portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-06-13 20:57:06 +01:00
Code Releases Activity

Merge branch 'develop' into feature/clover

Browse Source
This commit is contained in:
Guido Cossu
2017-08-04 12:19:57 +01:00
parent 62a64d9108 175f393f9d
commit fde71c3c52
296 changed files with 34674 additions and 9453 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
lib/DisableWarnings.h Normal file
View File

@ -0,0 +1,37 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/DisableWarnings.h
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 DISABLE_WARNINGS_H
#define DISABLE_WARNINGS_H
//disables and intel compiler specific warning (in json.hpp)
#pragma warning disable 488
#endif

2
lib/Grid.h
View File

@ -41,7 +41,9 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#include <Grid/GridCore.h>
#include <Grid/GridQCDcore.h>
#include <Grid/qcd/action/Action.h>
#include <Grid/qcd/utils/GaugeFix.h>
#include <Grid/qcd/smearing/Smearing.h>
#include <Grid/parallelIO/MetaData.h>
#include <Grid/qcd/hmc/HMC_aggregate.h>
#endif

23
lib/GridCore.h
View File

@ -38,28 +38,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#ifndef GRID_BASE_H
#define GRID_BASE_H
///////////////////
// Std C++ dependencies
///////////////////
#include <cassert>
#include <complex>
#include <vector>
#include <iostream>
#include <iomanip>
#include <random>
#include <functional>
#include <stdio.h>
#include <stdlib.h>
#include <stdio.h>
#include <signal.h>
#include <ctime>
#include <sys/time.h>
#include <chrono>
///////////////////
// Grid headers
///////////////////
#include "Config.h"
#include <Grid/GridStd.h>
#include <Grid/perfmon/Timer.h>
#include <Grid/perfmon/PerfCount.h>

29
lib/GridStd.h Normal file
View File

@ -0,0 +1,29 @@
#ifndef GRID_STD_H
#define GRID_STD_H
///////////////////
// Std C++ dependencies
///////////////////
#include <cassert>
#include <complex>
#include <vector>
#include <string>
#include <iostream>
#include <iomanip>
#include <random>
#include <functional>
#include <stdio.h>
#include <stdlib.h>
#include <stdio.h>
#include <signal.h>
#include <ctime>
#include <sys/time.h>
#include <chrono>
#include <zlib.h>
///////////////////
// Grid config
///////////////////
#include "Config.h"
#endif /* GRID_STD_H */

9
lib/Grid_Eigen_Dense.h Normal file
View File

@ -0,0 +1,9 @@
#pragma once
#if defined __GNUC__
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
#endif
#include <Grid/Eigen/Dense>
#if defined __GNUC__
#pragma GCC diagnostic pop
#endif

2
lib/algorithms/LinearOperator.h
View File

@ -235,7 +235,7 @@ namespace Grid {
Field tmp(in._grid);
_Mat.MeooeDag(in,tmp);
_Mat.MooeeInvDag(tmp,out);
_Mat.MooeeInvDag(tmp,out);
_Mat.MeooeDag(out,tmp);
_Mat.MooeeDag(in,out);

5
lib/algorithms/approx/Chebyshev.h
View File

@ -197,8 +197,9 @@ namespace Grid {
void operator() (LinearOperatorBase<Field> &Linop, const Field &in, Field &out) {
GridBase *grid=in._grid;
//std::cout << "Chevyshef(): in._grid="<<in._grid<<std::endl;
//<<" Linop.Grid()="<<Linop.Grid()<<"Linop.RedBlackGrid()="<<Linop.RedBlackGrid()<<std::endl;
// std::cout << "Chevyshef(): in._grid="<<in._grid<<std::endl;
//std::cout <<" Linop.Grid()="<<Linop.Grid()<<"Linop.RedBlackGrid()="<<Linop.RedBlackGrid()<<std::endl;
int vol=grid->gSites();

2
lib/algorithms/approx/Remez.h
View File

@ -16,7 +16,7 @@
#define INCLUDED_ALG_REMEZ_H
#include <stddef.h>
#include <Config.h>
#include <Grid/GridStd.h>
#ifdef HAVE_LIBGMP
#include "bigfloat.h"

137
lib/algorithms/densematrix/DenseMatrix.h
View File

@ -1,137 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/algorithms/iterative/DenseMatrix.h
Copyright (C) 2015
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_DENSE_MATRIX_H
#define GRID_DENSE_MATRIX_H
namespace Grid {
/////////////////////////////////////////////////////////////
// Matrix untils
/////////////////////////////////////////////////////////////
template<class T> using DenseVector = std::vector<T>;
template<class T> using DenseMatrix = DenseVector<DenseVector<T> >;
template<class T> void Size(DenseVector<T> & vec, int &N)
{
N= vec.size();
}
template<class T> void Size(DenseMatrix<T> & mat, int &N,int &M)
{
N= mat.size();
M= mat[0].size();
}
template<class T> void SizeSquare(DenseMatrix<T> & mat, int &N)
{
int M; Size(mat,N,M);
assert(N==M);
}
template<class T> void Resize(DenseVector<T > & mat, int N) {
mat.resize(N);
}
template<class T> void Resize(DenseMatrix<T > & mat, int N, int M) {
mat.resize(N);
for(int i=0;i<N;i++){
mat[i].resize(M);
}
}
template<class T> void Fill(DenseMatrix<T> & mat, T&val) {
int N,M;
Size(mat,N,M);
for(int i=0;i<N;i++){
for(int j=0;j<M;j++){
mat[i][j] = val;
}}
}
/** Transpose of a matrix **/
template<class T> DenseMatrix<T> Transpose(DenseMatrix<T> & mat){
int N,M;
Size(mat,N,M);
DenseMatrix<T> C; Resize(C,M,N);
for(int i=0;i<M;i++){
for(int j=0;j<N;j++){
C[i][j] = mat[j][i];
}}
return C;
}
/** Set DenseMatrix to unit matrix **/
template<class T> void Unity(DenseMatrix<T> &A){
int N; SizeSquare(A,N);
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
if ( i==j ) A[i][j] = 1;
else A[i][j] = 0;
}
}
}
/** Add C * I to matrix **/
template<class T>
void PlusUnit(DenseMatrix<T> & A,T c){
int dim; SizeSquare(A,dim);
for(int i=0;i<dim;i++){A[i][i] = A[i][i] + c;}
}
/** return the Hermitian conjugate of matrix **/
template<class T>
DenseMatrix<T> HermitianConj(DenseMatrix<T> &mat){
int dim; SizeSquare(mat,dim);
DenseMatrix<T> C; Resize(C,dim,dim);
for(int i=0;i<dim;i++){
for(int j=0;j<dim;j++){
C[i][j] = conj(mat[j][i]);
}
}
return C;
}
/**Get a square submatrix**/
template <class T>
DenseMatrix<T> GetSubMtx(DenseMatrix<T> &A,int row_st, int row_end, int col_st, int col_end)
{
DenseMatrix<T> H; Resize(H,row_end - row_st,col_end-col_st);
for(int i = row_st; i<row_end; i++){
for(int j = col_st; j<col_end; j++){
H[i-row_st][j-col_st]=A[i][j];
}}
return H;
}
}
#include "Householder.h"
#include "Francis.h"
#endif

