portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-17 23:37:06 +01:00
Code Releases Activity

Merge branch 'develop' into feature/gpu-port

Browse Source
This commit is contained in:
Peter Boyle
2018-12-13 05:11:34 +00:00
parent adbdc4e65b c509bd3fe2
commit b57a4d32aa
647 changed files with 49155 additions and 11160 deletions
Download Patch File Download Diff File Expand all files Collapse all files

124
Grid/tensors/Tensor_Ta.h Normal file
View File

@ -0,0 +1,124 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_Ta.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_TA_H
#define GRID_MATH_TA_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////
// Ta function for scalar, vector, matrix
///////////////////////////////////////////////
/*
accelerator_inline ComplexF Ta( const ComplexF &arg){ return arg;}
accelerator_inline ComplexD Ta( const ComplexD &arg){ return arg;}
accelerator_inline RealF Ta( const RealF &arg){ return arg;}
accelerator_inline RealD Ta( const RealD &arg){ return arg;}
*/
template<class vtype> accelerator_inline iScalar<vtype> Ta(const iScalar<vtype>&r)
{
iScalar<vtype> ret;
ret._internal = Ta(r._internal);
return ret;
}
template<class vtype,int N> accelerator_inline iVector<vtype,N> Ta(const iVector<vtype,N>&r)
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = Ta(r._internal[i]);
}
return ret;
}
template<class vtype,int N> accelerator_inline iMatrix<vtype,N> Ta(const iMatrix<vtype,N> &arg)
{
iMatrix<vtype,N> ret;
double factor = (1.0/(double)N);
ret= (arg - adj(arg))*0.5;
ret=ret - (trace(ret)*factor);
return ret;
}
///////////////////////////////////////////////
// ProjectOnGroup function for scalar, vector, matrix
// Projects on orthogonal, unitary group
///////////////////////////////////////////////
template<class vtype> accelerator_inline iScalar<vtype> ProjectOnGroup(const iScalar<vtype>&r)
{
iScalar<vtype> ret;
ret._internal = ProjectOnGroup(r._internal);
return ret;
}
template<class vtype,int N> accelerator_inline iVector<vtype,N> ProjectOnGroup(const iVector<vtype,N>&r)
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = ProjectOnGroup(r._internal[i]);
}
return ret;
}
template<class vtype,int N, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0 >::type * =nullptr>
accelerator_inline iMatrix<vtype,N> ProjectOnGroup(const iMatrix<vtype,N> &arg)
{
// need a check for the group type?
iMatrix<vtype,N> ret(arg);
vtype nrm;
vtype inner;
for(int c1=0;c1<N;c1++){
zeroit(inner);
for(int c2=0;c2<N;c2++)
inner += innerProduct(ret._internal[c1][c2],ret._internal[c1][c2]);
nrm = rsqrt(inner);
for(int c2=0;c2<N;c2++)
ret._internal[c1][c2]*= nrm;
for (int b=c1+1; b<N; ++b){
decltype(ret._internal[b][b]*ret._internal[b][b]) pr;
zeroit(pr);
for(int c=0; c<N; ++c)
pr += conjugate(ret._internal[c1][c])*ret._internal[b][c];
for(int c=0; c<N; ++c){
ret._internal[b][c] -= pr * ret._internal[c1][c];
}
}
}
// assuming the determinant is ok
return ret;
}
NAMESPACE_END(Grid);
#endif

38
Grid/tensors/Tensor_arith.h Normal file
View File

@ -0,0 +1,38 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_arith.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_ARITH_H
#define GRID_MATH_ARITH_H
#include "Tensor_arith_add.h"
#include "Tensor_arith_sub.h"
#include "Tensor_arith_mac.h"
#include "Tensor_arith_mul.h"
#include "Tensor_arith_scalar.h"
#endif

142
Grid/tensors/Tensor_arith_add.h Normal file
View File

@ -0,0 +1,142 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_arith_add.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_ARITH_ADD_H
#define GRID_MATH_ARITH_ADD_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////////////////////////////////////////////////////
/////////////////////////////////////////// ADD ///////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////////////////////
// ADD is simple for now; cannot mix types and straightforward template
// Scalar +/- Scalar
// Vector +/- Vector
// Matrix +/- Matrix
template<class vtype,class ltype,class rtype> accelerator_inline void add(iScalar<vtype> * __restrict__ ret,
const iScalar<ltype> * __restrict__ lhs,
const iScalar<rtype> * __restrict__ rhs)
{
add(&ret->_internal,&lhs->_internal,&rhs->_internal);
}
template<class vtype,class ltype,class rtype,int N> accelerator_inline void add(iVector<vtype,N> * __restrict__ ret,
const iVector<ltype,N> * __restrict__ lhs,
const iVector<rtype,N> * __restrict__ rhs)
{
for(int c=0;c<N;c++){
ret->_internal[c]=lhs->_internal[c]+rhs->_internal[c];
}
return;
}
template<class vtype,class ltype,class rtype, int N> accelerator_inline void add(iMatrix<vtype,N> * __restrict__ ret,
const iMatrix<ltype,N> * __restrict__ lhs,
const iMatrix<rtype,N> * __restrict__ rhs)
{
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
add(&ret->_internal[c1][c2],&lhs->_internal[c1][c2],&rhs->_internal[c1][c2]);
}}
return;
}
template<class vtype,class ltype,class rtype, int N> accelerator_inline void add(iMatrix<vtype,N> * __restrict__ ret,
const iScalar<ltype> * __restrict__ lhs,
const iMatrix<rtype,N> * __restrict__ rhs)
{
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
if ( c1==c2)
add(&ret->_internal[c1][c2],&lhs->_internal,&rhs->_internal[c1][c2]);
else
ret->_internal[c1][c2]=lhs->_internal[c1][c2];
}}
return;
}
template<class vtype,class ltype,class rtype, int N> accelerator_inline void add(iMatrix<vtype,N> * __restrict__ ret,
const iMatrix<ltype,N> * __restrict__ lhs,
const iScalar<rtype> * __restrict__ rhs)
{
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
if ( c1==c2)
add(&ret->_internal[c1][c2],&lhs->_internal[c1][c2],&rhs->_internal);
else
ret->_internal[c1][c2]=lhs->_internal[c1][c2];
}}
return;
}
// + operator for scalar, vector, matrix
template<class ltype,class rtype>
accelerator_inline auto operator + (const iScalar<ltype>& lhs,const iScalar<rtype>& rhs) -> iScalar<decltype(lhs._internal + rhs._internal)>
{
typedef iScalar<decltype(lhs._internal+rhs._internal)> ret_t;
ret_t ret ;
add(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator + (const iVector<ltype,N>& lhs,const iVector<rtype,N>& rhs) ->iVector<decltype(lhs._internal[0]+rhs._internal[0]),N>
{
typedef iVector<decltype(lhs._internal[0]+rhs._internal[0]),N> ret_t;
ret_t ret ;
add(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator + (const iMatrix<ltype,N>& lhs,const iMatrix<rtype,N>& rhs) ->iMatrix<decltype(lhs._internal[0][0]+rhs._internal[0][0]),N>
{
typedef iMatrix<decltype(lhs._internal[0][0]+rhs._internal[0][0]),N> ret_t;
ret_t ret ;
add(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator + (const iScalar<ltype>& lhs,const iMatrix<rtype,N>& rhs)->iMatrix<decltype(lhs._internal+rhs._internal[0][0]),N>
{
typedef iMatrix<decltype(lhs._internal+rhs._internal[0][0]),N> ret_t;
ret_t ret ;
add(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator + (const iMatrix<ltype,N>& lhs,const iScalar<rtype>& rhs)->iMatrix<decltype(lhs._internal[0][0]+rhs._internal),N>
{
typedef iMatrix<decltype(lhs._internal[0][0]+rhs._internal),N> ret_t;
ret_t ret ;
add(&ret,&lhs,&rhs);
return ret;
}
NAMESPACE_END(Grid);
#endif

108
Grid/tensors/Tensor_arith_mac.h Normal file
View File

@ -0,0 +1,108 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_arith_mac.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_ARITH_MAC_H
#define GRID_MATH_ARITH_MAC_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////////////////////////////////////////////////////
/////////////////////////////////////////// MAC ///////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////////////////////
///////////////////////////
///////////////////////////
// Legal multiplication table
///////////////////////////
// scal x scal = scal
// mat x mat = mat
// mat x scal = mat
// scal x mat = mat
// mat x vec = vec
// vec x scal = vec
// scal x vec = vec
///////////////////////////
template<class rtype,class vtype,class mtype>
accelerator_inline void mac(iScalar<rtype> * __restrict__ ret,const iScalar<vtype> * __restrict__ lhs,const iScalar<mtype> * __restrict__ rhs)
{
mac(&ret->_internal,&lhs->_internal,&rhs->_internal);
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mac(iMatrix<rrtype,N> * __restrict__ ret,const iMatrix<ltype,N> * __restrict__ lhs,const iMatrix<rtype,N> * __restrict__ rhs){
for(int c3=0;c3<N;c3++){
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mac(&ret->_internal[c1][c2],&lhs->_internal[c1][c3],&rhs->_internal[c3][c2]);
}}}
return;
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mac(iMatrix<rrtype,N> * __restrict__ ret,const iMatrix<ltype,N> * __restrict__ lhs,const iScalar<rtype> * __restrict__ rhs){
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mac(&ret->_internal[c1][c2],&lhs->_internal[c1][c2],&rhs->_internal);
}}
return;
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mac(iMatrix<rrtype,N> * __restrict__ ret,const iScalar<ltype> * __restrict__ lhs,const iMatrix<rtype,N> * __restrict__ rhs){
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mac(&ret->_internal[c1][c2],&lhs->_internal,&rhs->_internal[c1][c2]);
}}
return;
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mac(iVector<rrtype,N> * __restrict__ ret,const iMatrix<ltype,N> * __restrict__ lhs,const iVector<rtype,N> * __restrict__ rhs)
{
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mac(&ret->_internal[c1],&lhs->_internal[c1][c2],&rhs->_internal[c2]);
}}
return;
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mac(iVector<rrtype,N> * __restrict__ ret,const iScalar<ltype> * __restrict__ lhs,const iVector<rtype,N> * __restrict__ rhs)
{
for(int c1=0;c1<N;c1++){
mac(&ret->_internal[c1],&lhs->_internal,&rhs->_internal[c1]);
}
return;
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mac(iVector<rrtype,N> * __restrict__ ret,const iVector<ltype,N> * __restrict__ lhs,const iScalar<rtype> * __restrict__ rhs)
{
for(int c1=0;c1<N;c1++){
mac(&ret->_internal[c1],&lhs->_internal[c1],&rhs->_internal);
}
return;
}
NAMESPACE_END(Grid);
#endif

