int main(int argc,char *argv[])
{
int j,j_max,k,k_max = 2;
paramsXlfInfo *xlfInfo;
int ix, n, *nn,*mm,i;
double delta, deltamax;
spinor rsp;
int status = 0;
#ifdef MPI
DUM_DERI = 6;
DUM_SOLVER = DUM_DERI+2;
DUM_MATRIX = DUM_SOLVER+6;
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
MPI_Init(&argc, &argv);
#endif
g_rgi_C1 = 1.;
/* Read the input file */
read_input("hopping_test.input");
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 _INDEX_INDEP_GEOM
printf("# the code was compiled with index independent geometry\n");
#endif
#ifdef MPI
# ifdef _NON_BLOCKING
printf("# the code was compiled for non-blocking MPI calls (spinor and gauge)\n");
# endif
# ifdef _USE_TSPLITPAR
printf("# the code was compiled with tsplit parallelization\n");
# endif
#endif
printf("\n");
fflush(stdout);
}
#ifdef _GAUGE_COPY
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if(even_odd_flag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, 2*k_max+1);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, 2*k_max);
}
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
j = init_moment_field(VOLUME, VOLUMEPLUSRAND);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for moment fields! Aborting...\n");
exit(0);
}
if(g_proc_id == 0) {
fprintf(stdout,"The number of processes is %d \n",g_nproc);
printf("# The lattice size is %d x %d x %d x %d\n",
(int)(T*g_nproc_t), (int)(LX*g_nproc_x), (int)(LY*g_nproc_y), (int)(g_nproc_z*LZ));
printf("# The local lattice size is %d x %d x %d x %d\n",
(int)(T), (int)(LX), (int)(LY),(int) LZ);
if(even_odd_flag) {
printf("# testinging the even/odd preconditioned Dirac operator\n");
}
else {
printf("# testinging the standard Dirac operator\n");
}
fflush(stdout);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for halfspinor fields! Aborting...\n");
exit(0);
}
if(g_sloppy_precision_flag == 1) {
g_sloppy_precision = 1;
j = init_dirac_halfspinor32();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for 32-Bit halfspinor fields! Aborting...\n");
exit(0);
}
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
status = check_geometry();
if (status != 0) {
fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n");
exit(1);
}
#if (defined MPI && !(defined _USE_SHMEM))
check_xchange();
#endif
start_ranlux(1, 123456);
xlfInfo = construct_paramsXlfInfo(0.5, 0);
random_gauge_field(reproduce_randomnumber_flag);
if ( startoption == 2 ) { /* restart */
write_gauge_field(gauge_input_filename,gauge_precision_write_flag,xlfInfo);
} else if ( startoption == 0 ) { /* cold */
unit_g_gauge_field();
} else if (startoption == 3 ) { /* continue */
read_gauge_field(gauge_input_filename);
} else if ( startoption == 1 ) { /* hot */
}
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge();
#endif
#ifdef _GAUGE_COPY
update_backward_gauge();
#endif
if(even_odd_flag) {
/*initialize the pseudo-fermion fields*/
j_max=1;
for (k = 0; k < k_max; k++) {
random_spinor_field(g_spinor_field[k], VOLUME/2, 0);
}
if (read_source_flag == 2) { /* save */
/* even first, odd second */
write_spinorfield_cm_single(g_spinor_field[0],g_spinor_field[1],SourceInfo.basename);
} else if (read_source_flag == 1) { /* yes */
/* even first, odd second */
read_spinorfield_cm_single(g_spinor_field[0],g_spinor_field[1],SourceInfo.basename,-1,0);
# if (!defined MPI)
if (write_cp_flag == 1) {
strcat(SourceInfo.basename,".2");
read_spinorfield_cm_single(g_spinor_field[2],g_spinor_field[3],SourceInfo.basename,-1,0);
nn=(int*)calloc(VOLUME,sizeof(int));
if((void*)nn == NULL) return(100);
mm=(int*)calloc(VOLUME,sizeof(int));
if((void*)mm == NULL) return(100);
n=0;
deltamax=0.0;
for(ix=0;ix<VOLUME/2;ix++){
(rsp.s0).c0.re = (g_spinor_field[2][ix].s0).c0.re - (g_spinor_field[0][ix].s0).c0.re;
(rsp.s0).c0.im = (g_spinor_field[2][ix].s0).c0.im - (g_spinor_field[0][ix].s0).c0.im;
(rsp.s0).c1.re = (g_spinor_field[2][ix].s0).c1.re - (g_spinor_field[0][ix].s0).c1.re;
(rsp.s0).c1.im = (g_spinor_field[2][ix].s0).c1.im - (g_spinor_field[0][ix].s0).c1.im;
(rsp.s0).c2.re = (g_spinor_field[2][ix].s0).c2.re - (g_spinor_field[0][ix].s0).c2.re;
(rsp.s0).c2.im = (g_spinor_field[2][ix].s0).c2.im - (g_spinor_field[0][ix].s0).c2.im;
(rsp.s1).c0.re = (g_spinor_field[2][ix].s1).c0.re - (g_spinor_field[0][ix].s1).c0.re;
(rsp.s1).c0.im = (g_spinor_field[2][ix].s1).c0.im - (g_spinor_field[0][ix].s1).c0.im;
(rsp.s1).c1.re = (g_spinor_field[2][ix].s1).c1.re - (g_spinor_field[0][ix].s1).c1.re;
(rsp.s1).c1.im = (g_spinor_field[2][ix].s1).c1.im - (g_spinor_field[0][ix].s1).c1.im;
(rsp.s1).c2.re = (g_spinor_field[2][ix].s1).c2.re - (g_spinor_field[0][ix].s1).c2.re;
(rsp.s1).c2.im = (g_spinor_field[2][ix].s1).c2.im - (g_spinor_field[0][ix].s1).c2.im;
(rsp.s2).c0.re = (g_spinor_field[2][ix].s2).c0.re - (g_spinor_field[0][ix].s2).c0.re;
(rsp.s2).c0.im = (g_spinor_field[2][ix].s2).c0.im - (g_spinor_field[0][ix].s2).c0.im;
(rsp.s2).c1.re = (g_spinor_field[2][ix].s2).c1.re - (g_spinor_field[0][ix].s2).c1.re;
(rsp.s2).c1.im = (g_spinor_field[2][ix].s2).c1.im - (g_spinor_field[0][ix].s2).c1.im;
(rsp.s2).c2.re = (g_spinor_field[2][ix].s2).c2.re - (g_spinor_field[0][ix].s2).c2.re;