525
lib/algorithms/densematrix/Francis.h
View File

@ -1,525 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/algorithms/iterative/Francis.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 FRANCIS_H
#define FRANCIS_H
#include <cstdlib>
#include <string>
#include <cmath>
#include <iostream>
#include <sstream>
#include <stdexcept>
#include <fstream>
#include <complex>
#include <algorithm>
//#include <timer.h>
//#include <lapacke.h>
//#include <Eigen/Dense>
namespace Grid {
template <class T> int SymmEigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small);
template <class T> int Eigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small);
/**
Find the eigenvalues of an upper hessenberg matrix using the Francis QR algorithm.
H =
x x x x x x x x x
x x x x x x x x x
0 x x x x x x x x
0 0 x x x x x x x
0 0 0 x x x x x x
0 0 0 0 x x x x x
0 0 0 0 0 x x x x
0 0 0 0 0 0 x x x
0 0 0 0 0 0 0 x x
Factorization is P T P^H where T is upper triangular (mod cc blocks) and P is orthagonal/unitary.
**/
template <class T>
int QReigensystem(DenseMatrix<T> &Hin, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small)
{
DenseMatrix<T> H = Hin;
int N ; SizeSquare(H,N);
int M = N;
Fill(evals,0);
Fill(evecs,0);
T s,t,x=0,y=0,z=0;
T u,d;
T apd,amd,bc;
DenseVector<T> p(N,0);
T nrm = Norm(H); ///DenseMatrix Norm
int n, m;
int e = 0;
int it = 0;
int tot_it = 0;
int l = 0;
int r = 0;
DenseMatrix<T> P; Resize(P,N,N); Unity(P);
DenseVector<int> trows(N,0);
/// Check if the matrix is really hessenberg, if not abort
RealD sth = 0;
for(int j=0;j<N;j++){
for(int i=j+2;i<N;i++){
sth = abs(H[i][j]);
if(sth > small){
std::cout << "Non hessenberg H = " << sth << " > " << small << std::endl;
exit(1);
}
}
}
do{
std::cout << "Francis QR Step N = " << N << std::endl;
/** Check for convergence
x x x x x
0 x x x x
0 0 x x x
0 0 x x x
0 0 0 0 x
for this matrix l = 4
**/
do{
l = Chop_subdiag(H,nrm,e,small);
r = 0; ///May have converged on more than one eval
///Single eval
if(l == N-1){
evals[e] = H[l][l];
N--; e++; r++; it = 0;
}
///RealD eval
if(l == N-2){
trows[l+1] = 1; ///Needed for UTSolve
apd = H[l][l] + H[l+1][l+1];
amd = H[l][l] - H[l+1][l+1];
bc = (T)4.0*H[l+1][l]*H[l][l+1];
evals[e] = (T)0.5*( apd + sqrt(amd*amd + bc) );
evals[e+1] = (T)0.5*( apd - sqrt(amd*amd + bc) );
N-=2; e+=2; r++; it = 0;
}
} while(r>0);
if(N ==0) break;
DenseVector<T > ck; Resize(ck,3);
DenseVector<T> v; Resize(v,3);
for(int m = N-3; m >= l; m--){
///Starting vector essentially random shift.
if(it%10 == 0 && N >= 3 && it > 0){
s = (T)1.618033989*( abs( H[N-1][N-2] ) + abs( H[N-2][N-3] ) );
t = (T)0.618033989*( abs( H[N-1][N-2] ) + abs( H[N-2][N-3] ) );
x = H[m][m]*H[m][m] + H[m][m+1]*H[m+1][m] - s*H[m][m] + t;
y = H[m+1][m]*(H[m][m] + H[m+1][m+1] - s);
z = H[m+1][m]*H[m+2][m+1];
}
///Starting vector implicit Q theorem
else{
s = (H[N-2][N-2] + H[N-1][N-1]);
t = (H[N-2][N-2]*H[N-1][N-1] - H[N-2][N-1]*H[N-1][N-2]);
x = H[m][m]*H[m][m] + H[m][m+1]*H[m+1][m] - s*H[m][m] + t;
y = H[m+1][m]*(H[m][m] + H[m+1][m+1] - s);
z = H[m+1][m]*H[m+2][m+1];
}
ck[0] = x; ck[1] = y; ck[2] = z;
if(m == l) break;
/** Some stupid thing from numerical recipies, seems to work**/
// PAB.. for heaven's sake quote page, purpose, evidence it works.
// what sort of comment is that!?!?!?
u=abs(H[m][m-1])*(abs(y)+abs(z));
d=abs(x)*(abs(H[m-1][m-1])+abs(H[m][m])+abs(H[m+1][m+1]));
if ((T)abs(u+d) == (T)abs(d) ){
l = m; break;
}
//if (u < small){l = m; break;}
}
if(it > 100000){
std::cout << "QReigensystem: bugger it got stuck after 100000 iterations" << std::endl;
std::cout << "got " << e << " evals " << l << " " << N << std::endl;
exit(1);
}
normalize(ck); ///Normalization cancels in PHP anyway
T beta;
Householder_vector<T >(ck, 0, 2, v, beta);
Householder_mult<T >(H,v,beta,0,l,l+2,0);
Householder_mult<T >(H,v,beta,0,l,l+2,1);
///Accumulate eigenvector
Householder_mult<T >(P,v,beta,0,l,l+2,1);
int sw = 0; ///Are we on the last row?
for(int k=l;k<N-2;k++){
x = H[k+1][k];
y = H[k+2][k];
z = (T)0.0;
if(k+3 <= N-1){
z = H[k+3][k];
} else{
sw = 1;
v[2] = (T)0.0;
}
ck[0] = x; ck[1] = y; ck[2] = z;
normalize(ck);
Householder_vector<T >(ck, 0, 2-sw, v, beta);
Householder_mult<T >(H,v, beta,0,k+1,k+3-sw,0);
Householder_mult<T >(H,v, beta,0,k+1,k+3-sw,1);
///Accumulate eigenvector
Householder_mult<T >(P,v, beta,0,k+1,k+3-sw,1);
}
it++;
tot_it++;
}while(N > 1);
N = evals.size();
///Annoying - UT solves in reverse order;
DenseVector<T> tmp; Resize(tmp,N);
for(int i=0;i<N;i++){
tmp[i] = evals[N-i-1];
}
evals = tmp;
UTeigenvectors(H, trows, evals, evecs);
for(int i=0;i<evals.size();i++){evecs[i] = P*evecs[i]; normalize(evecs[i]);}
return tot_it;
}
template <class T>
int my_Wilkinson(DenseMatrix<T> &Hin, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small)
{
/**
Find the eigenvalues of an upper Hessenberg matrix using the Wilkinson QR algorithm.
H =
x x 0 0 0 0
x x x 0 0 0
0 x x x 0 0
0 0 x x x 0
0 0 0 x x x
0 0 0 0 x x
Factorization is P T P^H where T is upper triangular (mod cc blocks) and P is orthagonal/unitary. **/
return my_Wilkinson(Hin, evals, evecs, small, small);
}
template <class T>
int my_Wilkinson(DenseMatrix<T> &Hin, DenseVector<T> &evals, DenseMatrix<T> &evecs, RealD small, RealD tol)
{
int N; SizeSquare(Hin,N);
int M = N;
///I don't want to modify the input but matricies must be passed by reference
//Scale a matrix by its "norm"
//RealD Hnorm = abs( Hin.LargestDiag() ); H = H*(1.0/Hnorm);
DenseMatrix<T> H; H = Hin;
RealD Hnorm = abs(Norm(Hin));
H = H * (1.0 / Hnorm);
// TODO use openmp and memset
Fill(evals,0);
Fill(evecs,0);
T s, t, x = 0, y = 0, z = 0;
T u, d;
T apd, amd, bc;
DenseVector<T> p; Resize(p,N); Fill(p,0);
T nrm = Norm(H); ///DenseMatrix Norm
int n, m;
int e = 0;
int it = 0;
int tot_it = 0;
int l = 0;
int r = 0;
DenseMatrix<T> P; Resize(P,N,N);
Unity(P);
DenseVector<int> trows(N, 0);
/// Check if the matrix is really symm tridiag
RealD sth = 0;
for(int j = 0; j < N; ++j)
{
for(int i = j + 2; i < N; ++i)
{
if(abs(H[i][j]) > tol || abs(H[j][i]) > tol)
{
std::cout << "Non Tridiagonal H(" << i << ","<< j << ") = |" << Real( real( H[j][i] ) ) << "| > " << tol << std::endl;
std::cout << "Warning tridiagonalize and call again" << std::endl;
// exit(1); // see what is going on
//return;
}
}
}
do{
do{
//Jasper
//Check if the subdiagonal term is small enough (<small)
//if true then it is converged.
//check start from H.dim - e - 1
//How to deal with more than 2 are converged?
//What if Chop_symm_subdiag return something int the middle?
//--------------
l = Chop_symm_subdiag(H,nrm, e, small);
r = 0; ///May have converged on more than one eval
//Jasper
//In this case
// x x 0 0 0 0
// x x x 0 0 0
// 0 x x x 0 0
// 0 0 x x x 0
// 0 0 0 x x 0
// 0 0 0 0 0 x <- l
//--------------
///Single eval
if(l == N - 1)
{
evals[e] = H[l][l];
N--;
e++;
r++;
it = 0;
}
//Jasper
// x x 0 0 0 0
// x x x 0 0 0
// 0 x x x 0 0
// 0 0 x x 0 0
// 0 0 0 0 x x <- l
// 0 0 0 0 x x
//--------------
///RealD eval
if(l == N - 2)
{
trows[l + 1] = 1; ///Needed for UTSolve
apd = H[l][l] + H[l + 1][ l + 1];
amd = H[l][l] - H[l + 1][l + 1];
bc = (T) 4.0 * H[l + 1][l] * H[l][l + 1];
evals[e] = (T) 0.5 * (apd + sqrt(amd * amd + bc));
evals[e + 1] = (T) 0.5 * (apd - sqrt(amd * amd + bc));
N -= 2;
e += 2;
r++;
it = 0;
}
}while(r > 0);
//Jasper
//Already converged
//--------------
if(N == 0) break;
DenseVector<T> ck,v; Resize(ck,2); Resize(v,2);
for(int m = N - 3; m >= l; m--)
{
///Starting vector essentially random shift.
if(it%10 == 0 && N >= 3 && it > 0)
{
t = abs(H[N - 1][N - 2]) + abs(H[N - 2][N - 3]);
x = H[m][m] - t;
z = H[m + 1][m];
} else {
///Starting vector implicit Q theorem
d = (H[N - 2][N - 2] - H[N - 1][N - 1]) * (T) 0.5;
t = H[N - 1][N - 1] - H[N - 1][N - 2] * H[N - 1][N - 2]
/ (d + sign(d) * sqrt(d * d + H[N - 1][N - 2] * H[N - 1][N - 2]));
x = H[m][m] - t;
z = H[m + 1][m];
}
//Jasper
//why it is here????
//-----------------------
if(m == l)
break;
u = abs(H[m][m - 1]) * (abs(y) + abs(z));
d = abs(x) * (abs(H[m - 1][m - 1]) + abs(H[m][m]) + abs(H[m + 1][m + 1]));
if ((T)abs(u + d) == (T)abs(d))
{
l = m;
break;
}
}
//Jasper
if(it > 1000000)
{
std::cout << "Wilkinson: bugger it got stuck after 100000 iterations" << std::endl;
std::cout << "got " << e << " evals " << l << " " << N << std::endl;
exit(1);
}
//
T s, c;
Givens_calc<T>(x, z, c, s);
Givens_mult<T>(H, l, l + 1, c, -s, 0);
Givens_mult<T>(H, l, l + 1, c, s, 1);
Givens_mult<T>(P, l, l + 1, c, s, 1);
//
for(int k = l; k < N - 2; ++k)
{
x = H.A[k + 1][k];
z = H.A[k + 2][k];
Givens_calc<T>(x, z, c, s);
Givens_mult<T>(H, k + 1, k + 2, c, -s, 0);
Givens_mult<T>(H, k + 1, k + 2, c, s, 1);
Givens_mult<T>(P, k + 1, k + 2, c, s, 1);
}
it++;
tot_it++;
}while(N > 1);
N = evals.size();
///Annoying - UT solves in reverse order;
DenseVector<T> tmp(N);
for(int i = 0; i < N; ++i)
tmp[i] = evals[N-i-1];
evals = tmp;
//
UTeigenvectors(H, trows, evals, evecs);
//UTSymmEigenvectors(H, trows, evals, evecs);
for(int i = 0; i < evals.size(); ++i)
{
evecs[i] = P * evecs[i];
normalize(evecs[i]);
evals[i] = evals[i] * Hnorm;
}
// // FIXME this is to test
// Hin.write("evecs3", evecs);
// Hin.write("evals3", evals);
// // check rsd
// for(int i = 0; i < M; i++) {
// vector<T> Aevec = Hin * evecs[i];
// RealD norm2(0.);
// for(int j = 0; j < M; j++) {
// norm2 += (Aevec[j] - evals[i] * evecs[i][j]) * (Aevec[j] - evals[i] * evecs[i][j]);
// }
// }
return tot_it;
}
template <class T>
void Hess(DenseMatrix<T > &A, DenseMatrix<T> &Q, int start){
/**
turn a matrix A =
x x x x x
x x x x x
x x x x x
x x x x x
x x x x x
into
x x x x x
x x x x x
0 x x x x
0 0 x x x
0 0 0 x x
with householder rotations
Slow.
*/
int N ; SizeSquare(A,N);
DenseVector<T > p; Resize(p,N); Fill(p,0);
for(int k=start;k<N-2;k++){
//cerr << "hess" << k << std::endl;
DenseVector<T > ck,v; Resize(ck,N-k-1); Resize(v,N-k-1);
for(int i=k+1;i<N;i++){ck[i-k-1] = A(i,k);} ///kth column
normalize(ck); ///Normalization cancels in PHP anyway
T beta;
Householder_vector<T >(ck, 0, ck.size()-1, v, beta); ///Householder vector
Householder_mult<T>(A,v,beta,start,k+1,N-1,0); ///A -> PA
Householder_mult<T >(A,v,beta,start,k+1,N-1,1); ///PA -> PAP^H
///Accumulate eigenvector
Householder_mult<T >(Q,v,beta,start,k+1,N-1,1); ///Q -> QP^H
}
/*for(int l=0;l<N-2;l++){
for(int k=l+2;k<N;k++){
A(0,k,l);
}
}*/
}
template <class T>
void Tri(DenseMatrix<T > &A, DenseMatrix<T> &Q, int start){
///Tridiagonalize a matrix
int N; SizeSquare(A,N);
Hess(A,Q,start);
/*for(int l=0;l<N-2;l++){
for(int k=l+2;k<N;k++){
A(0,l,k);
}
}*/
}
template <class T>
void ForceTridiagonal(DenseMatrix<T> &A){
///Tridiagonalize a matrix
int N ; SizeSquare(A,N);
for(int l=0;l<N-2;l++){
for(int k=l+2;k<N;k++){
A[l][k]=0;
A[k][l]=0;
}
}
}
template <class T>
int my_SymmEigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
///Solve a symmetric eigensystem, not necessarily in tridiagonal form
int N; SizeSquare(Ain,N);
DenseMatrix<T > A; A = Ain;
DenseMatrix<T > Q; Resize(Q,N,N); Unity(Q);
Tri(A,Q,0);
int it = my_Wilkinson<T>(A, evals, evecs, small);
for(int k=0;k<N;k++){evecs[k] = Q*evecs[k];}
return it;
}
template <class T>
int Wilkinson(DenseMatrix<T> &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
return my_Wilkinson(Ain, evals, evecs, small);
}
template <class T>
int SymmEigensystem(DenseMatrix<T> &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
return my_SymmEigensystem(Ain, evals, evecs, small);
}
template <class T>
int Eigensystem(DenseMatrix<T > &Ain, DenseVector<T> &evals, DenseVector<DenseVector<T> > &evecs, RealD small){
///Solve a general eigensystem, not necessarily in tridiagonal form
int N = Ain.dim;
DenseMatrix<T > A(N); A = Ain;
DenseMatrix<T > Q(N);Q.Unity();
Hess(A,Q,0);
int it = QReigensystem<T>(A, evals, evecs, small);
for(int k=0;k<N;k++){evecs[k] = Q*evecs[k];}
return it;
}
}
#endif

