void gtrafo_eo_nd(spinor * const Even_s, spinor * const Odd_s, spinor * const Even_c, spinor * const Odd_c, spinor * const Even_new_s, spinor * const Odd_new_s, spinor * const Even_new_c, spinor * const Odd_new_c, GTRAFO_TYPE type){ /* initialize temporal gauge here */ int retval; double dret1, dret2; static double plaquette1 = 0.0; static double plaquette2 = 0.0; if(type==GTRAFO_APPLY){ /* need VOLUME here (not N=VOLUME/2)*/ if ((retval = init_temporalgauge_trafo(VOLUME, g_gauge_field)) != 0 ) { // initializes the transformation matrices if (g_proc_id == 0) printf("Error while gauge fixing to temporal gauge. Aborting...\n"); // g_tempgauge_field as a copy of g_gauge_field exit(200); } /* do trafo */ plaquette1 = measure_plaquette(g_gauge_field); apply_gtrafo(g_gauge_field, g_trafo); // transformation of the gauge field plaquette2 = measure_plaquette(g_gauge_field); if (g_proc_id == 0) printf("\tPlaquette before gauge fixing: %.16e\n", plaquette1/6./VOLUME); if (g_proc_id == 0) printf("\tPlaquette after gauge fixing: %.16e\n", plaquette2/6./VOLUME); /* do trafo to odd_s part of source */ dret1 = square_norm(Odd_s, VOLUME/2 , 1); apply_gtrafo_spinor_odd(Odd_s, g_trafo); // odd spinor transformation, strange dret2 = square_norm(Odd_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* do trafo to odd_c part of source */ dret1 = square_norm(Odd_c, VOLUME/2 , 1); apply_gtrafo_spinor_odd(Odd_c, g_trafo); // odd spinor transformation, charm dret2 = square_norm(Odd_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* do trafo to even_s part of source */ dret1 = square_norm(Even_s, VOLUME/2 , 1); apply_gtrafo_spinor_even(Even_s, g_trafo); // even spinor transformation, strange dret2 = square_norm(Even_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* do trafo to even_c part of source */ dret1 = square_norm(Even_c, VOLUME/2 , 1); apply_gtrafo_spinor_even(Even_c, g_trafo); // even spinor transformation, charm dret2 = square_norm(Even_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); } else { /* undo trafo */ /* apply_inv_gtrafo(g_gauge_field, g_trafo);*/ /* copy back the saved original field located in g_tempgauge_field -> update necessary*/ plaquette1 = measure_plaquette(g_gauge_field); copy_gauge_field(g_gauge_field, g_tempgauge_field); g_update_gauge_copy = 1; plaquette2 = measure_plaquette(g_gauge_field); if (g_proc_id == 0) printf("\tPlaquette before inverse gauge fixing: %.16e\n", plaquette1/6./VOLUME); if (g_proc_id == 0) printf("\tPlaquette after inverse gauge fixing: %.16e\n", plaquette2/6./VOLUME); /* undo trafo to source Even_s */ dret1 = square_norm(Even_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_s, g_trafo); dret2 = square_norm(Even_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* undo trafo to source Even_c */ dret1 = square_norm(Even_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_c, g_trafo); dret2 = square_norm(Even_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* undo trafo to source Odd_s */ dret1 = square_norm(Odd_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_s, g_trafo); dret2 = square_norm(Odd_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* undo trafo to source Odd_c */ dret1 = square_norm(Odd_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_c, g_trafo); dret2 = square_norm(Odd_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Even_new_s dret1 = square_norm(Even_new_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_new_s, g_trafo); dret2 = square_norm(Even_new_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Even_new_c dret1 = square_norm(Even_new_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_new_c, g_trafo); dret2 = square_norm(Even_new_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Odd_new_s dret1 = square_norm(Odd_new_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_new_s, g_trafo); dret2 = square_norm(Odd_new_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Odd_new_c dret1 = square_norm(Odd_new_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_new_c, g_trafo); dret2 = square_norm(Odd_new_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); finalize_temporalgauge(); } # ifdef TM_USE_MPI xchange_gauge(g_gauge_field); # endif }
int main(int argc,char *argv[]) { FILE *parameterfile = NULL; char datafilename[206]; char parameterfilename[206]; char conf_filename[50]; char scalar_filename[50]; char * input_filename = NULL; char * filename = NULL; double plaquette_energy; #ifdef _USE_HALFSPINOR #undef _USE_HALFSPINOR printf("# WARNING: USE_HALFSPINOR will be ignored (not supported here).\n"); #endif if(even_odd_flag) { even_odd_flag=0; printf("# WARNING: even_odd_flag will be ignored (not supported here).\n"); } int j,j_max,k,k_max = 2; _Complex double * drvsc; #ifdef HAVE_LIBLEMON paramsXlfInfo *xlfInfo; #endif int status = 0; static double t1,t2,dt,sdt,dts,qdt,sqdt; double antioptaway=0.0; #ifdef MPI static double dt2; DUM_DERI = 6; DUM_SOLVER = DUM_DERI+2; DUM_MATRIX = DUM_SOLVER+6; NO_OF_SPINORFIELDS = DUM_MATRIX+2; #ifdef OMP int mpi_thread_provided; MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_thread_provided); #else MPI_Init(&argc, &argv); #endif MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id); #else g_proc_id = 0; #endif g_rgi_C1 = 1.; process_args(argc,argv,&input_filename,&filename); set_default_filenames(&input_filename, &filename); /* Read the input file */ if( (j = read_input(input_filename)) != 0) { fprintf(stderr, "Could not find input file: %s\nAborting...\n", input_filename); exit(-1); } if(g_proc_id==0) { printf("parameter rho_BSM set to %f\n", rho_BSM); printf("parameter eta_BSM set to %f\n", eta_BSM); printf("parameter m0_BSM set to %f\n", m0_BSM); } #ifdef OMP init_openmp(); #endif tmlqcd_mpi_init(argc, argv); if(g_proc_id==0) { #ifdef SSE printf("# The code was compiled with SSE instructions\n"); #endif #ifdef SSE2 printf("# The code was compiled with SSE2 instructions\n"); #endif #ifdef SSE3 printf("# The code was compiled with SSE3 instructions\n"); #endif #ifdef P4 printf("# The code was compiled for Pentium4\n"); #endif #ifdef OPTERON printf("# The code was compiled for AMD Opteron\n"); #endif #ifdef _GAUGE_COPY printf("# The code was compiled with -D_GAUGE_COPY\n"); #endif #ifdef BGL printf("# The code was compiled for Blue Gene/L\n"); #endif #ifdef BGP printf("# The code was compiled for Blue Gene/P\n"); #endif #ifdef _USE_HALFSPINOR printf("# The code was compiled with -D_USE_HALFSPINOR\n"); #endif #ifdef _USE_SHMEM printf("# The code was compiled with -D_USE_SHMEM\n"); #ifdef _PERSISTENT printf("# The code was compiled for persistent MPI calls (halfspinor only)\n"); #endif #endif #ifdef MPI #ifdef _NON_BLOCKING printf("# The code was compiled for non-blocking MPI calls (spinor and gauge)\n"); #endif #endif printf("\n"); fflush(stdout); } #ifdef _GAUGE_COPY init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1); #else init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0); #endif init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand); j = init_bispinor_field(VOLUMEPLUSRAND, 12); if ( j!