portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-27 14:15:55 +01:00
Code Releases Activity

Format, NAMESPACE

Browse Source
This commit is contained in:
paboyle 2018-01-12 23:16:37 +00:00
parent 0a6168eef0
commit 62fcee72c5
1 changed files with 362 additions and 379 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

211
lib/simd/Grid_imci.h
View File

@ -1,4 +1,4 @@
/************************************************************************************* /*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid Grid physics library, www.github.com/paboyle/Grid
@ -24,16 +24,16 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/ *************************************************************************************/
/* END LEGAL */ /* END LEGAL */
#include <immintrin.h> #include <immintrin.h>
#include <zmmintrin.h> #include <zmmintrin.h>
namespace Grid{ NAMESPACE_BEGIN(Grid);
namespace Optimization { NAMESPACE_BEGIN(Optimization);
struct Vsplat{ struct Vsplat{
//Complex float //Complex float
inline __m512 operator()(float a, float b){ inline __m512 operator()(float a, float b){
return _mm512_set_ps(b,a,b,a,b,a,b,a,b,a,b,a,b,a,b,a); return _mm512_set_ps(b,a,b,a,b,a,b,a,b,a,b,a,b,a,b,a);
@ -54,9 +54,9 @@ namespace Optimization {
inline __m512i operator()(Integer a){ inline __m512i operator()(Integer a){
return _mm512_set1_epi32(a); return _mm512_set1_epi32(a);
} }
}; };
struct Vstore{ struct Vstore{
//Float //Float
inline void operator()(__m512 a, float* F){ inline void operator()(__m512 a, float* F){
_mm512_store_ps(F,a); _mm512_store_ps(F,a);
@ -70,10 +70,9 @@ namespace Optimization {
_mm512_store_si512((__m512i *)I,a); _mm512_store_si512((__m512i *)I,a);
} }
}; };
struct Vstream{
struct Vstream{
//Float //Float
inline void operator()(float * a, __m512 b){ inline void operator()(float * a, __m512 b){
_mm512_storenrngo_ps(a,b); _mm512_storenrngo_ps(a,b);
@ -84,11 +83,9 @@ namespace Optimization {
} }
}; };
struct Vset{
struct Vset{
// Complex float // Complex float
inline __m512 operator()(Grid::ComplexF *a){ inline __m512 operator()(Grid::ComplexF *a){
return _mm512_set_ps(a[7].imag(),a[7].real(),a[6].imag(),a[6].real(), return _mm512_set_ps(a[7].imag(),a[7].real(),a[6].imag(),a[6].real(),
@ -117,10 +114,10 @@ namespace Optimization {
} }
}; };
template <typename Out_type, typename In_type> template <typename Out_type, typename In_type>
struct Reduce{ struct Reduce{
//Need templated class to overload output type //Need templated class to overload output type
//General form must generate error if compiled //General form must generate error if compiled
inline Out_type operator()(In_type in){ inline Out_type operator()(In_type in){
@ -128,15 +125,12 @@ namespace Optimization {
exit(1); exit(1);
return 0; return 0;
} }
}; };
/////////////////////////////////////////////////////
// Arithmetic operations
/////////////////////////////////////////////////////
///////////////////////////////////////////////////// struct Sum{
// Arithmetic operations
/////////////////////////////////////////////////////
struct Sum{
//Complex/Real float //Complex/Real float
inline __m512 operator()(__m512 a, __m512 b){ inline __m512 operator()(__m512 a, __m512 b){
return _mm512_add_ps(a,b); return _mm512_add_ps(a,b);
@ -149,9 +143,9 @@ namespace Optimization {
inline __m512i operator()(__m512i a, __m512i b){ inline __m512i operator()(__m512i a, __m512i b){
return _mm512_add_epi32(a,b); return _mm512_add_epi32(a,b);
} }
}; };
struct Sub{ struct Sub{
//Complex/Real float //Complex/Real float
inline __m512 operator()(__m512 a, __m512 b){ inline __m512 operator()(__m512 a, __m512 b){
return _mm512_sub_ps(a,b); return _mm512_sub_ps(a,b);
@ -164,10 +158,9 @@ namespace Optimization {