252
Grid/tensors/Tensor_arith_mul.h Normal file
View File

@ -0,0 +1,252 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_arith_mul.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_ARITH_MUL_H
#define GRID_MATH_ARITH_MUL_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////////////////////////////////////////////////////
/////////////////////////////////////////// MUL ///////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////////////////////
template<class rtype,class vtype,class mtype>
accelerator_inline void mult(iScalar<rtype> * __restrict__ ret,const iScalar<mtype> * __restrict__ lhs,const iScalar<vtype> * __restrict__ rhs){
mult(&ret->_internal,&lhs->_internal,&rhs->_internal);
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mult(iMatrix<rrtype,N> * __restrict__ ret,const iMatrix<ltype,N> * __restrict__ lhs,const iMatrix<rtype,N> * __restrict__ rhs){
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mult(&ret->_internal[c1][c2],&lhs->_internal[c1][0],&rhs->_internal[0][c2]);
}
}
for(int c1=0;c1<N;c1++){
for(int c3=1;c3<N;c3++){
for(int c2=0;c2<N;c2++){
mac(&ret->_internal[c1][c2],&lhs->_internal[c1][c3],&rhs->_internal[c3][c2]);
}
}
}
return;
}
template<class rrtype,class ltype,class rtype,int N>
accelerator_inline void mult(iMatrix<rrtype,N> * __restrict__ ret,const iMatrix<ltype,N> * __restrict__ lhs,const iScalar<rtype> * __restrict__ rhs){
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
mult(&ret->_internal[c1][c2],&lhs->_internal[c1][c2],&rhs->_internal);
}}
return;
}
template<class rrtype,class ltype,class rtype, int N>
accelerator_inline void mult(iMatrix<rrtype,N> * __restrict__ ret,const iScalar<ltype> * __restrict__ lhs,const iMatrix<rtype,N> * __restrict__ rhs){
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
mult(&ret->_internal[c1][c2],&lhs->_internal,&rhs->_internal[c1][c2]);
}}
return;
}
// Matrix left multiplies vector
template<class rtype,class vtype,class mtype,int N>
accelerator_inline void mult(iVector<rtype,N> * __restrict__ ret,const iMatrix<mtype,N> * __restrict__ lhs,const iVector<vtype,N> * __restrict__ rhs)
{
for(int c1=0;c1<N;c1++){
mult(&ret->_internal[c1],&lhs->_internal[c1][0],&rhs->_internal[0]);
for(int c2=1;c2<N;c2++){
mac(&ret->_internal[c1],&lhs->_internal[c1][c2],&rhs->_internal[c2]);
}
}
return;
}
template<class rtype,class vtype,class mtype,int N>
accelerator_inline void mult(iVector<rtype,N> * __restrict__ ret,
const iScalar<mtype> * __restrict__ lhs,
const iVector<vtype,N> * __restrict__ rhs){
for(int c1=0;c1<N;c1++){
mult(&ret->_internal[c1],&lhs->_internal,&rhs->_internal[c1]);
}
}
template<class rtype,class vtype,class mtype,int N>
accelerator_inline void mult(iVector<rtype,N> * __restrict__ ret,
const iVector<vtype,N> * __restrict__ rhs,
const iScalar<mtype> * __restrict__ lhs){
for(int c1=0;c1<N;c1++){
mult(&ret->_internal[c1],&rhs->_internal[c1],&lhs->_internal);
}
}
template<class rtype,class vtype,class mtype,int N> accelerator_inline
iVector<rtype,N> operator * (const iMatrix<mtype,N>& lhs,const iVector<vtype,N>& rhs)
{
iVector<rtype,N> ret;
mult(&ret,&lhs,&rhs);
return ret;
}
template<class rtype,class vtype,class mtype,int N> accelerator_inline
iVector<rtype,N> operator * (const iScalar<mtype>& lhs,const iVector<vtype,N>& rhs)
{
iVector<rtype,N> ret;
mult(&ret,&lhs,&rhs);
return ret;
}
template<class rtype,class vtype,class mtype,int N> accelerator_inline
iVector<rtype,N> operator * (const iVector<mtype,N>& lhs,const iScalar<vtype>& rhs)
{
iVector<rtype,N> ret;
mult(&ret,&lhs,&rhs);
return ret;
}
//////////////////////////////////////////////////////////////////
// Divide by scalar
//////////////////////////////////////////////////////////////////
template<class rtype,class vtype> accelerator_inline
iScalar<rtype> operator / (const iScalar<rtype>& lhs,const iScalar<vtype>& rhs)
{
iScalar<rtype> ret;
ret._internal = lhs._internal/rhs._internal;
return ret;
}
template<class rtype,class vtype,int N> accelerator_inline
iVector<rtype,N> operator / (const iVector<rtype,N>& lhs,const iScalar<vtype>& rhs)
{
iVector<rtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = lhs._internal[i]/rhs._internal;
}
return ret;
}
template<class rtype,class vtype,int N> accelerator_inline
iMatrix<rtype,N> operator / (const iMatrix<rtype,N>& lhs,const iScalar<vtype>& rhs)
{
iMatrix<rtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = lhs._internal[i][j]/rhs._internal;
}}
return ret;
}
//////////////////////////////////////////////////////////////////
// Glue operators to mult routines. Must resolve return type cleverly from typeof(internal)
// since nesting matrix<scalar> x matrix<matrix>-> matrix<matrix>
// while matrix<scalar> x matrix<scalar>-> matrix<scalar>
// so return type depends on argument types in nasty way.
//////////////////////////////////////////////////////////////////
// scal x scal = scal
// mat x mat = mat
// mat x scal = mat
// scal x mat = mat
// mat x vec = vec
// vec x scal = vec
// scal x vec = vec
//
// We can special case scalar_type ??
template<class l,class r>
accelerator_inline auto operator * (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<decltype(lhs._internal * rhs._internal)>
{
typedef iScalar<decltype(lhs._internal*rhs._internal)> ret_t;
ret_t ret;
mult(&ret,&lhs,&rhs);
return ret;
}
template<class l,class r,int N> accelerator_inline
auto operator * (const iMatrix<l,N>& lhs,const iMatrix<r,N>& rhs) -> iMatrix<decltype(lhs._internal[0][0]*rhs._internal[0][0]),N>
{
typedef decltype(lhs._internal[0][0]*rhs._internal[0][0]) ret_t;
iMatrix<ret_t,N> ret;
mult(&ret,&lhs,&rhs);
return ret;
}
template<class l,class r, int N> accelerator_inline
auto operator * (const iMatrix<r,N>& lhs,const iScalar<l>& rhs) -> iMatrix<decltype(lhs._internal[0][0]*rhs._internal),N>
{
typedef decltype(lhs._internal[0][0]*rhs._internal) ret_t;
iMatrix<ret_t,N> ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mult(&ret._internal[c1][c2],&lhs._internal[c1][c2],&rhs._internal);
}}
return ret;
}
template<class l,class r,int N> accelerator_inline
auto operator * (const iScalar<l>& lhs,const iMatrix<r,N>& rhs) -> iMatrix<decltype(lhs._internal*rhs._internal[0][0]),N>
{
typedef decltype(lhs._internal*rhs._internal[0][0]) ret_t;
iMatrix<ret_t,N> ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
mult(&ret._internal[c1][c2],&lhs._internal,&rhs._internal[c1][c2]);
}}
return ret;
}
template<class l,class r,int N> accelerator_inline
auto operator * (const iMatrix<l,N>& lhs,const iVector<r,N>& rhs) -> iVector<decltype(lhs._internal[0][0]*rhs._internal[0]),N>
{
typedef decltype(lhs._internal[0][0]*rhs._internal[0]) ret_t;
iVector<ret_t,N> ret;
for(int c1=0;c1<N;c1++){
mult(&ret._internal[c1],&lhs._internal[c1][0],&rhs._internal[0]);
for(int c2=1;c2<N;c2++){
mac(&ret._internal[c1],&lhs._internal[c1][c2],&rhs._internal[c2]);
}
}
return ret;
}
template<class l,class r,int N> accelerator_inline
auto operator * (const iScalar<l>& lhs,const iVector<r,N>& rhs) -> iVector<decltype(lhs._internal*rhs._internal[0]),N>
{
typedef decltype(lhs._internal*rhs._internal[0]) ret_t;
iVector<ret_t,N> ret;
for(int c1=0;c1<N;c1++){
mult(&ret._internal[c1],&lhs._internal,&rhs._internal[c1]);
}
return ret;
}
template<class l,class r,int N> accelerator_inline
auto operator * (const iVector<l,N>& lhs,const iScalar<r>& rhs) -> iVector<decltype(lhs._internal[0]*rhs._internal),N>
{
typedef decltype(lhs._internal[0]*rhs._internal) ret_t;
iVector<ret_t,N> ret;
for(int c1=0;c1<N;c1++){
mult(&ret._internal[c1],&lhs._internal[c1],&rhs._internal);
}
return ret;
}
NAMESPACE_END(Grid);
#endif

287
Grid/tensors/Tensor_arith_scalar.h Normal file
View File

@ -0,0 +1,287 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_arith_scalar.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_ARITH_SCALAR_H
#define GRID_MATH_ARITH_SCALAR_H
NAMESPACE_BEGIN(Grid);
//////////////////////////////////////////////////////////////////////////////////////////
// Must support native C++ types Integer, Complex, Real
//////////////////////////////////////////////////////////////////////////////////////////
// multiplication by fundamental scalar type
template<class l> accelerator_inline iScalar<l> operator * (const iScalar<l>& lhs,const typename iScalar<l>::scalar_type rhs)
{
typename iScalar<l>::tensor_reduced srhs; srhs=rhs;
return lhs*srhs;
}
template<class l> accelerator_inline iScalar<l> operator * (const typename iScalar<l>::scalar_type lhs,const iScalar<l>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iVector<l,N> operator * (const iVector<l,N>& lhs,const typename iScalar<l>::scalar_type rhs)
{
typename iVector<l,N>::tensor_reduced srhs; srhs=rhs;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iVector<l,N> operator * (const typename iScalar<l>::scalar_type lhs,const iVector<l,N>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (const iMatrix<l,N>& lhs,const typename iScalar<l>::scalar_type &rhs)
{
typename iMatrix<l,N>::tensor_reduced srhs; srhs=rhs;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (const typename iScalar<l>::scalar_type & lhs,const iMatrix<l,N>& rhs) { return rhs*lhs; }
////////////////////////////////////////////////////////////////////
// Double support; cast to "scalar_type" through constructor
////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator * (const iScalar<l>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t; t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l> accelerator_inline iScalar<l> operator * (double lhs,const iScalar<l>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iVector<l,N> operator * (const iVector<l,N>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iVector<l,N> operator * (double lhs,const iVector<l,N>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (const iMatrix<l,N>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (double lhs,const iMatrix<l,N>& rhs) { return rhs*lhs; }
////////////////////////////////////////////////////////////////////
// Complex support; cast to "scalar_type" through constructor
////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator * (const iScalar<l>& lhs,ComplexD rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l> accelerator_inline iScalar<l> operator * (ComplexD lhs,const iScalar<l>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iVector<l,N> operator * (const iVector<l,N>& lhs,ComplexD rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iVector<l,N> operator * (ComplexD lhs,const iVector<l,N>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (const iMatrix<l,N>& lhs,ComplexD rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (ComplexD lhs,const iMatrix<l,N>& rhs) { return rhs*lhs; }
////////////////////////////////////////////////////////////////////
// Integer support; cast to "scalar_type" through constructor
////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator * (const iScalar<l>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t; t=rhs;
typename iScalar<l>::tensor_reduced srhs; srhs=t;
return lhs*srhs;
}
template<class l> accelerator_inline iScalar<l> operator * (Integer lhs,const iScalar<l>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iVector<l,N> operator * (const iVector<l,N>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iVector<l,N> operator * (Integer lhs,const iVector<l,N>& rhs) { return rhs*lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (const iMatrix<l,N>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs*srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator * (Integer lhs,const iMatrix<l,N>& rhs) { return rhs*lhs; }
///////////////////////////////////////////////////////////////////////////////////////////////
// addition by fundamental scalar type applies to matrix(down diag) and scalar
///////////////////////////////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator + (const iScalar<l>& lhs,const typename iScalar<l>::scalar_type rhs)
{
typename iScalar<l>::tensor_reduced srhs; srhs=rhs;
return lhs+srhs;
}
template<class l> accelerator_inline iScalar<l> operator + (const typename iScalar<l>::scalar_type lhs,const iScalar<l>& rhs) { return rhs+lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator + (const iMatrix<l,N>& lhs,const typename iScalar<l>::scalar_type rhs)
{
typename iMatrix<l,N>::tensor_reduced srhs; srhs=rhs;
return lhs+srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator + (const typename iScalar<l>::scalar_type lhs,const iMatrix<l,N>& rhs) { return rhs+lhs; }
////////////////////////////////////////////////////////////////////
// Double support; cast to "scalar_type" through constructor
////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator + (const iScalar<l>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t; t=rhs;
typename iScalar<l>::tensor_reduced srhs; srhs=t;
return lhs+srhs;
}
template<class l> accelerator_inline iScalar<l> operator + (double lhs,const iScalar<l>& rhs) { return rhs+lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator + (const iMatrix<l,N>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs+srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator + (double lhs,const iMatrix<l,N>& rhs) { return rhs+lhs; }
// Integer support cast to scalar type through constructor
template<class l> accelerator_inline iScalar<l> operator + (const iScalar<l>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t; t=rhs;
typename iScalar<l>::tensor_reduced srhs; srhs=t;
return lhs+srhs;
}
template<class l> accelerator_inline iScalar<l> operator + (Integer lhs,const iScalar<l>& rhs) { return rhs+lhs; }
template<class l,int N> accelerator_inline iMatrix<l,N> operator + (const iMatrix<l,N>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs+srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator + (Integer lhs,const iMatrix<l,N>& rhs) { return rhs+lhs; }
///////////////////////////////////////////////////////////////////////////////////////////////
// subtraction of fundamental scalar type applies to matrix(down diag) and scalar
///////////////////////////////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator - (const iScalar<l>& lhs,const typename iScalar<l>::scalar_type rhs)
{
typename iScalar<l>::tensor_reduced srhs; srhs=rhs;
return lhs-srhs;
}
template<class l> accelerator_inline iScalar<l> operator - (const typename iScalar<l>::scalar_type lhs,const iScalar<l>& rhs)
{
typename iScalar<l>::tensor_reduced slhs;slhs=lhs;
return slhs-rhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator - (const iMatrix<l,N>& lhs,const typename iScalar<l>::scalar_type rhs)
{
typename iScalar<l>::tensor_reduced srhs; srhs=rhs;
return lhs-srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator - (const typename iScalar<l>::scalar_type lhs,const iMatrix<l,N>& rhs)
{
typename iScalar<l>::tensor_reduced slhs;slhs=lhs;
return slhs-rhs;
}
////////////////////////////////////////////////////////////////////
// Double support; cast to "scalar_type" through constructor
////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator - (const iScalar<l>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t; t=rhs;
typename iScalar<l>::tensor_reduced srhs; srhs=t;
return lhs-srhs;
}
template<class l> accelerator_inline iScalar<l> operator - (double lhs,const iScalar<l>& rhs)
{
typename iScalar<l>::scalar_type t(lhs);
typename iScalar<l>::tensor_reduced slhs;slhs=t;
return slhs-rhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator - (const iMatrix<l,N>& lhs,double rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs-srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator - (double lhs,const iMatrix<l,N>& rhs)
{
typename iScalar<l>::scalar_type t(lhs);
typename iScalar<l>::tensor_reduced slhs;slhs=t;
return slhs-rhs;
}
////////////////////////////////////////////////////////////////////
// Integer support; cast to "scalar_type" through constructor
////////////////////////////////////////////////////////////////////
template<class l> accelerator_inline iScalar<l> operator - (const iScalar<l>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t; t=rhs;
typename iScalar<l>::tensor_reduced srhs; srhs=t;
return lhs-srhs;
}
template<class l> accelerator_inline iScalar<l> operator - (Integer lhs,const iScalar<l>& rhs)
{
typename iScalar<l>::scalar_type t;t=lhs;
typename iScalar<l>::tensor_reduced slhs;slhs=t;
return slhs-rhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator - (const iMatrix<l,N>& lhs,Integer rhs)
{
typename iScalar<l>::scalar_type t;t=rhs;
typename iScalar<l>::tensor_reduced srhs;srhs=t;
return lhs-srhs;
}
template<class l,int N> accelerator_inline iMatrix<l,N> operator - (Integer lhs,const iMatrix<l,N>& rhs)
{
typename iScalar<l>::scalar_type t;t=lhs;
typename iScalar<l>::tensor_reduced slhs;slhs=t;
return slhs-rhs;
}
NAMESPACE_END(Grid);
#endif