= 0) { fprintf(stderr, "Not enough memory for bispinor fields! Aborting...\n"); exit(0); } j = init_spinor_field(VOLUMEPLUSRAND, 12); if ( j!= 0) { fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n"); exit(0); } int numbScalarFields = 4; j = init_scalar_field(VOLUMEPLUSRAND, numbScalarFields); if ( j!= 0) { fprintf(stderr, "Not enough memory for scalar fields! Aborting...\n"); exit(0); } drvsc = malloc(18*VOLUMEPLUSRAND*sizeof(_Complex double)); if(g_proc_id == 0) { fprintf(stdout,"# The number of processes is %d \n",g_nproc); printf("# The lattice size is %d x %d x %d x %d\n", (int)(T*g_nproc_t), (int)(LX*g_nproc_x), (int)(LY*g_nproc_y), (int)(g_nproc_z*LZ)); printf("# The local lattice size is %d x %d x %d x %d\n", (int)(T), (int)(LX), (int)(LY),(int) LZ); fflush(stdout); } /* define the geometry */ geometry(); j = init_bsm_2hop_lookup(VOLUME); if ( j!= 0) { // this should not be reached since the init function calls fatal_error anyway fprintf(stderr, "Not enough memory for BSM2b 2hop lookup table! Aborting...\n"); exit(0); } /* define the boundary conditions for the fermion fields */ /* for the actual inversion, this is done in invert.c as the operators are iterated through */ // // For the BSM operator we don't use kappa normalisation, // as a result, when twisted boundary conditions are applied this needs to be unity. // In addition, unlike in the Wilson case, the hopping term comes with a plus sign. // However, in boundary(), the minus sign for the Wilson case is implicitly included. // We therefore use -1.0 here. boundary(-1.0); status = check_geometry(); if (status != 0) { fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n"); exit(1); } #if (defined MPI && !(defined _USE_SHMEM)) // fails, we're not using spinor fields // check_xchange(); #endif start_ranlux(1, 123456); // read gauge field if( strcmp(gauge_input_filename, "create_random_gaugefield") == 0 ) { random_gauge_field(reproduce_randomnumber_flag, g_gauge_field); } else { sprintf(conf_filename, "%s.%.4d", gauge_input_filename, nstore); if (g_cart_id == 0) { printf("#\n# Trying to read gauge field from file %s in %s precision.\n", conf_filename, (gauge_precision_read_flag == 32 ? "single" : "double")); fflush(stdout); } int i; if( (i = read_gauge_field(conf_filename,g_gauge_field)) !=0) { fprintf(stderr, "Error %d while reading gauge field from %s\n Aborting...\n", i, conf_filename); exit(-2); } if (g_cart_id == 0) { printf("# Finished reading gauge field.\n"); fflush(stdout); } } // read scalar field if( strcmp(scalar_input_filename, "create_random_scalarfield") == 0 ) { for( int s=0; s<numbScalarFields; s++ ) ranlxd(g_scalar_field[s], VOLUME); } else { sprintf(scalar_filename, "%s.%d", scalar_input_filename, nscalar); if (g_cart_id == 0) { printf("#\n# Trying to read scalar field from file %s in %s precision.\n", scalar_filename, (scalar_precision_read_flag == 32 ? "single" : "double")); fflush(stdout); } int i; if( (i = read_scalar_field(scalar_filename,g_scalar_field)) !=0) { fprintf(stderr, "Error %d while reading scalar field from %s\n Aborting...\n", i, scalar_filename); exit(-2); } if (g_cart_id == 0) { printf("# Finished reading scalar field.\n"); fflush(stdout); } } #ifdef MPI xchange_gauge(g_gauge_field); #endif /*compute the energy of the gauge field*/ plaquette_energy = measure_plaquette( (const su3**) g_gauge_field); if (g_cart_id == 0) { printf("# The computed plaquette value is %e.\n", plaquette_energy / (6.*VOLUME*g_nproc)); fflush(stdout); } #ifdef MPI for( int s=0; s<numbScalarFields; s++ ) generic_exchange(g_scalar_field[s], sizeof(scalar)); #endif /*initialize the bispinor fields*/ j_max=1; sdt=0.; // w random_spinor_field_lexic( (spinor*)(g_bispinor_field[4]), reproduce_randomnumber_flag, RN_GAUSS); random_spinor_field_lexic( (spinor*)(g_bispinor_field[4])+VOLUME, reproduce_randomnumber_flag, RN_GAUSS); // for the D^\dagger test: // v random_spinor_field_lexic( (spinor*)(g_bispinor_field[5]), reproduce_randomnumber_flag, RN_GAUSS); random_spinor_field_lexic( (spinor*)(g_bispinor_field[5])+VOLUME, reproduce_randomnumber_flag, RN_GAUSS); #if defined MPI generic_exchange(g_bispinor_field[4], sizeof(bispinor)); #endif // print L2-norm of source: double squarenorm = square_norm((spinor*)g_bispinor_field[4], 2*VOLUME, 1); if(g_proc_id==0) { printf("\n# square norm of the source: ||w||^2 = %e\n\n", squarenorm); fflush(stdout); } double t_MG, t_BK; /* inversion needs to be done first because it uses loads of the g_bispinor_fields internally */ #if TEST_INVERSION if(g_proc_id==1) printf("Testing inversion\n"); // Bartek's operator t1 = gettime(); cg_her_bi(g_bispinor_field[9], g_bispinor_field[4], 25000, 1.0e-14, 0, VOLUME, &Q2_psi_BSM2b); t_BK = gettime() - t1; // Marco's operator t1 = gettime(); cg_her_bi(g_bispinor_field[8], g_bispinor_field[4], 25000, 1.0e-14, 0, VOLUME, &Q2_psi_BSM2m); t_MG = gettime() - t1; if(g_proc_id==0) printf("Operator inversion time: t_MG = %f sec \t t_BK = %f sec\n\n", t_MG, t_BK); #endif /* now apply the operators to the same bispinor field and do various comparisons */ // Marco's operator #ifdef MPI MPI_Barrier(MPI_COMM_WORLD); #endif t_MG = 0.0; t1 = gettime(); D_psi_BSM2m(g_bispinor_field[0], g_bispinor_field[4]); t1 = gettime() - t1; #ifdef MPI MPI_Allreduce (&t1, &t_MG, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD); #else t_MG = t1; #endif // Bartek's operator #ifdef MPI MPI_Barrier(MPI_COMM_WORLD); #endif t_BK = 0.0; t1 = gettime(); D_psi_BSM2b(g_bispinor_field[1], g_bispinor_field[4]); t1 = gettime() - t1; #ifdef MPI MPI_Allreduce (&t1, &t_BK, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD); #else t_BK = t1; #endif if(g_proc_id==0) printf("Operator application time: t_MG = %f sec \t t_BK = %f sec\n\n", t_MG, t_BK); squarenorm = square_norm((spinor*)g_bispinor_field[0], 2*VOLUME, 1); if(g_proc_id==0) { printf("# || D_MG w ||^2 = %.16e\n", squarenorm); fflush(stdout); } squarenorm = square_norm((spinor*)g_bispinor_field[1], 2*VOLUME, 1); if(g_proc_id==0) { printf("# || D_BK w ||^2 = %.16e\n\n\n", squarenorm); fflush(stdout); } diff( (spinor*)g_bispinor_field[3], (spinor*)g_bispinor_field[0], (spinor*)g_bispinor_field[1], 