mirror of
https://github.com/paboyle/Grid.git
synced 2025-11-04 05:54:32 +00:00
first commit for QPX intrinsics
This commit is contained in:
@@ -125,11 +125,14 @@ case ${ax_cv_cxx_compiler_vendor} in
|
||||
AC_DEFINE([AVX512],[1],[AVX512 intrinsics])
|
||||
SIMD_FLAGS='-mavx512f -mavx512pf -mavx512er -mavx512cd';;
|
||||
IMCI|KNC)
|
||||
AC_DEFINE([IMCI],[1],[IMCI Intrinsics for Knights Corner])
|
||||
AC_DEFINE([IMCI],[1],[IMCI intrinsics for Knights Corner])
|
||||
SIMD_FLAGS='';;
|
||||
GEN)
|
||||
AC_DEFINE([GENERIC_VEC],[1],[generic vector code])
|
||||
SIMD_FLAGS='';;
|
||||
QPX|BGQ)
|
||||
AC_DEFINE([QPX],[1],[QPX intrinsics for BG/Q])
|
||||
SIMD_FLAGS='';;
|
||||
*)
|
||||
AC_MSG_ERROR(["SIMD option ${ac_SIMD} not supported by the GCC/Clang compiler"]);;
|
||||
esac;;
|
||||
|
||||
@@ -1,300 +1,312 @@
|
||||
/*************************************************************************************
|
||||
/*******************************************************************************
|
||||
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
Grid physics library, www.github.com/paboyle/Grid
|
||||
|
||||
Source file: ./lib/simd/Grid_qpx.h
|
||||
Source file: ./lib/simd/Grid_qpx.h
|
||||
|
||||
Copyright (C) 2015
|
||||
Copyright (C) 2016
|
||||
|
||||
Author: neo <cossu@post.kek.jp>
|
||||
Author: Antonin Portelli <antonin.portelli@me.com>
|
||||
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
This program is free software; you can redistribute it and/or modify
|
||||
it under the terms of the GNU General Public License as published by
|
||||
the Free Software Foundation; either version 2 of the License, or
|
||||
(at your option) any later version.
|
||||
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
This program is distributed in the hope that it will be useful,
|
||||
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
GNU General Public License for more details.
|
||||
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
You should have received a copy of the GNU General Public License along
|
||||
with this program; if not, write to the Free Software Foundation, Inc.,
|
||||
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
|
||||
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
*************************************************************************************/
|
||||
/* END LEGAL */
|
||||
//----------------------------------------------------------------------
|
||||
/*! @file Grid_qpx.h
|
||||
@brief Optimization libraries for QPX instructions set for BG/Q
|
||||
|
||||
Using intrinsics
|
||||
*/
|
||||
// Time-stamp: <2015-05-27 11:30:21 neo>
|
||||
//----------------------------------------------------------------------
|
||||
|
||||
// lot of undefined functions
|
||||
See the full license in the file "LICENSE" in the top level distribution directory
|
||||
******************************************************************************/
|
||||
|
||||
namespace Grid {
|
||||
namespace Optimization {
|
||||
inline std::ostream & operator<<(std::ostream& stream, const vector4double a)
|
||||
{
|
||||
stream << "{"<<vec_extract(a,0)<<","<<vec_extract(a,1)<<","<<vec_extract(a,2)<<","<<vec_extract(a,3)<<"}";
|
||||
return stream;
|
||||
};
|
||||
|
||||
struct Vsplat{
|
||||
//Complex float
|
||||
inline float operator()(float a, float b){
|
||||
return {a,b,a,b};
|
||||
inline vector4double operator()(float a, float b){
|
||||
return (vector4double){(double)a, (double)b, (double)a, (double)b};
|
||||
}
|
||||
// Real float
|
||||
inline float operator()(float a){
|
||||
return {a,a,a,a};
|
||||