242
lib/algorithms/densematrix/Householder.h
View File

@ -1,242 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/algorithms/iterative/Householder.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 HOUSEHOLDER_H
#define HOUSEHOLDER_H
#define TIMER(A) std::cout << GridLogMessage << __FUNC__ << " file "<< __FILE__ <<" line " << __LINE__ << std::endl;
#define ENTER() std::cout << GridLogMessage << "ENTRY "<<__FUNC__ << " file "<< __FILE__ <<" line " << __LINE__ << std::endl;
#define LEAVE() std::cout << GridLogMessage << "EXIT "<<__FUNC__ << " file "<< __FILE__ <<" line " << __LINE__ << std::endl;
#include <cstdlib>
#include <string>
#include <cmath>
#include <iostream>
#include <sstream>
#include <stdexcept>
#include <fstream>
#include <complex>
#include <algorithm>
namespace Grid {
/** Comparison function for finding the max element in a vector **/
template <class T> bool cf(T i, T j) {
return abs(i) < abs(j);
}
/**
Calculate a real Givens angle
**/
template <class T> inline void Givens_calc(T y, T z, T &c, T &s){
RealD mz = (RealD)abs(z);
if(mz==0.0){
c = 1; s = 0;
}
if(mz >= (RealD)abs(y)){
T t = -y/z;
s = (T)1.0 / sqrt ((T)1.0 + t * t);
c = s * t;
} else {
T t = -z/y;
c = (T)1.0 / sqrt ((T)1.0 + t * t);
s = c * t;
}
}
template <class T> inline void Givens_mult(DenseMatrix<T> &A, int i, int k, T c, T s, int dir)
{
int q ; SizeSquare(A,q);
if(dir == 0){
for(int j=0;j<q;j++){
T nu = A[i][j];
T w = A[k][j];
A[i][j] = (c*nu + s*w);
A[k][j] = (-s*nu + c*w);
}
}
if(dir == 1){
for(int j=0;j<q;j++){
T nu = A[j][i];
T w = A[j][k];
A[j][i] = (c*nu - s*w);
A[j][k] = (s*nu + c*w);
}
}
}
/**
from input = x;
Compute the complex Householder vector, v, such that
P = (I - b v transpose(v) )
b = 2/v.v
P | x | | x | k = 0
| x | | 0 |
| x | = | 0 |
| x | | 0 | j = 3
| x | | x |
These are the "Unreduced" Householder vectors.
**/
template <class T> inline void Householder_vector(DenseVector<T> input, int k, int j, DenseVector<T> &v, T &beta)
{
int N ; Size(input,N);
T m = *max_element(input.begin() + k, input.begin() + j + 1, cf<T> );
if(abs(m) > 0.0){
T alpha = 0;
for(int i=k; i<j+1; i++){
v[i] = input[i]/m;
alpha = alpha + v[i]*conj(v[i]);
}
alpha = sqrt(alpha);
beta = (T)1.0/(alpha*(alpha + abs(v[k]) ));
if(abs(v[k]) > 0.0) v[k] = v[k] + (v[k]/abs(v[k]))*alpha;
else v[k] = -alpha;
} else{
for(int i=k; i<j+1; i++){
v[i] = 0.0;
}
}
}
/**
from input = x;
Compute the complex Householder vector, v, such that
P = (I - b v transpose(v) )
b = 2/v.v
Px = alpha*e_dir
These are the "Unreduced" Householder vectors.
**/
template <class T> inline void Householder_vector(DenseVector<T> input, int k, int j, int dir, DenseVector<T> &v, T &beta)
{
int N = input.size();
T m = *max_element(input.begin() + k, input.begin() + j + 1, cf);
if(abs(m) > 0.0){
T alpha = 0;
for(int i=k; i<j+1; i++){
v[i] = input[i]/m;
alpha = alpha + v[i]*conj(v[i]);
}
alpha = sqrt(alpha);
beta = 1.0/(alpha*(alpha + abs(v[dir]) ));
if(abs(v[dir]) > 0.0) v[dir] = v[dir] + (v[dir]/abs(v[dir]))*alpha;
else v[dir] = -alpha;
}else{
for(int i=k; i<j+1; i++){
v[i] = 0.0;
}
}
}
/**
Compute the product PA if trans = 0
AP if trans = 1
P = (I - b v transpose(v) )
b = 2/v.v
start at element l of matrix A
v is of length j - k + 1 of v are nonzero
**/
template <class T> inline void Householder_mult(DenseMatrix<T> &A , DenseVector<T> v, T beta, int l, int k, int j, int trans)
{
int N ; SizeSquare(A,N);
if(abs(beta) > 0.0){
for(int p=l; p<N; p++){
T s = 0;
if(trans==0){
for(int i=k;i<j+1;i++) s += conj(v[i-k])*A[i][p];
s *= beta;
for(int i=k;i<j+1;i++){ A[i][p] = A[i][p]-s*conj(v[i-k]);}
} else {
for(int i=k;i<j+1;i++){ s += conj(v[i-k])*A[p][i];}
s *= beta;
for(int i=k;i<j+1;i++){ A[p][i]=A[p][i]-s*conj(v[i-k]);}
}
}
}
}
/**
Compute the product PA if trans = 0
AP if trans = 1
P = (I - b v transpose(v) )
b = 2/v.v
start at element l of matrix A
v is of length j - k + 1 of v are nonzero
A is tridiagonal
**/
template <class T> inline void Householder_mult_tri(DenseMatrix<T> &A , DenseVector<T> v, T beta, int l, int M, int k, int j, int trans)
{
if(abs(beta) > 0.0){
int N ; SizeSquare(A,N);
DenseMatrix<T> tmp; Resize(tmp,N,N); Fill(tmp,0);
T s;
for(int p=l; p<M; p++){
s = 0;
if(trans==0){
for(int i=k;i<j+1;i++) s = s + conj(v[i-k])*A[i][p];
}else{
for(int i=k;i<j+1;i++) s = s + v[i-k]*A[p][i];
}
s = beta*s;
if(trans==0){
for(int i=k;i<j+1;i++) tmp[i][p] = tmp(i,p) - s*v[i-k];
}else{
for(int i=k;i<j+1;i++) tmp[p][i] = tmp[p][i] - s*conj(v[i-k]);
}
}
for(int p=l; p<M; p++){
if(trans==0){
for(int i=k;i<j+1;i++) A[i][p] = A[i][p] + tmp[i][p];
}else{
for(int i=k;i<j+1;i++) A[p][i] = A[p][i] + tmp[p][i];
}
}
}
}
}
#endif