2*VOLUME); printf("element-wise difference between (D_BK w) and (D_MG w)\n"); printf("( D_MG w - M_BK w )->sp_up.s0.c0= %.16e + I*(%.16e)\n\n", creal(g_bispinor_field[3][0].sp_up.s0.c0), cimag(g_bispinor_field[3][0].sp_up.s0.c0) ); double diffnorm = square_norm( (spinor*) g_bispinor_field[3], 2*VOLUME, 1 ); if(g_proc_id==0){ printf("Square norm of the difference\n"); printf("|| D_MG w - D_BK w ||^2 = %.16e \n\n\n", diffnorm); } // < D w, v > printf("Check consistency of D and D^dagger\n"); _Complex double prod1_MG = scalar_prod( (spinor*)g_bispinor_field[0], (spinor*)g_bispinor_field[5], 2*VOLUME, 1 ); if(g_proc_id==0) printf("< D_MG w, v > = %.16e + I*(%.16e)\n", creal(prod1_MG), cimag(prod1_MG)); _Complex double prod1_BK = scalar_prod( (spinor*)g_bispinor_field[1], (spinor*)g_bispinor_field[5], 2*VOLUME, 1 ); if(g_proc_id==0) printf("< D_BK w, v > = %.16e + I*(%.16e)\n\n", creal(prod1_BK), cimag(prod1_BK)); // < w, D^\dagger v > t_MG = gettime(); D_psi_dagger_BSM2m(g_bispinor_field[6], g_bispinor_field[5]); t_MG = gettime()-t_MG; t_BK = gettime(); D_psi_dagger_BSM2b(g_bispinor_field[7], g_bispinor_field[5]); t_BK = gettime() - t_BK; if(g_proc_id==0) printf("Operator dagger application time: t_MG = %f sec \t t_BK = %f sec\n\n", t_MG, t_BK); _Complex double prod2_MG = scalar_prod((spinor*)g_bispinor_field[4], (spinor*)g_bispinor_field[6], 2*VOLUME, 1); _Complex double prod2_BK = scalar_prod((spinor*)g_bispinor_field[4], (spinor*)g_bispinor_field[7], 2*VOLUME, 1); if( g_proc_id == 0 ){ printf("< w, D_MG^dagger v > = %.16e + I*(%.16e)\n", creal(prod2_MG), cimag(prod2_MG)); printf("< w, D_BK^dagger v > = %.16e + I*(%.16e)\n", creal(prod2_BK), cimag(prod2_BK)); printf("\n| < D_MG w, v > - < w, D_MG^dagger v > | = %.16e\n",cabs(prod2_MG-prod1_MG)); printf("| < D_BK w, v > - < w, D_BK^dagger v > | = %.16e\n\n",cabs(prod2_BK-prod1_BK)); } #if TEST_INVERSION // check result of inversion Q2_psi_BSM2m(g_bispinor_field[10], g_bispinor_field[8]); Q2_psi_BSM2b(g_bispinor_field[11], g_bispinor_field[8]); assign_diff_mul((spinor*)g_bispinor_field[10], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME); assign_diff_mul((spinor*)g_bispinor_field[11], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME); double squarenorm_MGMG = square_norm((spinor*)g_bispinor_field[10], 2*VOLUME, 1); double squarenorm_BKMG = square_norm((spinor*)g_bispinor_field[11], 2*VOLUME, 1); if(g_proc_id==0) { printf("# ||Q2_MG*(Q2_MG)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_MGMG); printf("# ||Q2_BK*(Q2_MG)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_BKMG); fflush(stdout); } Q2_psi_BSM2b(g_bispinor_field[10], g_bispinor_field[9]); Q2_psi_BSM2m(g_bispinor_field[11], g_bispinor_field[9]); assign_diff_mul((spinor*)g_bispinor_field[10], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME); assign_diff_mul((spinor*)g_bispinor_field[11], (spinor*)g_bispinor_field[4], 1.0, 2*VOLUME); double squarenorm_BKBK = square_norm((spinor*)g_bispinor_field[10], 2*VOLUME, 1); double squarenorm_MGBK = square_norm((spinor*)g_bispinor_field[11], 2*VOLUME, 1); if(g_proc_id==0) { printf("# ||Q2_BK*(Q2_BK)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_BKBK); printf("# ||Q2_MG*(Q2_BK)^-1*(b)-b||^2 = %.16e\n\n", squarenorm_MGBK); fflush(stdout); } #endif #ifdef OMP free_omp_accumulators(); #endif free_gauge_field(); free_geometry_indices(); free_bispinor_field(); free_scalar_field(); #ifdef MPI MPI_Barrier(MPI_COMM_WORLD); MPI_Finalize(); #endif return(0); }
int update_tm(double *plaquette_energy, double *rectangle_energy, char * filename, const int return_check, const int acctest, const int traj_counter) { su3 *v, *w; int accept, i=0, j=0, iostatus=0; double yy[1]; double dh, expmdh, ret_dh=0., ret_gauge_diff=0., tmp; double atime=0., etime=0.; double ks = 0., kc = 0., ds, tr, ts, tt; char tmp_filename[50]; /* Energy corresponding to the Gauge part */ double new_plaquette_energy=0., new_rectangle_energy = 0.; /* Energy corresponding to the Momenta part */ double enep=0., enepx=0., ret_enep = 0.; /* Energy corresponding to the pseudo fermion part(s) */ FILE * datafile=NULL, * ret_check_file=NULL; hamiltonian_field_t hf; paramsXlfInfo *xlfInfo; hf.gaugefield = g_gauge_field; hf.momenta = moment; hf.derivative = df0; hf.update_gauge_copy = g_update_gauge_copy; hf.traj_counter = traj_counter; integrator_set_fields(&hf); sprintf(tmp_filename, ".conf.t%05d.tmp",traj_counter); atime = gettime(); /* * here the momentum and spinor fields are initialized * and their respective actions are calculated */ /* * copy the gauge field to gauge_tmp */ #ifdef TM_USE_OMP #pragma omp parallel for private(w,v) #endif for(int ix=0;ix<VOLUME;ix++) { for(int mu=0;mu<4;mu++) { v=&hf.gaugefield[ix][mu]; w=&gauge_tmp[ix][mu]; _su3_assign(*w,*v); } } #ifdef DDalphaAMG MG_reset(); #endif /* heatbath for all monomials */ for(i = 0; i < Integrator.no_timescales; i++) { for(j = 0; j < Integrator.no_mnls_per_ts[i]; j++) { monomial_list[ Integrator.mnls_per_ts[i][j] ].hbfunction(Integrator.mnls_per_ts[i][j], &hf); } } if(Integrator.monitor_forces) monitor_forces(&hf); /* initialize the momenta */ enep = random_su3adj_field(reproduce_randomnumber_flag, hf.momenta); g_sloppy_precision = 1; /* run the trajectory */ if(Integrator.n_int[Integrator.no_timescales-1] > 0) { Integrator.integrate[Integrator.no_timescales-1](Integrator.tau, Integrator.no_timescales-1, 1); } g_sloppy_precision = 0; /* compute the final energy contributions for all monomials */ dh = 0.; for(i = 0; i < Integrator.no_timescales; i++) { for(j = 0; j < Integrator.no_mnls_per_ts[i]; j++) { dh += monomial_list[ Integrator.mnls_per_ts[i][j] ].accfunction(Integrator.mnls_per_ts[i][j], &hf); } } enepx = moment_energy(hf.momenta); if (!bc_flag) { /* if PBC */ new_plaquette_energy = measure_plaquette( (const su3**) hf.gaugefield); if(g_rgi_C1 > 0. || g_rgi_C1 < 0.) { new_rectangle_energy = measure_rectangles( (const su3**) hf.gaugefield); } } if(g_proc_id == 0 && g_debug_level > 3) printf("called moment_energy: dh = %1.10e\n", (enepx - enep)); /* Compute the energy difference */ dh = dh + (enepx - enep); if(g_proc_id == 0 && g_debug_level > 3) { printf("called momenta_acc dH = %e\n", (enepx - enep)); } expmdh = exp(-dh); /* the random number is only taken at node zero and then distributed to the other sites */ ranlxd(yy,1); #ifdef TM_USE_MPI MPI_Bcast(&yy[0], 1, MPI_DOUBLE, 0, MPI_COMM_WORLD); #endif /* when acctest is 0 (i.e. do not perform acceptance test), the trajectory is accepted whatever the energy difference */ accept = (!acctest | (expmdh > yy[0])); if(g_proc_id == 0) { fprintf(stdout, "# Trajectory is %saccepted.