inline __m512i operator()(__m512i a, __m512i b){ inline __m512i operator()(__m512i a, __m512i b){
return _mm512_sub_epi32(a,b); return _mm512_sub_epi32(a,b);
} }
}; };
struct MultComplex{
struct MultComplex{
// Complex float // Complex float
inline __m512 operator()(__m512 a, __m512 b){ inline __m512 operator()(__m512 a, __m512 b){
__m512 vzero,ymm0,ymm1,real, imag; __m512 vzero,ymm0,ymm1,real, imag;
@ -210,9 +203,9 @@ namespace Optimization {
ymm0 = _mm512_swizzle_pd(b, _MM_SWIZ_REG_CDAB); // OK ymm0 = _mm512_swizzle_pd(b, _MM_SWIZ_REG_CDAB); // OK
return _mm512_fmadd_pd(ymm0,imag,ymm1); return _mm512_fmadd_pd(ymm0,imag,ymm1);
} }
}; };
struct Mult{ struct Mult{
inline void mac(__m512 &a, __m512 b, __m512 c){ inline void mac(__m512 &a, __m512 b, __m512 c){
a= _mm512_fmadd_ps( b, c, a); a= _mm512_fmadd_ps( b, c, a);
@ -234,9 +227,9 @@ namespace Optimization {
inline __m512i operator()(__m512i a, __m512i b){ inline __m512i operator()(__m512i a, __m512i b){
return _mm512_mullo_epi32(a,b); return _mm512_mullo_epi32(a,b);
} }
}; };
struct Div{ struct Div{
// Real float // Real float
inline __m512 operator()(__m512 a, __m512 b){ inline __m512 operator()(__m512 a, __m512 b){
return _mm512_div_ps(a,b); return _mm512_div_ps(a,b);
@ -245,10 +238,9 @@ namespace Optimization {
inline __m512d operator()(__m512d a, __m512d b){ inline __m512d operator()(__m512d a, __m512d b){
return _mm512_div_pd(a,b); return _mm512_div_pd(a,b);
} }
}; };
struct Conj{
struct Conj{
// Complex single // Complex single
inline __m512 operator()(__m512 in){ inline __m512 operator()(__m512 in){
return _mm512_mask_sub_ps(in,0xaaaa,_mm512_setzero_ps(),in); // Zero out 0+real 0-imag return _mm512_mask_sub_ps(in,0xaaaa,_mm512_setzero_ps(),in); // Zero out 0+real 0-imag
@ -258,9 +250,9 @@ namespace Optimization {
return _mm512_mask_sub_pd(in, 0xaa,_mm512_setzero_pd(), in); return _mm512_mask_sub_pd(in, 0xaa,_mm512_setzero_pd(), in);
} }
// do not define for integer input // do not define for integer input
}; };
struct TimesMinusI{ struct TimesMinusI{
//Complex single //Complex single
inline __m512 operator()(__m512 in, __m512 ret){ inline __m512 operator()(__m512 in, __m512 ret){
__m512 tmp = _mm512_mask_sub_ps(in,0xaaaa,_mm512_setzero_ps(),in); // real -imag __m512 tmp = _mm512_mask_sub_ps(in,0xaaaa,_mm512_setzero_ps(),in); // real -imag
@ -271,11 +263,9 @@ namespace Optimization {
__m512d tmp = _mm512_mask_sub_pd(in,0xaa,_mm512_setzero_pd(),in); // real -imag __m512d tmp = _mm512_mask_sub_pd(in,0xaa,_mm512_setzero_pd(),in); // real -imag
return _mm512_swizzle_pd(tmp, _MM_SWIZ_REG_CDAB);// OK return _mm512_swizzle_pd(tmp, _MM_SWIZ_REG_CDAB);// OK
} }
};
struct TimesI{
};
struct TimesI{
//Complex single //Complex single
inline __m512 operator()(__m512 in, __m512 ret){ inline __m512 operator()(__m512 in, __m512 ret){
__m512 tmp = _mm512_swizzle_ps(in, _MM_SWIZ_REG_CDAB);// OK __m512 tmp = _mm512_swizzle_ps(in, _MM_SWIZ_REG_CDAB);// OK
@ -288,10 +278,9 @@ namespace Optimization {
} }
}; };
struct Permute{
struct Permute{
static inline __m512 Permute0(__m512 in){ static inline __m512 Permute0(__m512 in){
return _mm512_permute4f128_ps(in,(_MM_PERM_ENUM)_MM_SELECT_FOUR_FOUR(1,0,3,2)); return _mm512_permute4f128_ps(in,(_MM_PERM_ENUM)_MM_SELECT_FOUR_FOUR(1,0,3,2));
@ -319,9 +308,9 @@ namespace Optimization {
return in; return in;
}; };
}; };
struct Rotate{ struct Rotate{
static inline __m512 rotate(__m512 in,int n){ static inline __m512 rotate(__m512 in,int n){
switch(n){ switch(n){
@ -367,82 +356,76 @@ namespace Optimization {
return (__m512d)_mm512_alignr_epi32((__m512i)in,(__m512i)in,2*n); return (__m512d)_mm512_alignr_epi32((__m512i)in,(__m512i)in,2*n);
}; };
}; };
//////////////////////////////////////////////
// Some Template specialization