(rsp.s2).c2.im = (g_spinor_field[2][ix].s2).c2.im - (g_spinor_field[0][ix].s2).c2.im;
(rsp.s3).c0.re = (g_spinor_field[2][ix].s3).c0.re - (g_spinor_field[0][ix].s3).c0.re;
(rsp.s3).c0.im = (g_spinor_field[2][ix].s3).c0.im - (g_spinor_field[0][ix].s3).c0.im;
(rsp.s3).c1.re = (g_spinor_field[2][ix].s3).c1.re - (g_spinor_field[0][ix].s3).c1.re;
(rsp.s3).c1.im = (g_spinor_field[2][ix].s3).c1.im - (g_spinor_field[0][ix].s3).c1.im;
(rsp.s3).c2.re = (g_spinor_field[2][ix].s3).c2.re - (g_spinor_field[0][ix].s3).c2.re;
(rsp.s3).c2.im = (g_spinor_field[2][ix].s3).c2.im - (g_spinor_field[0][ix].s3).c2.im;
_spinor_norm_sq(delta,rsp);
if (delta > 1.0e-12) {
nn[n] = g_eo2lexic[ix];
mm[n]=ix;
n++;
}
if(delta>deltamax) deltamax=delta;
}
if (n>0){
printf("mismatch in even spincolorfield in %d points:\n",n);
for(i=0; i< MIN(n,1000); i++){
printf("%d,(%d,%d,%d,%d):%f vs. %f\n",nn[i],g_coord[nn[i]][0],g_coord[nn[i]][1],g_coord[nn[i]][2],g_coord[nn[i]][3],(g_spinor_field[2][mm[i]].s0).c0.re, (g_spinor_field[0][mm[i]].s0).c0.re);fflush(stdout);
}
}
n = 0;
for(ix=0;ix<VOLUME/2;ix++){
(rsp.s0).c0.re = (g_spinor_field[3][ix].s0).c0.re - (g_spinor_field[1][ix].s0).c0.re;
(rsp.s0).c0.im = (g_spinor_field[3][ix].s0).c0.im - (g_spinor_field[1][ix].s0).c0.im;
(rsp.s0).c1.re = (g_spinor_field[3][ix].s0).c1.re - (g_spinor_field[1][ix].s0).c1.re;
(rsp.s0).c1.im = (g_spinor_field[3][ix].s0).c1.im - (g_spinor_field[1][ix].s0).c1.im;
(rsp.s0).c2.re = (g_spinor_field[3][ix].s0).c2.re - (g_spinor_field[1][ix].s0).c2.re;
(rsp.s0).c2.im = (g_spinor_field[3][ix].s0).c2.im - (g_spinor_field[1][ix].s0).c2.im;
(rsp.s1).c0.re = (g_spinor_field[3][ix].s1).c0.re - (g_spinor_field[1][ix].s1).c0.re;
(rsp.s1).c0.im = (g_spinor_field[3][ix].s1).c0.im - (g_spinor_field[1][ix].s1).c0.im;
(rsp.s1).c1.re = (g_spinor_field[3][ix].s1).c1.re - (g_spinor_field[1][ix].s1).c1.re;
(rsp.s1).c1.im = (g_spinor_field[3][ix].s1).c1.im - (g_spinor_field[1][ix].s1).c1.im;
(rsp.s1).c2.re = (g_spinor_field[3][ix].s1).c2.re - (g_spinor_field[1][ix].s1).c2.re;
(rsp.s1).c2.im = (g_spinor_field[3][ix].s1).c2.im - (g_spinor_field[1][ix].s1).c2.im;
(rsp.s2).c0.re = (g_spinor_field[3][ix].s2).c0.re - (g_spinor_field[1][ix].s2).c0.re;
(rsp.s2).c0.im = (g_spinor_field[3][ix].s2).c0.im - (g_spinor_field[1][ix].s2).c0.im;
(rsp.s2).c1.re = (g_spinor_field[3][ix].s2).c1.re - (g_spinor_field[1][ix].s2).c1.re;
(rsp.s2).c1.im = (g_spinor_field[3][ix].s2).c1.im - (g_spinor_field[1][ix].s2).c1.im;
(rsp.s2).c2.re = (g_spinor_field[3][ix].s2).c2.re - (g_spinor_field[1][ix].s2).c2.re;
(rsp.s2).c2.im = (g_spinor_field[3][ix].s2).c2.im - (g_spinor_field[1][ix].s2).c2.im;
(rsp.s3).c0.re = (g_spinor_field[3][ix].s3).c0.re - (g_spinor_field[1][ix].s3).c0.re;
(rsp.s3).c0.im = (g_spinor_field[3][ix].s3).c0.im - (g_spinor_field[1][ix].s3).c0.im;
(rsp.s3).c1.re = (g_spinor_field[3][ix].s3).c1.re - (g_spinor_field[1][ix].s3).c1.re;
(rsp.s3).c1.im = (g_spinor_field[3][ix].s3).c1.im - (g_spinor_field[1][ix].s3).c1.im;
(rsp.s3).c2.re = (g_spinor_field[3][ix].s3).c2.re - (g_spinor_field[1][ix].s3).c2.re;
(rsp.s3).c2.im = (g_spinor_field[3][ix].s3).c2.im - (g_spinor_field[1][ix].s3).c2.im;
_spinor_norm_sq(delta,rsp);
if (delta > 1.0e-12) {
nn[n]=g_eo2lexic[ix+(VOLUME+RAND)/2];
mm[n]=ix;
n++;
}
if(delta>deltamax) deltamax=delta;
}
if (n>0){
printf("mismatch in odd spincolorfield in %d points:\n",n);
for(i=0; i< MIN(n,1000); i++){
printf("%d,(%d,%d,%d,%d):%f vs. %f\n",nn[i],g_coord[nn[i]][0],g_coord[nn[i]][1],g_coord[nn[i]][2],g_coord[nn[i]][3],(g_spinor_field[3][mm[i]].s0).c0.re, (g_spinor_field[1][mm[i]].s0).c0.re);fflush(stdout);
}
}
printf("max delta=%e",deltamax);fflush(stdout);
}
# endif
}
if (read_source_flag > 0 && write_cp_flag == 0) { /* read-source yes or nobutsave; checkpoint no */
/* first spinorial arg is output, the second is input */
Hopping_Matrix(1, g_spinor_field[1], g_spinor_field[0]); /*ieo=1 M_{eo}*/
Hopping_Matrix(0, g_spinor_field[0], g_spinor_field[1]); /*ieo=0 M_{oe}*/
strcat(SourceInfo.basename,".out");
write_spinorfield_cm_single(g_spinor_field[0],g_spinor_field[1],SourceInfo.basename);
printf("Check-field printed. Exiting...\n");
fflush(stdout);
}
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
}
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
return(0);
}
int main(int argc,char *argv[]) {
double plaquette_energy;
paramsXlfInfo *xlfInfo;
#ifdef MPI
MPI_Init(&argc, &argv);
#endif
g_rgi_C1 = 1.;
/* Read the input file */
read_input("benchmark.input");
tmlqcd_mpi_init(argc, argv);
#ifdef _GAUGE_COPY
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if(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);
printf("# Testing IO routines for gauge-fields\n");
fflush(stdout);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
/* generate a random gauge field */
start_ranlux(1, 123456);
random_gauge_field(reproduce_randomnumber_flag, g_gauge_field);
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge(g_gauge_field);
#endif
plaquette_energy = measure_gauge_action(g_gauge_field) / (6.*VOLUME*g_nproc);
if(g_proc_id == 0) {
printf("# the first plaquette value is %e\n", plaquette_energy);
