inline double norm2(Float *v, int len) {
double sum=0.0;
for (int i=0; i<len; i++) sum += v[i]*v[i];
#ifdef MPI_COMMS
comm_allreduce(&sum);
#elif QMP_COMMS
QMP_sum_double(&sum);
#endif
return sum;
}
double
bench_inv(QOP_info_t *info, QOP_invert_arg_t *inv_arg,
QOP_resid_arg_t *res_arg, QDP_DiracFermion *out,
QDP_DiracFermion *in)
{
static QLA_Real r2s=-1, r2;
double sec=0, flop=0, mf=0;
int i, iter=0;
QOP_DiracFermion *qopout, *qopin;
QDP_D_eq_zero(out, QDP_all);
qopout = QOP_create_D_from_qdp(out);
qopin = QOP_create_D_from_qdp(in);
for(i=0; i<=nit; i++) {
QMP_barrier();
QOP_wilson_invert(info, flw, inv_arg, res_arg, kappa, qopout, qopin);
QMP_barrier();
printf("%i\t%i\t%g\t%i\n", i, res_arg->final_iter, info->final_sec, (int)info->final_flop);
if(i>0) {
iter += res_arg->final_iter;
sec += info->final_sec;
flop += info->final_flop;
//mf += info->final_flop/(1e6*info->final_sec);
}
}
QOP_destroy_D(qopout);
QOP_destroy_D(qopin);
QDP_r_eq_norm2_D(&r2, out, QDP_even);
if(r2s<0) r2s = r2;
if(fabs(1-r2/r2s)>1e-3) {
printf0("first norm = %g this norn = %g\n", r2s, r2);
}
mf = 1;
QMP_sum_double(&mf);
QMP_sum_double(&sec);
QMP_sum_double(&flop);
res_arg->final_iter = iter/nit;
info->final_sec = sec/(mf*nit);
info->final_flop = flop/(mf*nit);
mf = info->final_flop/(1e6*info->final_sec);
return mf;
}
QLA_Real
QOPPCV(relnorm2)(Vector2 **rsd, Vector2 **out, QDP_Subset subset, int nv)
{
g_rel = 0;
for(int i=0; i<nv; i++) {
g_r = expose_V(rsd[i]);
V_eq_funcit(out[i], relnorm_func, subset);
reset_V(rsd[i]);
}
QMP_sum_double(&g_rel);
QLA_Real rel = g_rel/QDP_subset_len(subset);
return rel;
}
void comm_allreduce(double* data)
{
QMP_CHECK( QMP_sum_double(data) );
}