\n", (accept ? "" : "not ")); } /* Here a reversibility test is performed */ /* The trajectory is integrated back */ if(return_check) { if(g_proc_id == 0) { fprintf(stdout, "# Performing reversibility check.\n"); } if(accept) { /* save gauge file to disk before performing reversibility check */ xlfInfo = construct_paramsXlfInfo((*plaquette_energy)/(6.*VOLUME*g_nproc), traj_counter); // Should write this to temporary file first, and then check if(g_proc_id == 0 && g_debug_level > 0) { fprintf(stdout, "# Writing gauge field to file %s.\n", tmp_filename); } if((iostatus = write_gauge_field( tmp_filename, 64, xlfInfo) != 0 )) { /* Writing failed directly */ fprintf(stderr, "Error %d while writing gauge field to %s\nAborting...\n", iostatus, tmp_filename); exit(-2); } /* There is double writing of the gauge field, also in hmc_tm.c in this case */ /* No reading back check needed here, as reading back is done further down */ if(g_proc_id == 0 && g_debug_level > 0) { fprintf(stdout, "# Writing done.\n"); } free(xlfInfo); } #ifdef DDalphaAMG MG_reset(); #endif g_sloppy_precision = 1; /* run the trajectory back */ Integrator.integrate[Integrator.no_timescales-1](-Integrator.tau, Integrator.no_timescales-1, 1); g_sloppy_precision = 0; /* compute the energy contributions from the pseudo-fermions */ ret_dh = 0.; for(i = 0; i < Integrator.no_timescales; i++) { for(j = 0; j < Integrator.no_mnls_per_ts[i]; j++) { ret_dh += monomial_list[ Integrator.mnls_per_ts[i][j] ].accfunction(Integrator.mnls_per_ts[i][j], &hf); } } ret_enep = moment_energy(hf.momenta); /* Compute the energy difference */ ret_dh += ret_enep - enep ; /* Compute Differences in the fields */ ks = 0.; kc = 0.; #ifdef TM_USE_OMP #pragma omp parallel private(w,v,tt,tr,ts,ds,ks,kc) { int thread_num = omp_get_thread_num(); #endif su3 ALIGN v0; #ifdef TM_USE_OMP #pragma omp for #endif for(int ix = 0; ix < VOLUME; ++ix) { for(int mu = 0; mu < 4; ++mu) { v=&hf.gaugefield[ix][mu]; w=&gauge_tmp[ix][mu]; _su3_minus_su3(v0, *v, *w); _su3_square_norm(ds, v0); tr = sqrt(ds) + kc; ts = tr + ks; tt = ts-ks; ks = ts; kc = tr-tt; } } kc=ks+kc; #ifdef TM_USE_OMP g_omp_acc_re[thread_num] = kc; } /* OpenMP parallel section closing brace */ /* sum up contributions from thread-local kahan summations */ for(int k = 0; k < omp_num_threads; ++k) ret_gauge_diff += g_omp_acc_re[k]; #else ret_gauge_diff = kc; #endif #ifdef TM_USE_MPI tmp = ret_gauge_diff; MPI_Reduce(&tmp, &ret_gauge_diff, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD); #endif /* compute the total H */ tmp = enep; for(i = 0; i < Integrator.no_timescales; i++) { for(j = 0; j < Integrator.no_mnls_per_ts[i]; j++) { tmp += monomial_list[ Integrator.mnls_per_ts[i][j] ].energy0; } } /* Output */ if(g_proc_id == 0) { ret_check_file = fopen("return_check.data","a"); fprintf(ret_check_file,"%08d ddh = %1.4e ddh/dh = %1.4e ddh/H = %1.4e ddU= %1.4e\n", traj_counter, ret_dh, ret_dh/dh, ret_dh/tmp, ret_gauge_diff/4./((double)(VOLUME*g_nproc))/3.); fclose(ret_check_file); } if(accept) { /* Read back gauge field FIXME unlike in hmc_tm we abort immediately if there is a failure */ if(g_proc_id == 0 && g_debug_level > 0) { fprintf(stdout, "# Trying to read gauge field from file %s.\n", tmp_filename); } if((iostatus = read_gauge_field(tmp_filename,g_gauge_field) != 0)) { fprintf(stderr, "Error %d while reading gauge field from %s\nAborting...\n", iostatus, tmp_filename); exit(-2); } if(g_proc_id == 0 && g_debug_level > 0) { fprintf(stdout, "# Reading done.\n"); } } if(g_proc_id == 0) { fprintf(stdout, "# Reversibility check done.\n"); } } /* end of reversibility check */ if(accept) { *plaquette_energy = new_plaquette_energy; *rectangle_energy = new_rectangle_energy; /* put the links back to SU(3) group */ if (!bc_flag) { /* periodic boundary conditions */ #ifdef TM_USE_OMP #pragma omp parallel for private(v) #endif for(int ix=0;ix<VOLUME;ix++) { for(int mu=0;mu<4;mu++) { v=&hf.gaugefield[ix][mu]; restoresu3_in_place(v); } } } } else { /* reject: copy gauge_tmp to hf.gaugefield */ #ifdef TM_USE_OMP #pragma omp parallel for private(w) private(v) #endif for(int ix=0;ix<VOLUME;ix++) { for(int mu=0;mu<4;mu++){ v=&hf.gaugefield[ix][mu]; w=&gauge_tmp[ix][mu]; _su3_assign(*v,*w); } } } hf.update_gauge_copy = 1; g_update_gauge_copy = 1; g_update_gauge_copy_32 = 1; #ifdef TM_USE_MPI xchange_gauge(hf.gaugefield); #endif /*Convert to a 32 bit gauge field, after xchange*/ convert_32_gauge_field(g_gauge_field_32, hf.gaugefield, VOLUMEPLUSRAND + g_dbw2rand); etime=gettime(); /* printing data in the .data file */ if(g_proc_id==0) { datafile = fopen(filename, "a"); if (!bc_flag) { /* if Periodic Boundary Conditions */ fprintf(datafile, "%.8d %14.12f %14.12f %e ", traj_counter, (*plaquette_energy)/(6.*VOLUME*g_nproc), dh, expmdh); } for(i = 0; i < Integrator.no_timescales; i++) { for(j = 0; j < Integrator.no_mnls_per_ts[i]; j++) { if(monomial_list[ Integrator.mnls_per_ts[i][j] ].type != GAUGE && monomial_list[ Integrator.mnls_per_ts[i][j] ].type != SFGAUGE && monomial_list[ Integrator.mnls_per_ts[i][j] ].type != NDPOLY && monomial_list[ Integrator.mnls_per_ts[i][j] ].type != NDCLOVER && monomial_list[ Integrator.mnls_per_ts[i][j] ].type != CLOVERNDTRLOG && monomial_list[ Integrator.mnls_per_ts[i][j] ].type != CLOVERTRLOG ) { fprintf(datafile,"%d %d ", monomial_list[ Integrator.mnls_per_ts[i][j] ].iter0, monomial_list[ Integrator.mnls_per_ts[i][j] ].iter1); } } } fprintf(datafile, "%d %e", accept, etime-atime); if(g_rgi_C1 > 0. || g_rgi_C1 < 0) { fprintf(datafile, " %e", (*rectangle_energy)/(12*VOLUME*g_nproc)); } fprintf(datafile, "\n"); fflush(datafile); fclose(datafile); } return(accept); }