142
Grid/tensors/Tensor_arith_sub.h Normal file
View File

@ -0,0 +1,142 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_arith_sub.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_ARITH_SUB_H
#define GRID_MATH_ARITH_SUB_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////////////////////////////////////////////////////
/////////////////////////////////////////// SUB ///////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////////////////////
// SUB is simple for now; cannot mix types and straightforward template
// Scalar +/- Scalar
// Vector +/- Vector
// Matrix +/- Matrix
// Matrix /- scalar
template<class vtype,class ltype,class rtype> accelerator_inline void sub(iScalar<vtype> * __restrict__ ret,
const iScalar<ltype> * __restrict__ lhs,
const iScalar<rtype> * __restrict__ rhs)
{
sub(&ret->_internal,&lhs->_internal,&rhs->_internal);
}
template<class vtype,class ltype,class rtype,int N> accelerator_inline void sub(iVector<vtype,N> * __restrict__ ret,
const iVector<ltype,N> * __restrict__ lhs,
const iVector<rtype,N> * __restrict__ rhs)
{
for(int c=0;c<N;c++){
ret->_internal[c]=lhs->_internal[c]-rhs->_internal[c];
}
return;
}
template<class vtype,class ltype,class rtype, int N> accelerator_inline void sub(iMatrix<vtype,N> * __restrict__ ret,
const iMatrix<ltype,N> * __restrict__ lhs,
const iMatrix<rtype,N> * __restrict__ rhs){
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
sub(&ret->_internal[c1][c2],&lhs->_internal[c1][c2],&rhs->_internal[c1][c2]);
}}
return;
}
template<class vtype,class ltype,class rtype, int N> accelerator_inline void sub(iMatrix<vtype,N> * __restrict__ ret,
const iScalar<ltype> * __restrict__ lhs,
const iMatrix<rtype,N> * __restrict__ rhs){
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
if ( c1==c2) {
sub(&ret->_internal[c1][c2],&lhs->_internal,&rhs->_internal[c1][c2]);
} else {
// Fails -- need unary minus. Catalogue other unops?
ret->_internal[c1][c2]=Zero();
ret->_internal[c1][c2]=ret->_internal[c1][c2]-rhs->_internal[c1][c2];
}
}}
return;
}
template<class vtype,class ltype,class rtype, int N> accelerator_inline void sub(iMatrix<vtype,N> * __restrict__ ret,
const iMatrix<ltype,N> * __restrict__ lhs,
const iScalar<rtype> * __restrict__ rhs){
for(int c2=0;c2<N;c2++){
for(int c1=0;c1<N;c1++){
if ( c1==c2)
sub(&ret->_internal[c1][c2],&lhs->_internal[c1][c2],&rhs->_internal);
else
ret->_internal[c1][c2]=lhs->_internal[c1][c2];
}}
return;
}
// - operator for scalar, vector, matrix
template<class ltype,class rtype> accelerator_inline auto
operator - (const iScalar<ltype>& lhs, const iScalar<rtype>& rhs) -> iScalar<decltype(lhs._internal - rhs._internal)>
{
typedef iScalar<decltype(lhs._internal-rhs._internal)> ret_t;
ret_t ret;
sub(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator - (const iVector<ltype,N>& lhs,const iVector<rtype,N>& rhs) ->iVector<decltype(lhs._internal[0]-rhs._internal[0]),N>
{
typedef iVector<decltype(lhs._internal[0]-rhs._internal[0]),N> ret_t;
ret_t ret;
sub(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator - (const iMatrix<ltype,N>& lhs,const iMatrix<rtype,N>& rhs) ->iMatrix<decltype(lhs._internal[0][0]-rhs._internal[0][0]),N>
{
typedef iMatrix<decltype(lhs._internal[0][0]-rhs._internal[0][0]),N> ret_t;
ret_t ret;
sub(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator - (const iScalar<ltype>& lhs,const iMatrix<rtype,N>& rhs)->iMatrix<decltype(lhs._internal-rhs._internal[0][0]),N>
{
typedef iMatrix<decltype(lhs._internal-rhs._internal[0][0]),N> ret_t;
ret_t ret;
sub(&ret,&lhs,&rhs);
return ret;
}
template<class ltype,class rtype,int N>
accelerator_inline auto operator - (const iMatrix<ltype,N>& lhs,const iScalar<rtype>& rhs)->iMatrix<decltype(lhs._internal[0][0]-rhs._internal),N>
{
typedef iMatrix<decltype(lhs._internal[0][0]-rhs._internal),N> ret_t;
ret_t ret;
sub(&ret,&lhs,&rhs);
return ret;
}
NAMESPACE_END(Grid);
#endif

448
Grid/tensors/Tensor_class.h Normal file
View File

@ -0,0 +1,448 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_class.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_TENSORS_H
#define GRID_MATH_TENSORS_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////
// Scalar, Vector, Matrix objects.
// These can be composed to form tensor products of internal indices.
///////////////////////////////////////////////////
// It is useful to NOT have any constructors
// so that these classes assert "is_pod<class> == true"
// because then the standard C++ valarray container eliminates fill overhead on
// new allocation and
// non-move copying.
//
// However note that doing this eliminates some syntactical sugar such as
// calling the constructor explicitly or implicitly
//
class GridTensorBase {};
template <class vtype>
class iScalar {
public:
vtype _internal;
typedef vtype element;
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::vector_type vector_type;
typedef typename GridTypeMapper<vtype>::vector_typeD vector_typeD;
typedef typename GridTypeMapper<vtype>::tensor_reduced tensor_reduced_v;
typedef typename GridTypeMapper<vtype>::scalar_object recurse_scalar_object;
typedef iScalar<tensor_reduced_v> tensor_reduced;
typedef iScalar<recurse_scalar_object> scalar_object;
// substitutes a real or complex version with same tensor structure
typedef iScalar<typename GridTypeMapper<vtype>::Complexified> Complexified;
typedef iScalar<typename GridTypeMapper<vtype>::Realified> Realified;
// get double precision version
typedef iScalar<typename GridTypeMapper<vtype>::DoublePrecision> DoublePrecision;
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1 };
// Scalar no action
accelerator iScalar() = default;
friend accelerator_inline void zeroit(iScalar<vtype> &that){
zeroit(that._internal);
}
accelerator_inline iScalar(scalar_type s) : _internal(s){}; // recurse down and hit the constructor for vector_type
accelerator_inline iScalar(const Zero &z) { zeroit(*this); };
accelerator_inline iScalar<vtype> &operator=(const Zero &hero) {
zeroit(*this); return *this;
}
friend accelerator_inline void vstream(iScalar<vtype> &out, const iScalar<vtype> &in) {
vstream(out._internal, in._internal);
}
friend accelerator_inline void vbroadcast(iScalar<vtype> &out,const iScalar<vtype> &in,int lane){
vbroadcast(out._internal,in._internal,lane);
}
friend accelerator_inline void prefetch(iScalar<vtype> &that) {
prefetch(that._internal);
}
friend accelerator_inline void permute(iScalar<vtype> &out, const iScalar<vtype> &in, int permutetype) {
permute(out._internal, in._internal, permutetype);
}
friend accelerator_inline void rotate(iScalar<vtype> &out,const iScalar<vtype> &in,int rot){
rotate(out._internal,in._internal,rot);
}
friend accelerator_inline void exchange(iScalar<vtype> &out1,iScalar<vtype> &out2,
const iScalar<vtype> &in1,const iScalar<vtype> &in2,int type)
{
exchange(out1._internal,out2._internal,in1._internal, in2._internal,type);
}
// Unary negation
friend accelerator_inline iScalar<vtype> operator-(const iScalar<vtype> &r) {
iScalar<vtype> ret;
ret._internal = -r._internal;
return ret;
}
// *=,+=,-= operators inherit from corresponding "*,-,+" behaviour
accelerator_inline iScalar<vtype> &operator*=(const iScalar<vtype> &r) {
*this = (*this) * r;
return *this;
}
accelerator_inline iScalar<vtype> &operator-=(const iScalar<vtype> &r) {
*this = (*this) - r;
return *this;
}
accelerator_inline iScalar<vtype> &operator+=(const iScalar<vtype> &r) {
*this = (*this) + r;
return *this;
}
accelerator_inline vtype &operator()(void) { return _internal; }
accelerator_inline const vtype &operator()(void) const { return _internal; }
// Type casts meta programmed, must be pure scalar to match TensorRemove
template <class U = vtype, class V = scalar_type, IfComplex<V> = 0, IfNotSimd<U> = 0> accelerator_inline
operator ComplexF() const {
return (TensorRemove(_internal));
}
template <class U = vtype, class V = scalar_type, IfComplex<V> = 0, IfNotSimd<U> = 0> accelerator_inline
operator ComplexD() const {
return (TensorRemove(_internal));
}
template <class U = vtype, class V = scalar_type, IfReal<V> = 0,IfNotSimd<U> = 0> accelerator_inline
operator RealD() const {
return TensorRemove(_internal);
}
template <class U = vtype, class V = scalar_type, IfInteger<V> = 0, IfNotSimd<U> = 0> accelerator_inline
operator Integer() const {
return Integer(TensorRemove(_internal));
}
// convert from a something to a scalar via constructor of something arg
template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type * = nullptr>
accelerator_inline iScalar<vtype> operator=(T arg) {
_internal = arg;
return *this;
}
// Convert elements
template <class ttype>
accelerator_inline iScalar<vtype> operator=(iScalar<ttype> &&arg) {
_internal = arg._internal;
return *this;
}
// Host only
friend std::ostream &operator<<(std::ostream &stream,const iScalar<vtype> &o) {
stream << "S {" << o._internal << "}";
return stream;
};
};
///////////////////////////////////////////////////////////
// Allows to turn scalar<scalar<scalar<double>>>> back to double.
///////////////////////////////////////////////////////////
template <class T>
accelerator_inline typename std::enable_if<!isGridTensor<T>::value, T>::type
TensorRemove(T arg) {
return arg;
}
template <class vtype>
accelerator_inline auto TensorRemove(iScalar<vtype> arg)
-> decltype(TensorRemove(arg._internal)) {
return TensorRemove(arg._internal);
}
template <class vtype, int N>
class iVector {
public:
vtype _internal[N];
typedef vtype element;
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::vector_type vector_type;
typedef typename GridTypeMapper<vtype>::vector_typeD vector_typeD;
typedef typename GridTypeMapper<vtype>::tensor_reduced tensor_reduced_v;
typedef typename GridTypeMapper<vtype>::scalar_object recurse_scalar_object;
typedef iScalar<tensor_reduced_v> tensor_reduced;
typedef iVector<recurse_scalar_object, N> scalar_object;
// substitutes a real or complex version with same tensor structure
typedef iVector<typename GridTypeMapper<vtype>::Complexified, N> Complexified;
typedef iVector<typename GridTypeMapper<vtype>::Realified, N> Realified;
// get double precision version
typedef iVector<typename GridTypeMapper<vtype>::DoublePrecision, N> DoublePrecision;
template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type * = nullptr>
accelerator_inline auto operator=(T arg) -> iVector<vtype, N> {
zeroit(*this);
for (int i = 0; i < N; i++) _internal[i] = arg;
return *this;
}
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1 };
accelerator_inline iVector(const Zero &z) { zeroit(*this); };
accelerator iVector() = default;
accelerator_inline iVector<vtype, N> &operator=(const Zero &hero) {
zeroit(*this);
return *this;
}
friend accelerator_inline void zeroit(iVector<vtype, N> &that) {
for (int i = 0; i < N; i++) {
zeroit(that._internal[i]);
}
}
friend accelerator_inline void prefetch(iVector<vtype, N> &that) {
for (int i = 0; i < N; i++) prefetch(that._internal[i]);
}
friend accelerator_inline void vstream(iVector<vtype, N> &out, const iVector<vtype, N> &in) {
for (int i = 0; i < N; i++) {
vstream(out._internal[i], in._internal[i]);
}
}
friend accelerator_inline void vbroadcast(iVector<vtype,N> &out,const iVector<vtype,N> &in,int lane){
for(int i=0;i<N;i++){
vbroadcast(out._internal[i],in._internal[i],lane);
}
}
friend accelerator_inline void permute(iVector<vtype,N> &out,const iVector<vtype,N> &in,int permutetype){
for(int i=0;i<N;i++){
permute(out._internal[i],in._internal[i],permutetype);
}
}
friend accelerator_inline void rotate(iVector<vtype,N> &out,const iVector<vtype,N> &in,int rot){
for(int i=0;i<N;i++){
rotate(out._internal[i],in._internal[i],rot);
}
}
friend accelerator_inline void exchange(iVector<vtype,N> &out1,iVector<vtype,N> &out2,
const iVector<vtype,N> &in1,const iVector<vtype,N> &in2,int type){
for(int i=0;i<N;i++){
exchange(out1._internal[i],out2._internal[i],in1._internal[i], in2._internal[i],type);
}
}
// Unary negation
friend accelerator_inline iVector<vtype, N> operator-(const iVector<vtype, N> &r) {
iVector<vtype, N> ret;
for (int i = 0; i < N; i++) ret._internal[i] = -r._internal[i];
return ret;
}
// *=,+=,-= operators inherit from corresponding "*,-,+" behaviour
accelerator_inline iVector<vtype, N> &operator*=(const iScalar<vtype> &r) {
*this = (*this) * r;
return *this;
}
accelerator_inline iVector<vtype, N> &operator-=(const iVector<vtype, N> &r) {
*this = (*this) - r;
return *this;
}
accelerator_inline iVector<vtype, N> &operator+=(const iVector<vtype, N> &r) {
*this = (*this) + r;
return *this;
}
accelerator_inline vtype &operator()(int i) { return _internal[i]; }
accelerator_inline const vtype &operator()(int i) const { return _internal[i]; }
// Host
friend std::ostream &operator<<(std::ostream &stream, const iVector<vtype, N> &o) {
stream << "V<" << N << ">{";
for (int i = 0; i < N; i++) {
stream << o._internal[i];
if (i < N - 1) stream << ",";
}
stream << "}";
return stream;
};
};
template <class vtype, int N>
class iMatrix {
public:
vtype _internal[N][N];
typedef vtype element;
typedef typename GridTypeMapper<vtype>::scalar_type scalar_type;
typedef typename GridTypeMapper<vtype>::vector_type vector_type;
typedef typename GridTypeMapper<vtype>::vector_typeD vector_typeD;
typedef typename GridTypeMapper<vtype>::tensor_reduced tensor_reduced_v;
typedef typename GridTypeMapper<vtype>::scalar_object recurse_scalar_object;
// substitutes a real or complex version with same tensor structure
typedef iMatrix<typename GridTypeMapper<vtype>::Complexified, N> Complexified;
typedef iMatrix<typename GridTypeMapper<vtype>::Realified, N> Realified;
// get double precision version
typedef iMatrix<typename GridTypeMapper<vtype>::DoublePrecision, N> DoublePrecision;
// Tensor removal
typedef iScalar<tensor_reduced_v> tensor_reduced;
typedef iMatrix<recurse_scalar_object, N> scalar_object;
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1 };
accelerator_inline iMatrix(const Zero &z) { zeroit(*this); };
accelerator iMatrix() = default;
accelerator_inline iMatrix &operator=(const iMatrix &rhs) {
for (int i = 0; i < N; i++)
for (int j = 0; j < N; j++)
vstream(_internal[i][j], rhs._internal[i][j]);
return *this;
};
accelerator_inline iMatrix(scalar_type s) {
(*this) = s;
}; // recurse down and hit the constructor for vector_type
accelerator_inline iMatrix<vtype, N> &operator=(const Zero &hero) {
zeroit(*this);
return *this;
}
template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type * = nullptr>
accelerator_inline auto operator=(T arg) -> iMatrix<vtype, N> {
zeroit(*this);
for (int i = 0; i < N; i++) _internal[i][i] = arg;
return *this;
}
friend accelerator_inline void zeroit(iMatrix<vtype,N> &that){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
zeroit(that._internal[i][j]);
}}
}
friend accelerator_inline void prefetch(iMatrix<vtype,N> &that){
for(int i=0;i<N;i++) {
for(int j=0;j<N;j++) {
prefetch(that._internal[i][j]);
}}
}
friend accelerator_inline void vstream(iMatrix<vtype,N> &out,const iMatrix<vtype,N> &in){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
vstream(out._internal[i][j],in._internal[i][j]);
}}
}
friend accelerator_inline void vbroadcast(iMatrix<vtype,N> &out,const iMatrix<vtype,N> &in,int lane){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
vbroadcast(out._internal[i][j],in._internal[i][j],lane);
}}
}
friend accelerator_inline void permute(iMatrix<vtype,N> &out,const iMatrix<vtype,N> &in,int permutetype){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
permute(out._internal[i][j],in._internal[i][j],permutetype);
}}
}
friend accelerator_inline void rotate(iMatrix<vtype,N> &out,const iMatrix<vtype,N> &in,int rot){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
rotate(out._internal[i][j],in._internal[i][j],rot);
}}
}
friend accelerator_inline void exchange(iMatrix<vtype,N> &out1,iMatrix<vtype,N> &out2,
const iMatrix<vtype,N> &in1,const iMatrix<vtype,N> &in2,int type){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
exchange(out1._internal[i][j],out2._internal[i][j],in1._internal[i][j], in2._internal[i][j],type);
}}
}
// Unary negation
friend accelerator_inline iMatrix<vtype, N> operator-(const iMatrix<vtype, N> &r) {
iMatrix<vtype, N> ret;
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
ret._internal[i][j] = -r._internal[i][j];
}}
return ret;
}
// *=,+=,-= operators inherit from corresponding "*,-,+" behaviour
template <class T>
accelerator_inline iMatrix<vtype, N> &operator*=(const T &r) {
*this = (*this) * r;
return *this;
}
template <class T>
accelerator_inline iMatrix<vtype, N> &operator-=(const T &r) {
*this = (*this) - r;
return *this;
}
template <class T>
accelerator_inline iMatrix<vtype, N> &operator+=(const T &r) {
*this = (*this) + r;
return *this;
}
// returns an lvalue reference
accelerator_inline vtype &operator()(int i, int j) { return _internal[i][j]; }
accelerator_inline const vtype &operator()(int i, int j) const {
return _internal[i][j];
}
// Host function only
friend std::ostream &operator<<(std::ostream &stream, const iMatrix<vtype, N> &o) {
stream << "M<" << N << ">{";
for (int i = 0; i < N; i++) {
stream << "{";
for (int j = 0; j < N; j++) {
stream << o._internal[i][j];
if (i < N - 1) stream << ",";
}
stream << "}";
if (i != N - 1) stream << "\n\t\t";
}
stream << "}";
return stream;
};
};
template <class v> accelerator_inline
void vprefetch(const iScalar<v> &vv) {
vprefetch(vv._internal);
}
template <class v, int N> accelerator_inline
void vprefetch(const iVector<v, N> &vv) {
for (int i = 0; i < N; i++) {
vprefetch(vv._internal[i]);
}
}
template <class v, int N> accelerator_inline
void vprefetch(const iMatrix<v, N> &vv) {
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
vprefetch(vv._internal[i][j]);
}
}
}
NAMESPACE_END(Grid);
#endif

