portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

reduce acle works, clean up

Browse Source
This commit is contained in:
nmeyer-ur 2020-04-03 20:40:44 +02:00
parent b27e31957a
commit 15238e8d5e
1 changed files with 10 additions and 95 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

105
Grid/simd/Grid_a64fx-2.h
View File

@ -251,17 +251,17 @@ struct Vstore{
}
};
struct Vstream{
// Real
template <typename T>
inline void operator()(T * a, vec<T> b){
struct Vstream{
// Real
template <typename T>
inline void operator()(T * a, vec<T> b){
svbool_t pg1 = acle<T>::pg1();
typename acle<T>::vt b_v = svld1(pg1, b.v);
svstnt1(pg1, a, b_v);
//svst1(pg1, a, b_v);
}
};
svbool_t pg1 = acle<T>::pg1();
typename acle<T>::vt b_v = svld1(pg1, b.v);
svstnt1(pg1, a, b_v);
//svst1(pg1, a, b_v);
}
};
struct Vset{
// Complex
@ -293,7 +293,6 @@ struct Vstore{
// Arithmetic operations
/////////////////////////////////////////////////////
struct Sum{
template <typename T>
inline vec<T> operator()(vec<T> a, vec<T> b){
@ -324,7 +323,6 @@ struct Sub{
}
};
struct Mult{
template <typename T>
inline vec<T> operator()(vec<T> a, vec<T> b){
@ -431,7 +429,6 @@ struct Conj{
}
};
struct TimesMinusI{
// Complex
template <typename T>
@ -472,7 +469,6 @@ struct TimesI{
}
};
struct PrecisionChange {
static inline vech StoH (const vecf &sa,const vecf &sb) {
@ -578,7 +574,6 @@ struct PrecisionChange {
}
};
struct Exchange{
// Exchange0 is valid for arbitrary SVE vector length
@ -646,7 +641,6 @@ struct Exchange{
}
};
struct Permute{
// Permute0 is valid for any SVE vector width
@ -771,9 +765,6 @@ struct Rotate{
}
};
// =======================================================================
// SVE ACLE reduce does not compile, check later
// tree-based reduction
#define svred(pg, v)\
svaddv(pg, v);
@ -857,82 +848,6 @@ inline Integer Reduce<Integer, veci>::operator()(veci in){
#undef svred
// =======================================================================
/*
#define acc(v, a, off, step, n)\
for (unsigned int i = off; i < n; i += step)\
{\
a += v[i];\
}
template <typename Out_type, typename In_type>
struct Reduce{
//Need templated class to overload output type
//General form must generate error if compiled
inline Out_type operator()(In_type in){
printf("Error, using wrong Reduce function\n");
exit(1);
return 0;
}
};
//Complex float Reduce
template <>
inline Grid::ComplexF Reduce<Grid::ComplexF, vecf>::operator()(vecf in){
float a = 0.f, b = 0.f;
acc(in.v, a, 0, 2, W<float>::r);
acc(in.v, b, 1, 2, W<float>::r);
return Grid::ComplexF(a, b);
}
//Real float Reduce
template<>
inline Grid::RealF Reduce<Grid::RealF, vecf>::operator()(vecf in){
float a = 0.;
acc(in.v, a, 0, 1, W<float>::r);
return a;
}
//Complex double Reduce
template<>
inline Grid::ComplexD Reduce<Grid::ComplexD, vecd>::operator()(vecd in){
double a = 0., b = 0.;
acc(in.v, a, 0, 2, W<double>::r);
acc(in.v, b, 1, 2, W<double>::r);
return Grid::ComplexD(a, b);
}
//Real double Reduce
template<>
inline Grid::RealD Reduce<Grid::RealD, vecd>::operator()(vecd in){
double a = 0.f;
acc(in.v, a, 0, 1, W<double>::r);
return a;
}
//Integer Reduce
template<>
inline Integer Reduce<Integer, veci>::operator()(veci in){
Integer a = 0;
acc(in.v, a, 0, 1, W<Integer>::r);
return a;
}
#undef acc // EIGEN compatibility
*/
NAMESPACE_END(Optimization)
//////////////////////////////////////////////////////////////////////////////////////