qdp_sse_fused_spin_recon_evaluates_wrapper.h

Go to the documentation of this file.

#ifndef QDP_SSE_FUSED_SPIN_RECON_EVALUATES_WRAPPER_H
#define QDP_SSE_FUSED_SPIN_RECON_EVALUATES_WRAPPER_H

// Threading evaluates wrappers
//
// by Xu Guo, EPCC, 21 Oct, 2008
 
// user arg for evaluate having order
struct ordered_sse_fused_spin_recon_user_arg{
  const OLattice< SU3Mat32 >& u;
  const OLattice< HVec32 >& a;
  OLattice< FVec32 >& d;
  int base;
  void (*func)(const REAL32*, REAL32*, unsigned int);
};
 
 
// user func for evaluate having order
inline
void ordered_sse_fused_spin_recon_evaluate_function (int lo, int hi, int myId, ordered_sse_fused_spin_recon_user_arg* arg){
 
  const OLattice< SU3Mat32 >& u = arg->u;
  const OLattice< HVec32 >& a = arg->a ;
  OLattice< FVec32 >& d = arg->d;
  int base = arg->base;
  void (*func)(const REAL32*, REAL32*, unsigned int) = arg->func; 
 
  int low = lo + base;
  int high = hi + base;
 
  for (int site = low; site < high; site++){
 
      HVec32 tmp ;
 
      su3_matrixf* um = (su3_matrixf *)&(u.elem(site).elem().elem(0,0).real());
      half_wilson_vectorf *ah = (half_wilson_vectorf *)&( a.elem(site).elem(0).elem(0).real());
      half_wilson_vectorf *tmph = (half_wilson_vectorf *)&( tmp.elem(0).elem(0).real());
 
      intrin_sse_mult_su3_mat_hwvec(um, ah, tmph);
      
      
      func( (REAL32 *)&(tmp.elem(0).elem(0).real()),
            (REAL32 *)&(d.elem(site).elem(0).elem(0).real()),
            1);
 
  }
 
}
 
 
// user arg for evaluate NOT having order
struct unordered_sse_fused_spin_recon_user_arg{
  const OLattice< SU3Mat32 >& u;
  const OLattice< HVec32 >& a;
  OLattice< FVec32 >& d;
  const int *tab;
  void (*func)(const REAL32*, REAL32*, unsigned int);
};
 
 
// user func for evaluate NOT having order
inline
void unordered_sse_fused_spin_recon_evaluate_function (int lo, int hi, int myId, unordered_sse_fused_spin_recon_user_arg* arg){
 
  const OLattice< SU3Mat32 >& u = arg->u;
  const OLattice< HVec32 >& a = arg->a ;
  OLattice< FVec32 >& d = arg->d;
  const int *tab = arg->tab;
  void (*func)(const REAL32*, REAL32*, unsigned int) = arg->func; 
 
  for (int j = lo; j < hi; j++){
 
    int site=tab[j];
      
    HVec32 tmp ;
 
    su3_matrixf* um = (su3_matrixf *)&(u.elem(site).elem().elem(0,0).real());
    half_wilson_vectorf *ah = (half_wilson_vectorf *)&( a.elem(site).elem(0).elem(0).real());
    half_wilson_vectorf *tmph = (half_wilson_vectorf *)&( tmp.elem(0).elem(0).real());
 
    intrin_sse_mult_su3_mat_hwvec(um, ah, tmph);
      
      
    func( (REAL32 *)&(tmp.elem(0).elem(0).real()),
          (REAL32 *)&(d.elem(site).elem(0).elem(0).real()),
          1);
  }
 
}
 
 
#endif