printf("# writing with lime first to conf.lime\n");
}
/* write with lime first */
xlfInfo = construct_paramsXlfInfo(plaquette_energy, 0);
write_lime_gauge_field( "conf.lime", 64, xlfInfo);
#ifdef HAVE_LIBLEMON
if(g_proc_id == 0) {
printf("Now we do write with lemon to conf.lemon...\n");
}
write_lemon_gauge_field_parallel( "conf.lemon", 64, xlfInfo);
if(g_proc_id == 0) {
printf("# now we read with lemon from conf.lime\n");
}
read_lemon_gauge_field_parallel("conf.lime", NULL, NULL, NULL);
plaquette_energy = measure_gauge_action(g_gauge_field) / (6.*VOLUME*g_nproc);
if(g_proc_id == 0) {
printf("# the plaquette value after lemon read of conf.lime is %e\n", plaquette_energy);
}
if(g_proc_id == 0) {
printf("# now we read with lemon from conf.lemon\n");
}
read_lemon_gauge_field_parallel("conf.lemon", NULL, NULL, NULL);
plaquette_energy = measure_gauge_action(g_gauge_field) / (6.*VOLUME*g_nproc);
if(g_proc_id == 0) {
printf("# the plaquette value after lemon read of conf.lemon is %e\n", plaquette_energy);
}
if(g_proc_id == 0) {
printf("# now we read with lime from conf.lemon\n");
}
read_lime_gauge_field("conf.lemon");
plaquette_energy = measure_gauge_action(g_gauge_field) / (6.*VOLUME*g_nproc);
if(g_proc_id == 0) {
printf("# the plaquette value after lime read of conf.lemon is %e\n", plaquette_energy);
}
free(xlfInfo);
if(g_proc_id==0) {
printf("done ...\n");
}
#endif
if(g_proc_id == 0) {
printf("# now we read with lime from conf.lime\n");
}
read_lime_gauge_field("conf.lime", NULL, NULL, NULL);
plaquette_energy = measure_gauge_action(g_gauge_field) / (6.*VOLUME*g_nproc);
if(g_proc_id == 0) {
printf("# the plaquette value after lime read of conf.lime is %e\n", plaquette_energy);
}
#ifdef MPI
MPI_Finalize();
#endif
free_gauge_field();
free_geometry_indices();
return(0);
}
int main(int argc,char *argv[])
{
#ifdef _USE_HALFSPINOR
#undef _USE_HALFSPINOR
printf("# WARNING: USE_HALFSPINOR will be ignored (not supported here).\n");
#endif
if(even_odd_flag)
{
even_odd_flag=0;
printf("# WARNING: even_odd_flag will be ignored (not supported here).\n");
}
int j,j_max,k,k_max = 1;
#ifdef HAVE_LIBLEMON
paramsXlfInfo *xlfInfo;
#endif
int status = 0;
static double t1,t2,dt,sdt,dts,qdt,sqdt;
double antioptaway=0.0;
#ifdef MPI
static double dt2;
DUM_DERI = 6;
DUM_SOLVER = DUM_DERI+2;
DUM_MATRIX = DUM_SOLVER+6;
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
# ifdef OMP
int mpi_thread_provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_thread_provided);
# else
MPI_Init(&argc, &argv);
# endif
MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id);
#else
g_proc_id = 0;
#endif
g_rgi_C1 = 1.;
/* Read the input file */
if((status = read_input("test_Dslash.input")) != 0) {
fprintf(stderr, "Could not find input file: test_Dslash.input\nAborting...\n");
exit(-1);
}
#ifdef OMP
init_openmp();
#endif
tmlqcd_mpi_init(argc, argv);
if(g_proc_id==0) {
#ifdef SSE
printf("# The code was compiled with SSE instructions\n");
#endif
#ifdef SSE2
printf("# The code was compiled with SSE2 instructions\n");
#endif
#ifdef SSE3
printf("# The code was compiled with SSE3 instructions\n");
#endif
#ifdef P4
printf("# The code was compiled for Pentium4\n");
#endif
#ifdef OPTERON
printf("# The code was compiled for AMD Opteron\n");
#endif
#ifdef _GAUGE_COPY
printf("# The code was compiled with -D_GAUGE_COPY\n");
#endif
#ifdef BGL
printf("# The code was compiled for Blue Gene/L\n");
#endif
#ifdef BGP
printf("# The code was compiled for Blue Gene/P\n");
#endif
#ifdef _USE_HALFSPINOR
printf("# The code was compiled with -D_USE_HALFSPINOR\n");
#endif
#ifdef _USE_SHMEM
printf("# The code was compiled with -D_USE_SHMEM\n");
# ifdef _PERSISTENT
printf("# The code was compiled for persistent MPI calls (halfspinor only)\n");
# endif
#endif
#ifdef MPI
# ifdef _NON_BLOCKING
printf("# The code was compiled for non-blocking MPI calls (spinor and gauge)\n");
# endif
#endif
printf("\n");
fflush(stdout);
}
#ifdef _GAUGE_COPY
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if(even_odd_flag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, 2*k_max+1);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, 2*k_max);
}
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
j = init_moment_field(VOLUME, VOLUMEPLUSRAND + g_dbw2rand);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for moment fields! Aborting...\n");
exit(0);
}
if(g_proc_id == 0) {
fprintf(stdout,"# The number of processes is %d \n",g_nproc);
printf("# The lattice size is %d x %d x %d x %d\n",
(int)(T*g_nproc_t), (int)(LX*g_nproc_x), (int)(LY*g_nproc_y), (int)(g_nproc_z*LZ));
printf("# The local lattice size is %d x %d x %d x %d\n",
(int)(T), (int)(LX), (int)(LY),(int) LZ);
// if(even_odd_flag) {
// printf("# benchmarking the even/odd preconditioned Dirac operator\n");
// }
// else {
// printf("# benchmarking the standard Dirac operator\n");
// }
fflush(stdout);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for halfspinor fields! Aborting...\n");
exit(0);
}
if(g_sloppy_precision_flag == 1) {
g_sloppy_precision = 1;
j = init_dirac_halfspinor32();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for 32-Bit halfspinor fields! Aborting...\n");