307
lib/algorithms/iterative/BlockConjugateGradient.h
View File

@ -33,6 +33,8 @@ directory
namespace Grid {
enum BlockCGtype { BlockCG, BlockCGrQ, CGmultiRHS };
//////////////////////////////////////////////////////////////////////////
// Block conjugate gradient. Dimension zero should be the block direction
//////////////////////////////////////////////////////////////////////////
@ -40,25 +42,273 @@ template <class Field>
class BlockConjugateGradient : public OperatorFunction<Field> {
public:
typedef typename Field::scalar_type scomplex;
const int blockDim = 0;
int blockDim ;
int Nblock;
BlockCGtype CGtype;
bool ErrorOnNoConverge; // throw an assert when the CG fails to converge.
// Defaults true.
RealD Tolerance;
Integer MaxIterations;
Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
BlockConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
: Tolerance(tol),
MaxIterations(maxit),
ErrorOnNoConverge(err_on_no_conv){};
BlockConjugateGradient(BlockCGtype cgtype,int _Orthog,RealD tol, Integer maxit, bool err_on_no_conv = true)
: Tolerance(tol), CGtype(cgtype), blockDim(_Orthog), MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv)
{};
////////////////////////////////////////////////////////////////////////////////////////////////////
// Thin QR factorisation (google it)
////////////////////////////////////////////////////////////////////////////////////////////////////
void ThinQRfact (Eigen::MatrixXcd &m_rr,
Eigen::MatrixXcd &C,
Eigen::MatrixXcd &Cinv,
Field & Q,
const Field & R)
{
int Orthog = blockDim; // First dimension is block dim; this is an assumption
////////////////////////////////////////////////////////////////////////////////////////////////////
//Dimensions
// R_{ferm x Nblock} = Q_{ferm x Nblock} x C_{Nblock x Nblock} -> ferm x Nblock
//
// Rdag R = m_rr = Herm = L L^dag <-- Cholesky decomposition (LLT routine in Eigen)
//
// Q C = R => Q = R C^{-1}
//
// Want Ident = Q^dag Q = C^{-dag} R^dag R C^{-1} = C^{-dag} L L^dag C^{-1} = 1_{Nblock x Nblock}
//
// Set C = L^{dag}, and then Q^dag Q = ident
//
// Checks:
// Cdag C = Rdag R ; passes.
// QdagQ = 1 ; passes
////////////////////////////////////////////////////////////////////////////////////////////////////
sliceInnerProductMatrix(m_rr,R,R,Orthog);
////////////////////////////////////////////////////////////////////////////////////////////////////
// Cholesky from Eigen
// There exists a ldlt that is documented as more stable
////////////////////////////////////////////////////////////////////////////////////////////////////
Eigen::MatrixXcd L = m_rr.llt().matrixL();
C = L.adjoint();
Cinv = C.inverse();
////////////////////////////////////////////////////////////////////////////////////////////////////
// Q = R C^{-1}
//
// Q_j = R_i Cinv(i,j)
//
// NB maddMatrix conventions are Right multiplication X[j] a[j,i] already
////////////////////////////////////////////////////////////////////////////////////////////////////
// FIXME:: make a sliceMulMatrix to avoid zero vector
sliceMulMatrix(Q,Cinv,R,Orthog);
}
////////////////////////////////////////////////////////////////////////////////////////////////////
// Call one of several implementations
////////////////////////////////////////////////////////////////////////////////////////////////////
void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
{
int Orthog = 0; // First dimension is block dim
if ( CGtype == BlockCGrQ ) {
BlockCGrQsolve(Linop,Src,Psi);
} else if (CGtype == BlockCG ) {
BlockCGsolve(Linop,Src,Psi);
} else if (CGtype == CGmultiRHS ) {
CGmultiRHSsolve(Linop,Src,Psi);
} else {
assert(0);
}
}
////////////////////////////////////////////////////////////////////////////
// BlockCGrQ implementation:
//--------------------------
// X is guess/Solution
// B is RHS
// Solve A X_i = B_i ; i refers to Nblock index
////////////////////////////////////////////////////////////////////////////
void BlockCGrQsolve(LinearOperatorBase<Field> &Linop, const Field &B, Field &X)
{
int Orthog = blockDim; // First dimension is block dim; this is an assumption
Nblock = B._grid->_fdimensions[Orthog];
std::cout<<GridLogMessage<<" Block Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
X.checkerboard = B.checkerboard;
conformable(X, B);
Field tmp(B);
Field Q(B);
Field D(B);
Field Z(B);
Field AD(B);
Eigen::MatrixXcd m_DZ = Eigen::MatrixXcd::Identity(Nblock,Nblock);
Eigen::MatrixXcd m_M = Eigen::MatrixXcd::Identity(Nblock,Nblock);
Eigen::MatrixXcd m_rr = Eigen::MatrixXcd::Zero(Nblock,Nblock);
Eigen::MatrixXcd m_C = Eigen::MatrixXcd::Zero(Nblock,Nblock);
Eigen::MatrixXcd m_Cinv = Eigen::MatrixXcd::Zero(Nblock,Nblock);
Eigen::MatrixXcd m_S = Eigen::MatrixXcd::Zero(Nblock,Nblock);
Eigen::MatrixXcd m_Sinv = Eigen::MatrixXcd::Zero(Nblock,Nblock);
Eigen::MatrixXcd m_tmp = Eigen::MatrixXcd::Identity(Nblock,Nblock);
Eigen::MatrixXcd m_tmp1 = Eigen::MatrixXcd::Identity(Nblock,Nblock);
// Initial residual computation & set up
std::vector<RealD> residuals(Nblock);
std::vector<RealD> ssq(Nblock);
sliceNorm(ssq,B,Orthog);
RealD sssum=0;
for(int b=0;b<Nblock;b++) sssum+=ssq[b];
sliceNorm(residuals,B,Orthog);
for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
sliceNorm(residuals,X,Orthog);
for(int b=0;b<Nblock;b++){ assert(std::isnan(residuals[b])==0); }
/************************************************************************
* Block conjugate gradient rQ (Sebastien Birk Thesis, after Dubrulle 2001)
************************************************************************
* Dimensions:
*
* X,B==(Nferm x Nblock)
* A==(Nferm x Nferm)
*
* Nferm = Nspin x Ncolour x Ncomplex x Nlattice_site
*
* QC = R = B-AX, D = Q ; QC => Thin QR factorisation (google it)
* for k:
* Z = AD
* M = [D^dag Z]^{-1}
* X = X + D MC
* QS = Q - ZM
* D = Q + D S^dag
* C = S C
*/
///////////////////////////////////////
// Initial block: initial search dir is guess
///////////////////////////////////////
std::cout << GridLogMessage<<"BlockCGrQ algorithm initialisation " <<std::endl;
//1. QC = R = B-AX, D = Q ; QC => Thin QR factorisation (google it)
Linop.HermOp(X, AD);
tmp = B - AD;
ThinQRfact (m_rr, m_C, m_Cinv, Q, tmp);
D=Q;
std::cout << GridLogMessage<<"BlockCGrQ computed initial residual and QR fact " <<std::endl;
///////////////////////////////////////
// Timers
///////////////////////////////////////
GridStopWatch sliceInnerTimer;
GridStopWatch sliceMaddTimer;
GridStopWatch QRTimer;
GridStopWatch MatrixTimer;
GridStopWatch SolverTimer;
SolverTimer.Start();
int k;
for (k = 1; k <= MaxIterations; k++) {
//3. Z = AD
MatrixTimer.Start();
Linop.HermOp(D, Z);
MatrixTimer.Stop();
//4. M = [D^dag Z]^{-1}
sliceInnerTimer.Start();
sliceInnerProductMatrix(m_DZ,D,Z,Orthog);
sliceInnerTimer.Stop();
m_M = m_DZ.inverse();
//5. X = X + D MC
m_tmp = m_M * m_C;
sliceMaddTimer.Start();
sliceMaddMatrix(X,m_tmp, D,X,Orthog);
sliceMaddTimer.Stop();
//6. QS = Q - ZM
sliceMaddTimer.Start();
sliceMaddMatrix(tmp,m_M,Z,Q,Orthog,-1.0);
sliceMaddTimer.Stop();
QRTimer.Start();
ThinQRfact (m_rr, m_S, m_Sinv, Q, tmp);
QRTimer.Stop();
//7. D = Q + D S^dag
m_tmp = m_S.adjoint();
sliceMaddTimer.Start();
sliceMaddMatrix(D,m_tmp,D,Q,Orthog);
sliceMaddTimer.Stop();
//8. C = S C
m_C = m_S*m_C;
/*********************
* convergence monitor
*********************
*/
m_rr = m_C.adjoint() * m_C;
RealD max_resid=0;
RealD rrsum=0;
RealD rr;
for(int b=0;b<Nblock;b++) {
rrsum+=real(m_rr(b,b));
rr = real(m_rr(b,b))/ssq[b];
if ( rr > max_resid ) max_resid = rr;
}
std::cout << GridLogIterative << "\titeration "<<k<<" rr_sum "<<rrsum<<" ssq_sum "<< sssum
<<" ave "<<std::sqrt(rrsum/sssum) << " max "<< max_resid <<std::endl;
if ( max_resid < Tolerance*Tolerance ) {
SolverTimer.Stop();
std::cout << GridLogMessage<<"BlockCGrQ converged in "<<k<<" iterations"<<std::endl;
for(int b=0;b<Nblock;b++){
std::cout << GridLogMessage<< "\t\tblock "<<b<<" computed resid "
<< std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
}
std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
Linop.HermOp(X, AD);
AD = AD-B;
std::cout << GridLogMessage <<"\t True residual is " << std::sqrt(norm2(AD)/norm2(B)) <<std::endl;
std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
std::cout << GridLogMessage << "\tElapsed " << SolverTimer.Elapsed() <<std::endl;
std::cout << GridLogMessage << "\tMatrix " << MatrixTimer.Elapsed() <<std::endl;
std::cout << GridLogMessage << "\tInnerProd " << sliceInnerTimer.Elapsed() <<std::endl;
std::cout << GridLogMessage << "\tMaddMatrix " << sliceMaddTimer.Elapsed() <<std::endl;
std::cout << GridLogMessage << "\tThinQRfact " << QRTimer.Elapsed() <<std::endl;
IterationsToComplete = k;
return;
}
}
std::cout << GridLogMessage << "BlockConjugateGradient(rQ) did NOT converge" << std::endl;
if (ErrorOnNoConverge) assert(0);
IterationsToComplete = k;
}
//////////////////////////////////////////////////////////////////////////
// Block conjugate gradient; Original O'Leary Dimension zero should be the block direction
//////////////////////////////////////////////////////////////////////////
void BlockCGsolve(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
{
int Orthog = blockDim; // First dimension is block dim; this is an assumption
Nblock = Src._grid->_fdimensions[Orthog];
std::cout<<GridLogMessage<<" Block Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
@ -162,8 +412,9 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
*********************
*/
RealD max_resid=0;
RealD rr;
for(int b=0;b<Nblock;b++){
RealD rr = real(m_rr(b,b))/ssq[b];
rr = real(m_rr(b,b))/ssq[b];
if ( rr > max_resid ) max_resid = rr;
}
@ -173,13 +424,14 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
std::cout << GridLogMessage<<"BlockCG converged in "<<k<<" iterations"<<std::endl;
for(int b=0;b<Nblock;b++){
std::cout << GridLogMessage<< "\t\tblock "<<b<<" resid "<< std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
std::cout << GridLogMessage<< "\t\tblock "<<b<<" computed resid "
<< std::sqrt(real(m_rr(b,b))/ssq[b])<<std::endl;
}
std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
Linop.HermOp(Psi, AP);
AP = AP-Src;
std::cout << GridLogMessage <<"\tTrue residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<std::endl;
std::cout << GridLogMessage <<"\t True residual is " << std::sqrt(norm2(AP)/norm2(Src)) <<std::endl;
std::cout << GridLogMessage << "Time Breakdown "<<std::endl;
std::cout << GridLogMessage << "\tElapsed " << SolverTimer.Elapsed() <<std::endl;
@ -197,35 +449,13 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
if (ErrorOnNoConverge) assert(0);
IterationsToComplete = k;
}
};
//////////////////////////////////////////////////////////////////////////
// multiRHS conjugate gradient. Dimension zero should be the block direction
// Use this for spread out across nodes
//////////////////////////////////////////////////////////////////////////
template <class Field>
class MultiRHSConjugateGradient : public OperatorFunction<Field> {
public:
typedef typename Field::scalar_type scomplex;
const int blockDim = 0;
int Nblock;
bool ErrorOnNoConverge; // throw an assert when the CG fails to converge.
// Defaults true.
RealD Tolerance;
Integer MaxIterations;
Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion
MultiRHSConjugateGradient(RealD tol, Integer maxit, bool err_on_no_conv = true)
: Tolerance(tol),
MaxIterations(maxit),
ErrorOnNoConverge(err_on_no_conv){};
void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
void CGmultiRHSsolve(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
{
int Orthog = 0; // First dimension is block dim
int Orthog = blockDim; // First dimension is block dim
Nblock = Src._grid->_fdimensions[Orthog];
std::cout<<GridLogMessage<<"MultiRHS Conjugate Gradient : Orthog "<<Orthog<<" Nblock "<<Nblock<<std::endl;
@ -285,12 +515,10 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
MatrixTimer.Stop();
// Alpha
// sliceInnerProductVectorTest(v_pAp_test,P,AP,Orthog);
sliceInnerTimer.Start();
sliceInnerProductVector(v_pAp,P,AP,Orthog);
sliceInnerTimer.Stop();
for(int b=0;b<Nblock;b++){
// std::cout << " "<< v_pAp[b]<<" "<< v_pAp_test[b]<<std::endl;
v_alpha[b] = v_rr[b]/real(v_pAp[b]);
}
@ -332,7 +560,7 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
std::cout << GridLogMessage<<"MultiRHS solver converged in " <<k<<" iterations"<<std::endl;
for(int b=0;b<Nblock;b++){
std::cout << GridLogMessage<< "\t\tBlock "<<b<<" resid "<< std::sqrt(v_rr[b]/ssq[b])<<std::endl;
std::cout << GridLogMessage<< "\t\tBlock "<<b<<" computed resid "<< std::sqrt(v_rr[b]/ssq[b])<<std::endl;
}
std::cout << GridLogMessage<<"\tMax residual is "<<std::sqrt(max_resid)<<std::endl;
@ -358,9 +586,8 @@ void operator()(LinearOperatorBase<Field> &Linop, const Field &Src, Field &Psi)
if (ErrorOnNoConverge) assert(0);
IterationsToComplete = k;
}
};
}
#endif

