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[])
{
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[])
{
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);
}