exit(0);
}
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
status = check_geometry();
if (status != 0) {
fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n");
exit(1);
}
#if (defined MPI && !(defined _USE_SHMEM))
check_xchange();
#endif
start_ranlux(1, 123456);
random_gauge_field(reproduce_randomnumber_flag, g_gauge_field);
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge(g_gauge_field);
#endif
/* the non even/odd case now */
/*initialize the pseudo-fermion fields*/
j_max=1;
sdt=0.;
for (k=0;k<k_max;k++) {
random_spinor_field_lexic(g_spinor_field[k], reproduce_randomnumber_flag, RN_GAUSS);
}
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
/* here the actual Dslash */
D_psi(g_spinor_field[0], g_spinor_field[1]);
t2 = gettime();
dt=t2-t1;
#ifdef MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
if(g_proc_id==0) {
printf("# Time for Dslash %e sec.\n", sdt);
printf("\n");
fflush(stdout);
}
#ifdef HAVE_LIBLEMON
if(g_proc_id==0) {
printf("# Performing parallel IO test ...\n");
}
xlfInfo = construct_paramsXlfInfo(0.5, 0);
write_gauge_field( "conf.test", 64, xlfInfo);
free(xlfInfo);
if(g_proc_id==0) {
printf("# done ...\n");
}
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
#endif
return(0);
}
int main(int argc,char *argv[]) {
FILE *parameterfile=NULL, *countfile=NULL;
char *filename = NULL;
char datafilename[50];
char parameterfilename[50];
char gauge_filename[50];
char nstore_filename[50];
char tmp_filename[50];
char *input_filename = NULL;
int status = 0, accept = 0;
int j,ix,mu, trajectory_counter=1;
struct timeval t1;
/* Energy corresponding to the Gauge part */
double plaquette_energy = 0., rectangle_energy = 0.;
/* Acceptance rate */
int Rate=0;
/* Do we want to perform reversibility checks */
/* See also return_check_flag in read_input.h */
int return_check = 0;
/* For getopt */
int c;
paramsXlfInfo *xlfInfo;
/* For online measurements */
measurement * meas;
int imeas;
#ifdef _KOJAK_INST
#pragma pomp inst init
#pragma pomp inst begin(main)
#endif
#if (defined SSE || defined SSE2 || SSE3)
signal(SIGILL,&catch_ill_inst);
#endif
strcpy(gauge_filename,"conf.save");
strcpy(nstore_filename,".nstore_counter");
strcpy(tmp_filename, ".conf.tmp");
verbose = 1;
g_use_clover_flag = 0;
#ifdef MPI
# ifdef OMP
int mpi_thread_provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_thread_provided);
# else
MPI_Init(&argc, &argv);
# endif
MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id);
#else
g_proc_id = 0;
#endif
while ((c = getopt(argc, argv, "h?vVf:o:")) != -1) {
switch (c) {
case 'f':
input_filename = calloc(200, sizeof(char));
strcpy(input_filename,optarg);
break;
case 'o':
filename = calloc(200, sizeof(char));
strcpy(filename,optarg);
break;
case 'v':
verbose = 1;
break;
case 'V':
fprintf(stdout,"%s %s\n",PACKAGE_STRING,git_hash);
exit(0);
break;
case 'h':
case '?':
default:
usage();
break;
}
}
if(input_filename == NULL){
input_filename = "hmc.input";
}
if(filename == NULL){
filename = "output";
}
/* Read the input file */
if( (status = read_input(input_filename)) != 0) {
fprintf(stderr, "Could not find input file: %s\nAborting...\n", input_filename);
exit(-1);
}
/* set number of omp threads to be used */
#ifdef OMP
if(omp_num_threads > 0)
{
omp_set_num_threads(omp_num_threads);
}
else {
if( g_proc_id == 0 )
printf("# No value provided for OmpNumThreads, running in single-threaded mode!\n");
omp_num_threads = 1;
omp_set_num_threads(omp_num_threads);
}
init_omp_accumulators(omp_num_threads);
#endif
DUM_DERI = 4;
DUM_SOLVER = DUM_DERI+1;
DUM_MATRIX = DUM_SOLVER+6;
if(g_running_phmc) {
NO_OF_SPINORFIELDS = DUM_MATRIX+8;
}
else {
NO_OF_SPINORFIELDS = DUM_MATRIX+6;
}
DUM_BI_DERI = 6;
DUM_BI_SOLVER = DUM_BI_DERI+7;
DUM_BI_MATRIX = DUM_BI_SOLVER+6;
NO_OF_BISPINORFIELDS = DUM_BI_MATRIX+6;
tmlqcd_mpi_init(argc, argv);
if(nstore == -1) {
countfile = fopen(nstore_filename, "r");
if(countfile != NULL) {
j = fscanf(countfile, "%d %d %s\n", &nstore, &trajectory_counter, gauge_input_filename);
if(j < 1) nstore = 0;
if(j < 2) trajectory_counter = 0;
fclose(countfile);
}
else {
nstore = 0;
trajectory_counter = 0;
}
}
#ifndef MPI
g_dbw2rand = 0;
#endif
g_mu = g_mu1;
#ifdef _GAUGE_COPY
status = init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
status = init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
if (status != 0) {
fprintf(stderr, "Not enough memory for gauge_fields! Aborting...\n");
exit(0);
}
j = init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if (j != 0) {
fprintf(stderr, "Not enough memory for geometry_indices! Aborting...\n");
exit(0);
}
if(even_odd_flag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, NO_OF_SPINORFIELDS);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, NO_OF_SPINORFIELDS);
}
if (j != 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
if(even_odd_flag) {
j = init_csg_field(VOLUMEPLUSRAND/2);
}
else {
j = init_csg_field(VOLUMEPLUSRAND);
}
if (j != 0) {
fprintf(stderr, "Not enough memory for csg fields! Aborting...\n");
exit(0);
}
j = init_moment_field(VOLUME, VOLUMEPLUSRAND + g_dbw2rand);
if (j != 0) {