7
lib/algorithms/iterative/ConjugateGradient.h
View File

@ -123,8 +123,11 @@ class ConjugateGradient : public OperatorFunction<Field> {
p = p * b + r;
LinalgTimer.Stop();
std::cout << GridLogIterative << "ConjugateGradient: Iteration " << k
<< " residual " << cp << " target " << rsq << std::endl;
std::cout << GridLogDebug << "a = "<< a << " b_pred = "<< b_pred << " b = "<< b << std::endl;
std::cout << GridLogDebug << "qq = "<< qq << " d = "<< d << " c = "<< c << std::endl;
// Stopping condition
if (cp <= rsq) {
@ -132,8 +135,6 @@ class ConjugateGradient : public OperatorFunction<Field> {
Linop.HermOpAndNorm(psi, mmp, d, qq);
p = mmp - src;
RealD mmpnorm = sqrt(norm2(mmp));
RealD psinorm = sqrt(norm2(psi));
RealD srcnorm = sqrt(norm2(src));
RealD resnorm = sqrt(norm2(p));
RealD true_residual = resnorm / srcnorm;
@ -157,8 +158,10 @@ class ConjugateGradient : public OperatorFunction<Field> {
}
std::cout << GridLogMessage << "ConjugateGradient did NOT converge"
<< std::endl;
if (ErrorOnNoConverge) assert(0);
IterationsToComplete = k;
}
};
}

81
lib/algorithms/iterative/EigenSort.h
View File

@ -1,81 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/algorithms/iterative/EigenSort.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_EIGENSORT_H
#define GRID_EIGENSORT_H
namespace Grid {
/////////////////////////////////////////////////////////////
// Eigen sorter to begin with
/////////////////////////////////////////////////////////////
template<class Field>
class SortEigen {
private:
//hacking for testing for now
private:
static bool less_lmd(RealD left,RealD right){
return left > right;
}
static bool less_pair(std::pair<RealD,Field const*>& left,
std::pair<RealD,Field const*>& right){
return left.first > (right.first);
}
public:
void push(DenseVector<RealD>& lmd,
DenseVector<Field>& evec,int N) {
DenseVector<Field> cpy(lmd.size(),evec[0]._grid);
for(int i=0;i<lmd.size();i++) cpy[i] = evec[i];
DenseVector<std::pair<RealD, Field const*> > emod(lmd.size());
for(int i=0;i<lmd.size();++i)
emod[i] = std::pair<RealD,Field const*>(lmd[i],&cpy[i]);
partial_sort(emod.begin(),emod.begin()+N,emod.end(),less_pair);
typename DenseVector<std::pair<RealD, Field const*> >::iterator it = emod.begin();
for(int i=0;i<N;++i){
lmd[i]=it->first;
evec[i]=*(it->second);
++it;
}
}
void push(DenseVector<RealD>& lmd,int N) {
std::partial_sort(lmd.begin(),lmd.begin()+N,lmd.end(),less_lmd);
}
bool saturated(RealD lmd, RealD thrs) {
return fabs(lmd) > fabs(thrs);
}
};
}
#endif

1635
lib/algorithms/iterative/ImplicitlyRestartedLanczos.h
View File

File diff suppressed because it is too large Load Diff

16
lib/allocator/AlignedAllocator.h
View File

@ -98,7 +98,14 @@ public:
#else
if ( ptr == (_Tp *) NULL ) ptr = (_Tp *) memalign(128,bytes);
#endif
// First touch optimise in threaded loop
uint8_t *cp = (uint8_t *)ptr;
#ifdef GRID_OMP
#pragma omp parallel for
#endif
for(size_type n=0;n<bytes;n+=4096){
cp[n]=0;
}
return ptr;
}
@ -186,6 +193,13 @@ public:
#else
_Tp * ptr = (_Tp *) memalign(128,__n*sizeof(_Tp));
#endif
size_type bytes = __n*sizeof(_Tp);
uint8_t *cp = (uint8_t *)ptr;
// One touch per 4k page, static OMP loop to catch same loop order
#pragma omp parallel for schedule(static)
for(size_type n=0;n<bytes;n+=4096){
cp[n]=0;
}
return ptr;
}
void deallocate(pointer __p, size_type) {

74
lib/cartesian/Cartesian_base.h
View File

@ -6,8 +6,9 @@
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@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
@ -49,7 +50,6 @@ public:
GridBase(const std::vector<int> & processor_grid) : CartesianCommunicator(processor_grid) {};
// Physics Grid information.
std::vector<int> _simd_layout;// Which dimensions get relayed out over simd lanes.
std::vector<int> _fdimensions;// (full) Global dimensions of array prior to cb removal
@ -62,13 +62,12 @@ public:
int _isites;
int _fsites; // _isites*_osites = product(dimensions).
int _gsites;
std::vector<int> _slice_block; // subslice information
std::vector<int> _slice_block;// subslice information
std::vector<int> _slice_stride;
std::vector<int> _slice_nblock;
// Might need these at some point
// std::vector<int> _lstart; // local start of array in gcoors. _processor_coor[d]*_ldimensions[d]
// std::vector<int> _lend; // local end of array in gcoors _processor_coor[d]*_ldimensions[d]+_ldimensions_[d]-1
std::vector<int> _lstart; // local start of array in gcoors _processor_coor[d]*_ldimensions[d]
std::vector<int> _lend ; // local end of array in gcoors _processor_coor[d]*_ldimensions[d]+_ldimensions_[d]-1
public:
@ -99,7 +98,7 @@ public:
virtual int oIndex(std::vector<int> &coor)
{
int idx=0;
// Works with either global or local coordinates
// Works with either global or local coordinates
for(int d=0;d<_ndimension;d++) idx+=_ostride[d]*(coor[d]%_rdimensions[d]);
return idx;
}
@ -121,6 +120,12 @@ public:
Lexicographic::CoorFromIndex(coor,Oindex,_rdimensions);
}
inline void InOutCoorToLocalCoor (std::vector<int> &ocoor, std::vector<int> &icoor, std::vector<int> &lcoor) {
lcoor.resize(_ndimension);
for (int d = 0; d < _ndimension; d++)
lcoor[d] = ocoor[d] + _rdimensions[d] * icoor[d];
}
//////////////////////////////////////////////////////////
// SIMD lane addressing
//////////////////////////////////////////////////////////
@ -128,6 +133,7 @@ public:
{
Lexicographic::CoorFromIndex(coor,lane,_simd_layout);
}
inline int PermuteDim(int dimension){
return _simd_layout[dimension]>1;
}
@ -145,15 +151,15 @@ public:
// Distance should be either 0,1,2..
//
if ( _simd_layout[dimension] > 2 ) {
for(int d=0;d<_ndimension;d++){
if ( d != dimension ) assert ( (_simd_layout[d]==1) );
}
permute_type = RotateBit; // How to specify distance; this is not just direction.
return permute_type;
for(int d=0;d<_ndimension;d++){
if ( d != dimension ) assert ( (_simd_layout[d]==1) );
}
permute_type = RotateBit; // How to specify distance; this is not just direction.
return permute_type;
}
for(int d=_ndimension-1;d>dimension;d--){
if (_simd_layout[d]>1 ) permute_type++;
if (_simd_layout[d]>1 ) permute_type++;
}
return permute_type;
}
@ -168,11 +174,30 @@ public:
inline int gSites(void) const { return _isites*_osites*_Nprocessors; };
inline int Nd (void) const { return _ndimension;};
inline const std::vector<int> LocalStarts(void) { return _lstart; };
inline const std::vector<int> &FullDimensions(void) { return _fdimensions;};
inline const std::vector<int> &GlobalDimensions(void) { return _gdimensions;};
inline const std::vector<int> &LocalDimensions(void) { return _ldimensions;};
inline const std::vector<int> &VirtualLocalDimensions(void) { return _ldimensions;};
////////////////////////////////////////////////////////////////
// Utility to print the full decomposition details
////////////////////////////////////////////////////////////////
void show_decomposition(){
std::cout << GridLogMessage << "Full Dimensions : " << _fdimensions << std::endl;
std::cout << GridLogMessage << "Global Dimensions : " << _gdimensions << std::endl;
std::cout << GridLogMessage << "Local Dimensions : " << _ldimensions << std::endl;
std::cout << GridLogMessage << "Reduced Dimensions : " << _rdimensions << std::endl;
std::cout << GridLogMessage << "Outer strides : " << _ostride << std::endl;
std::cout << GridLogMessage << "Inner strides : " << _istride << std::endl;
std::cout << GridLogMessage << "iSites : " << _isites << std::endl;
std::cout << GridLogMessage << "oSites : " << _osites << std::endl;
std::cout << GridLogMessage << "lSites : " << lSites() << std::endl;
std::cout << GridLogMessage << "gSites : " << gSites() << std::endl;
std::cout << GridLogMessage << "Nd : " << _ndimension << std::endl;
}
////////////////////////////////////////////////////////////////
// Global addressing
////////////////////////////////////////////////////////////////
@ -184,12 +209,15 @@ public:
assert(lidx<lSites());
Lexicographic::CoorFromIndex(lcoor,lidx,_ldimensions);
}
void GlobalCoorToGlobalIndex(const std::vector<int> & gcoor,int & gidx){
gidx=0;
int mult=1;
for(int mu=0;mu<_ndimension;mu++) {
gidx+=mult*gcoor[mu];
mult*=_gdimensions[mu];
gidx+=mult*gcoor[mu];
mult*=_gdimensions[mu];
}
}
void GlobalCoorToProcessorCoorLocalCoor(std::vector<int> &pcoor,std::vector<int> &lcoor,const std::vector<int> &gcoor)
@ -197,9 +225,9 @@ public:
pcoor.resize(_ndimension);
lcoor.resize(_ndimension);
for(int mu=0;mu<_ndimension;mu++){
int _fld = _fdimensions[mu]/_processors[mu];
pcoor[mu] = gcoor[mu]/_fld;
lcoor[mu] = gcoor[mu]%_fld;
int _fld = _fdimensions[mu]/_processors[mu];
pcoor[mu] = gcoor[mu]/_fld;
lcoor[mu] = gcoor[mu]%_fld;
}
}
void GlobalCoorToRankIndex(int &rank, int &o_idx, int &i_idx ,const std::vector<int> &gcoor)
@ -211,9 +239,9 @@ public:
/*
std::vector<int> cblcoor(lcoor);
for(int d=0;d<cblcoor.size();d++){
if( this->CheckerBoarded(d) ) {
cblcoor[d] = lcoor[d]/2;
}
if( this->CheckerBoarded(d) ) {
cblcoor[d] = lcoor[d]/2;
}
}
*/
i_idx= iIndex(lcoor);
@ -239,7 +267,7 @@ public:
{
RankIndexToGlobalCoor(rank,o_idx,i_idx ,fcoor);
if(CheckerBoarded(0)){
fcoor[0] = fcoor[0]*2+cb;
fcoor[0] = fcoor[0]*2+cb;
}
}
void ProcessorCoorLocalCoorToGlobalCoor(std::vector<int> &Pcoor,std::vector<int> &Lcoor,std::vector<int> &gcoor)