inline vector4double operator()(float a){
|
||||
return (vector4double){(double)a, (double)a, (double)a, (double)a};
|
||||
}
|
||||
//Complex double
|
||||
inline vector4double operator()(double a, double b){
|
||||
return {a,b,a,b};
|
||||
return (vector4double){a, b, a, b};
|
||||
}
|
||||
//Real double
|
||||
inline vector4double operator()(double a){
|
||||
return {a,a,a,a};
|
||||
return (vector4double){a, a, a, a};
|
||||
}
|
||||
//Integer
|
||||
inline int operator()(Integer a){
|
||||
#error
|
||||
return a;
|
||||
}
|
||||
};
|
||||
|
||||
struct Vstore{
|
||||
//Float
|
||||
inline void operator()(float a, float* F){
|
||||
assert(0);
|
||||
inline void operator()(vector4double a, float *f){
|
||||
vec_st(a, 0, f);
|
||||
}
|
||||
//Double
|
||||
inline void operator()(vector4double a, double* D){
|
||||
assert(0);
|
||||
inline void operator()(vector4double a, double *d){
|
||||
vec_st(a, 0, d);
|
||||
}
|
||||
//Integer
|
||||
inline void operator()(int a, Integer* I){
|
||||
assert(0);
|
||||
inline void operator()(int a, Integer *i){
|
||||
i[0] = a;
|
||||
}
|
||||
|
||||
};
|
||||
|
||||
|
||||
struct Vstream{
|
||||
//Float
|
||||
inline void operator()(float * a, float b){
|
||||
assert(0);
|
||||
inline void operator()(float *f, vector4double a){
|
||||
vec_st(a, 0, f);
|
||||
}
|
||||
//Double
|
||||
inline void operator()(double * a, vector4double b){
|
||||
assert(0);
|
||||
inline void operator()(double *d, vector4double a){
|
||||
vec_st(a, 0, d);
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
|
||||
|
||||
struct Vset{
|
||||
// Complex float
|
||||
inline float operator()(Grid::ComplexF *a){
|
||||
return {a[0].real(),a[0].imag(),a[1].real(),a[1].imag(),a[2].real(),a[2].imag(),a[3].real(),a[3].imag()};
|
||||
inline vector4double operator()(Grid::ComplexF *a){
|
||||
return vec_ld(0, (float *)a);
|
||||
}
|
||||
// Complex double
|
||||
inline vector4double operator()(Grid::ComplexD *a){
|
||||
return {a[0].real(),a[0].imag(),a[1].real(),a[1].imag(),a[2].real(),a[2].imag(),a[3].real(),a[3].imag()};
|
||||
return vec_ld(0, (double *)a);
|
||||
}
|
||||
// Real float
|
||||
inline float operator()(float *a){
|
||||
return {a[0],a[1],a[2],a[3],a[4],a[5],a[6],a[7]};
|
||||
inline vector4double operator()(float *a){
|
||||
return vec_ld(0, a);
|
||||
}
|
||||
// Real double
|
||||
inline vector4double operator()(double *a){
|
||||
return {a[0],a[1],a[2],a[3],a[4],a[5],a[6],a[7]};
|
||||
return vec_ld(0, a);
|
||||
}
|
||||
// Integer
|
||||
inline int operator()(Integer *a){
|
||||
#error
|
||||
return a[0];
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
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;
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
|
||||
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;
|
||||
}
|
||||
};
|
||||
|
||||
/////////////////////////////////////////////////////
|
||||
// Arithmetic operations
|
||||
/////////////////////////////////////////////////////
|
||||
struct Sum{
|
||||
//Complex/Real float
|
||||
inline float operator()(float a, float b){
|
||||
#error
|
||||
}
|
||||
//Complex/Real double
|
||||
inline vector4double operator()(vector4double a, vector4double b){
|
||||
return vec_add(a,b);
|
||||
return vec_add(a, b);
|
||||
}
|
||||
//Integer
|
||||
inline int operator()(int a, int b){
|
||||
#error
|
||||
return a + b;
|
||||
}
|
||||
};
|
||||
|
||||
struct Sub{
|
||||
//Complex/Real float
|
||||
inline float operator()(float a, float b){
|
||||
#error
|
||||
}
|
||||
//Complex/Real double
|
||||
inline vector4double operator()(vector4double a, vector4double b){
|
||||
#error