fprintf(stderr, "Not enough memory for moment fields! Aborting...\n");
exit(0);
}
if(g_running_phmc) {
j = init_bispinor_field(VOLUME/2, NO_OF_BISPINORFIELDS);
if (j!= 0) {
fprintf(stderr, "Not enough memory for bi-spinor fields! Aborting...\n");
exit(0);
}
}
/* list and initialize measurements*/
if(g_proc_id == 0) {
printf("\n");
for(j = 0; j < no_measurements; j++) {
printf("# measurement id %d, type = %d: Frequency %d\n", j, measurement_list[j].type, measurement_list[j].freq);
}
}
init_measurements();
/*construct the filenames for the observables and the parameters*/
strcpy(datafilename,filename);
strcat(datafilename,".data");
strcpy(parameterfilename,filename);
strcat(parameterfilename,".para");
if(g_proc_id == 0){
parameterfile = fopen(parameterfilename, "a");
write_first_messages(parameterfile, "hmc", git_hash);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
status = check_geometry();
if (status != 0) {
fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n");
exit(1);
}
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if (j!= 0) {
fprintf(stderr, "Not enough memory for halffield! Aborting...\n");
exit(-1);
}
if(g_sloppy_precision_flag == 1) {
init_dirac_halfspinor32();
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
/* Initialise random number generator */
start_ranlux(rlxd_level, random_seed^nstore );
/* Set up the gauge field */
/* continue and restart */
if(startoption==3 || startoption == 2) {
if(g_proc_id == 0) {
printf("# Trying to read gauge field from file %s in %s precision.\n",
gauge_input_filename, (gauge_precision_read_flag == 32 ? "single" : "double"));
fflush(stdout);
}
if( (status = read_gauge_field(gauge_input_filename)) != 0) {
fprintf(stderr, "Error %d while reading gauge field from %s\nAborting...\n", status, gauge_input_filename);
exit(-2);
}
if (g_proc_id == 0){
printf("# Finished reading gauge field.\n");
fflush(stdout);
}
}
else if (startoption == 1) {
/* hot */
random_gauge_field(reproduce_randomnumber_flag, g_gauge_field);
}
else if(startoption == 0) {
/* cold */
unit_g_gauge_field();
}
/*For parallelization: exchange the gaugefield */
#ifdef MPI
xchange_gauge(g_gauge_field);
#endif
if(even_odd_flag) {
j = init_monomials(VOLUMEPLUSRAND/2, even_odd_flag);
}
else {
j = init_monomials(VOLUMEPLUSRAND, even_odd_flag);
}
if (j != 0) {
fprintf(stderr, "Not enough memory for monomial pseudo fermion fields! Aborting...\n");
exit(0);
}
init_integrator();
if(g_proc_id == 0) {
for(j = 0; j < no_monomials; j++) {
printf("# monomial id %d type = %d timescale %d\n", j, monomial_list[j].type, monomial_list[j].timescale);
}
}
plaquette_energy = measure_gauge_action( (const su3**) g_gauge_field);
if(g_rgi_C1 > 0. || g_rgi_C1 < 0.) {
rectangle_energy = measure_rectangles( (const su3**) g_gauge_field);
if(g_proc_id == 0){
fprintf(parameterfile,"# Computed rectangle value: %14.12f.\n",rectangle_energy/(12.*VOLUME*g_nproc));
}
}
//eneg = g_rgi_C0 * plaquette_energy + g_rgi_C1 * rectangle_energy;
if(g_proc_id == 0) {
fprintf(parameterfile,"# Computed plaquette value: %14.12f.\n", plaquette_energy/(6.*VOLUME*g_nproc));
printf("# Computed plaquette value: %14.12f.\n", plaquette_energy/(6.*VOLUME*g_nproc));
fclose(parameterfile);
}
/* set ddummy to zero */
for(ix = 0; ix < VOLUMEPLUSRAND; ix++){
for(mu=0; mu<4; mu++){
ddummy[ix][mu].d1=0.;
ddummy[ix][mu].d2=0.;
ddummy[ix][mu].d3=0.;
ddummy[ix][mu].d4=0.;
ddummy[ix][mu].d5=0.;
ddummy[ix][mu].d6=0.;
ddummy[ix][mu].d7=0.;
ddummy[ix][mu].d8=0.;
}
}
if(g_proc_id == 0) {
gettimeofday(&t1,NULL);
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "!!! Timestamp %ld, Nsave = %d, g_mu = %e, g_mu1 = %e, g_mu_2 = %e, g_mu3 = %e, beta = %f, kappa = %f, C1 = %f, ",
t1.tv_sec, Nsave, g_mu, g_mu1, g_mu2, g_mu3, g_beta, g_kappa, g_rgi_C1);
for(j = 0; j < Integrator.no_timescales; j++) {
fprintf(countfile, "n_int[%d] = %d ", j, Integrator.no_mnls_per_ts[j]);
}
fprintf(countfile, "\n");
fclose(countfile);
}
/* Loop for measurements */
for(j = 0; j < Nmeas; j++) {
if(g_proc_id == 0) {
printf("#\n# Starting trajectory no %d\n", trajectory_counter);
}
return_check = return_check_flag && (trajectory_counter%return_check_interval == 0);
accept = update_tm(&plaquette_energy, &rectangle_energy, datafilename,
return_check, Ntherm<trajectory_counter, trajectory_counter);
Rate += accept;
/* Save gauge configuration all Nsave times */
if((Nsave !=0) && (trajectory_counter%Nsave == 0) && (trajectory_counter!=0)) {
sprintf(gauge_filename,"conf.%.4d", nstore);
if(g_proc_id == 0) {
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "%.4d, measurement %d of %d, Nsave = %d, Plaquette = %e, trajectory nr = %d\n",
nstore, j, Nmeas, Nsave, plaquette_energy/(6.*VOLUME*g_nproc),
trajectory_counter);
fclose(countfile);
}
nstore ++;
}
else {
sprintf(gauge_filename,"conf.save");
}
if(((Nsave !=0) && (trajectory_counter%Nsave == 0) && (trajectory_counter!=0)) || (write_cp_flag == 1) || (j >= (Nmeas - 1))) {
/* If a reversibility check was performed this trajectory, and the trajectory was accepted,
* then the configuration is currently stored in .conf.tmp, written out by update_tm.
