int
main(int argc, char** argv)
{
double *A;
int n, ret, event;
double startTime;
double endTime;
long long value;
n = atoi(argv[2]);
A = load_matrix(argv[1], n);
event = atoi(argv[3]);
if (event != 5) {
papi_init(event);
papi_start();
} else {
startTime = dclock();
}
ret = chol(A, n);
if (event != 5) {
value = papi_stop();
printf("%lld\n", value);
} else {
endTime = dclock();
printf("%lf\n", endTime - startTime);
}
fprintf(stderr, "RET:%d\n", ret);
check(A,n);
free(A);
return 0;
}
int main( int argc, const char* argv[] )
{
int i,j,iret;
double first[SIZE][SIZE];
double second[SIZE][SIZE];
double multiply[SIZE][SIZE];
double dtime;
for (i = 0; i < SIZE; i++) { //rows in first
for (j = 0; j < SIZE; j++) { //columns in first
first[i][j]=i+j;
second[i][j]=i-j;
}
}
dtime = dclock();
iret=mm(first,second,multiply);
dtime = dclock()-dtime;
printf( "Time: %le \n", dtime);
fflush( stdout );
double check=0.0;
for(i=0;i<SIZE;i++){
for(j=0;j<SIZE;j++){
check+=multiply[i][j];
}
}
printf("check %le \n",check);
fflush( stdout );
return iret;
}
void contract_light_twopt(complex *corr, field_offset q_zonked,
field_offset q_sequential,
int zonked_pt, int spect_pt)
{
double t_start ;
int base_pt, q_stride, op_stride;
t_start = dclock() ;
/* Compute partial offset for storage of result in corr[] */
base_pt = TWOPT_FORM_WHERE(0,zonked_pt,spect_pt,0,0 ) ;
q_stride = TWOPT_FORM_WHERE(0,0, 0, 1,0 ) ;
op_stride = TWOPT_FORM_WHERE(0,0, 0, 0,1 ) ;
meson_cont_mom(corr , q_zonked, q_sequential,
base_pt, q_stride, op_stride,
two_pt, MAX_TWOPT);
IF_VERBOSE_ON(1)
printf("Time to Wick contract light 2pt correlators = %g sec\n",
dclock() - t_start) ;
}
int
main()
{
srand((unsigned int)time(NULL));
double *A;
double dtime;
int i, j;
A = generateSPDmatrix();
for(i = 0; i < SIZE; i++){
for(j = 0; j < SIZE; j++){
printf("%le \t", A[IDX(i, j, SIZE)]);
}
printf("\n");
}
dtime = dclock();
chol_left_looking(A, SIZE);
dtime = dclock()-dtime;
double gflops = ((1.0/3.0) * SIZE * SIZE * SIZE * 10e-9) / dtime;
printf( "Time: %le \n", dtime);
printf("Gflops: %le \n", gflops);
return 0;
}
void contract_LL2(complex *corr, field_offset q_zonked, field_offset q_spectator,
int zonked_pt, int spect_pt)
{
int base_pt, q_stride, op_stride;
double t_start ;
t_start = dclock() ;
/************************************************************/
/* Compute partial offset for storage of result in corr[] */
base_pt = LL_TWOPT_FORM_WHERE(0,zonked_pt,spect_pt,0,0 ) ;
q_stride = LL_TWOPT_FORM_WHERE(0,0, 0, 1,0 ) ;
op_stride = LL_TWOPT_FORM_WHERE(0,0, 0, 0,1 ) ;
meson_cont_mom_lean2(corr , q_zonked, q_spectator,
base_pt, q_stride, op_stride,
w_meson_store_t,w_meson_my_t,w_meson_nstore,
no_k_values,k_momstore,
MAX_TWOPT, two_pt,
F_OFFSET(QTMP),DIMQTMP);
IF_VERBOSE_ON(1)
printf("Time to Wick contract light-light 2pt correlators = %g sec\n",
dclock() - t_start) ;
}
void
create_hisq_links_milc(info_t *info, fn_links_t **fn, fn_links_t **fn_deps,