8
lib/cartesian/Cartesian_full.h
View File

@ -76,6 +76,8 @@ public:
_ldimensions.resize(_ndimension);
_rdimensions.resize(_ndimension);
_simd_layout.resize(_ndimension);
_lstart.resize(_ndimension);
_lend.resize(_ndimension);
_ostride.resize(_ndimension);
_istride.resize(_ndimension);
@ -94,8 +96,10 @@ public:
// Use a reduced simd grid
_ldimensions[d]= _gdimensions[d]/_processors[d]; //local dimensions
_rdimensions[d]= _ldimensions[d]/_simd_layout[d]; //overdecomposition
_osites *= _rdimensions[d];
_isites *= _simd_layout[d];
_lstart[d] = _processor_coor[d]*_ldimensions[d];
_lend[d] = _processor_coor[d]*_ldimensions[d]+_ldimensions[d]-1;
_osites *= _rdimensions[d];
_isites *= _simd_layout[d];
// Addressing support
if ( d==0 ) {

4
lib/cartesian/Cartesian_red_black.h
View File

@ -151,6 +151,8 @@ public:
_ldimensions.resize(_ndimension);
_rdimensions.resize(_ndimension);
_simd_layout.resize(_ndimension);
_lstart.resize(_ndimension);
_lend.resize(_ndimension);
_ostride.resize(_ndimension);
_istride.resize(_ndimension);
@ -169,6 +171,8 @@ public:
_gdimensions[d] = _gdimensions[d]/2; // Remove a checkerboard
}
_ldimensions[d] = _gdimensions[d]/_processors[d];
_lstart[d] = _processor_coor[d]*_ldimensions[d];
_lend[d] = _processor_coor[d]*_ldimensions[d]+_ldimensions[d]-1;
// Use a reduced simd grid
_simd_layout[d] = simd_layout[d];

2
lib/communicator/Communicator_base.cc
View File

@ -60,6 +60,7 @@ void CartesianCommunicator::ShmBufferFreeAll(void) {
/////////////////////////////////
// Grid information queries
/////////////////////////////////
int CartesianCommunicator::Dimensions(void) { return _ndimension; };
int CartesianCommunicator::IsBoss(void) { return _processor==0; };
int CartesianCommunicator::BossRank(void) { return 0; };
int CartesianCommunicator::ThisRank(void) { return _processor; };
@ -91,6 +92,7 @@ void CartesianCommunicator::GlobalSumVector(ComplexD *c,int N)
#if !defined( GRID_COMMS_MPI3) && !defined (GRID_COMMS_MPI3L)
int CartesianCommunicator::NodeCount(void) { return ProcessorCount();};
int CartesianCommunicator::RankCount(void) { return ProcessorCount();};
double CartesianCommunicator::StencilSendToRecvFromBegin(std::vector<CommsRequest_t> &list,
void *xmit,

4
lib/communicator/Communicator_base.h
View File

@ -148,6 +148,7 @@ class CartesianCommunicator {
int RankFromProcessorCoor(std::vector<int> &coor);
void ProcessorCoorFromRank(int rank,std::vector<int> &coor);
int Dimensions(void) ;
int IsBoss(void) ;
int BossRank(void) ;
int ThisRank(void) ;
@ -155,6 +156,7 @@ class CartesianCommunicator {
const std::vector<int> & ProcessorGrid(void) ;
int ProcessorCount(void) ;
int NodeCount(void) ;
int RankCount(void) ;
////////////////////////////////////////////////////////////////////////////////
// very VERY rarely (Log, serial RNG) we need world without a grid
@ -175,6 +177,8 @@ class CartesianCommunicator {
void GlobalSumVector(ComplexF *c,int N);
void GlobalSum(ComplexD &c);
void GlobalSumVector(ComplexD *c,int N);
void GlobalXOR(uint32_t &);
void GlobalXOR(uint64_t &);
template<class obj> void GlobalSum(obj &o){
typedef typename obj::scalar_type scalar_type;

8
lib/communicator/Communicator_mpi.cc
View File

@ -83,6 +83,14 @@ void CartesianCommunicator::GlobalSum(uint64_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalXOR(uint32_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_BXOR,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalXOR(uint64_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_BXOR,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSum(float &f){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
assert(ierr==0);

33
lib/communicator/Communicator_mpi3.cc
View File

@ -37,7 +37,10 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#include <sys/ipc.h>
#include <sys/shm.h>
#include <sys/mman.h>
//#include <zlib.h>
#include <zlib.h>
#ifdef HAVE_NUMAIF_H
#include <numaif.h>
#endif
#ifndef SHM_HUGETLB
#define SHM_HUGETLB 04000
#endif
@ -65,6 +68,7 @@ std::vector<int> CartesianCommunicator::MyGroup;
std::vector<void *> CartesianCommunicator::ShmCommBufs;
int CartesianCommunicator::NodeCount(void) { return GroupSize;};
int CartesianCommunicator::RankCount(void) { return WorldSize;};
#undef FORCE_COMMS
@ -213,6 +217,25 @@ void CartesianCommunicator::Init(int *argc, char ***argv) {
void * ptr = mmap(NULL,size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
if ( ptr == MAP_FAILED ) { perror("failed mmap"); assert(0); }
assert(((uint64_t)ptr&0x3F)==0);
// Try to force numa domain on the shm segment if we have numaif.h
#ifdef HAVE_NUMAIF_H
int status;
int flags=MPOL_MF_MOVE;
#ifdef KNL
int nodes=1; // numa domain == MCDRAM
// Find out if in SNC2,SNC4 mode ?
#else
int nodes=r; // numa domain == MPI ID
#endif
unsigned long count=1;
for(uint64_t page=0;page<size;page+=4096){
void *pages = (void *) ( page + (uint64_t)ptr );
uint64_t *cow_it = (uint64_t *)pages; *cow_it = 1;
ierr= move_pages(0,count, &pages,&nodes,&status,flags);
if (ierr && (page==0)) perror("numa relocate command failed");
}
#endif
ShmCommBufs[r] =ptr;
}
@ -509,6 +532,14 @@ void CartesianCommunicator::GlobalSum(uint64_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_SUM,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalXOR(uint32_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_BXOR,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalXOR(uint64_t &u){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT64_T,MPI_BXOR,communicator);
assert(ierr==0);
}
void CartesianCommunicator::GlobalSum(float &f){
int ierr=MPI_Allreduce(MPI_IN_PLACE,&f,1,MPI_FLOAT,MPI_SUM,communicator);
assert(ierr==0);

2
lib/communicator/Communicator_none.cc
View File

@ -59,6 +59,8 @@ void CartesianCommunicator::GlobalSum(double &){}
void CartesianCommunicator::GlobalSum(uint32_t &){}
void CartesianCommunicator::GlobalSum(uint64_t &){}
void CartesianCommunicator::GlobalSumVector(double *,int N){}
void CartesianCommunicator::GlobalXOR(uint32_t &){}
void CartesianCommunicator::GlobalXOR(uint64_t &){}
void CartesianCommunicator::SendRecvPacket(void *xmit,
void *recv,

55
lib/cshift/Cshift_common.h
View File

@ -30,21 +30,11 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
namespace Grid {
template<class vobj>
class SimpleCompressor {
public:
void Point(int) {};
vobj operator() (const vobj &arg) {
return arg;
}
};
///////////////////////////////////////////////////////////////////
// Gather for when there is no need to SIMD split with compression
// Gather for when there is no need to SIMD split
///////////////////////////////////////////////////////////////////
template<class vobj,class cobj,class compressor> void
Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimension,int plane,int cbmask,compressor &compress, int off=0)
template<class vobj> void
Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer,int dimension,int plane,int cbmask, int off=0)
{
int rd = rhs._grid->_rdimensions[dimension];
@ -62,7 +52,7 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimen
for(int b=0;b<e2;b++){
int o = n*stride;
int bo = n*e2;
buffer[off+bo+b]=compress(rhs._odata[so+o+b]);
buffer[off+bo+b]=rhs._odata[so+o+b];
}
}
} else {
@ -78,17 +68,16 @@ Gather_plane_simple (const Lattice<vobj> &rhs,commVector<cobj> &buffer,int dimen
}
}
parallel_for(int i=0;i<table.size();i++){
buffer[off+table[i].first]=compress(rhs._odata[so+table[i].second]);
buffer[off+table[i].first]=rhs._odata[so+table[i].second];
}
}
}
///////////////////////////////////////////////////////////////////
// Gather for when there *is* need to SIMD split with compression
// Gather for when there *is* need to SIMD split
///////////////////////////////////////////////////////////////////
template<class cobj,class vobj,class compressor> void
Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_object *> pointers,int dimension,int plane,int cbmask,compressor &compress)
template<class vobj> void
Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
{
int rd = rhs._grid->_rdimensions[dimension];
@ -109,8 +98,8 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_
int o = n*n1;
int offset = b+n*e2;
cobj temp =compress(rhs._odata[so+o+b]);
extract<cobj>(temp,pointers,offset);
vobj temp =rhs._odata[so+o+b];
extract<vobj>(temp,pointers,offset);
}
}
@ -127,32 +116,14 @@ Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename cobj::scalar_
int offset = b+n*e2;
if ( ocb & cbmask ) {
cobj temp =compress(rhs._odata[so+o+b]);
extract<cobj>(temp,pointers,offset);
vobj temp =rhs._odata[so+o+b];
extract<vobj>(temp,pointers,offset);
}
}
}
}
}
//////////////////////////////////////////////////////
// Gather for when there is no need to SIMD split
//////////////////////////////////////////////////////
template<class vobj> void Gather_plane_simple (const Lattice<vobj> &rhs,commVector<vobj> &buffer, int dimension,int plane,int cbmask)
{
SimpleCompressor<vobj> dontcompress;
Gather_plane_simple (rhs,buffer,dimension,plane,cbmask,dontcompress);
}
//////////////////////////////////////////////////////
// Gather for when there *is* need to SIMD split
//////////////////////////////////////////////////////
template<class vobj> void Gather_plane_extract(const Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
{
SimpleCompressor<vobj> dontcompress;
Gather_plane_extract<vobj,vobj,decltype(dontcompress)>(rhs,pointers,dimension,plane,cbmask,dontcompress);
}
//////////////////////////////////////////////////////
// Scatter for when there is no need to SIMD split
//////////////////////////////////////////////////////
@ -200,7 +171,7 @@ template<class vobj> void Scatter_plane_simple (Lattice<vobj> &rhs,commVector<vo
//////////////////////////////////////////////////////
// Scatter for when there *is* need to SIMD split
//////////////////////////////////////////////////////
template<class vobj,class cobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<cobj *> pointers,int dimension,int plane,int cbmask)
template<class vobj> void Scatter_plane_merge(Lattice<vobj> &rhs,std::vector<typename vobj::scalar_object *> pointers,int dimension,int plane,int cbmask)
{
int rd = rhs._grid->_rdimensions[dimension];