* In that case also a readback was performed, so no need to test .conf.tmp
* In all other cases the gauge configuration still needs to be written out here. */
if (!(return_check && accept)) {
xlfInfo = construct_paramsXlfInfo(plaquette_energy/(6.*VOLUME*g_nproc), trajectory_counter);
if (g_proc_id == 0) {
fprintf(stdout, "# Writing gauge field to %s.\n", tmp_filename);
}
if((status = write_gauge_field( tmp_filename, gauge_precision_write_flag, xlfInfo) != 0 )) {
/* Writing the gauge field failed directly */
fprintf(stderr, "Error %d while writing gauge field to %s\nAborting...\n", status, tmp_filename);
exit(-2);
}
if (!g_disable_IO_checks) {
#ifdef HAVE_LIBLEMON
/* Read gauge field back to verify the writeout */
if (g_proc_id == 0) {
fprintf(stdout, "# Write completed, verifying write...\n");
}
if( (status = read_gauge_field(tmp_filename)) != 0) {
fprintf(stderr, "WARNING, writeout of %s returned no error, but verification discovered errors.\n", tmp_filename);
fprintf(stderr, "Potential disk or MPI I/O error. Aborting...\n");
exit(-3);
}
if (g_proc_id == 0) {
fprintf(stdout, "# Write successfully verified.\n");
}
#else
if (g_proc_id == 0) {
fprintf(stdout, "# Write completed successfully.\n");
}
#endif
}
free(xlfInfo);
}
/* Now move .conf.tmp into place */
if(g_proc_id == 0) {
fprintf(stdout, "# Renaming %s to %s.\n", tmp_filename, gauge_filename);
if (rename(tmp_filename, gauge_filename) != 0) {
/* Errno can be inspected here for more descriptive error reporting */
fprintf(stderr, "Error while trying to rename temporary file %s to %s. Unable to proceed.\n", tmp_filename, gauge_filename);
exit(-2);
}
countfile = fopen(nstore_filename, "w");
fprintf(countfile, "%d %d %s\n", nstore, trajectory_counter+1, gauge_filename);
fclose(countfile);
}
}
/* online measurements */
for(imeas = 0; imeas < no_measurements; imeas++){
meas = &measurement_list[imeas];
if(trajectory_counter%meas->freq == 0){
if (g_proc_id == 0) {
fprintf(stdout, "#\n# Beginning online measurement.\n");
}
meas->measurefunc(trajectory_counter, imeas, even_odd_flag);
}
}
if(g_proc_id == 0) {
verbose = 1;
}
ix = reread_input("hmc.reread");
if(g_proc_id == 0) {
verbose = 0;
}
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
if(ix == 0 && g_proc_id == 0) {
countfile = fopen("history_hmc_tm", "a");
fprintf(countfile, "# Changed input parameters according to hmc.reread: measurement %d of %d\n", j, Nmeas);
fclose(countfile);
printf("# Changed input parameters according to hmc.reread (see stdout): measurement %d of %d\n", j, Nmeas);
remove("hmc.reread");
}
trajectory_counter++;
} /* end of loop over trajectories */
if(g_proc_id == 0 && Nmeas != 0) {
printf("# Acceptance rate was %3.2f percent, %d out of %d trajectories accepted.\n", 100.*(double)Rate/(double)Nmeas, Rate, Nmeas);
fflush(stdout);
parameterfile = fopen(parameterfilename, "a");
fprintf(parameterfile, "# Acceptance rate was %3.2f percent, %d out of %d trajectories accepted.\n", 100.*(double)Rate/(double)Nmeas, Rate, Nmeas);
fclose(parameterfile);
}
#ifdef MPI
MPI_Finalize();
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_gauge_tmp();
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
free_monomials();
if(g_running_phmc) {
free_bispinor_field();
free_chi_spinor_field();
}
return(0);
#ifdef _KOJAK_INST
#pragma pomp inst end(main)
#endif
}
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;
static int ini_g_tmp = 0;
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.update_gauge_energy = g_update_gauge_energy;
hf.update_rectangle_energy = g_update_rectangle_energy;
hf.traj_counter = traj_counter;
integrator_set_fields(&hf);
strcpy(tmp_filename, ".conf.tmp");
if(ini_g_tmp == 0) {
ini_g_tmp = init_gauge_tmp(VOLUME);
if(ini_g_tmp != 0) {
exit(-1);
}
ini_g_tmp = 1;
}
atime = gettime();
/*
* here the momentum and spinor fields are initialized
* and their respective actions are calculated
*/
/*
* copy the gauge field to gauge_tmp
*/
#ifdef 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);
}
}
/* 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_gauge_action( (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);
if(g_proc_id==0) {
#ifdef MPI
for(i = 1; i < g_nproc; i++) {
MPI_Send(&yy[0], 1, MPI_DOUBLE, i, 31, MPI_COMM_WORLD);
}
#endif
}
#ifdef MPI
else{
MPI_Recv(&yy[0], 1, MPI_DOUBLE, 0, 31, MPI_COMM_WORLD, &status);
}
#endif
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), -1);
// 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);
}
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 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 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 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 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,"ddh = %1.4e ddU= %1.4e ddh/H = %1.4e\n",