hisq_auxiliary_t **aux, ks_action_paths_hisq *ap,
su3_matrix *links, int want_deps, int want_back){
//char myname[] = "create_hisq_links_milc";
int n_naiks = ap->n_naiks;
int i;
double final_flop = 0.;
double dtime = -dclock();
*aux = create_hisq_auxiliary_t(ap, links);
load_hisq_aux_links(info, ap, *aux, links);
final_flop += info->final_flop;
for(i = 0; i < n_naiks; i++)
fn[i] = create_fn_links();
if(want_deps)
*fn_deps = create_fn_links();
else
*fn_deps = NULL;
load_hisq_fn_links(info, fn, *fn_deps, *aux, ap, links,
want_deps, want_back);
final_flop += info->final_flop;
dtime += dclock();
info->final_sec = dtime;
}
static QOP_FermionLinksWilson *
create_qop_wilson_fermion_links( Real clov )
{
QOP_FermionLinksWilson *qop_links = NULL;
QOP_info_t info;
QOP_GaugeField *links;
QOP_wilson_coeffs_t coeffs;
double remaptime;
/* Load coeffs structure */
load_qop_wilson_coeffs(&coeffs, clov);
/* Map SU(3) gauge field to G type */
remaptime = -dclock();
links = create_G_from_site4(F_OFFSET(link),EVENANDODD);
remaptime += dclock();
/* Create links */
qop_links = QOP_wilson_create_L_from_G(&info, &coeffs, links);
QOP_destroy_G(links);
#ifdef FFTIME
#ifdef REMAP
node0_printf("FFREMAP: time = %e\n",remaptime);
#endif
node0_printf("FFTIME: time = %e (cl_qop) terms = 1 mflops = %e\n",
info.final_sec, (Real)info.final_flop/(1e6*info.final_sec) );
#endif
return qop_links;
}
// Generate the rational approximation x^(pnum/pden)
void AlgRemez::generateApprox()
{
char *fname = "generateApprox()";
Float time = -dclock();
iter = 0;
spread = 1.0e37;
if (approx_type == RATIONAL_APPROX_ZERO_POLE) {
n--;
neq--;
}
initialGuess();
stpini(step);
while (spread > tolerance) { //iterate until convergance
if (iter++%100==0)
VRB.Result(cname,fname,"Iteration %d, spread %e delta %e\n", iter-1,(Float)spread,(Float)delta);
equations();
if (delta < tolerance)
ERR.General(cname, fname,"Delta too small, try increasing precision\n");
search(step);
}
int sign;
Float error = (Float)getErr(mm[0],&sign);
VRB.Result(cname,fname,"Converged at %d iterations, error = %e\n",
iter,error);
//!< Once the approximation has been generated, calculate the roots
if(!root()) ERR.General(cname,fname,"Root finding failed\n");
if (approx_type == RATIONAL_APPROX_ZERO_POLE) {
roots[n] = (bigfloat)0.0;
n++;
neq++;
}
//!< Now find the partial fraction expansions
if (remez_arg->field_type == BOSON) {
getPFE(remez_arg->residue, remez_arg->pole, &(remez_arg->norm));
getIPFE(remez_arg->residue_inv, remez_arg->pole_inv, &(remez_arg->norm_inv));
} else {
getIPFE(remez_arg->residue, remez_arg->pole, &(remez_arg->norm));
getPFE(remez_arg->residue_inv, remez_arg->pole_inv, &(remez_arg->norm_inv));
}
remez_arg->error = error;
time += dclock();
print_time(cname,fname,time);
}
CPS_START_NAMESPACE
/*!\file
\brief Definitions of functions that perform operations on complex matrices
and vectors.