73
Grid/tensors/Tensor_determinant.h Normal file
View File

@ -0,0 +1,73 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_determinant.h
Copyright (C) 2015
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_DET_H
#define GRID_MATH_DET_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////
// Determinant function for scalar, vector, matrix
///////////////////////////////////////////////
accelerator_inline ComplexF Determinant( const ComplexF &arg){ return arg;}
accelerator_inline ComplexD Determinant( const ComplexD &arg){ return arg;}
accelerator_inline RealF Determinant( const RealF &arg){ return arg;}
accelerator_inline RealD Determinant( const RealD &arg){ return arg;}
template<class vtype> accelerator_inline auto Determinant(const iScalar<vtype>&r) -> iScalar<decltype(Determinant(r._internal))>
{
iScalar<decltype(Determinant(r._internal))> ret;
ret._internal = Determinant(r._internal);
return ret;
}
template<class vtype,int N, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0 >::type * =nullptr>
accelerator_inline iScalar<vtype> Determinant(const iMatrix<vtype,N> &arg)
{
iMatrix<vtype,N> ret(arg);
iScalar<vtype> det = vtype(1.0);
/* Conversion of matrix to upper triangular */
for(int i = 0; i < N; i++){
for(int j = 0; j < N; j++){
if(j>i){
vtype ratio = ret._internal[j][i]/ret._internal[i][i];
for(int k = 0; k < N; k++){
ret._internal[j][k] -= ratio * ret._internal[i][k];
}
}
}
}
for(int i = 0; i < N; i++)
det *= ret._internal[i][i];
return det;
}
NAMESPACE_END(Grid);
#endif

141
Grid/tensors/Tensor_exp.h Normal file
View File

@ -0,0 +1,141 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_exp.h
Copyright (C) 2015
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_EXP_H
#define GRID_MATH_EXP_H
#define DEFAULT_MAT_EXP 12
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////
// Exponentiate function for scalar, vector, matrix
///////////////////////////////////////////////
template<class vtype> accelerator_inline iScalar<vtype> Exponentiate(const iScalar<vtype>&r, RealD alpha , Integer Nexp = DEFAULT_MAT_EXP)
{
iScalar<vtype> ret;
ret._internal = Exponentiate(r._internal, alpha, Nexp);
return ret;
}
template<class vtype, int N> accelerator_inline iVector<vtype, N> Exponentiate(const iVector<vtype,N>&r, RealD alpha , Integer Nexp = DEFAULT_MAT_EXP)
{
iVector<vtype, N> ret;
for (int i = 0; i < N; i++)
ret._internal[i] = Exponentiate(r._internal[i], alpha, Nexp);
return ret;
}
// Specialisation: Cayley-Hamilton exponential for SU(3)
template<class vtype, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0>::type * =nullptr>
accelerator_inline iMatrix<vtype,3> Exponentiate(const iMatrix<vtype,3> &arg, RealD alpha , Integer Nexp = DEFAULT_MAT_EXP )
{
// for SU(3) 2x faster than the std implementation using Nexp=12
// notice that it actually computes
// exp ( input matrix )
// the i sign is coming from outside
// input matrix is anti-hermitian NOT hermitian
typedef iMatrix<vtype,3> mat;
typedef iScalar<vtype> scalar;
mat unit(1.0);
const Complex one_over_three = 1.0 / 3.0;
const Complex one_over_two = 1.0 / 2.0;
scalar c0, c1, tmp, c0max, theta, u, w;
scalar xi0, u2, w2, cosw;
scalar fden, h0, h1, h2;
scalar e2iu, emiu, ixi0;
scalar f0, f1, f2;
scalar unity(1.0);
mat iQ2 = arg*arg*alpha*alpha;
mat iQ3 = arg*iQ2*alpha;
// sign in c0 from the conventions on the Ta
scalar imQ3, reQ2;
imQ3 = imag( trace(iQ3) );
reQ2 = real( trace(iQ2) );
c0 = -imQ3 * one_over_three;
c1 = -reQ2 * one_over_two;
// Cayley Hamilton checks to machine precision, tested
tmp = c1 * one_over_three;
c0max = 2.0 * pow(tmp, 1.5);
theta = acos(c0 / c0max) * one_over_three;
u = sqrt(tmp) * cos(theta);
w = sqrt(c1) * sin(theta);
xi0 = sin(w) / w;
u2 = u * u;
w2 = w * w;
cosw = cos(w);
ixi0 = timesI(xi0);
emiu = cos(u) - timesI(sin(u));
e2iu = cos(2.0 * u) + timesI(sin(2.0 * u));
h0 = e2iu * (u2 - w2) +
emiu * ((8.0 * u2 * cosw) + (2.0 * u * (3.0 * u2 + w2) * ixi0));
h1 = e2iu * (2.0 * u) - emiu * ((2.0 * u * cosw) - (3.0 * u2 - w2) * ixi0);
h2 = e2iu - emiu * (cosw + (3.0 * u) * ixi0);
fden = unity / (9.0 * u2 - w2); // reals
f0 = h0 * fden;
f1 = h1 * fden;
f2 = h2 * fden;
return (f0 * unit + timesMinusI(f1) * arg*alpha - f2 * iQ2);
}
// General exponential
template<class vtype,int N, typename std::enable_if< GridTypeMapper<vtype>::TensorLevel == 0 >::type * =nullptr>
accelerator_inline iMatrix<vtype,N> Exponentiate(const iMatrix<vtype,N> &arg, RealD alpha , Integer Nexp = DEFAULT_MAT_EXP )
{
// notice that it actually computes
// exp ( input matrix )
// the i sign is coming from outside
// input matrix is anti-hermitian NOT hermitian
typedef iMatrix<vtype,N> mat;
mat unit(1.0);
mat temp(unit);
for(int i=Nexp; i>=1;--i){
temp *= alpha/RealD(i);
temp = unit + temp*arg;
}
return temp;
}
NAMESPACE_END(Grid);
#endif