|
||||
return vec_sub(a, b);
|
||||
}
|
||||
//Integer
|
||||
inline floati operator()(int a, int b){
|
||||
#error
|
||||
inline int operator()(int a, int b){
|
||||
return a - b;
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
struct MultComplex{
|
||||
// Complex float
|
||||
inline float operator()(float a, float b){
|
||||
#error
|
||||
}
|
||||
// Complex double
|
||||
inline vector4double operator()(vector4double a, vector4double b){
|
||||
#error
|
||||
return vec_xxnpmadd(a, b, vec_xmul(b, a));
|
||||
}
|
||||
};
|
||||
|
||||
struct Mult{
|
||||
// Real float
|
||||
inline float operator()(float a, float b){
|
||||
#error
|
||||
}
|
||||
// Real double
|
||||
inline vector4double operator()(vector4double a, vector4double b){
|
||||
#error
|
||||
return vec_mul(a, b);
|
||||
}
|
||||
// Integer
|
||||
inline int operator()(int a, int b){
|
||||
#error
|
||||
return a*b;
|
||||
}
|
||||
};
|
||||
|
||||
|
||||
struct Conj{
|
||||
// Complex single
|
||||
inline float operator()(float in){
|
||||
assert(0);
|
||||
}
|
||||
// Complex double
|
||||
inline vector4double operator()(vector4double in){
|
||||
assert(0);
|
||||
inline vector4double operator()(vector4double v){
|
||||
return vec_mul(v, (vector4double){1., -1., 1., -1.});
|
||||
}
|
||||
// do not define for integer input
|
||||
};
|
||||
|
||||
struct TimesMinusI{
|
||||
//Complex single
|
||||
inline float operator()(float in, float ret){
|
||||
assert(0);
|
||||
}
|
||||
//Complex double
|
||||
inline vector4double operator()(vector4double in, vector4double ret){
|
||||
assert(0);
|
||||
inline vector4double operator()(vector4double v, vector4double ret){
|
||||
return vec_xxcpnmadd(v, (vector4double){1., 1., 1., 1.},
|
||||
(vector4double){0., 0., 0., 0.});
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
struct TimesI{
|
||||
//Complex single
|
||||
inline float operator()(float in, float ret){
|
||||
|
||||
}
|
||||
//Complex double
|
||||
inline vector4double operator()(vector4double in, vector4double ret){
|
||||
|
||||
inline vector4double operator()(vector4double v, vector4double ret){
|
||||
return vec_xxcpnmadd(v, (vector4double){-1., -1., -1., -1.},
|
||||
(vector4double){0., 0., 0., 0.});
|
||||
}
|
||||
|
||||
|
||||
};
|
||||
|
||||
struct Permute{
|
||||
static inline vector4double Permute0(vector4double v){ //0123 -> 2301
|
||||
return vec_perm(v, v, vec_gpci(02301));
|
||||
};
|
||||
static inline vector4double Permute1(vector4double v){ //0123 -> 1032
|
||||
return vec_perm(v, v, vec_gpci(01032));
|
||||
};
|
||||
static inline vector4double Permute2(vector4double v){
|
||||
return v;
|
||||
};
|
||||
static inline vector4double Permute3(vector4double v){
|
||||
return v;
|
||||
};
|
||||
};
|
||||
|
||||
|
||||
|
||||
|
||||
//////////////////////////////////////////////
|
||||
// Some Template specialization
|
||||
struct Rotate{
|
||||
static inline vector4double rotate(vector4double v, int n){
|
||||
switch(n){
|
||||
case 0:
|
||||
return v;
|
||||
break;
|
||||
case 1:
|
||||
return vec_perm(v, v, vec_gpci(01230));
|
||||
break;
|
||||
case 2:
|
||||
return vec_perm(v, v, vec_gpci(02301));
|
||||
break;
|
||||
case 3:
|
||||
return vec_perm(v, v, vec_gpci(03012));
|
||||
break;
|
||||
default: assert(0);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
//Complex float Reduce
|
||||
template<>
|
||||
inline Grid::ComplexF Reduce<Grid::ComplexF, float>::operator()(float in){
|
||||
assert(0);
|
||||
inline Grid::ComplexF
|
||||
Reduce<Grid::ComplexF, vector4double>::operator()(vector4double v) { //2 complex
|
||||
vector4double v1,v2;
|
||||
|