$Id: vector_util.C,v 1.10 2013-04-19 20:25:52 chulwoo Exp $
*/
//--------------------------------------------------------------------
// CVS keywords
//
// $Author: chulwoo $
// $Date: 2013-04-19 20:25:52 $
// $Header: /home/chulwoo/CPS/repo/CVS/cps_only/cps_pp/src/util/vector/comsrc/vector_util.C,v 1.10 2013-04-19 20:25:52 chulwoo Exp $
// $Id: vector_util.C,v 1.10 2013-04-19 20:25:52 chulwoo Exp $
// $Name: not supported by cvs2svn $
// $Locker: $
// $Revision: 1.10 $
// $Source: /home/chulwoo/CPS/repo/CVS/cps_only/cps_pp/src/util/vector/comsrc/vector_util.C,v $
// $State: Exp $
//
//--------------------------------------------------------------------
/*------------------------------------------------------------------*/
/*
For these functions there exists optimized assembly
code.
*/
/*------------------------------------------------------------------*/
CPS_END_NAMESPACE
#include <string.h> /* memcpy */
#include <util/vector.h>
#include <util/time_cps.h>
//#include<omp.h>
CPS_START_NAMESPACE
/*!
\param b The vector to be copied to
\param a The vector to be copied from.
\param len The number of bytes to be copied.
The arrays \a c and \a b must not alias each other.
*/
//---------------------------------------------------------------//
void moveMem(void *b, const void *a, int len)
{
#undef PROFILE
#ifdef PROFILE
double time = -dclock();
#endif
memcpy(b, a, len);
#ifdef PROFILE
time += dclock();
print_flops("","moveMem",len,time);
#endif
}
//Parallel transport of a vector through one hop
void PT::vec(int n, IFloat **vout, IFloat **vin, const int *dir){
int i;
static int call_num=0;
SCUDirArgIR *SCUarg_p[2*n];
call_num++;
//for(int s = 0; s < GJP.VolNodeSites(); s++)
// {
// for(int t = 0; t < 4; t++)
// {
// printf("site = %d, direction = %d\n",s,t);
// for(int u = 0; u < 9; u++)
// printf("%e %e\n",*(gauge_field_addr+4*GAUGE_LEN*s + GAUGE_LEN*t + 2*u),*(gauge_field_addr+4*GAUGE_LEN*s + GAUGE_LEN*t + 2*u+1));
// }
// }
#ifdef PROFILE
Float dtime = - dclock();
#endif
int wire[n];
SCUDirArgMulti SCUmulti;
char *fname="pt_1vec";
// VRB.Func("",fname);
int non_local_dir=0;
for(i=0;i<n;i++) wire[i] = dir[i]; // from (x,y,z,t) to (t,x,y,z)
// for(i=0;i<n;i++) printf("wire[%d]=%d\n",i,dir[i]);
for(i=0;i<n;i++)
if (!local[wire[i]/2]){
IFloat * addr = (vin[i]+VECT_LEN*offset[wire[i]]);
SCUarg_p[2*non_local_dir] = SCUarg[0][2*wire[i]];
SCUarg_p[2*non_local_dir+1] = SCUarg[0][2*wire[i]+1];
SCUarg_p[2*non_local_dir+1]->Addr((void *)addr);
non_local_dir++;
}
if(non_local_dir){
SCUmulti.Init(SCUarg_p,non_local_dir*2);
SCUmulti.SlowStartTrans();
}
for(i=0;i<n;i++)
partrans_cmv_agg(local_chi[wire[i]],(long)uc_l[wire[i]], (long)vin[i],(long)vout[i]);
if(non_local_dir){ SCUmulti.TransComplete(); }
for(i=0;i<n;i++)
partrans_cmv_agg(non_local_chi[wire[i]],(long)uc_nl[wire[i]], (long)rcv_buf[wire[i]],(long)vout[i]);