15
lib/cshift/Cshift_mpi.h
View File

@ -154,13 +154,7 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
recv_from_rank,
bytes);
grid->Barrier();
/*
for(int i=0;i<send_buf.size();i++){
assert(recv_buf.size()==buffer_size);
assert(send_buf.size()==buffer_size);
std::cout << "SendRecv_Cshift_comms ["<<i<<" "<< dimension<<"] snd "<<send_buf[i]<<" rcv " << recv_buf[i] << " 0x" << cbmask<<std::endl;
}
*/
Scatter_plane_simple (ret,recv_buf,dimension,x,cbmask);
}
}
@ -246,13 +240,6 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
(void *)&recv_buf_extract[i][0],
recv_from_rank,
bytes);
/*
for(int w=0;w<recv_buf_extract[i].size();w++){
assert(recv_buf_extract[i].size()==buffer_size);
assert(send_buf_extract[i].size()==buffer_size);
std::cout << "SendRecv_Cshift_comms ["<<w<<" "<< dimension<<"] recv "<<recv_buf_extract[i][w]<<" send " << send_buf_extract[nbr_lane][w] << cbmask<<std::endl;
}
*/
grid->Barrier();
rpointers[i] = &recv_buf_extract[i][0];
} else {

12276
lib/json/json.hpp Normal file
View File

File diff suppressed because it is too large Load Diff

112
lib/lattice/Lattice_base.h
View File

@ -235,64 +235,74 @@ public:
}
};
//////////////////////////////////////////////////////////////////
// Constructor requires "grid" passed.
// what about a default grid?
//////////////////////////////////////////////////////////////////
Lattice(GridBase *grid) : _odata(grid->oSites()) {
_grid = grid;
//////////////////////////////////////////////////////////////////
// Constructor requires "grid" passed.
// what about a default grid?
//////////////////////////////////////////////////////////////////
Lattice(GridBase *grid) : _odata(grid->oSites()) {
_grid = grid;
// _odata.reserve(_grid->oSites());
// _odata.resize(_grid->oSites());
// std::cout << "Constructing lattice object with Grid pointer "<<_grid<<std::endl;
assert((((uint64_t)&_odata[0])&0xF) ==0);
checkerboard=0;
}
Lattice(const Lattice& r){ // copy constructor
_grid = r._grid;
checkerboard = r.checkerboard;
_odata.resize(_grid->oSites());// essential
parallel_for(int ss=0;ss<_grid->oSites();ss++){
_odata[ss]=r._odata[ss];
}
}
virtual ~Lattice(void) = default;
assert((((uint64_t)&_odata[0])&0xF) ==0);
checkerboard=0;
}
Lattice(const Lattice& r){ // copy constructor
_grid = r._grid;
checkerboard = r.checkerboard;
_odata.resize(_grid->oSites());// essential
parallel_for(int ss=0;ss<_grid->oSites();ss++){
_odata[ss]=r._odata[ss];
}
}
virtual ~Lattice(void) = default;
template<class sobj> strong_inline Lattice<vobj> & operator = (const sobj & r){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r;
}
return *this;
}
template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
this->checkerboard = r.checkerboard;
conformable(*this,r);
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];
}
return *this;
void reset(GridBase* grid) {
if (_grid != grid) {
_grid = grid;
_odata.resize(grid->oSites());
checkerboard = 0;
}
}
// *=,+=,-= operators inherit behvour from correspond */+/- operation
template<class T> strong_inline Lattice<vobj> &operator *=(const T &r) {
*this = (*this)*r;
return *this;
template<class sobj> strong_inline Lattice<vobj> & operator = (const sobj & r){
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r;
}
template<class T> strong_inline Lattice<vobj> &operator -=(const T &r) {
*this = (*this)-r;
return *this;
return *this;
}
template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
this->checkerboard = r.checkerboard;
conformable(*this,r);
parallel_for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];
}
template<class T> strong_inline Lattice<vobj> &operator +=(const T &r) {
*this = (*this)+r;
return *this;
}
}; // class Lattice
return *this;
}
// *=,+=,-= operators inherit behvour from correspond */+/- operation
template<class T> strong_inline Lattice<vobj> &operator *=(const T &r) {
*this = (*this)*r;
return *this;
}
template<class T> strong_inline Lattice<vobj> &operator -=(const T &r) {
*this = (*this)-r;
return *this;
}
template<class T> strong_inline Lattice<vobj> &operator +=(const T &r) {
*this = (*this)+r;
return *this;
}
}; // class Lattice
template<class vobj> std::ostream& operator<< (std::ostream& stream, const Lattice<vobj> &o){
std::vector<int> gcoor;
typedef typename vobj::scalar_object sobj;
@ -310,7 +320,7 @@ public:
}
return stream;
}
}

250
lib/lattice/Lattice_reduction.h
View File

@ -1,45 +1,37 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/lattice/Lattice_reduction.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_LATTICE_REDUCTION_H
#define GRID_LATTICE_REDUCTION_H
#include <Grid/Eigen/Dense>
#include <Grid/Grid_Eigen_Dense.h>
namespace Grid {
#ifdef GRID_WARN_SUBOPTIMAL
#warning "Optimisation alert all these reduction loops are NOT threaded "
#endif
////////////////////////////////////////////////////////////////////////////////////////////////////
// Deterministic Reduction operations
////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////////////////////
// Deterministic Reduction operations
////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
ComplexD nrm = innerProduct(arg,arg);
return std::real(nrm);
@ -336,6 +328,8 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
typedef typename vobj::vector_type vector_type;
typedef typename vobj::tensor_reduced tensor_reduced;
scalar_type zscale(scale);
GridBase *grid = X._grid;
int Nsimd =grid->Nsimd();
@ -361,7 +355,7 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
grid->iCoorFromIindex(icoor,l);
int ldx =r+icoor[orthogdim]*rd;
scalar_type *as =(scalar_type *)&av;
as[l] = scalar_type(a[ldx])*scale;
as[l] = scalar_type(a[ldx])*zscale;
}
tensor_reduced at; at=av;
@ -375,74 +369,6 @@ static void sliceMaddVector(Lattice<vobj> &R,std::vector<RealD> &a,const Lattice
}
};
/*
template<class vobj>
static void sliceMaddVectorSlow (Lattice<vobj> &R,std::vector<RealD> &a,const Lattice<vobj> &X,const Lattice<vobj> &Y,
int Orthog,RealD scale=1.0)
{
// FIXME: Implementation is slow
// Best base the linear combination by constructing a
// set of vectors of size grid->_rdimensions[Orthog].
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
int Nblock = X._grid->GlobalDimensions()[Orthog];
GridBase *FullGrid = X._grid;
GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
Lattice<vobj> Xslice(SliceGrid);
Lattice<vobj> Rslice(SliceGrid);
// If we based this on Cshift it would work for spread out
// but it would be even slower
for(int i=0;i<Nblock;i++){
ExtractSlice(Rslice,Y,i,Orthog);
ExtractSlice(Xslice,X,i,Orthog);
Rslice = Rslice + Xslice*(scale*a[i]);
InsertSlice(Rslice,R,i,Orthog);
}
};
template<class vobj>
static void sliceInnerProductVectorSlow( std::vector<ComplexD> & vec, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog)
{
// FIXME: Implementation is slow
// Look at localInnerProduct implementation,
// and do inside a site loop with block strided iterators
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
typedef typename vobj::tensor_reduced scalar;
typedef typename scalar::scalar_object scomplex;
int Nblock = lhs._grid->GlobalDimensions()[Orthog];
vec.resize(Nblock);
std::vector<scomplex> sip(Nblock);
Lattice<scalar> IP(lhs._grid);
IP=localInnerProduct(lhs,rhs);
sliceSum(IP,sip,Orthog);
for(int ss=0;ss<Nblock;ss++){
vec[ss] = TensorRemove(sip[ss]);
}
}
*/
//////////////////////////////////////////////////////////////////////////////////////////
// FIXME: Implementation is slow
// If we based this on Cshift it would work for spread out
// but it would be even slower
//
// Repeated extract slice is inefficient
//
// Best base the linear combination by constructing a
// set of vectors of size grid->_rdimensions[Orthog].
//////////////////////////////////////////////////////////////////////////////////////////
inline GridBase *makeSubSliceGrid(const GridBase *BlockSolverGrid,int Orthog)
{
int NN = BlockSolverGrid->_ndimension;
@ -462,7 +388,6 @@ inline GridBase *makeSubSliceGrid(const GridBase *BlockSolverGrid,int Or
return (GridBase *)new GridCartesian(latt_phys,simd_phys,mpi_phys);
}
template<class vobj>
static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,const Lattice<vobj> &Y,int Orthog,RealD scale=1.0)
{
@ -471,28 +396,103 @@ static void sliceMaddMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice
typedef typename vobj::vector_type vector_type;
int Nblock = X._grid->GlobalDimensions()[Orthog];
GridBase *FullGrid = X._grid;
GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
Lattice<vobj> Xslice(SliceGrid);
Lattice<vobj> Rslice(SliceGrid);
for(int i=0;i<Nblock;i++){
ExtractSlice(Rslice,Y,i,Orthog);
for(int j=0;j<Nblock;j++){
ExtractSlice(Xslice,X,j,Orthog);
Rslice = Rslice + Xslice*(scale*aa(j,i));
}
InsertSlice(Rslice,R,i,Orthog);
assert( FullGrid->_simd_layout[Orthog]==1);
int nh = FullGrid->_ndimension;
int nl = SliceGrid->_ndimension;
//FIXME package in a convenient iterator
//Should loop over a plane orthogonal to direction "Orthog"
int stride=FullGrid->_slice_stride[Orthog];
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
#pragma omp parallel
{
std::vector<vobj> s_x(Nblock);
#pragma omp for collapse(2)
for(int n=0;n<nblock;n++){
for(int b=0;b<block;b++){
int o = n*stride + b;
for(int i=0;i<Nblock;i++){
s_x[i] = X[o+i*ostride];
}
vobj dot;
for(int i=0;i<Nblock;i++){
dot = Y[o+i*ostride];
for(int j=0;j<Nblock;j++){
dot = dot + s_x[j]*(scale*aa(j,i));
}
R[o+i*ostride]=dot;
}
}}
}
};
template<class vobj>
static void sliceMulMatrix (Lattice<vobj> &R,Eigen::MatrixXcd &aa,const Lattice<vobj> &X,int Orthog,RealD scale=1.0)
{
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
int Nblock = X._grid->GlobalDimensions()[Orthog];
GridBase *FullGrid = X._grid;
GridBase *SliceGrid = makeSubSliceGrid(FullGrid,Orthog);
Lattice<vobj> Xslice(SliceGrid);
Lattice<vobj> Rslice(SliceGrid);
assert( FullGrid->_simd_layout[Orthog]==1);
int nh = FullGrid->_ndimension;
int nl = SliceGrid->_ndimension;
//FIXME package in a convenient iterator
//Should loop over a plane orthogonal to direction "Orthog"
int stride=FullGrid->_slice_stride[Orthog];
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
#pragma omp parallel
{
std::vector<vobj> s_x(Nblock);
#pragma omp for collapse(2)
for(int n=0;n<nblock;n++){
for(int b=0;b<block;b++){
int o = n*stride + b;
for(int i=0;i<Nblock;i++){
s_x[i] = X[o+i*ostride];
}
vobj dot;
for(int i=0;i<Nblock;i++){
dot = s_x[0]*(scale*aa(0,i));
for(int j=1;j<Nblock;j++){
dot = dot + s_x[j]*(scale*aa(j,i));
}
R[o+i*ostride]=dot;
}
}}
}
};
template<class vobj>
static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice<vobj> &lhs,const Lattice<vobj> &rhs,int Orthog)
{
// FIXME: Implementation is slow
// Not sure of best solution.. think about it
typedef typename vobj::scalar_object sobj;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
@ -506,25 +506,55 @@ static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice<vobj>
Lattice<vobj> Rslice(SliceGrid);
mat = Eigen::MatrixXcd::Zero(Nblock,Nblock);
for(int i=0;i<Nblock;i++){
ExtractSlice(Lslice,lhs,i,Orthog);
for(int j=0;j<Nblock;j++){
ExtractSlice(Rslice,rhs,j,Orthog);
mat(i,j) = innerProduct(Lslice,Rslice);
}
assert( FullGrid->_simd_layout[Orthog]==1);
int nh = FullGrid->_ndimension;
int nl = SliceGrid->_ndimension;
//FIXME package in a convenient iterator
//Should loop over a plane orthogonal to direction "Orthog"
int stride=FullGrid->_slice_stride[Orthog];
int block =FullGrid->_slice_block [Orthog];
int nblock=FullGrid->_slice_nblock[Orthog];
int ostride=FullGrid->_ostride[Orthog];
typedef typename vobj::vector_typeD vector_typeD;
#pragma omp parallel
{
std::vector<vobj> Left(Nblock);
std::vector<vobj> Right(Nblock);
Eigen::MatrixXcd mat_thread = Eigen::MatrixXcd::Zero(Nblock,Nblock);
#pragma omp for collapse(2)
for(int n=0;n<nblock;n++){
for(int b=0;b<block;b++){
int o = n*stride + b;
for(int i=0;i<Nblock;i++){
Left [i] = lhs[o+i*ostride];
Right[i] = rhs[o+i*ostride];
}
for(int i=0;i<Nblock;i++){
for(int j=0;j<Nblock;j++){
auto tmp = innerProduct(Left[i],Right[j]);
// vector_typeD rtmp = TensorRemove(tmp);
auto rtmp = TensorRemove(tmp);
mat_thread(i,j) += Reduce(rtmp);
}}
}}
#pragma omp critical
{
mat += mat_thread;
}
}
#undef FORCE_DIAG
#ifdef FORCE_DIAG
for(int i=0;i<Nblock;i++){
for(int j=0;j<Nblock;j++){
if ( i != j ) mat(i,j)=0.0;
}
}
#endif
return;
}
} /*END NAMESPACE GRID*/
#endif