int main(int argc, char *argv[]) { FILE *parameterfile = NULL; int j, i, ix = 0, isample = 0, op_id = 0; char datafilename[206]; char parameterfilename[206]; char conf_filename[50]; char * input_filename = NULL; char * filename = NULL; double plaquette_energy; struct stout_parameters params_smear; #ifdef _KOJAK_INST #pragma pomp inst init #pragma pomp inst begin(main) #endif #if (defined SSE || defined SSE2 || SSE3) signal(SIGILL, &catch_ill_inst); #endif DUM_DERI = 8; DUM_MATRIX = DUM_DERI + 5; NO_OF_SPINORFIELDS = DUM_MATRIX + 4; //4 extra fields (corresponding to DUM_MATRIX+0..5) for deg. and ND matrix mult. NO_OF_SPINORFIELDS_32 = 6; verbose = 0; g_use_clover_flag = 0; process_args(argc,argv,&input_filename,&filename); set_default_filenames(&input_filename, &filename); init_parallel_and_read_input(argc, argv, input_filename); /* this DBW2 stuff is not needed for the inversion ! */ if (g_dflgcr_flag == 1) { even_odd_flag = 0; } g_rgi_C1 = 0; if (Nsave == 0) { Nsave = 1; } if (g_running_phmc) { NO_OF_SPINORFIELDS = DUM_MATRIX + 8; } tmlqcd_mpi_init(argc, argv); g_dbw2rand = 0; /* starts the single and double precision random number */ /* generator */ start_ranlux(rlxd_level, random_seed^nstore); /* we need to make sure that we don't have even_odd_flag = 1 */ /* if any of the operators doesn't use it */ /* in this way even/odd can still be used by other operators */ for(j = 0; j < no_operators; j++) if(!operator_list[j].even_odd_flag) even_odd_flag = 0; #ifndef TM_USE_MPI g_dbw2rand = 0; #endif #ifdef _GAUGE_COPY j = init_gauge_field(VOLUMEPLUSRAND, 1); j += init_gauge_field_32(VOLUMEPLUSRAND, 1); #else j = init_gauge_field(VOLUMEPLUSRAND, 0); j += init_gauge_field_32(VOLUMEPLUSRAND, 0); #endif if (j != 0) { fprintf(stderr, "Not enough memory for gauge_fields! Aborting...\n"); exit(-1); } j = init_geometry_indices(VOLUMEPLUSRAND); if (j != 0) { fprintf(stderr, "Not enough memory for geometry indices! Aborting...\n"); exit(-1); } if (no_monomials > 0) { if (even_odd_flag) { j = init_monomials(VOLUMEPLUSRAND / 2, even_odd_flag); } else { j = init_monomials(VOLUMEPLUSRAND, even_odd_flag); } if (j != 0) { fprintf(stderr, "Not enough memory for monomial pseudo fermion fields! Aborting...\n"); exit(-1); } } if (even_odd_flag) { j = init_spinor_field(VOLUMEPLUSRAND / 2, NO_OF_SPINORFIELDS); j += init_spinor_field_32(VOLUMEPLUSRAND / 2, NO_OF_SPINORFIELDS_32); } else { j = init_spinor_field(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS); j += init_spinor_field_32(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS_32); } if (j != 0) { fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n"); exit(-1); } if (g_running_phmc) { j = init_chi_spinor_field(VOLUMEPLUSRAND / 2, 20); if (j != 0) { fprintf(stderr, "Not enough memory for PHMC Chi fields! Aborting...\n"); exit(-1); } } g_mu = g_mu1; if (g_cart_id == 0) { /*construct the filenames for the observables and the parameters*/ strncpy(datafilename, filename, 200); strcat(datafilename, ".data"); strncpy(parameterfilename, filename, 200); strcat(parameterfilename, ".para"); parameterfile = fopen(parameterfilename, "w"); write_first_messages(parameterfile, "invert", git_hash); fclose(parameterfile); } /* define the geometry */ geometry(); /* define the boundary conditions for the fermion fields */ boundary(g_kappa); phmc_invmaxev = 1.; init_operators(); /* list and initialize measurements*/ if(g_proc_id == 0) { printf("\n"); for(int j = 0; j < no_measurements; j++) { printf("# measurement id %d, type = %d\n", j, measurement_list[j].type); } } init_measurements(); /* this could be maybe moved to init_operators */ #ifdef _USE_HALFSPINOR j = init_dirac_halfspinor(); if (j != 0) { fprintf(stderr, "Not enough memory for halffield! Aborting...\n"); exit(-1); } /* for mixed precision solvers, the 32 bit halfspinor field must always be there */ j = init_dirac_halfspinor32(); if (j != 0) { fprintf(stderr, "Not enough memory for 32-bit halffield! Aborting...\n"); exit(-1); } # if (defined _PERSISTENT) if (even_odd_flag) init_xchange_halffield(); # endif #endif for (j = 0; j < Nmeas; j++) { sprintf(conf_filename, "%s.%.4d", gauge_input_filename, nstore); if (g_cart_id == 0) { printf("#\n# Trying to read gauge field from file %s in %s precision.\n", conf_filename, (gauge_precision_read_flag == 32 ? "single" : "double")); fflush(stdout); } if( (i = read_gauge_field(conf_filename,g_gauge_field)) !=0) { fprintf(stderr, "Error %d while reading gauge field from %s\n Aborting...\n", i, conf_filename); exit(-2); } if (g_cart_id == 0) { printf("# Finished reading gauge field.\n"); fflush(stdout); } #ifdef TM_USE_MPI xchange_gauge(g_gauge_field); #endif /*Convert to a 32 bit gauge field, after xchange*/ convert_32_gauge_field(g_gauge_field_32, g_gauge_field, VOLUMEPLUSRAND); /*compute the energy of the gauge field*/ plaquette_energy = measure_plaquette( (const su3**) g_gauge_field); if (g_cart_id == 0) { printf("# The computed plaquette value is %e.\n", plaquette_energy / (6.*VOLUME*g_nproc)); fflush(stdout); } if (use_stout_flag == 1){ params_smear.rho = stout_rho; params_smear.iterations = stout_no_iter; /* if (stout_smear((su3_tuple*)(g_gauge_field[0]), ¶ms_smear, (su3_tuple*)(g_gauge_field[0])) != 0) */ /* exit(1) ; */ g_update_gauge_copy = 1; plaquette_energy = measure_plaquette( (const su3**) g_gauge_field); if (g_cart_id == 0) { printf("# The plaquette value after stouting is %e\n", plaquette_energy / (6.*VOLUME*g_nproc)); fflush(stdout); } } /* if any measurements are defined in the input file, do them here */ measurement * meas; for(int imeas = 0; imeas < no_measurements; imeas++){ meas = &measurement_list[imeas]; if (g_proc_id == 0) { fprintf(stdout, "#\n# Beginning online measurement.\n"); } meas->measurefunc(nstore, imeas, even_odd_flag); } if (reweighting_flag == 1) { reweighting_factor(reweighting_samples, nstore); } /* Compute minimal eigenvalues, if wanted */ if (compute_evs != 0) { eigenvalues(&no_eigenvalues, 5000, eigenvalue_precision, 0, compute_evs, nstore, even_odd_flag); } if (phmc_compute_evs != 0) { #ifdef TM_USE_MPI MPI_Finalize(); #endif return(0); } /* Compute the mode number or topological susceptibility using spectral projectors, if wanted*/ if(compute_modenumber != 0 || compute_topsus !=0){ invert_compute_modenumber(); } // set up blocks if Deflation is used if (g_dflgcr_flag) init_blocks(nblocks_t, nblocks_x, nblocks_y, nblocks_z); if(SourceInfo.type == SRC_TYPE_VOL || SourceInfo.type == SRC_TYPE_PION_TS || SourceInfo.type == SRC_TYPE_GEN_PION_TS) { index_start = 0; index_end = 1; } g_precWS=NULL; if(use_preconditioning == 1){ /* todo load fftw wisdom */ #if (defined HAVE_FFTW ) && !( defined TM_USE_MPI) loadFFTWWisdom(g_spinor_field[0],g_spinor_field[1],T,LX); #else use_preconditioning=0; #endif } if (g_cart_id == 0) { fprintf(stdout, "#\n"); /*Indicate starting of the operator part*/ } for(op_id = 0; op_id < no_operators; op_id++) { boundary(operator_list[op_id].kappa); g_kappa = operator_list[op_id].kappa; g_mu = operator_list[op_id].mu; g_c_sw = operator_list[op_id].c_sw; // DFLGCR and DFLFGMRES if(operator_list[op_id].solver == DFLGCR || operator_list[op_id].solver == DFLFGMRES) { generate_dfl_subspace(g_N_s, VOLUME, reproduce_randomnumber_flag); } if(use_preconditioning==1 && PRECWSOPERATORSELECT[operator_list[op_id].solver]!