212
Grid/tensors/Tensor_extract_merge.h Normal file
View File

@ -0,0 +1,212 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_extract_merge.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Christopher Kelly <ckelly@phys.columbia.edu>
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.
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 */
#pragma once
#include <string.h>
//#pragma GCC optimize("no-strict-aliasing")
NAMESPACE_BEGIN(Grid);
/////////////////////////////////////////////////////////////////
// Generic extract/merge/permute
/////////////////////////////////////////////////////////////////
template<class __T> using ExtractPointerArray = AcceleratorVector<__T *,GRID_MAX_SIMD>;
template<class __T> using ExtractBuffer = AcceleratorVector<__T ,GRID_MAX_SIMD>;
//void extract(const vobj &vec,ExtractBuffer<typename vobj::scalar_object> &extracted);
//void extract(const vobj &vec,ExtractPointerArray<sobj> &extracted, int offset);
//void merge(vobj &vec,ExtractBuffer<typename vobj::scalar_object> &extracted)
//void merge(vobj &vec,ExtractPointerArray<typename vobj::scalar_object> &extracted)
////////////////////////////////////////////////////////////////////////
// Extract to contiguous array scalar object
////////////////////////////////////////////////////////////////////////
template<class vobj,class sobj> accelerator
void extract(const vobj &vec,ExtractBuffer<sobj> &extracted)
{
typedef typename GridTypeMapper<sobj>::scalar_type sobj_scalar_type;
typedef typename GridTypeMapper<vobj>::scalar_type scalar_type;
typedef typename GridTypeMapper<vobj>::vector_type vector_type;
const int words=sizeof(vobj)/sizeof(vector_type);
const int Nsimd=vector_type::Nsimd();
const int Nextr=extracted.size();
const int s=Nsimd/Nextr;
sobj_scalar_type *sp = (sobj_scalar_type *) &extracted[0];
scalar_type *vp = (scalar_type *)&vec;
scalar_type vtmp;
sobj_scalar_type stmp;
for(int w=0;w<words;w++){
for(int i=0;i<Nextr;i++){
memcpy((char *)&vtmp,(char *)&vp[w*Nsimd+i*s],sizeof(vtmp));
stmp = vtmp;
memcpy((char *)&sp[i*words+w],(char *)&stmp,sizeof(stmp));
}
}
return;
}
////////////////////////////////////////////////////////////////////////
// Merge a contiguous array of scalar objects
////////////////////////////////////////////////////////////////////////
template<class vobj,class sobj> accelerator
void merge(vobj &vec,ExtractBuffer<sobj> &extracted)
{
typedef typename GridTypeMapper<sobj>::scalar_type sobj_scalar_type;
typedef typename GridTypeMapper<vobj>::scalar_type scalar_type;
typedef typename GridTypeMapper<vobj>::vector_type vector_type;
const int words=sizeof(vobj)/sizeof(vector_type);
const int Nsimd=vector_type::Nsimd();
const int Nextr = extracted.size();
const int s=Nsimd/Nextr;
sobj_scalar_type *sp = (sobj_scalar_type *)&extracted[0];
scalar_type *vp = (scalar_type *)&vec;
scalar_type vtmp;
sobj_scalar_type stmp;
for(int w=0;w<words;w++){
for(int i=0;i<Nextr;i++){
for(int ii=0;ii<s;ii++){
memcpy((char *)&stmp,(char *)&sp[i*words+w],sizeof(stmp));
vtmp = stmp;
memcpy((char *)&vp[w*Nsimd+i*s+ii],(char *)&vtmp,sizeof(vtmp));
}
}
}
}
////////////////////////////////////////////////////////////////////////
// Extract/Insert a single lane
////////////////////////////////////////////////////////////////////////
template<class vobj> accelerator_inline
typename vobj::scalar_object extractLane(int lane, const vobj & __restrict__ vec)
{
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::scalar_object scalar_object;
typedef typename vobj::vector_type vector_type;
typedef typename ExtractTypeMap<scalar_type>::extract_type extract_type;
typedef extract_type * pointer;
constexpr int words=sizeof(vobj)/sizeof(vector_type);
constexpr int Nsimd=vector_type::Nsimd();
scalar_object extracted;
pointer __restrict__ sp = (pointer)&extracted; // Type pun
pointer __restrict__ vp = (pointer)&vec;
for(int w=0;w<words;w++){
sp[w]=vp[w*Nsimd+lane];
}
return extracted;
}
template<class vobj> accelerator_inline
void insertLane(int lane, vobj & __restrict__ vec,const typename vobj::scalar_object & __restrict__ extracted)
{
typedef typename vobj::vector_type vector_type;
typedef typename vector_type::scalar_type scalar_type;
typedef typename ExtractTypeMap<scalar_type>::extract_type extract_type;
typedef extract_type * pointer;
constexpr int words=sizeof(vobj)/sizeof(vector_type);
constexpr int Nsimd=vector_type::Nsimd();
pointer __restrict__ sp = (pointer)&extracted;
pointer __restrict__ vp = (pointer)&vec;
for(int w=0;w<words;w++){
vp[w*Nsimd+lane]=sp[w];
}
}
////////////////////////////////////////////////////////////////////////
// Extract to a bunch of scalar object pointers of different scalar type, with offset. Useful for precision change
////////////////////////////////////////////////////////////////////////
template<class vobj, class sobj> accelerator
void extract(const vobj &vec,ExtractPointerArray<sobj> &extracted, int offset)
{
typedef typename GridTypeMapper<sobj>::scalar_type sobj_scalar_type;
typedef typename GridTypeMapper<vobj>::scalar_type scalar_type;
typedef typename GridTypeMapper<vobj>::vector_type vector_type;
const int words=sizeof(vobj)/sizeof(vector_type);
const int Nsimd=vector_type::Nsimd();
const int Nextr=extracted.size();
const int s = Nsimd/Nextr;
scalar_type * vp = (scalar_type *)&vec;
scalar_type vtmp;
sobj_scalar_type stmp;
for(int w=0;w<words;w++){
for(int i=0;i<Nextr;i++){
sobj_scalar_type * pointer = (sobj_scalar_type *)& extracted[i][offset];
memcpy((char *)&vtmp,(char *)&vp[w*Nsimd+i*s],sizeof(vtmp));
stmp = vtmp;
memcpy((char *)&pointer[w],(char *)&stmp,sizeof(stmp)); // may do a precision conversion
}
}
}
////////////////////////////////////////////////////////////////////////
// Merge bunch of scalar object pointers of different scalar type, with offset. Useful for precision change
////////////////////////////////////////////////////////////////////////
template<class vobj, class sobj> accelerator
void merge(vobj &vec,ExtractPointerArray<sobj> &extracted, int offset)
{
typedef typename GridTypeMapper<sobj>::scalar_type sobj_scalar_type;
typedef typename GridTypeMapper<vobj>::scalar_type scalar_type;
typedef typename GridTypeMapper<vobj>::vector_type vector_type;
const int words=sizeof(vobj)/sizeof(vector_type);
const int Nsimd=vector_type::Nsimd();
const int Nextr=extracted.size();
const int s = Nsimd/Nextr;
scalar_type * vp = (scalar_type *)&vec;
scalar_type vtmp;
sobj_scalar_type stmp;
for(int w=0;w<words;w++){
for(int i=0;i<Nextr;i++){
sobj_scalar_type * pointer = (sobj_scalar_type *)& extracted[i][offset];
for(int ii=0;ii<s;ii++){
memcpy((char *)&stmp,(char *)&pointer[w],sizeof(stmp));
vtmp=stmp;
memcpy((char *)&vp[w*Nsimd+i*s+ii],(char *)&vtmp,sizeof(vtmp));
}
}
}
}
NAMESPACE_END(Grid);