ret_dh, ret_gauge_diff/4./((double)(VOLUME*g_nproc))/3., ret_dh/tmp);
fclose(ret_check_file);
}
if(accept) {
/* Read back gauge field */
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) != 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 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 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;
hf.update_gauge_energy = 1;
g_update_gauge_energy = 1;
hf.update_rectangle_energy = 1;
g_update_rectangle_energy = 1;
#ifdef MPI
xchange_gauge(hf.gaugefield);
#endif
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[])
{
int j,j_max,k,k_max = 1;
#ifdef HAVE_LIBLEMON
paramsXlfInfo *xlfInfo;
#endif
int status = 0;
static double t1,t2,dt,sdt,dts,qdt,sqdt;
double antioptaway=0.0;
#ifdef MPI
static double dt2;
DUM_DERI = 6;
DUM_SOLVER = DUM_DERI+2;
DUM_MATRIX = DUM_SOLVER+6;
NO_OF_SPINORFIELDS = DUM_MATRIX+2;
# ifdef OMP
int mpi_thread_provided;
MPI_Init_thread(&argc, &argv, MPI_THREAD_SERIALIZED, &mpi_thread_provided);
# else
MPI_Init(&argc, &argv);
# endif
MPI_Comm_rank(MPI_COMM_WORLD, &g_proc_id);
#else
g_proc_id = 0;
#endif
g_rgi_C1 = 1.;
/* Read the input file */
if((status = read_input("benchmark.input")) != 0) {
fprintf(stderr, "Could not find input file: benchmark.input\nAborting...\n");
exit(-1);
}
#ifdef OMP
if(omp_num_threads > 0)
{
omp_set_num_threads(omp_num_threads);
}
else {
if( g_proc_id == 0 )
printf("# No value provided for OmpNumThreads, running in single-threaded mode!\n");
omp_num_threads = 1;
omp_set_num_threads(omp_num_threads);
}
init_omp_accumulators(omp_num_threads);
#endif
tmlqcd_mpi_init(argc, argv);
if(g_proc_id==0) {
#ifdef SSE
printf("# The code was compiled with SSE instructions\n");
#endif
#ifdef SSE2
printf("# The code was compiled with SSE2 instructions\n");
#endif
#ifdef SSE3
printf("# The code was compiled with SSE3 instructions\n");
#endif
#ifdef P4
printf("# The code was compiled for Pentium4\n");
#endif
#ifdef OPTERON
printf("# The code was compiled for AMD Opteron\n");
#endif
#ifdef _GAUGE_COPY
printf("# The code was compiled with -D_GAUGE_COPY\n");
#endif
#ifdef BGL
printf("# The code was compiled for Blue Gene/L\n");
#endif
#ifdef BGP
printf("# The code was compiled for Blue Gene/P\n");
#endif
#ifdef _USE_HALFSPINOR
printf("# The code was compiled with -D_USE_HALFSPINOR\n");
#endif
#ifdef _USE_SHMEM
printf("# The code was compiled with -D_USE_SHMEM\n");
# ifdef _PERSISTENT
printf("# The code was compiled for persistent MPI calls (halfspinor only)\n");
# endif
#endif
#ifdef MPI
# ifdef _NON_BLOCKING
printf("# The code was compiled for non-blocking MPI calls (spinor and gauge)\n");
# endif
#endif
printf("\n");
fflush(stdout);
}
#ifdef _GAUGE_COPY
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 1);
#else
init_gauge_field(VOLUMEPLUSRAND + g_dbw2rand, 0);
#endif
init_geometry_indices(VOLUMEPLUSRAND + g_dbw2rand);
if(even_odd_flag) {
j = init_spinor_field(VOLUMEPLUSRAND/2, 2*k_max+1);
}
else {
j = init_spinor_field(VOLUMEPLUSRAND, 2*k_max);
}
if ( j!= 0) {
fprintf(stderr, "Not enough memory for spinor fields! Aborting...\n");
exit(0);
}
j = init_moment_field(VOLUME, VOLUMEPLUSRAND + g_dbw2rand);
if ( j!= 0) {
fprintf(stderr, "Not enough memory for moment fields! Aborting...\n");
exit(0);
}
if(g_proc_id == 0) {
fprintf(stdout,"# The number of processes is %d \n",g_nproc);
printf("# The lattice size is %d x %d x %d x %d\n",
(int)(T*g_nproc_t), (int)(LX*g_nproc_x), (int)(LY*g_nproc_y), (int)(g_nproc_z*LZ));
printf("# The local lattice size is %d x %d x %d x %d\n",
(int)(T), (int)(LX), (int)(LY),(int) LZ);
if(even_odd_flag) {
printf("# benchmarking the even/odd preconditioned Dirac operator\n");
}
else {
printf("# benchmarking the standard Dirac operator\n");
}
fflush(stdout);
}
/* define the geometry */
geometry();
/* define the boundary conditions for the fermion fields */
boundary(g_kappa);
#ifdef _USE_HALFSPINOR
j = init_dirac_halfspinor();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for halfspinor fields! Aborting...\n");
exit(0);
}
if(g_sloppy_precision_flag == 1) {
g_sloppy_precision = 1;
j = init_dirac_halfspinor32();
if ( j!= 0) {
fprintf(stderr, "Not enough memory for 32-Bit halfspinor fields! Aborting...\n");
exit(0);
}
}
# if (defined _PERSISTENT)
init_xchange_halffield();
# endif
#endif
status = check_geometry();
if (status != 0) {
fprintf(stderr, "Checking of geometry failed. Unable to proceed.\nAborting....\n");
exit(1);
}
#if (defined MPI && !(defined _USE_SHMEM))
check_xchange();
#endif
start_ranlux(1, 123456);
random_gauge_field(reproduce_randomnumber_flag);
#ifdef MPI