=PRECWS_NO ){ printf("# Using preconditioning with treelevel preconditioning operator: %s \n", precWSOpToString(PRECWSOPERATORSELECT[operator_list[op_id].solver])); /* initial preconditioning workspace */ operator_list[op_id].precWS=(spinorPrecWS*)malloc(sizeof(spinorPrecWS)); spinorPrecWS_Init(operator_list[op_id].precWS, operator_list[op_id].kappa, operator_list[op_id].mu/2./operator_list[op_id].kappa, -(0.5/operator_list[op_id].kappa-4.), PRECWSOPERATORSELECT[operator_list[op_id].solver]); g_precWS = operator_list[op_id].precWS; if(PRECWSOPERATORSELECT[operator_list[op_id].solver] == PRECWS_D_DAGGER_D) { fitPrecParams(op_id); } } for(isample = 0; isample < no_samples; isample++) { for (ix = index_start; ix < index_end; ix++) { if (g_cart_id == 0) { fprintf(stdout, "#\n"); /*Indicate starting of new index*/ } /* we use g_spinor_field[0-7] for sources and props for the moment */ /* 0-3 in case of 1 flavour */ /* 0-7 in case of 2 flavours */ prepare_source(nstore, isample, ix, op_id, read_source_flag, source_location, random_seed); //randmize initial guess for eigcg if needed-----experimental if( (operator_list[op_id].solver == INCREIGCG) && (operator_list[op_id].solver_params.eigcg_rand_guess_opt) ){ //randomize the initial guess gaussian_volume_source( operator_list[op_id].prop0, operator_list[op_id].prop1,isample,ix,0); //need to check this } operator_list[op_id].inverter(op_id, index_start, 1); } } if(use_preconditioning==1 && operator_list[op_id].precWS!=NULL ){ /* free preconditioning workspace */ spinorPrecWS_Free(operator_list[op_id].precWS); free(operator_list[op_id].precWS); } if(operator_list[op_id].type == OVERLAP){ free_Dov_WS(); } } nstore += Nsave; } #ifdef TM_USE_OMP free_omp_accumulators(); #endif free_blocks(); free_dfl_subspace(); free_gauge_field(); free_gauge_field_32(); free_geometry_indices(); free_spinor_field(); free_spinor_field_32(); free_moment_field(); free_chi_spinor_field(); free(filename); free(input_filename); free(SourceInfo.basename); free(PropInfo.basename); #ifdef TM_USE_QUDA _endQuda(); #endif #ifdef TM_USE_MPI MPI_Barrier(MPI_COMM_WORLD); MPI_Finalize(); #endif return(0); #ifdef _KOJAK_INST #pragma pomp inst end(main) #endif }
int invert_doublet_eo(spinor * const Even_new_s, spinor * const Odd_new_s, spinor * const Even_new_c, spinor * const Odd_new_c, spinor * const Even_s, spinor * const Odd_s, spinor * const Even_c, spinor * const Odd_c, const double precision, const int max_iter, const int solver_flag, const int rel_prec) { int iter = 0; #ifdef HAVE_GPU # ifdef TEMPORALGAUGE /* initialize temporal gauge here */ int retval; double dret1, dret2; double plaquette1 = 0.0; double plaquette2 = 0.0; if (usegpu_flag) { /* need VOLUME here (not N=VOLUME/2)*/ if ((retval = init_temporalgauge_trafo(VOLUME, g_gauge_field)) != 0 ) { // initializes the transformation matrices if (g_proc_id == 0) printf("Error while gauge fixing to temporal gauge. Aborting...\n"); // g_tempgauge_field as a copy of g_gauge_field exit(200); } /* do trafo */ plaquette1 = measure_plaquette(g_gauge_field); apply_gtrafo(g_gauge_field, g_trafo); // transformation of the gauge field plaquette2 = measure_plaquette(g_gauge_field); if (g_proc_id == 0) printf("\tPlaquette before gauge fixing: %.16e\n", plaquette1/6./VOLUME); if (g_proc_id == 0) printf("\tPlaquette after gauge fixing: %.16e\n", plaquette2/6./VOLUME); /* do trafo to odd_s part of source */ dret1 = square_norm(Odd_s, VOLUME/2 , 1); apply_gtrafo_spinor_odd(Odd_s, g_trafo); // odd spinor transformation, strange dret2 = square_norm(Odd_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* do trafo to odd_c part of source */ dret1 = square_norm(Odd_c, VOLUME/2 , 1); apply_gtrafo_spinor_odd(Odd_c, g_trafo); // odd spinor transformation, charm dret2 = square_norm(Odd_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* do trafo to even_s part of source */ dret1 = square_norm(Even_s, VOLUME/2 , 1); apply_gtrafo_spinor_even(Even_s, g_trafo); // even spinor transformation, strange dret2 = square_norm(Even_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* do trafo to even_c part of source */ dret1 = square_norm(Even_c, VOLUME/2 , 1); apply_gtrafo_spinor_even(Even_c, g_trafo); // even spinor transformation, charm dret2 = square_norm(Even_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); # ifdef MPI xchange_gauge(g_gauge_field); # endif } # endif #endif /* HAVE_GPU*/ /* here comes the inversion using even/odd preconditioning */ if(g_proc_id == 0) {printf("# Using even/odd preconditioning!\n"); fflush(stdout);} M_ee_inv_ndpsi(Even_new_s, Even_new_c, Even_s, Even_c, g_mubar, g_epsbar); Hopping_Matrix(OE, g_spinor_field[DUM_DERI], Even_new_s); Hopping_Matrix(OE, g_spinor_field[DUM_DERI+1], Even_new_c); /* The sign is plus, since in Hopping_Matrix */ /* the minus is missing */ assign_mul_add_r(g_spinor_field[DUM_DERI], +1., Odd_s, VOLUME/2); assign_mul_add_r(g_spinor_field[DUM_DERI+1], +1., Odd_c, VOLUME/2); /* Do the inversion with the preconditioned */ /* matrix to get the odd sites */ /* Here we invert the hermitean operator squared */ if(g_proc_id == 0) { printf("# Using CG for TMWILSON flavour doublet!\n"); fflush(stdout); } gamma5(g_spinor_field[DUM_DERI], g_spinor_field[DUM_DERI], VOLUME/2); gamma5(g_spinor_field[DUM_DERI+1], g_spinor_field[DUM_DERI+1], VOLUME/2); #ifdef HAVE_GPU if (usegpu_flag) { // GPU, mixed precision solver # if defined(MPI) && defined(PARALLELT) iter = mixedsolve_eo_nd(Odd_new_s, Odd_new_c, g_spinor_field[DUM_DERI], g_spinor_field[DUM_DERI+1], max_iter, precision, rel_prec); # elif !defined(MPI) && !defined(PARALLELT) iter = mixedsolve_eo_nd(Odd_new_s, Odd_new_c, g_spinor_field[DUM_DERI], g_spinor_field[DUM_DERI+1], max_iter, precision, rel_prec); # else printf("MPI and/or PARALLELT are not appropriately set for the GPU implementation. Aborting...