411
Grid/tensors/Tensor_index.h Normal file
View File

@ -0,0 +1,411 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_index.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_TENSOR_INDEX_H
#define GRID_TENSOR_INDEX_H
////////////////////////////////////////////////////////////////////////////////////////
// Recursion for trace, transpose, peek, poke a specific index
////////////////////////////////////////////////////////////////////////////////////////
// Allow trace to recurse if vector, but never terminate on a vector
// trace of a different index can distribute across the vector index in a replicated way
// but we do not trace a vector index.
NAMESPACE_BEGIN(Grid);
/* Needed?
template<int Level> accelerator_inline ComplexF traceIndex(const ComplexF arg) { return arg;}
template<int Level> accelerator_inline ComplexD traceIndex(const ComplexD arg) { return arg;}
template<int Level> accelerator_inline RealF traceIndex(const RealF arg) { return arg;}
template<int Level> accelerator_inline RealD traceIndex(const RealD arg) { return arg;}
*/
template<int Level>
class TensorIndexRecursion {
public:
////////////////////////////////////////////////////
// Type Queries
////////////////////////////////////////////////////
template<class vtype> static accelerator_inline int indexRank(const iScalar<vtype> tmp) { return TensorIndexRecursion<Level-1>::indexRank(tmp._internal); }
template<class vtype,int N> static accelerator_inline int indexRank(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::indexRank(tmp._internal[0]); }
template<class vtype,int N> static accelerator_inline int indexRank(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::indexRank(tmp._internal[0][0]); }
template<class vtype> static accelerator_inline int isScalar(const iScalar<vtype> tmp) { return TensorIndexRecursion<Level-1>::isScalar(tmp._internal); }
template<class vtype,int N> static accelerator_inline int isScalar(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isScalar(tmp._internal[0]); }
template<class vtype,int N> static accelerator_inline int isScalar(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isScalar(tmp._internal[0][0]); }
template<class vtype> static accelerator_inline int isVector(const iScalar<vtype> tmp) { return TensorIndexRecursion<Level-1>::isVector(tmp._internal); }
template<class vtype,int N> static accelerator_inline int isVector(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isVector(tmp._internal[0]); }
template<class vtype,int N> static accelerator_inline int isVector(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isVector(tmp._internal[0][0]); }
template<class vtype> static accelerator_inline int isMatrix(const iScalar<vtype> tmp) { return TensorIndexRecursion<Level-1>::isMatrix(tmp._internal); }
template<class vtype,int N> static accelerator_inline int isMatrix(const iVector<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isMatrix(tmp._internal[0]); }
template<class vtype,int N> static accelerator_inline int isMatrix(const iMatrix<vtype,N> tmp){ return TensorIndexRecursion<Level-1>::isMatrix(tmp._internal[0][0]); }
////////////////////////////////////////////////////
// Trace
////////////////////////////////////////////////////
template<class vtype>
static accelerator_inline auto traceIndex(const iScalar<vtype> arg) -> iScalar<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal))>
{
iScalar<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal))> ret;
ret._internal = TensorIndexRecursion<Level-1>::traceIndex(arg._internal);
return ret;
}
template<class vtype,int N>
static accelerator_inline auto traceIndex(const iVector<vtype,N> arg) -> iVector<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal[0])),N>
{
iVector<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal[0])),N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = TensorIndexRecursion<Level-1>::traceIndex(arg._internal[i]);
}
return ret;
}
template<class vtype,int N>
static accelerator_inline auto traceIndex(const iMatrix<vtype,N> arg) -> iMatrix<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal[0][0])),N>
{
iMatrix<decltype(TensorIndexRecursion<Level-1>::traceIndex(arg._internal[0][0])),N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = TensorIndexRecursion<Level-1>::traceIndex(arg._internal[i][j]);
}}
return ret;
}
////////////////////////////////////////////
// Recursion for peeking a specific index
////////////////////////////////////////////
template<class vtype>
static accelerator_inline auto peekIndex(const iScalar<vtype> arg,int i) -> iScalar<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal,0))>
{
iScalar<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal,0))> ret;
ret._internal = TensorIndexRecursion<Level-1>::peekIndex(arg._internal,i);
return ret;
}
template<class vtype>
static accelerator_inline auto peekIndex(const iScalar<vtype> arg,int i,int j) -> iScalar<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal,0,0))>
{
iScalar<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal,0,0))> ret;
ret._internal = TensorIndexRecursion<Level-1>::peekIndex(arg._internal,i,j);
return ret;
}
template<class vtype,int N>
static accelerator_inline auto peekIndex(const iVector<vtype,N> arg,int ii) -> iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0],0)),N>
{
iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0],0)),N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = TensorIndexRecursion<Level-1>::peekIndex(arg._internal[i],ii);
}
return ret;
}
template<class vtype,int N>
static accelerator_inline auto peekIndex(const iVector<vtype,N> arg,int ii,int jj)
-> iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0],0,0)),N>
{
iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0],0,0)),N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = TensorIndexRecursion<Level-1>::peekIndex(arg._internal[i],ii,jj);
}
return ret;
}
template<class vtype,int N>
static accelerator_inline auto peekIndex(const iMatrix<vtype,N> arg,int ii) -> iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0][0],0)),N>
{
iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0][0],0)),N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = TensorIndexRecursion<Level-1>::peekIndex(arg._internal[i][j],ii);
}}
return ret;
}
template<class vtype,int N>
static accelerator_inline auto peekIndex(const iMatrix<vtype,N> arg,int ii,int jj)
-> iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0][0],0,0)),N>
{
iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(arg._internal[0][0],0,0)),N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = TensorIndexRecursion<Level-1>::peekIndex(arg._internal[i][j],ii,jj);
}}
return ret;
}
////////////////////////////////////////////
// Recursion for poking a specific index
////////////////////////////////////////////
template<class vtype> accelerator_inline static
void pokeIndex(iScalar<vtype> &ret, const iScalar<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal,0))> &arg, int i)
{
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal,arg._internal,i);
}
template<class vtype> accelerator_inline static
void pokeIndex(iScalar<vtype> &ret, const iScalar<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal,0,0))> &arg, int i,int j)
{
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal,arg._internal,i,j);
}
template<class vtype,int N> accelerator_inline static
void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0],0)),N> &arg, int i)
{
for(int ii=0;ii<N;ii++){
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal[ii],arg._internal[ii],i);
}
}
template<class vtype,int N> accelerator_inline static
void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0],0,0)),N> &arg, int i,int j)
{
for(int ii=0;ii<N;ii++){
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal[ii],arg._internal[ii],i,j);
}
}
template<class vtype,int N> accelerator_inline static
void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0][0],0)),N> &arg, int i)
{
for(int ii=0;ii<N;ii++){
for(int jj=0;jj<N;jj++){
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal[ii][jj],arg._internal[ii][jj],i);
}}
}
template<class vtype,int N> accelerator_inline static
void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0][0],0,0)),N> &arg, int i,int j)
{
for(int ii=0;ii<N;ii++){
for(int jj=0;jj<N;jj++){
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal[ii][jj],arg._internal[ii][jj],i,j);
}}
}
////////////////////////////////////////////
// Recursion for transposing a specific index
////////////////////////////////////////////
template<class vtype> accelerator_inline
static auto transposeIndex(const iScalar<vtype> arg) -> iScalar<vtype>
{
iScalar<vtype> ret;
ret._internal = TensorIndexRecursion<Level-1>::transposeIndex(arg._internal);
return ret;
}
template<class vtype,int N> accelerator_inline
static auto transposeIndex(const iVector<vtype,N> arg) -> iVector<vtype,N>
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = TensorIndexRecursion<Level-1>::transposeIndex(arg._internal[i]);
}
return ret;
}
template<class vtype,int N> accelerator_inline
static auto transposeIndex(const iMatrix<vtype,N> arg) -> iMatrix<vtype,N>
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = TensorIndexRecursion<Level-1>::transposeIndex(arg._internal[i][j]);
}}
return ret;
}
};
////////////////////////////
// strip const & ref quali's
////////////////////////////
#define RemoveCRV(a) typename std::remove_const<typename std::remove_reference<decltype(a)>::type>::type
template<>
class TensorIndexRecursion<0> {
public:
////////////////////////////////////////////////////
// Type Queries
////////////////////////////////////////////////////
template<class vtype> static accelerator_inline int indexRank(const iScalar<vtype> tmp) { return 1; }
template<class vtype,int N> static accelerator_inline int indexRank(const iVector<vtype,N> tmp){ return N; }
template<class vtype,int N> static accelerator_inline int indexRank(const iMatrix<vtype,N> tmp){ return N; }
template<class vtype> static accelerator_inline int isScalar(const iScalar<vtype> tmp) { return true;}
template<class vtype,int N> static accelerator_inline int isScalar(const iVector<vtype,N> tmp){ return false;}
template<class vtype,int N> static accelerator_inline int isScalar(const iMatrix<vtype,N> tmp){ return false;}
template<class vtype> static accelerator_inline int isVector(const iScalar<vtype> tmp) { return false;}
template<class vtype,int N> static accelerator_inline int isVector(const iVector<vtype,N> tmp){ return true;}
template<class vtype,int N> static accelerator_inline int isVector(const iMatrix<vtype,N> tmp){ return false;}
template<class vtype> static accelerator_inline int isMatrix(const iScalar<vtype> tmp) { return false;}
template<class vtype,int N> static accelerator_inline int isMatrix(const iVector<vtype,N> tmp){ return false;}
template<class vtype,int N> static accelerator_inline int isMatrix(const iMatrix<vtype,N> tmp){ return true;}
/////////////////////////////////////////
// Ends recursion for trace (scalar/vector/matrix)
/////////////////////////////////////////
template<class vtype> accelerator_inline
static auto traceIndex(const iScalar<vtype> arg) -> iScalar<RemoveCRV(arg._internal)>
{
iScalar<RemoveCRV(arg._internal)> ret;
ret._internal = arg._internal;
return ret;
}
template<class vtype,int N> accelerator_inline
static auto traceIndex(const iVector<vtype,N> arg) -> iScalar<RemoveCRV(arg._internal[0])>
{
iScalar<RemoveCRV(arg._internal[0])> ret;
ret._internal=Zero();
for(int i=0;i<N;i++){
ret._internal = ret._internal+ arg._internal[i];
}
return ret;
}
template<class vtype,int N> accelerator_inline
static auto traceIndex(const iMatrix<vtype,N> arg) -> iScalar<RemoveCRV(arg._internal[0][0])>
{
iScalar<RemoveCRV(arg._internal[0][0])> ret;
zeroit(ret);
for(int i=0;i<N;i++){
ret._internal = ret._internal+arg._internal[i][i];
}
return ret;
}
/////////////////////////////////////////
// Ends recursion for transpose scalar/matrix ; no way to terminate on vector
/////////////////////////////////////////
template<class vtype> accelerator_inline
static auto transposeIndex(const iScalar<vtype> arg) -> iScalar<vtype>
{
iScalar<vtype> ret;
ret._internal = arg._internal;
return ret;
}
template<class vtype,int N> accelerator_inline
static auto transposeIndex(const iMatrix<vtype,N> arg) -> iMatrix<vtype,N>
{
iMatrix<vtype,N> ret;
ret=Zero();
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = ret._internal[i][j]+arg._internal[i][j];
}}
return ret;
}
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// End recursion for peeking a specific index; single index on vector, double index on matrix
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vtype,int N> accelerator_inline
static auto peekIndex(const iVector<vtype,N> arg,int ii) -> iScalar<vtype>
{
iScalar<vtype> ret;
ret._internal = arg._internal[ii];
return ret;
}
template<class vtype,int N> accelerator_inline
static auto peekIndex(const iMatrix<vtype,N> arg,int ii,int jj) -> iScalar<vtype>
{
iScalar<vtype> ret;
ret._internal = arg._internal[ii][jj];
return ret;
}
// Vector poke, one index
template<class vtype,int N> accelerator_inline static
void pokeIndex(iVector<vtype,N> &ret, const iScalar<vtype> &arg,int i)
{
ret._internal[i] = arg._internal;
}
// Matrix poke two indices
template<class vtype,int N> accelerator_inline static
void pokeIndex(iMatrix<vtype,N> &ret, const iScalar<vtype> &arg,int i,int j)
{
ret._internal[i][j] = arg._internal;
}
};
////////////////////////////////////////////////////////////////////////////////////////////////////////
// External wrappers
////////////////////////////////////////////////////////////////////////////////////////////////////////
template<int Level,class vtype> accelerator_inline int indexRank(void)
{
vtype tmp;
return TensorIndexRecursion<Level>::indexRank(tmp);
}
template<int Level,class vtype> accelerator_inline int isScalar(void)
{
vtype tmp;
return TensorIndexRecursion<Level>::isScalar(tmp);
}
template<int Level,class vtype> accelerator_inline int isVector(void)
{
vtype tmp;
return TensorIndexRecursion<Level>::isVector(tmp);
}
template<int Level,class vtype> accelerator_inline int isMatrix(void)
{
vtype tmp;
return TensorIndexRecursion<Level>::isMatrix(tmp);
}
template<int Level,class vtype> accelerator_inline auto traceIndex (const vtype &arg) -> RemoveCRV(TensorIndexRecursion<Level>::traceIndex(arg))
{
RemoveCRV(TensorIndexRecursion<Level>::traceIndex(arg)) ret;
ret=TensorIndexRecursion<Level>::traceIndex(arg);
return ret;
}
template<int Level,class vtype> accelerator_inline auto transposeIndex (const vtype &arg) -> RemoveCRV(TensorIndexRecursion<Level>::transposeIndex(arg))
{
RemoveCRV(TensorIndexRecursion<Level>::transposeIndex(arg)) ret;
ret=TensorIndexRecursion<Level>::transposeIndex(arg);
return ret;
}
template<int Level,class vtype> accelerator_inline auto peekIndex (const vtype &arg,int i) -> RemoveCRV(TensorIndexRecursion<Level>::peekIndex(arg,0))
{
RemoveCRV(TensorIndexRecursion<Level>::peekIndex(arg,0)) ret;
ret=TensorIndexRecursion<Level>::peekIndex(arg,i);
return ret;
}
template<int Level,class vtype> accelerator_inline auto peekIndex (const vtype &arg,int i,int j) -> RemoveCRV(TensorIndexRecursion<Level>::peekIndex(arg,0,0))
{
RemoveCRV(TensorIndexRecursion<Level>::peekIndex(arg,0,0)) ret;
ret=TensorIndexRecursion<Level>::peekIndex(arg,i,j);
return ret;
}
template<int Level,class vtype> accelerator_inline
void pokeIndex (vtype &ret,const decltype(TensorIndexRecursion<Level>::peekIndex(ret,0)) &arg,int i)
{
TensorIndexRecursion<Level>::pokeIndex(ret,arg,i);
}
template<int Level,class vtype> accelerator_inline
void pokeIndex (vtype &ret,const decltype(TensorIndexRecursion<Level>::peekIndex(ret,0,0)) &arg,int i,int j)
{
TensorIndexRecursion<Level>::pokeIndex(ret,arg,i,j);
}
#undef RemoveCRV
NAMESPACE_END(Grid);
#endif

139
Grid/tensors/Tensor_inner.h Normal file
View File

@ -0,0 +1,139 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_inner.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_INNER_H
#define GRID_MATH_INNER_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////////////////////////////////////////
// innerProduct Scalar x Scalar -> Scalar
// innerProduct Vector x Vector -> Scalar
// innerProduct Matrix x Matrix -> Scalar
///////////////////////////////////////////////////////////////////////////////////////
template<class sobj> accelerator_inline RealD norm2(const sobj &arg){
auto nrm = innerProductD(arg,arg);
RealD ret = real(nrm);
return ret;
}
//////////////////////////////////////
// If single promote to double and sum 2x
//////////////////////////////////////
accelerator_inline ComplexD innerProductD(const ComplexF &l,const ComplexF &r){ return innerProduct(l,r); }
accelerator_inline ComplexD innerProductD(const ComplexD &l,const ComplexD &r){ return innerProduct(l,r); }
accelerator_inline RealD innerProductD(const RealD &l,const RealD &r){ return innerProduct(l,r); }
accelerator_inline RealD innerProductD(const RealF &l,const RealF &r){ return innerProduct(l,r); }
accelerator_inline vComplexD innerProductD(const vComplexD &l,const vComplexD &r){ return innerProduct(l,r); }
accelerator_inline vRealD innerProductD(const vRealD &l,const vRealD &r){ return innerProduct(l,r); }
accelerator_inline vComplexD innerProductD(const vComplexF &l,const vComplexF &r){
vComplexD la,lb;
vComplexD ra,rb;
Optimization::PrecisionChange::StoD(l.v,la.v,lb.v);
Optimization::PrecisionChange::StoD(r.v,ra.v,rb.v);
return innerProduct(la,ra) + innerProduct(lb,rb);
}
accelerator_inline vRealD innerProductD(const vRealF &l,const vRealF &r){
vRealD la,lb;
vRealD ra,rb;
Optimization::PrecisionChange::StoD(l.v,la.v,lb.v);
Optimization::PrecisionChange::StoD(r.v,ra.v,rb.v);
return innerProduct(la,ra) + innerProduct(lb,rb);
}
template<class l,class r,int N> accelerator_inline
auto innerProductD (const iVector<l,N>& lhs,const iVector<r,N>& rhs) -> iScalar<decltype(innerProductD(lhs._internal[0],rhs._internal[0]))>
{
typedef decltype(innerProductD(lhs._internal[0],rhs._internal[0])) ret_t;
iScalar<ret_t> ret;
zeroit(ret);
for(int c1=0;c1<N;c1++){
ret._internal += innerProductD(lhs._internal[c1],rhs._internal[c1]);
}
return ret;
}
template<class l,class r,int N> accelerator_inline
auto innerProductD (const iMatrix<l,N>& lhs,const iMatrix<r,N>& rhs) -> iScalar<decltype(innerProductD(lhs._internal[0][0],rhs._internal[0][0]))>
{
typedef decltype(innerProductD(lhs._internal[0][0],rhs._internal[0][0])) ret_t;
iScalar<ret_t> ret;
ret=Zero();
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal+=innerProductD(lhs._internal[c1][c2],rhs._internal[c1][c2]);
}}
return ret;
}
template<class l,class r> accelerator_inline
auto innerProductD (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<decltype(innerProductD(lhs._internal,rhs._internal))>
{
typedef decltype(innerProductD(lhs._internal,rhs._internal)) ret_t;
iScalar<ret_t> ret;
ret._internal = innerProductD(lhs._internal,rhs._internal);
return ret;
}
//////////////////////
// Keep same precison
//////////////////////
template<class l,class r,int N> accelerator_inline
auto innerProduct (const iVector<l,N>& lhs,const iVector<r,N>& rhs) -> iScalar<decltype(innerProduct(lhs._internal[0],rhs._internal[0]))>
{
typedef decltype(innerProduct(lhs._internal[0],rhs._internal[0])) ret_t;
iScalar<ret_t> ret;
ret=Zero();
for(int c1=0;c1<N;c1++){
ret._internal += innerProduct(lhs._internal[c1],rhs._internal[c1]);
}
return ret;
}
template<class l,class r,int N> accelerator_inline
auto innerProduct (const iMatrix<l,N>& lhs,const iMatrix<r,N>& rhs) -> iScalar<decltype(innerProduct(lhs._internal[0][0],rhs._internal[0][0]))>
{
typedef decltype(innerProduct(lhs._internal[0][0],rhs._internal[0][0])) ret_t;
iScalar<ret_t> ret;
iScalar<ret_t> tmp;
ret=Zero();
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal+=innerProduct(lhs._internal[c1][c2],rhs._internal[c1][c2]);
}}
return ret;
}
template<class l,class r> accelerator_inline
auto innerProduct (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<decltype(innerProduct(lhs._internal,rhs._internal))>
{
typedef decltype(innerProduct(lhs._internal,rhs._internal)) ret_t;
iScalar<ret_t> ret;
ret._internal = innerProduct(lhs._internal,rhs._internal);
return ret;
}
NAMESPACE_END(Grid);
#endif

