy%.2dz%.2d", Nconf, sx0, sx1, sx2, sx3);
}
sprintf(contype, "cvc - lvc in position space, all 16 components");
status = write_lime_contraction(conn, filename, 64, 16, contype, Nconf, 0);
if(status != 0) {
fprintf(stderr, "[] Error from write_lime_contractions, status was %d\n", status);
exit(16);
}
if(outfile_prefix_set) {
sprintf(filename, "%s/lvc_lvc_x.%.4d.t%.2dx%.2dy%.2dz%.2d", outfile_prefix, Nconf, sx0, sx1, sx2, sx3);
} else {
sprintf(filename, "lvc_lvc_x.%.4d.t%.2dx%.2dy%.2dz%.2d", Nconf, sx0, sx1, sx2, sx3);
}
sprintf(contype, "lvc - lvc in position space, all 16 components");
status = write_lime_contraction(conn2, filename, 64, 16, contype, Nconf, 0);
if(status != 0) {
fprintf(stderr, "[] Error from write_lime_contractions, status was %d\n", status);
exit(17);
}
#ifndef MPI
if(write_ascii) {
if(outfile_prefix_set) {
sprintf(filename, "%s/cvc_lvc_x.%.4d.ascii", outfile_prefix, Nconf);
} else {
sprintf(filename, "cvc_lvc_x.%.4d.ascii", Nconf);
}
write_contraction(conn, NULL, filename, 16, 2, 0);
if(outfile_prefix_set) {
sprintf(filename, "%s/lvc_lvc_x.%.4d.ascii", outfile_prefix, Nconf);
} else {
sprintf(filename, "lvc_lvc_x.%.4d.ascii", Nconf);
}
write_contraction(conn2, NULL, filename, 16, 2, 0);
}
#endif
#ifdef MPI
retime = MPI_Wtime();
#else
retime = (double)clock() / CLOCKS_PER_SEC;
#endif
if(g_cart_id==0) fprintf(stdout, "saved position space results in %e seconds\n", retime-ratime);
#ifndef MPI
// check the Ward identity in position space
if(check_position_space_WI) {
sprintf(filename, "WI_X.%.4d", Nconf);
ofs = fopen(filename,"w");
fprintf(stdout, "\n# [avc_exact2_lowmem_xspace] checking Ward identity in position space ...\n");
for(x0=0; x0<T; x0++) {
for(x1=0; x1<LX; x1++) {
for(x2=0; x2<LY; x2++) {
for(x3=0; x3<LZ; x3++) {
fprintf(ofs, "# t=%2d x=%2d y=%2d z=%2d\n", x0, x1, x2, x3);
ix=g_ipt[x0][x1][x2][x3];
for(nu=0; nu<4; nu++) {
w.re = conn[_GWI(4*0+nu,ix,VOLUME)] + conn[_GWI(4*1+nu,ix,VOLUME)]
+ conn[_GWI(4*2+nu,ix,VOLUME)] + conn[_GWI(4*3+nu,ix,VOLUME)]
- conn[_GWI(4*0+nu,g_idn[ix][0],VOLUME)] - conn[_GWI(4*1+nu,g_idn[ix][1],VOLUME)]
- conn[_GWI(4*2+nu,g_idn[ix][2],VOLUME)] - conn[_GWI(4*3+nu,g_idn[ix][3],VOLUME)];
w.im = conn[_GWI(4*0+nu,ix,VOLUME)+1] + conn[_GWI(4*1+nu,ix,VOLUME)+1]
+ conn[_GWI(4*2+nu,ix,VOLUME)+1] + conn[_GWI(4*3+nu,ix,VOLUME)+1]
- conn[_GWI(4*0+nu,g_idn[ix][0],VOLUME)+1] - conn[_GWI(4*1+nu,g_idn[ix][1],VOLUME)+1]
- conn[_GWI(4*2+nu,g_idn[ix][2],VOLUME)+1] - conn[_GWI(4*3+nu,g_idn[ix][3],VOLUME)+1];
fprintf(ofs, "\t%3d%25.16e%25.16e\n", nu, w.re, w.im);
}
}}}}
fclose(ofs);
}
#endif
/****************************************
* free the allocated memory, finalize
****************************************/
free(g_gauge_field);
for(i=0; i<no_fields; i++) free(g_spinor_field[i]);
free(g_spinor_field);
free_geometry();
if(conn != NULL) free(conn);
if(conn2 != NULL) free(conn2);
#ifdef MPI
free(status);
MPI_Finalize();
#endif
if(g_cart_id==0) {
g_the_time = time(NULL);
fprintf(stdout, "\n# [cvc_lvc_exact2_lowmem_xspace] %s# [cvc_lvc_exact2_lowmem_xspace] end of run\n", ctime(&g_the_time));
fprintf(stderr, "\n# [cvc_lvc_exact2_lowmem_xspace] %s# [cvc_lvc_exact2_lowmem_xspace] end of run\n", ctime(&g_the_time));
}
return(0);
}