\n"); exit(-1); # endif } else { // CPU, conjugate gradient iter = cg_her_nd(Odd_new_s, Odd_new_c, g_spinor_field[DUM_DERI], g_spinor_field[DUM_DERI+1], max_iter, precision, rel_prec, VOLUME/2, &Qtm_pm_ndpsi); } #else // CPU, conjugate gradient iter = cg_her_nd(Odd_new_s, Odd_new_c, g_spinor_field[DUM_DERI], g_spinor_field[DUM_DERI+1], max_iter, precision, rel_prec, VOLUME/2, &Qtm_pm_ndpsi); #endif Qtm_dagger_ndpsi(Odd_new_s, Odd_new_c, Odd_new_s, Odd_new_c); /* Reconstruct the even sites */ Hopping_Matrix(EO, g_spinor_field[DUM_DERI], Odd_new_s); Hopping_Matrix(EO, g_spinor_field[DUM_DERI+1], Odd_new_c); M_ee_inv_ndpsi(g_spinor_field[DUM_DERI+2], g_spinor_field[DUM_DERI+3], g_spinor_field[DUM_DERI], g_spinor_field[DUM_DERI+1], g_mubar, g_epsbar); /* The sign is plus, since in Hopping_Matrix */ /* the minus is missing */ assign_add_mul_r(Even_new_s, g_spinor_field[DUM_DERI+2], +1., VOLUME/2); assign_add_mul_r(Even_new_c, g_spinor_field[DUM_DERI+3], +1., VOLUME/2); #ifdef HAVE_GPU /* return from temporal gauge again */ # ifdef TEMPORALGAUGE if (usegpu_flag) { /* undo trafo */ /* apply_inv_gtrafo(g_gauge_field, g_trafo);*/ /* copy back the saved original field located in g_tempgauge_field -> update necessary*/ plaquette1 = measure_plaquette(g_gauge_field); copy_gauge_field(g_gauge_field, g_tempgauge_field); g_update_gauge_copy = 1; plaquette2 = measure_plaquette(g_gauge_field); if (g_proc_id == 0) printf("\tPlaquette before inverse gauge fixing: %.16e\n", plaquette1/6./VOLUME); if (g_proc_id == 0) printf("\tPlaquette after inverse gauge fixing: %.16e\n", plaquette2/6./VOLUME); /* undo trafo to source Even_s */ dret1 = square_norm(Even_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_s, g_trafo); dret2 = square_norm(Even_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* undo trafo to source Even_c */ dret1 = square_norm(Even_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_c, g_trafo); dret2 = square_norm(Even_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* undo trafo to source Odd_s */ dret1 = square_norm(Odd_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_s, g_trafo); dret2 = square_norm(Odd_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); /* undo trafo to source Odd_c */ dret1 = square_norm(Odd_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_c, g_trafo); dret2 = square_norm(Odd_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Even_new_s dret1 = square_norm(Even_new_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_new_s, g_trafo); dret2 = square_norm(Even_new_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Even_new_c dret1 = square_norm(Even_new_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_even(Even_new_c, g_trafo); dret2 = square_norm(Even_new_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Odd_new_s dret1 = square_norm(Odd_new_s, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_new_s, g_trafo); dret2 = square_norm(Odd_new_s, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); // Odd_new_c dret1 = square_norm(Odd_new_c, VOLUME/2 , 1); apply_inv_gtrafo_spinor_odd(Odd_new_c, g_trafo); dret2 = square_norm(Odd_new_c, VOLUME/2, 1); if (g_proc_id == 0) printf("\tsquare norm before gauge fixing: %.16e\n", dret1); if (g_proc_id == 0) printf("\tsquare norm after gauge fixing: %.16e\n", dret2); finalize_temporalgauge(); # ifdef MPI xchange_gauge(g_gauge_field); # endif } # endif #endif return(iter); }
int main(int argc, char *argv[]) { FILE *parameterfile = NULL; int j, i, ix = 0, isample = 0, op_id = 0; char datafilename[206]; char parameterfilename[206]; char conf_filename[50]; char * input_filename = NULL; char * filename = NULL; double plaquette_energy; struct stout_parameters params_smear; spinor **s, *s_; #ifdef _KOJAK_INST #pragma pomp inst init #pragma pomp inst begin(main) #endif #if (defined SSE || defined SSE2 || SSE3) signal(SIGILL, &catch_ill_inst); #endif DUM_DERI = 8; DUM_MATRIX = DUM_DERI + 5; #if ((defined BGL && defined XLC) || defined _USE_TSPLITPAR) NO_OF_SPINORFIELDS = DUM_MATRIX + 3; #else NO_OF_SPINORFIELDS = DUM_MATRIX + 3; #endif verbose = 0; g_use_clover_flag = 0; #ifdef MPI # ifdef OMP int mpi_thread_provided; MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_thread_provided); # else MPI_Init(&argc, &argv); # endif MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id); #else g_proc_id = 0; #endif process_args(argc,argv,&input_filename,&filename); set_default_filenames(&input_filename, &filename); /* Read the input file */ if( (j = read_input(input_filename)) != 0) { fprintf(stderr, "Could not find input file: %s\nAborting...\n", input_filename); exit(-1); } #ifdef OMP init_openmp(); #endif /* this DBW2 stuff is not needed for the inversion ! */ if (g_dflgcr_flag == 1) { even_odd_flag = 0; } g_rgi_C1 = 0; if (Nsave == 0) { Nsave = 1; } if (g_running_phmc) { NO_OF_SPINORFIELDS = DUM_MATRIX + 8; } tmlqcd_mpi_init(argc, argv); g_dbw2rand = 0; /* starts the single and double precision random number */ /* generator */ start_ranlux(rlxd_level, random_seed); /* we need to make sure that we don't have even_odd_flag = 1 */ /* if any of the operators doesn't use it */ /* in this way even/odd can still be used by other operators */ for(j = 0; j < no_operators; j++) if(!operator_list[j].even_odd_flag) even_odd_flag = 0; #ifndef MPI g_dbw2rand = 0; #endif #ifdef _GAUGE_COPY j = init_gauge_field(VOLUMEPLUSRAND, 1); #else j = init_gauge_field(VOLUMEPLUSRAND, 0); #endif if (j != 0) { fprintf(stderr, "Not enough memory for gauge_fields! Aborting...\n"); exit(-1); } j = init_geometry_indices(VOLUMEPLUSRAND); if (j != 0) { fprintf(stderr, "Not enough memory for geometry indices! Aborting...\n"); exit(-1); } if (no_monomials > 0) { if (even_odd_flag) { j = init_monomials(VOLUMEPLUSRAND / 2, even_odd_flag); } else { j = init_monomials(VOLUMEPLUSRAND, even_odd_flag); } if (j != 0) { fprintf(stderr, "Not enough memory for monomial pseudo fermion fields! Aborting...\n"); exit(-1); } } if (even_odd_flag) { j = init_spinor_field(VOLUMEPLUSRAND / 2, NO_OF_SPINORFIELDS); } else { j = init_spinor_field(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS); } if (j != 0) { fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n"); exit(-1); } if (g_running_phmc) { j = init_chi_spinor_field(VOLUMEPLUSRAND / 2, 20); if (j != 0) { fprintf(stderr, "Not enough memory for PHMC Chi fields! Aborting...\n"); exit(-1); } } g_mu = g_mu1; if (g_cart_id == 0) { /*construct the filenames for the observables and the parameters*/ strncpy(datafilename, filename, 200); strcat(datafilename, ".data"); strncpy(parameterfilename, filename, 200); strcat(parameterfilename, ".para"); parameterfile = fopen(parameterfilename, "w"); write_first_messages(parameterfile, "invert", git_hash); fclose(parameterfile); } /* define the geometry */ geometry(); /* define the boundary conditions for the fermion fields */ boundary(g_kappa); phmc_invmaxev = 1.; init_operators(); /* list and initialize measurements*/ if(g_proc_id == 0) { printf("\n"); for(int j = 0; j < no_measurements; j++) { printf("# measurement id %d, type = %d\n", j, measurement_list[j].type); } } init_measurements(); /* this could be maybe moved to init_operators */ #ifdef _USE_HALFSPINOR j = init_dirac_halfspinor(); if (j != 0) { fprintf(stderr, "Not enough memory for halffield! Aborting...\n"); exit(-1); } if (g_sloppy_precision_flag == 1) { j = init_dirac_halfspinor32(); if (j != 0) { fprintf(stderr, "Not enough memory for 32-bit halffield! Aborting...