92
Grid/tensors/Tensor_logical.h Normal file
View File

@ -0,0 +1,92 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_logical.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
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.
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 */
#pragma once
NAMESPACE_BEGIN(Grid);
#define LOGICAL_BINOP(Op) \
template<class v> accelerator_inline iScalar<v> operator Op (const iScalar<v>& lhs,const iScalar<v>& rhs) \
{ \
iScalar<v> ret; \
ret._internal = lhs._internal Op rhs._internal ; \
return ret; \
} \
template<class l> accelerator_inline iScalar<l> operator Op (const iScalar<l>& lhs,Integer rhs) \
{ \
typename iScalar<l>::scalar_type t; t=rhs; \
typename iScalar<l>::tensor_reduced srhs; srhs=t; \
return lhs Op srhs; \
} \
template<class l> accelerator_inline iScalar<l> operator Op (Integer lhs,const iScalar<l>& rhs) \
{ \
typename iScalar<l>::scalar_type t;t=lhs; \
typename iScalar<l>::tensor_reduced slhs;slhs=t; \
return slhs Op rhs; \
}
LOGICAL_BINOP(|);
LOGICAL_BINOP(&);
LOGICAL_BINOP(||);
LOGICAL_BINOP(&&);
template <class T>
strong_inline bool operator==(const iScalar<T> &t1, const iScalar<T> &t2)
{
return (t1._internal == t2._internal);
}
template <class T, int N>
strong_inline bool operator==(const iVector<T, N> &t1, const iVector<T, N> &t2)
{
bool res = true;
for (unsigned int i = 0; i < N; ++i)
{
res = (res && (t1._internal[i] == t2._internal[i]));
}
return res;
}
template <class T, int N>
strong_inline bool operator==(const iMatrix<T, N> &t1, const iMatrix<T, N> &t2)
{
bool res = true;
for (unsigned int i = 0; i < N; ++i)
for (unsigned int j = 0; j < N; ++j)
{
res = (res && (t1._internal[i][j] == t2._internal[i][j]));
}
return res;
}
NAMESPACE_END(Grid);

80
Grid/tensors/Tensor_outer.h Normal file
View File

@ -0,0 +1,80 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_outer.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_OUTER_H
#define GRID_MATH_OUTER_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////////////////////////////////////////////
// outerProduct Scalar x Scalar -> Scalar
// Vector x Vector -> Matrix
///////////////////////////////////////////////////////////////////////////////////////
template<class l,class r,int N> accelerator_inline
auto outerProduct (const iVector<l,N>& lhs,const iVector<r,N>& rhs) -> iMatrix<decltype(outerProduct(lhs._internal[0],rhs._internal[0])),N>
{
typedef decltype(outerProduct(lhs._internal[0],rhs._internal[0])) ret_t;
iMatrix<ret_t,N> ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal[c1][c2] = outerProduct(lhs._internal[c1],rhs._internal[c2]);
}}
return ret;
}
template<class l,class r> accelerator_inline
auto outerProduct (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<decltype(outerProduct(lhs._internal,rhs._internal))>
{
typedef decltype(outerProduct(lhs._internal,rhs._internal)) ret_t;
iScalar<ret_t> ret;
ret._internal = outerProduct(lhs._internal,rhs._internal);
return ret;
}
accelerator_inline ComplexF outerProduct(const ComplexF &l, const ComplexF& r)
{
return l*conj(r);
}
accelerator_inline ComplexD outerProduct(const ComplexD &l, const ComplexD& r)
{
return l*conj(r);
}
accelerator_inline RealF outerProduct(const RealF &l, const RealF& r)
{
return l*r;
}
accelerator_inline RealD outerProduct(const RealD &l, const RealD& r)
{
return l*r;
}
NAMESPACE_END(Grid);
#endif

236
Grid/tensors/Tensor_reality.h Normal file
View File

@ -0,0 +1,236 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_reality.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_REALITY_H
#define GRID_MATH_REALITY_H
NAMESPACE_BEGIN(Grid);
///////////////////////////////////////////////
// multiply by I; make recursive.
///////////////////////////////////////////////
template<class vtype> accelerator_inline iScalar<vtype> timesI(const iScalar<vtype>&r)
{
iScalar<vtype> ret;
timesI(ret._internal,r._internal);
return ret;
}
template<class vtype,int N> accelerator_inline iVector<vtype,N> timesI(const iVector<vtype,N>&r)
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
timesI(ret._internal[i],r._internal[i]);
}
return ret;
}
template<class vtype,int N> accelerator_inline iMatrix<vtype,N> timesI(const iMatrix<vtype,N>&r)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
timesI(ret._internal[i][j],r._internal[i][j]);
}}
return ret;
}
template<class vtype> accelerator_inline void timesI(iScalar<vtype> &ret,const iScalar<vtype>&r)
{
timesI(ret._internal,r._internal);
}
template<class vtype,int N> accelerator_inline void timesI(iVector<vtype,N> &ret,const iVector<vtype,N>&r)
{
for(int i=0;i<N;i++){
timesI(ret._internal[i],r._internal[i]);
}
}
template<class vtype,int N> accelerator_inline void timesI(iMatrix<vtype,N> &ret,const iMatrix<vtype,N>&r)
{
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
timesI(ret._internal[i][j],r._internal[i][j]);
}}
}
template<class vtype> accelerator_inline iScalar<vtype> timesMinusI(const iScalar<vtype>&r)
{
iScalar<vtype> ret;
timesMinusI(ret._internal,r._internal);
return ret;
}
template<class vtype,int N> accelerator_inline iVector<vtype,N> timesMinusI(const iVector<vtype,N>&r)
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
timesMinusI(ret._internal[i],r._internal[i]);
}
return ret;
}
template<class vtype,int N> accelerator_inline iMatrix<vtype,N> timesMinusI(const iMatrix<vtype,N>&r)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
timesMinusI(ret._internal[i][j],r._internal[i][j]);
}}
return ret;
}
template<class vtype> accelerator_inline void timesMinusI(iScalar<vtype> &ret,const iScalar<vtype>&r)
{
timesMinusI(ret._internal,r._internal);
}
template<class vtype,int N> accelerator_inline void timesMinusI(iVector<vtype,N> &ret,const iVector<vtype,N>&r)
{
for(int i=0;i<N;i++){
timesMinusI(ret._internal[i],r._internal[i]);
}
}
template<class vtype,int N> accelerator_inline void timesMinusI(iMatrix<vtype,N> &ret,const iMatrix<vtype,N>&r)
{
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
timesMinusI(ret._internal[i][j],r._internal[i][j]);
}}
}
///////////////////////////////////////////////
// Conj function for scalar, vector, matrix
///////////////////////////////////////////////
template<class vtype> accelerator_inline iScalar<vtype> conjugate(const iScalar<vtype>&r)
{
iScalar<vtype> ret;
ret._internal = conjugate(r._internal);
return ret;
}
template<class vtype,int N> accelerator_inline iVector<vtype,N> conjugate(const iVector<vtype,N>&r)
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = conjugate(r._internal[i]);
}
return ret;
}
template<class vtype,int N> accelerator_inline iMatrix<vtype,N> conjugate(const iMatrix<vtype,N>&r)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = conjugate(r._internal[i][j]);
}}
return ret;
}
///////////////////////////////////////////////
// Adj function for scalar, vector, matrix
///////////////////////////////////////////////
template<class vtype> accelerator_inline iScalar<vtype> adj(const iScalar<vtype>&r)
{
iScalar<vtype> ret;
ret._internal = adj(r._internal);
return ret;
}
template<class vtype,int N> accelerator_inline iVector<vtype,N> adj(const iVector<vtype,N>&r)
{
iVector<vtype,N> ret;
for(int i=0;i<N;i++){
ret._internal[i] = adj(r._internal[i]);
}
return ret;
}
template<class vtype,int N> accelerator_inline iMatrix<vtype,N> adj(const iMatrix<vtype,N> &arg)
{
iMatrix<vtype,N> ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal[c1][c2]=adj(arg._internal[c2][c1]);
}}
return ret;
}
/////////////////////////////////////////////////////////////////
// Can only take the real/imag part of scalar objects, since
// lattice objects of different complex nature are non-conformable.
/////////////////////////////////////////////////////////////////
template<class itype> accelerator_inline auto real(const iScalar<itype> &z) -> iScalar<decltype(real(z._internal))>
{
iScalar<decltype(real(z._internal))> ret;
ret._internal = real(z._internal);
return ret;
}
template<class itype,int N> accelerator_inline auto real(const iMatrix<itype,N> &z) -> iMatrix<decltype(real(z._internal[0][0])),N>
{
iMatrix<decltype(real(z._internal[0][0])),N> ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal[c1][c2] = real(z._internal[c1][c2]);
}}
return ret;
}
template<class itype,int N> accelerator_inline auto real(const iVector<itype,N> &z) -> iVector<decltype(real(z._internal[0])),N>
{
iVector<decltype(real(z._internal[0])),N> ret;
for(int c1=0;c1<N;c1++){
ret._internal[c1] = real(z._internal[c1]);
}
return ret;
}
template<class itype> accelerator_inline auto imag(const iScalar<itype> &z) -> iScalar<decltype(imag(z._internal))>
{
iScalar<decltype(imag(z._internal))> ret;
ret._internal = imag(z._internal);
return ret;
}
template<class itype,int N> accelerator_inline auto imag(const iMatrix<itype,N> &z) -> iMatrix<decltype(imag(z._internal[0][0])),N>
{
iMatrix<decltype(imag(z._internal[0][0])),N> ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal[c1][c2] = imag(z._internal[c1][c2]);
}}
return ret;
}
template<class itype,int N> accelerator_inline auto imag(const iVector<itype,N> &z) -> iVector<decltype(imag(z._internal[0])),N>
{
iVector<decltype(imag(z._internal[0])),N> ret;
for(int c1=0;c1<N;c1++){
ret._internal[c1] = imag(z._internal[c1]);
}
return ret;
}
NAMESPACE_END(Grid);
#endif

75
Grid/tensors/Tensor_trace.h Normal file
View File

@ -0,0 +1,75 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_trace.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_TRACE_H
#define GRID_MATH_TRACE_H
NAMESPACE_BEGIN(Grid);
//////////////////////////////////////////////////////////////////
// Traces: both all indices and a specific index. Indices must be
// either scalar or matrix
/////////////////////////////////////////////////////////////////
accelerator_inline ComplexF trace( const ComplexF &arg){ return arg;}
accelerator_inline ComplexD trace( const ComplexD &arg){ return arg;}
accelerator_inline RealF trace( const RealF &arg){ return arg;}
accelerator_inline RealD trace( const RealD &arg){ return arg;}
template<class vtype,int N>
accelerator_inline auto trace(const iMatrix<vtype,N> &arg) -> iScalar<decltype(trace(arg._internal[0][0]))>
{
iScalar<decltype( trace(arg._internal[0][0] )) > ret;
zeroit(ret._internal);
for(int i=0;i<N;i++){
ret._internal=ret._internal+trace(arg._internal[i][i]);
}
return ret;
}
template<class vtype>
accelerator_inline auto trace(const iScalar<vtype> &arg) -> iScalar<decltype(trace(arg._internal))>
{
iScalar<decltype(trace(arg._internal))> ret;
ret._internal=trace(arg._internal);
return ret;
}
template<class vtype,int N>
accelerator_inline auto trace(const iVector<vtype,N> &arg) -> iVector<decltype(trace(arg._internal[0])),N>
{
iVector<decltype(trace(arg._internal[0])),N> ret;
for(int i=0;i<N;i++){
ret._internal[i]=trace(arg._internal[i]);
}
return ret;
}
NAMESPACE_END(Grid);
#endif