/*For parallelization: exchange the gaugefield */
xchange_gauge(g_gauge_field);
#endif
if(even_odd_flag) {
/*initialize the pseudo-fermion fields*/
j_max=2048;
sdt=0.;
for (k = 0; k < k_max; k++) {
random_spinor_field(g_spinor_field[k], VOLUME/2, 0);
}
while(sdt < 30.) {
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
antioptaway=0.0;
for (j=0;j<j_max;j++) {
for (k=0;k<k_max;k++) {
Hopping_Matrix(0, g_spinor_field[k+k_max], g_spinor_field[k]);
Hopping_Matrix(1, g_spinor_field[2*k_max], g_spinor_field[k+k_max]);
antioptaway+=creal(g_spinor_field[2*k_max][0].s0.c0);
}
}
t2 = gettime();
dt = t2-t1;
#ifdef MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
qdt=dt*dt;
#ifdef MPI
MPI_Allreduce (&qdt, &sqdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sqdt = qdt;
#endif
sdt=sdt/((double)g_nproc);
sqdt=sqrt(sqdt/g_nproc-sdt*sdt);
j_max*=2;
}
j_max=j_max/2;
dts=dt;
sdt=1.0e6f*sdt/((double)(k_max*j_max*(VOLUME)));
sqdt=1.0e6f*sqdt/((double)(k_max*j_max*(VOLUME)));
if(g_proc_id==0) {
printf("# The following result is just to make sure that the calculation is not optimized away: %e\n", antioptaway);
printf("# Total compute time %e sec, variance of the time %e sec. (%d iterations).\n", sdt, sqdt, j_max);
printf("# Communication switched on:\n# (%d Mflops [%d bit arithmetic])\n", (int)(1608.0f/sdt),(int)sizeof(spinor)/3);
#ifdef OMP
printf("# Mflops per OpenMP thread ~ %d\n",(int)(1608.0f/(omp_num_threads*sdt)));
#endif
printf("\n");
fflush(stdout);
}
#ifdef MPI
/* isolated computation */
t1 = gettime();
antioptaway=0.0;
for (j=0;j<j_max;j++) {
for (k=0;k<k_max;k++) {
Hopping_Matrix_nocom(0, g_spinor_field[k+k_max], g_spinor_field[k]);
Hopping_Matrix_nocom(1, g_spinor_field[2*k_max], g_spinor_field[k+k_max]);
antioptaway += creal(g_spinor_field[2*k_max][0].s0.c0);
}
}
t2 = gettime();
dt2 = t2-t1;
/* compute the bandwidth */
dt=dts-dt2;
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
sdt=sdt/((double)g_nproc);
MPI_Allreduce (&dt2, &dt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
dt=dt/((double)g_nproc);
dt=1.0e6f*dt/((double)(k_max*j_max*(VOLUME)));
if(g_proc_id==0) {
printf("# The following result is printed just to make sure that the calculation is not optimized away: %e\n",antioptaway);
printf("# Communication switched off: \n# (%d Mflops [%d bit arithmetic])\n", (int)(1608.0f/dt),(int)sizeof(spinor)/3);
#ifdef OMP
printf("# Mflops per OpenMP thread ~ %d\n",(int)(1608.0f/(omp_num_threads*dt)));
#endif
printf("\n");
fflush(stdout);
}
sdt=sdt/((double)k_max);
sdt=sdt/((double)j_max);
sdt=sdt/((double)(2*SLICE));
if(g_proc_id==0) {
printf("# The size of the package is %d bytes.\n",(SLICE)*192);
#ifdef _USE_HALFSPINOR
printf("# The bandwidth is %5.2f + %5.2f MB/sec\n", 192./sdt/1024/1024, 192./sdt/1024./1024);
#else
printf("# The bandwidth is %5.2f + %5.2f MB/sec\n", 2.*192./sdt/1024/1024, 2.*192./sdt/1024./1024);
#endif
}
#endif
fflush(stdout);
}
else {
/* the non even/odd case now */
/*initialize the pseudo-fermion fields*/
j_max=1;
sdt=0.;
for (k=0;k<k_max;k++) {
random_spinor_field(g_spinor_field[k], VOLUME, 0);
}
while(sdt < 3.) {
#ifdef MPI
MPI_Barrier(MPI_COMM_WORLD);
#endif
t1 = gettime();
for (j=0;j<j_max;j++) {
for (k=0;k<k_max;k++) {
D_psi(g_spinor_field[k+k_max], g_spinor_field[k]);
antioptaway+=creal(g_spinor_field[k+k_max][0].s0.c0);
}
}
t2 = gettime();
dt=t2-t1;
#ifdef MPI
MPI_Allreduce (&dt, &sdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sdt = dt;
#endif
qdt=dt*dt;
#ifdef MPI
MPI_Allreduce (&qdt, &sqdt, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
#else
sqdt = qdt;
#endif
sdt=sdt/((double)g_nproc);
sqdt=sqrt(sqdt/g_nproc-sdt*sdt);
j_max*=2;
}
j_max=j_max/2;
dts=dt;
sdt=1.0e6f*sdt/((double)(k_max*j_max*(VOLUME)));
sqdt=1.0e6f*sqdt/((double)(k_max*j_max*(VOLUME)));
if(g_proc_id==0) {
printf("# The following result is just to make sure that the calculation is not optimized away: %e\n", antioptaway);
printf("# Total compute time %e sec, variance of the time %e sec. (%d iterations).\n", sdt, sqdt, j_max);
printf("\n# (%d Mflops [%d bit arithmetic])\n", (int)(1680.0f/sdt),(int)sizeof(spinor)/3);
#ifdef OMP
printf("# Mflops per OpenMP thread ~ %d\n",(int)(1680.0f/(omp_num_threads*sdt)));
#endif
printf("\n");
fflush(stdout);
}
}
#ifdef HAVE_LIBLEMON
if(g_proc_id==0) {
printf("# Performing parallel IO test ...\n");
}
xlfInfo = construct_paramsXlfInfo(0.5, 0);
write_gauge_field( "conf.test", 64, xlfInfo);
free(xlfInfo);
if(g_proc_id==0) {
printf("# done ...\n");
}
#endif
#ifdef MPI
MPI_Finalize();
#endif
#ifdef OMP
free_omp_accumulators();
#endif
free_gauge_field();
free_geometry_indices();
free_spinor_field();
free_moment_field();
return(0);
}