\n"); exit(-1); } } # if (defined _PERSISTENT) if (even_odd_flag) init_xchange_halffield(); # endif #endif for (j = 0; j < Nmeas; j++) { sprintf(conf_filename, "%s.%.4d", gauge_input_filename, nstore); if (g_cart_id == 0) { printf("#\n# Trying to read gauge field from file %s in %s precision.\n", conf_filename, (gauge_precision_read_flag == 32 ? "single" : "double")); fflush(stdout); } if( (i = read_gauge_field(conf_filename,g_gauge_field)) !=0) { fprintf(stderr, "Error %d while reading gauge field from %s\n Aborting...\n", i, conf_filename); exit(-2); } if (g_cart_id == 0) { printf("# Finished reading gauge field.\n"); fflush(stdout); } #ifdef MPI xchange_gauge(g_gauge_field); #endif /*compute the energy of the gauge field*/ plaquette_energy = measure_plaquette( (const su3**) g_gauge_field); if (g_cart_id == 0) { printf("# The computed plaquette value is %e.\n", plaquette_energy / (6.*VOLUME*g_nproc)); fflush(stdout); } if (use_stout_flag == 1){ params_smear.rho = stout_rho; params_smear.iterations = stout_no_iter; /* if (stout_smear((su3_tuple*)(g_gauge_field[0]), ¶ms_smear, (su3_tuple*)(g_gauge_field[0])) != 0) */ /* exit(1) ; */ g_update_gauge_copy = 1; plaquette_energy = measure_plaquette( (const su3**) g_gauge_field); if (g_cart_id == 0) { printf("# The plaquette value after stouting is %e\n", plaquette_energy / (6.*VOLUME*g_nproc)); fflush(stdout); } } /* if any measurements are defined in the input file, do them here */ measurement * meas; for(int imeas = 0; imeas < no_measurements; imeas++){ meas = &measurement_list[imeas]; if (g_proc_id == 0) { fprintf(stdout, "#\n# Beginning online measurement.\n"); } meas->measurefunc(nstore, imeas, even_odd_flag); } if (reweighting_flag == 1) { reweighting_factor(reweighting_samples, nstore); } /* Compute minimal eigenvalues, if wanted */ if (compute_evs != 0) { eigenvalues(&no_eigenvalues, 5000, eigenvalue_precision, 0, compute_evs, nstore, even_odd_flag); } if (phmc_compute_evs != 0) { #ifdef MPI MPI_Finalize(); #endif return(0); } /* Compute the mode number or topological susceptibility using spectral projectors, if wanted*/ if(compute_modenumber != 0 || compute_topsus !=0){ s_ = calloc(no_sources_z2*VOLUMEPLUSRAND+1, sizeof(spinor)); s = calloc(no_sources_z2, sizeof(spinor*)); if(s_ == NULL) { printf("Not enough memory in %s: %d",__FILE__,__LINE__); exit(42); } if(s == NULL) { printf("Not enough memory in %s: %d",__FILE__,__LINE__); exit(42); } for(i = 0; i < no_sources_z2; i++) { #if (defined SSE3 || defined SSE2 || defined SSE) s[i] = (spinor*)(((unsigned long int)(s_)+ALIGN_BASE)&~ALIGN_BASE)+i*VOLUMEPLUSRAND; #else s[i] = s_+i*VOLUMEPLUSRAND; #endif random_spinor_field_lexic(s[i], reproduce_randomnumber_flag,RN_Z2); /* what is this here needed for?? */ /* spinor *aux_,*aux; */ /* #if ( defined SSE || defined SSE2 || defined SSE3 ) */ /* aux_=calloc(VOLUMEPLUSRAND+1, sizeof(spinor)); */ /* aux = (spinor *)(((unsigned long int)(aux_)+ALIGN_BASE)&~ALIGN_BASE); */ /* #else */ /* aux_=calloc(VOLUMEPLUSRAND, sizeof(spinor)); */ /* aux = aux_; */ /* #endif */ if(g_proc_id == 0) { printf("source %d \n", i); } if(compute_modenumber != 0){ mode_number(s[i], mstarsq); } if(compute_topsus !=0) { top_sus(s[i], mstarsq); } } free(s); free(s_); } /* move to operators as well */ if (g_dflgcr_flag == 1) { /* set up deflation blocks */ init_blocks(nblocks_t, nblocks_x, nblocks_y, nblocks_z); /* the can stay here for now, but later we probably need */ /* something like init_dfl_solver called somewhere else */ /* create set of approximate lowest eigenvectors ("global deflation subspace") */ /* g_mu = 0.; */ /* boundary(0.125); */ generate_dfl_subspace(g_N_s, VOLUME, reproduce_randomnumber_flag); /* boundary(g_kappa); */ /* g_mu = g_mu1; */ /* Compute little Dirac operators */ /* alt_block_compute_little_D(); */ if (g_debug_level > 0) { check_projectors(reproduce_randomnumber_flag); check_local_D(reproduce_randomnumber_flag); } if (g_debug_level > 1) { check_little_D_inversion(reproduce_randomnumber_flag); } } if(SourceInfo.type == 1) { index_start = 0; index_end = 1; } g_precWS=NULL; if(use_preconditioning == 1){ /* todo load fftw wisdom */ #if (defined HAVE_FFTW ) && !( defined MPI) loadFFTWWisdom(g_spinor_field[0],g_spinor_field[1],T,LX); #else use_preconditioning=0; #endif } if (g_cart_id == 0) { fprintf(stdout, "#\n"); /*Indicate starting of the operator part*/ } for(op_id = 0; op_id < no_operators; op_id++) { boundary(operator_list[op_id].kappa); g_kappa = operator_list[op_id].kappa; g_mu = 0.; if(use_preconditioning==1 && PRECWSOPERATORSELECT[operator_list[op_id].solver]!=PRECWS_NO ){ printf("# Using preconditioning with treelevel preconditioning operator: %s \n", precWSOpToString(PRECWSOPERATORSELECT[operator_list[op_id].solver])); /* initial preconditioning workspace */ operator_list[op_id].precWS=(spinorPrecWS*)malloc(sizeof(spinorPrecWS)); spinorPrecWS_Init(operator_list[op_id].precWS, operator_list[op_id].kappa, operator_list[op_id].mu/2./operator_list[op_id].kappa, -(0.5/operator_list[op_id].kappa-4.), PRECWSOPERATORSELECT[operator_list[op_id].solver]); g_precWS = operator_list[op_id].precWS; if(PRECWSOPERATORSELECT[operator_list[op_id].solver] == PRECWS_D_DAGGER_D) { fitPrecParams(op_id); } } for(isample = 0; isample < no_samples; isample++) { for (ix = index_start; ix < index_end; ix++) { if (g_cart_id == 0) { fprintf(stdout, "#\n"); /*Indicate starting of new index*/ } /* we use g_spinor_field[0-7] for sources and props for the moment */ /* 0-3 in case of 1 flavour */ /* 0-7 in case of 2 flavours */ prepare_source(nstore, isample, ix, op_id, read_source_flag, source_location); //randmize initial guess for eigcg if needed-----experimental if( (operator_list[op_id].solver == INCREIGCG) && (operator_list[op_id].solver_params.eigcg_rand_guess_opt) ){ //randomize the initial guess gaussian_volume_source( operator_list[op_id].prop0, operator_list[op_id].prop1,isample,ix,0); //need to check this } operator_list[op_id].inverter(op_id, index_start, 1); } } if(use_preconditioning==1 && operator_list[op_id].precWS!=NULL ){ /* free preconditioning workspace */ spinorPrecWS_Free(operator_list[op_id].precWS); free(operator_list[op_id].precWS); } if(operator_list[op_id].type == OVERLAP){ free_Dov_WS(); } } nstore += Nsave; } #ifdef OMP free_omp_accumulators(); #endif free_blocks(); free_dfl_subspace(); free_gauge_field(); free_geometry_indices(); free_spinor_field(); free_moment_field(); free_chi_spinor_field(); free(filename); free(input_filename); #ifdef MPI MPI_Barrier(MPI_COMM_WORLD); MPI_Finalize(); #endif return(0); #ifdef _KOJAK_INST #pragma pomp inst end(main) #endif }