296
Grid/tensors/Tensor_traits.h Normal file
View File

@ -0,0 +1,296 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_traits.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Christopher Kelly <ckelly@phys.columbia.edu>
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.
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 */
#ifndef GRID_MATH_TRAITS_H
#define GRID_MATH_TRAITS_H
#include <type_traits>
NAMESPACE_BEGIN(Grid);
//////////////////////////////////////////////////////////////////////////////////
// Want to recurse: GridTypeMapper<Matrix<vComplexD> >::scalar_type == ComplexD.
// Use of a helper class like this allows us to template specialise and "dress"
// other classes such as RealD == double, ComplexD == std::complex<double> with these
// traits.
//
// It is possible that we could do this more elegantly if I introduced a
// queryable trait in iScalar, iMatrix and iVector and used the query on vtype in
// place of the type mapper?
//
// Not sure how to do this, but probably could be done with a research effort
// to study C++11's type_traits.h file. (std::enable_if<isGridTensorType<vtype> >)
//
//////////////////////////////////////////////////////////////////////////////////
template <class T> class GridTypeMapper {
public:
typedef typename T::scalar_type scalar_type;
typedef typename T::vector_type vector_type;
typedef typename T::vector_typeD vector_typeD;
typedef typename T::tensor_reduced tensor_reduced;
typedef typename T::scalar_object scalar_object;
typedef typename T::Complexified Complexified;
typedef typename T::Realified Realified;
typedef typename T::DoublePrecision DoublePrecision;
enum { TensorLevel = T::TensorLevel };
};
//////////////////////////////////////////////////////////////////////////////////
// Recursion stops with these template specialisations
//////////////////////////////////////////////////////////////////////////////////
template<> class GridTypeMapper<RealF> {
public:
typedef RealF scalar_type;
typedef RealF vector_type;
typedef RealD vector_typeD;
typedef RealF tensor_reduced ;
typedef RealF scalar_object;
typedef ComplexF Complexified;
typedef RealF Realified;
typedef RealD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<RealD> {
public:
typedef RealD scalar_type;
typedef RealD vector_type;
typedef RealD vector_typeD;
typedef RealD tensor_reduced;
typedef RealD scalar_object;
typedef ComplexD Complexified;
typedef RealD Realified;
typedef RealD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<ComplexF> {
public:
typedef ComplexF scalar_type;
typedef ComplexF vector_type;
typedef ComplexD vector_typeD;
typedef ComplexF tensor_reduced;
typedef ComplexF scalar_object;
typedef ComplexF Complexified;
typedef RealF Realified;
typedef ComplexD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<ComplexD> {
public:
typedef ComplexD scalar_type;
typedef ComplexD vector_type;
typedef ComplexD vector_typeD;
typedef ComplexD tensor_reduced;
typedef ComplexD scalar_object;
typedef ComplexD Complexified;
typedef RealD Realified;
typedef ComplexD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<Integer> {
public:
typedef Integer scalar_type;
typedef Integer vector_type;
typedef Integer vector_typeD;
typedef Integer tensor_reduced;
typedef Integer scalar_object;
typedef void Complexified;
typedef void Realified;
typedef void DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vRealF> {
public:
typedef RealF scalar_type;
typedef vRealF vector_type;
typedef vRealD vector_typeD;
typedef vRealF tensor_reduced;
typedef RealF scalar_object;
typedef vComplexF Complexified;
typedef vRealF Realified;
typedef vRealD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vRealD> {
public:
typedef RealD scalar_type;
typedef vRealD vector_type;
typedef vRealD vector_typeD;
typedef vRealD tensor_reduced;
typedef RealD scalar_object;
typedef vComplexD Complexified;
typedef vRealD Realified;
typedef vRealD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vComplexH> {
public:
typedef ComplexF scalar_type;
typedef vComplexH vector_type;
typedef vComplexD vector_typeD;
typedef vComplexH tensor_reduced;
typedef ComplexF scalar_object;
typedef vComplexH Complexified;
typedef vRealH Realified;
typedef vComplexD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vComplexF> {
public:
typedef ComplexF scalar_type;
typedef vComplexF vector_type;
typedef vComplexD vector_typeD;
typedef vComplexF tensor_reduced;
typedef ComplexF scalar_object;
typedef vComplexF Complexified;
typedef vRealF Realified;
typedef vComplexD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vComplexD> {
public:
typedef ComplexD scalar_type;
typedef vComplexD vector_type;
typedef vComplexD vector_typeD;
typedef vComplexD tensor_reduced;
typedef ComplexD scalar_object;
typedef vComplexD Complexified;
typedef vRealD Realified;
typedef vComplexD DoublePrecision;
enum { TensorLevel = 0 };
};
template<> class GridTypeMapper<vInteger> {
public:
typedef Integer scalar_type;
typedef vInteger vector_type;
typedef vInteger vector_typeD;
typedef vInteger tensor_reduced;
typedef Integer scalar_object;
typedef void Complexified;
typedef void Realified;
typedef void DoublePrecision;
enum { TensorLevel = 0 };
};
// First some of my own traits
template<typename T> struct isGridTensor {
static const bool value = true;
static const bool notvalue = false;
};
template<> struct isGridTensor<int > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<RealD > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<RealF > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<ComplexD > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<ComplexF > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<Integer > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<vRealD > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<vRealF > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<vComplexD > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<vComplexF > {
static const bool value = false;
static const bool notvalue = true;
};
template<> struct isGridTensor<vInteger > {
static const bool value = false;
static const bool notvalue = true;
};
// Match the index
template<typename T,int Level> struct matchGridTensorIndex {
static const bool value = (Level==T::TensorLevel);
static const bool notvalue = (Level!=T::TensorLevel);
};
// What is the vtype
template<typename T> struct isComplex {
static const bool value = false;
};
template<> struct isComplex<ComplexF> {
static const bool value = true;
};
template<> struct isComplex<ComplexD> {
static const bool value = true;
};
//Get the SIMD vector type from a Grid tensor or Lattice<Tensor>
template<typename T>
struct getVectorType{
typedef T type;
};
//Query if a tensor or Lattice<Tensor> is SIMD vector or scalar
template<typename T>
class isSIMDvectorized{
template<typename U>
static typename std::enable_if< !std::is_same< typename GridTypeMapper<typename getVectorType<U>::type>::scalar_type,
typename GridTypeMapper<typename getVectorType<U>::type>::vector_type>::value, char>::type test(void *);
template<typename U>
static double test(...);
public:
enum {value = sizeof(test<T>(0)) == sizeof(char) };
};
//Get the precision of a Lattice, tensor or scalar type in units of sizeof(float)
template<typename T>
class getPrecision{
public:
//get the vector_obj (i.e. a grid Tensor) if its a Lattice<vobj>, do nothing otherwise (i.e. if fundamental or grid Tensor)
typedef typename getVectorType<T>::type vector_obj;
typedef typename GridTypeMapper<vector_obj>::scalar_type scalar_type; //get the associated scalar type. Works on fundamental and tensor types
typedef typename GridTypeMapper<scalar_type>::Realified real_scalar_type; //remove any std::complex wrapper, should get us to the fundamental type
enum { value = sizeof(real_scalar_type)/sizeof(float) };
};
NAMESPACE_END(Grid);
#endif

130
Grid/tensors/Tensor_transpose.h Normal file
View File

@ -0,0 +1,130 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_transpose.h
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_MATH_TRANSPOSE_H
#define GRID_MATH_TRANSPOSE_H
NAMESPACE_BEGIN(Grid);
/////////////////////////////////////////////////////////////////
// Transpose all indices
/////////////////////////////////////////////////////////////////
accelerator_inline ComplexD transpose(ComplexD &rhs){ return rhs;}
accelerator_inline ComplexF transpose(ComplexF &rhs){ return rhs;}
accelerator_inline RealD transpose(RealD &rhs){ return rhs;}
accelerator_inline RealF transpose(RealF &rhs){ return rhs;}
template<class vtype,int N>
accelerator_inline typename std::enable_if<isGridTensor<vtype>::value, iMatrix<vtype,N> >::type
transpose(iMatrix<vtype,N> arg)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = transpose(arg._internal[j][i]); // NB recurses
}}
return ret;
}
template<class vtype,int N>
accelerator_inline typename std::enable_if<isGridTensor<vtype>::notvalue, iMatrix<vtype,N> >::type
transpose(iMatrix<vtype,N> arg)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = arg._internal[j][i]; // Stop recursion if not a tensor type
}}
return ret;
}
template<class vtype>
accelerator_inline typename std::enable_if<isGridTensor<vtype>::value, iScalar<vtype> >::type
transpose(iScalar<vtype> arg)
{
iScalar<vtype> ret;
ret._internal = transpose(arg._internal); // NB recurses
return ret;
}
template<class vtype>
accelerator_inline typename std::enable_if<isGridTensor<vtype>::notvalue, iScalar<vtype> >::type
transpose(iScalar<vtype> arg)
{
iScalar<vtype> ret;
ret._internal = arg._internal; // NB recursion stops
return ret;
}
////////////////////////////////////////////////////////////////////////////////////////////
// Transpose a specific index; instructive to compare this style of recursion termination
// to that of adj; which is easiers?
////////////////////////////////////////////////////////////////////////////////////////////
#if 0
template<int Level,class vtype,int N> accelerator_inline
typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,Level>::value, iMatrix<vtype,N> >::type
transposeIndex (const iMatrix<vtype,N> &arg)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = arg._internal[j][i];
}}
return ret;
}
// or not
template<int Level,class vtype,int N> accelerator_inline
typename std::enable_if<matchGridTensorIndex<iMatrix<vtype,N>,Level>::notvalue, iMatrix<vtype,N> >::type
transposeIndex (const iMatrix<vtype,N> &arg)
{
iMatrix<vtype,N> ret;
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
ret._internal[i][j] = transposeIndex<Level>(arg._internal[i][j]);
}}
return ret;
}
template<int Level,class vtype> accelerator_inline
typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,Level>::notvalue, iScalar<vtype> >::type
transposeIndex (const iScalar<vtype> &arg)
{
iScalar<vtype> ret;
ret._internal=transposeIndex<Level>(arg._internal);
return ret;
}
template<int Level,class vtype> accelerator_inline
typename std::enable_if<matchGridTensorIndex<iScalar<vtype>,Level>::value, iScalar<vtype> >::type
transposeIndex (const iScalar<vtype> &arg)
{
return arg;
}
#endif
NAMESPACE_END(Grid);
#endif

155
Grid/tensors/Tensor_unary.h Normal file
View File

@ -0,0 +1,155 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/tensors/Tensor_unary.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
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.
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 */
#ifndef GRID_TENSOR_UNARY_H
#define GRID_TENSOR_UNARY_H
NAMESPACE_BEGIN(Grid);
#define UNARY(func) \
template<class obj> accelerator_inline auto func(const iScalar<obj> &z) -> iScalar<obj> \
{ \
iScalar<obj> ret; \
ret._internal = func( (z._internal)); \
return ret; \
} \
template<class obj,int N> accelerator_inline auto func(const iVector<obj,N> &z) -> iVector<obj,N> \
{ \
iVector<obj,N> ret; \
for(int c1=0;c1<N;c1++){ \
ret._internal[c1] = func( (z._internal[c1])); \
} \
return ret; \
} \
template<class obj,int N> accelerator_inline auto func(const iMatrix<obj,N> &z) -> iMatrix<obj,N> \
{ \
iMatrix<obj,N> ret; \
for(int c1=0;c1<N;c1++){ \
for(int c2=0;c2<N;c2++){ \
ret._internal[c1][c2] = func( (z._internal[c1][c2])); \
}} \
return ret; \
}
#define BINARY_RSCALAR(func,scal) \
template<class obj> accelerator_inline iScalar<obj> func(const iScalar<obj> &z,scal y) \
{ \
iScalar<obj> ret; \
ret._internal = func(z._internal,y); \
return ret; \
} \
template<class obj,int N> accelerator_inline iVector<obj,N> func(const iVector<obj,N> &z,scal y) \
{ \
iVector<obj,N> ret; \
for(int c1=0;c1<N;c1++){ \
ret._internal[c1] = func(z._internal[c1],y); \
} \
return ret; \
} \
template<class obj,int N> accelerator_inline iMatrix<obj,N> func(const iMatrix<obj,N> &z, scal y) \
{ \
iMatrix<obj,N> ret; \
for(int c1=0;c1<N;c1++){ \
for(int c2=0;c2<N;c2++){ \
ret._internal[c1][c2] = func(z._internal[c1][c2],y); \
}} \
return ret; \
}
UNARY(sqrt);
UNARY(rsqrt);
UNARY(sin);
UNARY(cos);
UNARY(asin);
UNARY(acos);
UNARY(log);
UNARY(exp);
UNARY(abs);
UNARY(Not);
template<class obj> accelerator_inline auto toReal(const iScalar<obj> &z) -> typename iScalar<obj>::Realified
{
typename iScalar<obj>::Realified ret;
ret._internal = toReal(z._internal);
return ret;
}
template<class obj,int N> accelerator_inline auto toReal(const iVector<obj,N> &z) -> typename iVector<obj,N>::Realified
{
typename iVector<obj,N>::Realified ret;
for(int c1=0;c1<N;c1++){
ret._internal[c1] = toReal(z._internal[c1]);
}
return ret;
}
template<class obj,int N> accelerator_inline auto toReal(const iMatrix<obj,N> &z) -> typename iMatrix<obj,N>::Realified
{
typename iMatrix<obj,N>::Realified ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal[c1][c2] = toReal(z._internal[c1][c2]);
}}
return ret;
}
template<class obj> accelerator_inline auto toComplex(const iScalar<obj> &z) -> typename iScalar<obj>::Complexified
{
typename iScalar<obj>::Complexified ret;
ret._internal = toComplex(z._internal);
return ret;
}
template<class obj,int N> accelerator_inline auto toComplex(const iVector<obj,N> &z) -> typename iVector<obj,N>::Complexified
{
typename iVector<obj,N>::Complexified ret;
for(int c1=0;c1<N;c1++){
ret._internal[c1] = toComplex(z._internal[c1]);
}
return ret;
}
template<class obj,int N> accelerator_inline auto toComplex(const iMatrix<obj,N> &z) -> typename iMatrix<obj,N>::Complexified
{
typename iMatrix<obj,N>::Complexified ret;
for(int c1=0;c1<N;c1++){
for(int c2=0;c2<N;c2++){
ret._internal[c1][c2] = toComplex(z._internal[c1][c2]);
}}
return ret;
}
BINARY_RSCALAR(div,Integer);
BINARY_RSCALAR(mod,Integer);
BINARY_RSCALAR(pow,RealD);
#undef UNARY
#undef BINARY_RSCALAR
NAMESPACE_END(Grid);
#endif

51
Grid/tensors/Tensors.h Normal file
View File

@ -0,0 +1,51 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/Tensors.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
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.
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 */
#ifndef GRID_MATH_H
#define GRID_MATH_H
#include <Grid/tensors/Tensor_traits.h>
#include <Grid/tensors/Tensor_class.h>
#include <Grid/tensors/Tensor_arith.h>
#include <Grid/tensors/Tensor_inner.h>
#include <Grid/tensors/Tensor_outer.h>
#include <Grid/tensors/Tensor_transpose.h>
#include <Grid/tensors/Tensor_trace.h>
#include <Grid/tensors/Tensor_index.h>
#include <Grid/tensors/Tensor_Ta.h>
#include <Grid/tensors/Tensor_determinant.h>
#include <Grid/tensors/Tensor_exp.h>
//#include <Grid/tensors/Tensor_peek.h>
//#include <Grid/tensors/Tensor_poke.h>
#include <Grid/tensors/Tensor_reality.h>
#include <Grid/tensors/Tensor_unary.h>
#include <Grid/tensors/Tensor_extract_merge.h>
#include <Grid/tensors/Tensor_logical.h>
#endif