#ifdef PROFILE
dtime +=dclock();
print_flops("",fname,66*n*vol,dtime);
#endif
Flops +=66*n*vol;
}
void moveVec(Float *b, const Float *a, int len) {
#undef PROFILE
#ifdef PROFILE
double time = -dclock();
#endif
// for(int i =0;i<len*6;i++) *b++ = *a++;
memcpy(b, a, len*sizeof(Vector));
#ifdef PROFILE
time += dclock();
print_flops("","moveVec",len*sizeof(Float),time);
#endif
}
ForceArg GimprRect::EvolveMomGforce(Matrix *mom, Float dt){
char *fname = "EvolveMomGforce(M*,F)";
VRB.Func(cname,fname);
Float L1=0.0;
Float L2=0.0;
Float Linf=0.0;
#ifdef PROFILE
Float time = -dclock();
ForceFlops = 0;
#endif
setCbufCntrlReg(4, CBUF_MODE4);
int x[4];
for(x[0] = 0; x[0] < GJP.XnodeSites(); ++x[0])
for(x[1] = 0; x[1] < GJP.YnodeSites(); ++x[1])
for(x[2] = 0; x[2] < GJP.ZnodeSites(); ++x[2])
for(x[3] = 0; x[3] < GJP.TnodeSites(); ++x[3]) {
int uoff = GsiteOffset(x);
for (int mu = 0; mu < 4; ++mu) {
GforceSite(*mp0, x, mu);
IFloat *ihp = (IFloat *)(mom+uoff+mu);
IFloat *dotp = (IFloat *)mp0;
fTimesV1PlusV2(ihp, dt, dotp, ihp, 18);
Float norm = ((Matrix*)dotp)->norm();
Float tmp = sqrt(norm);
L1 += tmp;
L2 += norm;
Linf = (tmp>Linf ? tmp : Linf);
}
}
ForceFlops +=GJP.VolNodeSites()*4*18*2;
#ifdef PROFILE
time += dclock();
print_flops(cname,fname,ForceFlops,time);
#endif
glb_sum(&L1);
glb_sum(&L2);
glb_max(&Linf);
L1 /= 4.0*GJP.VolSites();
L2 /= 4.0*GJP.VolSites();
VRB.FuncEnd(cname,fname);
return ForceArg(dt*L1, dt*sqrt(L2), dt*Linf);
}
//!< Calculate gauge contribution to the Hamiltonian
Float AlgMomentum::energy() {
Float dtime = -dclock();
const char *fname = "energy()";
Lattice &lat = LatticeFactory::Create(F_CLASS_NONE, G_CLASS_NONE);
Float h = lat.MomHamiltonNode(mom);
LatticeFactory::Destroy();
dtime += dclock();
print_flops(cname, fname, 0, dtime);
return h;
}
//!< evolve method evolves the gauge field due to the momentum
void AlgMomentum::evolve(Float dt, int steps)
{
const char *fname = "evolve()";
Float dtime = -dclock();
Lattice &lat = LatticeFactory::Create(F_CLASS_NONE, G_CLASS_NONE);
for (int i=0; i<steps; i++) lat.EvolveGfield(mom, dt);
lat.MdTimeInc(dt*steps);
VRB.Flow(cname,fname,"%s%f\n", md_time_str, IFloat(lat.MdTime()));
LatticeFactory::Destroy();
dtime += dclock();
print_flops(cname, fname, 1968. * 4. * GJP.VolNodeSites() * steps, dtime);
}
int main(int argc, char *argv[]) {
unsigned n;
int evt;
double *A;
int i, j;
double checksum = 0;
double startTime, endTime;
long long counter;
if (argc < 2) {
return -1;
}
n = atoi(argv[1]);
evt = (argc > 2) ? atoi(argv[2]) : -1;
A = randomMatrix(n);
assert(A != NULL);
if (evt == -1) {
startTime = dclock();
} else {
papi_init(evt);
papi_start();
}
if (chol(A, n)) {
fprintf(stderr, "Error: matrix is either not symmetric or not positive definite.\n");
} else {
for (i = 0; i < n; i++) {
for (j = i; j < n; j++) {
checksum += A[IDX(i, j, n)];
}
}