||||
v1 = Optimization::Permute::Permute0(v);
|
||||
v1 = vec_add(v1, v);
|
||||
|
||||
return Grid::ComplexF((float)vec_extract(v1, 0), (float)vec_extract(v1, 1));
|
||||
}
|
||||
//Real float Reduce
|
||||
template<>
|
||||
inline Grid::RealF Reduce<Grid::RealF, float>::operator()(float in){
|
||||
assert(0);
|
||||
inline Grid::RealF
|
||||
Reduce<Grid::RealF, vector4double>::operator()(vector4double v){ //4 floats
|
||||
vector4double v1,v2;
|
||||
|
||||
v1 = Optimization::Permute::Permute0(v);
|
||||
v1 = vec_add(v1, v);
|
||||
v2 = Optimization::Permute::Permute1(v1);
|
||||
v1 = vec_add(v1, v2);
|
||||
|
||||
return (float)vec_extract(v1, 0);
|
||||
}
|
||||
|
||||
|
||||
//Complex double Reduce
|
||||
template<>
|
||||
inline Grid::ComplexD Reduce<Grid::ComplexD, vector4double>::operator()(vector4double in){
|
||||
assert(0);
|
||||
inline Grid::ComplexD
|
||||
Reduce<Grid::ComplexD, vector4double>::operator()(vector4double v){ //2 complex
|
||||
vector4double v1;
|
||||
|
||||
v1 = Optimization::Permute::Permute0(v);
|
||||
v1 = vec_add(v1, v);
|
||||
|
||||
return Grid::ComplexD(vec_extract(v1, 0), vec_extract(v1, 1));
|
||||
}
|
||||
|
||||
//Real double Reduce
|
||||
template<>
|
||||
inline Grid::RealD Reduce<Grid::RealD, vector4double>::operator()(vector4double in){
|
||||
assert(0);
|
||||
inline Grid::RealD
|
||||
Reduce<Grid::RealD, vector4double>::operator()(vector4double v){ //4 doubles
|
||||
vector4double v1,v2;
|
||||
|
||||
v1 = Optimization::Permute::Permute0(v);
|
||||
v1 = vec_add(v1, v);
|
||||
v2 = Optimization::Permute::Permute1(v1);
|
||||
v1 = vec_add(v1, v2);
|
||||
|
||||
return vec_extract(v1, 0);
|
||||
}
|
||||
|
||||
//Integer Reduce
|
||||
template<>
|
||||
inline Integer Reduce<Integer, floati>::operator()(float in){
|
||||
inline Integer Reduce<Integer, int>::operator()(int in){
|
||||
// FIXME unimplemented
|
||||
printf("Reduce : Missing integer implementation -> FIX\n");
|
||||
assert(0);
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
|
||||
//////////////////////////////////////////////////////////////////////////////////////
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// Here assign types
|
||||
namespace Grid {
|
||||
typedef float SIMD_Ftype __attribute__ ((vector_size (16))); // Single precision type
|
||||
typedef vector4double SIMD_Dtype; // Double precision type
|
||||
typedef int SIMD_Itype; // Integer type
|
||||
|
||||
inline void v_prefetch0(int size, const char *ptr){};
|
||||
typedef vector4double SIMD_Ftype; // Single precision type
|
||||
typedef vector4double SIMD_Dtype; // Double precision type
|
||||
typedef int SIMD_Itype; // Integer type
|
||||
|
||||
// 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>;
|
||||
// prefetch utilities
|
||||
inline void v_prefetch0(int size, const char *ptr){};
|
||||
inline void prefetch_HINT_T0(const char *ptr){};
|
||||
|
||||
|
||||
// Arithmetic operations
|
||||
typedef Optimization::Sum SumSIMD;
|
||||
typedef Optimization::Sub SubSIMD;
|
||||
typedef Optimization::Mult MultSIMD;
|
||||
typedef Optimization::MultComplex MultComplexSIMD;
|
||||
typedef Optimization::Conj ConjSIMD;
|
||||
typedef Optimization::TimesMinusI TimesMinusISIMD;
|
||||
typedef Optimization::TimesI TimesISIMD;
|
||||
// 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::Mult MultSIMD;
|
||||
typedef Optimization::MultComplex MultComplexSIMD;
|
||||
typedef Optimization::Conj ConjSIMD;
|
||||
typedef Optimization::TimesMinusI TimesMinusISIMD;
|
||||
typedef Optimization::TimesI TimesISIMD;
|
||||
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user