portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-10 14:10:46 +01:00
Code Releases Activity

Added integer reduce functionality

Browse Source
This commit is contained in:
Nils Meyer 2017-07-24 11:12:59 +02:00
parent 0933aeefd4
commit 7a53dc3715
1 changed files with 20 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
lib/simd/Grid_neon.h
View File

@ -27,7 +27,7 @@ Author: neo <cossu@post.kek.jp>
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
//----------------------------------------------------------------------
/*
ARMv8 NEON intrinsics layer by
@ -37,9 +37,6 @@ Author: neo <cossu@post.kek.jp>
SFB/TRR55
*/
//----------------------------------------------------------------------
//#ifndef ARM_NEON
//#define ARM_NEON
#ifndef GEN_SIMD_WIDTH
#define GEN_SIMD_WIDTH 16u
@ -85,11 +82,11 @@ namespace Optimization {
double tmp[2]={a,b};
return vld1q_f64(tmp);
}
//Real double // N:tbc
//Real double
inline float64x2_t operator()(double a){
return vdupq_n_f64(a);
}
//Integer // N:tbc
//Integer
inline uint32x4_t operator()(Integer a){
return vdupq_n_u32(a);
}
@ -127,33 +124,32 @@ namespace Optimization {
// Nils: Vset untested; not used currently in Grid at all;
// git commit 4a8c4ccfba1d05159348d21a9698028ea847e77b
struct Vset{
// Complex float // N:ok
// Complex float
inline float32x4_t operator()(Grid::ComplexF *a){
float tmp[4]={a[1].imag(),a[1].real(),a[0].imag(),a[0].real()};
return vld1q_f32(tmp);
}
// Complex double // N:ok
// Complex double
inline float64x2_t operator()(Grid::ComplexD *a){
double tmp[2]={a[0].imag(),a[0].real()};
return vld1q_f64(tmp);
}
// Real float // N:ok
// Real float
inline float32x4_t operator()(float *a){
float tmp[4]={a[3],a[2],a[1],a[0]};
return vld1q_f32(tmp);
}
// Real double // N:ok
// Real double
inline float64x2_t operator()(double *a){
double tmp[2]={a[1],a[0]};
return vld1q_f64(tmp);
}
// Integer // N:ok
// Integer
inline uint32x4_t operator()(Integer *a){
return vld1q_dup_u32(a);
}
};
// N:leaving as is
template <typename Out_type, typename In_type>
struct Reduce{
//Need templated class to overload output type
@ -424,11 +420,6 @@ namespace Optimization {
}
}
// working, but no restriction on n
// template<int n> static inline float32x4_t tRotate(float32x4_t in){ return vextq_f32(in,in,n); };
// template<int n> static inline float64x2_t tRotate(float64x2_t in){ return vextq_f64(in,in,n); };
// restriction on n
template<int n> static inline float32x4_t tRotate(float32x4_t in){ return vextq_f32(in,in,n%4); };
template<int n> static inline float64x2_t tRotate(float64x2_t in){ return vextq_f64(in,in,n%2); };
@ -550,7 +541,7 @@ namespace Optimization {
//Complex double Reduce
template<> // N:by Boyle
template<>
inline Grid::ComplexD Reduce<Grid::ComplexD, float64x2_t>::operator()(float64x2_t in){
u128d conv; conv.v = in;
return Grid::ComplexD(conv.f[0],conv.f[1]);
@ -565,9 +556,7 @@ namespace Optimization {
//Integer Reduce
template<>
inline Integer Reduce<Integer, uint32x4_t>::operator()(uint32x4_t in){
// FIXME unimplemented
printf("Reduce : Missing integer implementation -> FIX\n");
assert(0);
return vaddvq_u32(in);
}
}
@ -607,5 +596,3 @@ namespace Optimization {
typedef Optimization::TimesI TimesISIMD;
}
//#endif // ARM_NEON