printf("Checksum: %f \n", checksum);
}
if (evt == -1) {
endTime = dclock();
fprintf(stderr, "%f\n", endTime - startTime);
} else {
counter = papi_stop();
fprintf(stderr, "%lld\n", counter);
}
free(A);
return 0;
}
void wilson_vector_hqet_src(field_offset out, field_offset in, int spin, int tB)
{
register int i;
register site *s;
int colour ;
double t_start ;
t_start = dclock() ;
FORALLSITES(i,s)
{
/*** zero the hqet source **/
zero_zu3_matrix( (su3_matrix *)F_PT(s,out) );
if( s->t == tB)
{
for(colour = 0 ; colour < 3 ; ++colour)
((su3_matrix *)F_PT(s,out))->e[colour][colour]
= ((wilson_vector *)F_PT(s,in))->d[spin].c[colour] ;
}
} /** end of the loop over lattice sites ****/
void hopping(field_offset src, field_offset temp,
field_offset light_quark,
int nhop, Real kappa_c, int parity_of_source,
int color, int spin, int wallflag, FILE * fp_m_out, int fb_m_out)
{
double dtime ;
/** double dtime1; ****/
int N_iter;
register int i;
register site *s;
Real size_src, size_r;
int old_parity, new_parity = 0x00, channel;
double **meson_prop;
wilson_vector *light_wall = NULL, *heavy_wall = NULL;
/* Start Hopping */
dtime = -dclock();
/* Normalisation */
size_src = 0.0;
FORSOMEPARITY(i, s, parity_of_source)
{
size_src += magsq_wvec(((wilson_vector *) F_PT(s, src)));
}
void contract_hqet_to_light(complex *corr,
field_offset q_zonked,
field_offset q_zonked_rot,
field_offset q_sequential,
int vel_pt, int zonked_pt, int spect_pt)
{
double t_start ;
int base_pt, q_stride, op_stride ;
t_start = dclock() ;
/* Compute partial offset for storage of result in corr[] */
base_pt = HQET_FORM_WHERE(0,zonked_pt,spect_pt,0,vel_pt, 0 ) ;
q_stride = HQET_FORM_WHERE(0,0, 0, 1,0, 0 ) ;
op_stride = HQET_FORM_WHERE(0,0, 0, 0,0, 1 ) ;
/* First, contract zonked and sequential */
meson_cont_mom(corr , q_zonked, q_sequential,
base_pt, q_stride, op_stride,
hqet_to_light, MAX_THREEPT) ;
/* Second, contract rotated zonked and sequential
Results go to second half of corr */
base_pt += op_stride*MAX_THREEPT;
meson_cont_mom(corr , q_zonked_rot, q_sequential,
base_pt, q_stride, op_stride,
hqet_to_light, MAX_THREEPT) ;
IF_VERBOSE_ON(1)
printf("contract_hqet_to_light::Time to Wick contract hqet-->light correlators = %g sec\n",dclock() - t_start) ;
}
void update_time(void)
{
double c;
c = dclock();
runtime += (c-cprev);
cprev = c;
}
SPDP dtime()
{
SPDP q;
q = dclock();
return q;
}
//!< Heat Bath for the conjugate momentum
void AlgMomentum::heatbath() {
const char *fname = "heatbath()";
Float dtime = -dclock();
Lattice &lat = LatticeFactory::Create(F_CLASS_NONE, G_CLASS_NONE);
lat.RandGaussAntiHermMatrix(mom, 1.0);
//!< reset MD time in Lattice (a momentum refresh means a new trajectory)
lat.MdTime(0.0);
VRB.Flow(cname,fname,"%s%f\n", md_time_str, IFloat(lat.MdTime()));
LatticeFactory::Destroy();
dtime += dclock();
print_flops(cname, fname, 0, dtime);
}