//Complex float Reduce
////////////////////////////////////////////// template<>
// Some Template specialization inline Grid::ComplexF Reduce<Grid::ComplexF, __m512>::operator()(__m512 in){
//Complex float Reduce
template<>
inline Grid::ComplexF Reduce<Grid::ComplexF, __m512>::operator()(__m512 in){
return Grid::ComplexF(_mm512_mask_reduce_add_ps(0x5555, in),_mm512_mask_reduce_add_ps(0xAAAA, in)); return Grid::ComplexF(_mm512_mask_reduce_add_ps(0x5555, in),_mm512_mask_reduce_add_ps(0xAAAA, in));
}
//Real float Reduce
template<>
inline Grid::RealF Reduce<Grid::RealF, __m512>::operator()(__m512 in){
return _mm512_reduce_add_ps(in);
}
//Complex double Reduce
template<>
inline Grid::ComplexD Reduce<Grid::ComplexD, __m512d>::operator()(__m512d in){
return Grid::ComplexD(_mm512_mask_reduce_add_pd(0x55, in),_mm512_mask_reduce_add_pd(0xAA, in));
}
//Real double Reduce
template<>
inline Grid::RealD Reduce<Grid::RealD, __m512d>::operator()(__m512d in){
return _mm512_reduce_add_pd(in);
}
//Integer Reduce
template<>
inline Integer Reduce<Integer, __m512i>::operator()(__m512i in){
return _mm512_reduce_add_epi32(in);
}
} }
//Real float Reduce
template<>
inline Grid::RealF Reduce<Grid::RealF, __m512>::operator()(__m512 in){
return _mm512_reduce_add_ps(in);
}
//Complex double Reduce
template<>
inline Grid::ComplexD Reduce<Grid::ComplexD, __m512d>::operator()(__m512d in){
return Grid::ComplexD(_mm512_mask_reduce_add_pd(0x55, in),_mm512_mask_reduce_add_pd(0xAA, in));
}
//Real double Reduce
template<>
inline Grid::RealD Reduce<Grid::RealD, __m512d>::operator()(__m512d in){
return _mm512_reduce_add_pd(in);
}
//Integer Reduce
template<>
inline Integer Reduce<Integer, __m512i>::operator()(__m512i in){
return _mm512_reduce_add_epi32(in);
}
NAMESPACE_END(Grid);
////////////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////////////
// Here assign types // Here assign types
typedef __m512 SIMD_Ftype; // Single precision type typedef __m512 SIMD_Ftype; // Single precision type
typedef __m512d SIMD_Dtype; // Double precision type typedef __m512d SIMD_Dtype; // Double precision type
typedef __m512i SIMD_Itype; // Integer type typedef __m512i SIMD_Itype; // Integer type
// prefecth // prefecth
inline void v_prefetch0(int size, const char *ptr){ inline void v_prefetch0(int size, const char *ptr){
for(int i=0;i<size;i+=64){ // Define L1 linesize above for(int i=0;i<size;i+=64){ // Define L1 linesize above
_mm_prefetch(ptr+i+4096,_MM_HINT_T1); _mm_prefetch(ptr+i+4096,_MM_HINT_T1);
_mm_prefetch(ptr+i+512,_MM_HINT_T0); _mm_prefetch(ptr+i+512,_MM_HINT_T0);
} }
}
inline void prefetch_HINT_T0(const char *ptr){
_mm_prefetch(ptr,_MM_HINT_T0);
}
// Function name aliases
typedef Optimization::Vsplat VsplatSIMD;
typedef Optimization::Vstore VstoreSIMD;
typedef Optimization::Vset VsetSIMD;
typedef Optimization::Vstream VstreamSIMD;
template <typename S, typename T> using ReduceSIMD = Optimization::Reduce<S,T>;
// Arithmetic operations
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;
} }
inline void prefetch_HINT_T0(const char *ptr){
_mm_prefetch(ptr,_MM_HINT_T0);
}
// Function name aliases
typedef Optimization::Vsplat VsplatSIMD;
typedef Optimization::Vstore VstoreSIMD;
typedef Optimization::Vset VsetSIMD;
typedef Optimization::Vstream VstreamSIMD;
template <typename S, typename T> using ReduceSIMD = Optimization::Reduce<S,T>;
// Arithmetic operations
typedef Optimization::Sum SumSIMD;
typedef Optimization::Sub SubSIMD;
typedef Optimization::Div DivSIMD;
typedef Optimization::Mult MultSIMD;
typedef Optimization::MultComplex MultComplexSIMD;
typedef Optimization::Conj ConjSIMD;
typedef Optimization::TimesMinusI TimesMinusISIMD;
typedef Optimization::TimesI TimesISIMD;
NAMESPACE_END(Grid);