double dtime()
{
double q;
q = dclock();
return q;
}
void moveFloat(Float *b, const Float *a, int len) {
#undef PROFILE
#ifdef PROFILE
double time = -dclock();
#endif
#ifdef USE_OMP
#pragma omp parallel for
for(int i =0;i<len;i++) b[i] = a[i];
#else
memcpy(b, a, len*sizeof(Float));
#endif
#ifdef PROFILE
time += dclock();
print_flops("","moveFloat",len*sizeof(Float),time);
#endif
}
/*--------------------------------------------------------------------*/
void print_timing(double dtime, char *str){
#ifdef PRTIME
dtime += dclock();
node0_printf("Time for %s %e\n",str, dtime); fflush(stdout);
#endif
}
static void
create_qop_links_from_milc_fn(ferm_links_t *fn)
{
double remaptime;
char myname[] = "create_qop_links_from_milc";
remaptime = -dclock();
DESTROY_QOP_ASQTAD_FERMION_LINKS(fn);
fn->QOP_L = CREATE_L_FROM_FIELDS(fn->fat, fn->lng, EVENANDODD);
remaptime += dclock();
#ifdef LLTIME
#ifdef REMAP
node0_printf("LLREMAP: time = %e\n",remaptime);
#endif
#endif
}
void load_fn_links_gpu(info_t *info, fn_links_t *fn, ks_action_paths *ap,
su3_matrix *links, int want_back)
{
ks_component_paths *p = &ap->p;
double final_flop = 0;
double dtime = -dclock();
load_fatlonglinks_gpu(info, fn->fat, fn->lng, p, links);
if(want_back)
load_fn_backlinks(fn);
else
destroy_fn_backlinks(fn);
dtime += dclock();
info->final_sec = dtime;
info->final_flop = final_flop;
}
static void __timer_reset(mtimer_t * timer)
{
#ifdef XT3
timer->starttime = timer->stoptime = dclock();
#else
gettimeofday(&timer->start_time, 0);
timer->stop_time = timer->start_time;
#endif
}
int main( int argc, const char* argv[] )
{
FILE *fp = init_file(argv[0] + 2);
int iret;
for (SIZE = 8; SIZE <= 512; SIZE += 8) {
int i,j;
double first[SIZE][SIZE];
double second[SIZE][SIZE];
double multiply[SIZE][SIZE];
double dtime;
double gflops;
for (i = 0; i < SIZE; i++) { //rows in first
for (j = 0; j < SIZE; j++) { //columns in first
first[i][j]=i+j;
second[i][j]=i-j;
multiply[i][j]=0.0;
}
}
papi_init();
dtime = dclock();
iret = mm(first,second,multiply);
dtime = dclock()-dtime;
fprintf(fp, "%d, ", SIZE);
papi_results(fp);
gflops = 2.0 * SIZE * SIZE * SIZE * 1e-9 / dtime;
printf( "%d, %le, %f\n", SIZE, dtime, gflops);
//double check=0.0;
//for(i=0;i<SIZE;i++){
// for(j=0;j<SIZE;j++){
// check+=multiply[i][j];
// }
//}
//printf("check %le \n",check);
fflush( stdout );
}
return iret;
}
/* cray timers */
void metric_read_craytimers(int tid, int idx, double values[]) {
#ifdef CRAY_TIMERS
#ifdef TAU_CATAMOUNT /* for Cray XT3 */
values[idx] = dclock() * 1.0e6;
#else /* for Cray X1 */
long long tick = _rtc();
values[idx] = (double)tick / HZ;
#endif /* TAU_CATAMOUNT */
#endif /* CRAY_TIMERS */
}
/*--------------------------------------------------------------------*/
double start_timing(void){
double dtime;
#ifdef PRTIME
dtime = -dclock();
#else
dtime = 0;
#endif
return dtime;
}