131
lib/lattice/Lattice_rng.h
View File

@ -6,8 +6,8 @@
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@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
@ -75,6 +75,55 @@ namespace Grid {
return multiplicity;
}
// merge of April 11 2017
//<<<<<<< HEAD
// this function is necessary for the LS vectorised field
inline int RNGfillable_general(GridBase *coarse,GridBase *fine)
{
int rngdims = coarse->_ndimension;
// trivially extended in higher dims, with locality guaranteeing RNG state is local to node
int lowerdims = fine->_ndimension - coarse->_ndimension; assert(lowerdims >= 0);
// assumes that the higher dimensions are not using more processors
// all further divisions are local
for(int d=0;d<lowerdims;d++) assert(fine->_processors[d]==1);
for(int d=0;d<rngdims;d++) assert(coarse->_processors[d] == fine->_processors[d+lowerdims]);
// then divide the number of local sites
// check that the total number of sims agree, meanse the iSites are the same
assert(fine->Nsimd() == coarse->Nsimd());
// check that the two grids divide cleanly
assert( (fine->lSites() / coarse->lSites() ) * coarse->lSites() == fine->lSites() );
return fine->lSites() / coarse->lSites();
}
/*
// Wrap seed_seq to give common interface with random_device
class fixedSeed {
public:
typedef std::seed_seq::result_type result_type;
std::seed_seq src;
fixedSeed(const std::vector<int> &seeds) : src(seeds.begin(),seeds.end()) {};
result_type operator () (void){
std::vector<result_type> list(1);
src.generate(list.begin(),list.end());
return list[0];
}
};
=======
>>>>>>> develop
*/
// real scalars are one component
template<class scalar,class distribution,class generator>
void fillScalar(scalar &s,distribution &dist,generator & gen)
@ -109,7 +158,7 @@ namespace Grid {
#ifdef RNG_SITMO
typedef sitmo::prng_engine RngEngine;
typedef uint64_t RngStateType;
static const int RngStateCount = 4;
static const int RngStateCount = 13;
#endif
std::vector<RngEngine> _generators;
@ -164,7 +213,7 @@ namespace Grid {
ss<<eng;
ss.seekg(0,ss.beg);
for(int i=0;i<RngStateCount;i++){
ss>>saved[i];
ss>>saved[i];
}
}
void GetState(std::vector<RngStateType> & saved,int gen) {
@ -174,7 +223,7 @@ namespace Grid {
assert(saved.size()==RngStateCount);
std::stringstream ss;
for(int i=0;i<RngStateCount;i++){
ss<< saved[i]<<" ";
ss<< saved[i]<<" ";
}
ss.seekg(0,ss.beg);
ss>>eng;
@ -215,7 +264,7 @@ namespace Grid {
dist[0].reset();
for(int idx=0;idx<words;idx++){
fillScalar(buf[idx],dist[0],_generators[0]);
fillScalar(buf[idx],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
@ -247,7 +296,7 @@ namespace Grid {
RealF *pointer=(RealF *)&l;
dist[0].reset();
for(int i=0;i<2*vComplexF::Nsimd();i++){
fillScalar(pointer[i],dist[0],_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
@ -255,7 +304,7 @@ namespace Grid {
RealD *pointer=(RealD *)&l;
dist[0].reset();
for(int i=0;i<2*vComplexD::Nsimd();i++){
fillScalar(pointer[i],dist[0],_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
@ -263,7 +312,7 @@ namespace Grid {
RealF *pointer=(RealF *)&l;
dist[0].reset();
for(int i=0;i<vRealF::Nsimd();i++){
fillScalar(pointer[i],dist[0],_generators[0]);
fillScalar(pointer[i],dist[0],_generators[0]);
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
@ -275,7 +324,7 @@ namespace Grid {
}
CartesianCommunicator::BroadcastWorld(0,(void *)&l,sizeof(l));
}
void SeedFixedIntegers(const std::vector<int> &seeds){
CartesianCommunicator::BroadcastWorld(0,(void *)&seeds[0],sizeof(int)*seeds.size());
std::seed_seq src(seeds.begin(),seeds.end());
@ -284,18 +333,20 @@ namespace Grid {
};
class GridParallelRNG : public GridRNGbase {
double _time_counter;
public:
GridBase *_grid;
int _vol;
public:
unsigned int _vol;
int generator_idx(int os,int is){
int generator_idx(int os,int is) {
return is*_grid->oSites()+os;
}
GridParallelRNG(GridBase *grid) : GridRNGbase() {
_grid=grid;
_vol =_grid->iSites()*_grid->oSites();
_grid = grid;
_vol =_grid->iSites()*_grid->oSites();
_generators.resize(_vol);
_uniform.resize(_vol,std::uniform_real_distribution<RealD>{0,1});
@ -309,33 +360,34 @@ namespace Grid {
typedef typename vobj::scalar_object scalar_object;
typedef typename vobj::scalar_type scalar_type;
typedef typename vobj::vector_type vector_type;
int multiplicity = RNGfillable(_grid,l._grid);
int Nsimd =_grid->Nsimd();
int osites=_grid->oSites();
int words=sizeof(scalar_object)/sizeof(scalar_type);
double inner_time_counter = usecond();
int multiplicity = RNGfillable_general(_grid, l._grid); // l has finer or same grid
int Nsimd = _grid->Nsimd(); // guaranteed to be the same for l._grid too
int osites = _grid->oSites(); // guaranteed to be <= l._grid->oSites() by a factor multiplicity
int words = sizeof(scalar_object) / sizeof(scalar_type);
parallel_for(int ss=0;ss<osites;ss++){
std::vector<scalar_object> buf(Nsimd);
for (int m = 0; m < multiplicity; m++) { // Draw from same generator multiplicity times
std::vector<scalar_object> buf(Nsimd);
for(int m=0;m<multiplicity;m++) {// Draw from same generator multiplicity times
int sm = multiplicity * ss + m; // Maps the generator site to the fine site
int sm=multiplicity*ss+m; // Maps the generator site to the fine site
for(int si=0;si<Nsimd;si++){
int gdx = generator_idx(ss,si); // index of generator state
scalar_type *pointer = (scalar_type *)&buf[si];
dist[gdx].reset();
for(int idx=0;idx<words;idx++){
fillScalar(pointer[idx],dist[gdx],_generators[gdx]);
}
}
// merge into SIMD lanes
merge(l._odata[sm],buf);
}
for (int si = 0; si < Nsimd; si++) {
int gdx = generator_idx(ss, si); // index of generator state
scalar_type *pointer = (scalar_type *)&buf[si];
dist[gdx].reset();
for (int idx = 0; idx < words; idx++)
fillScalar(pointer[idx], dist[gdx], _generators[gdx]);
}
// merge into SIMD lanes, FIXME suboptimal implementation
merge(l._odata[sm], buf);
}
}
_time_counter += usecond()- inner_time_counter;
};
void SeedFixedIntegers(const std::vector<int> &seeds){
@ -412,6 +464,12 @@ namespace Grid {
}
#endif
}
void Report(){
std::cout << GridLogMessage << "Time spent in the fill() routine by GridParallelRNG: "<< _time_counter/1e3 << " ms" << std::endl;
}
////////////////////////////////////////////////////////////////////////
// Support for rigorous test of RNG's
// Return uniform random uint32_t from requested site generator
@ -419,7 +477,6 @@ namespace Grid {
uint32_t GlobalU01(int gsite){
uint32_t the_number;
// who
std::vector<int> gcoor;
int rank,o_idx,i_idx;

55
lib/lattice/Lattice_transfer.h
View File

@ -551,7 +551,10 @@ void Replicate(Lattice<vobj> &coarse,Lattice<vobj> & fine)
//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in){
typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>::value, void>::type
unvectorizeToLexOrdArray(std::vector<sobj> &out, const Lattice<vobj> &in)
{
typedef typename vobj::vector_type vtype;
GridBase* in_grid = in._grid;
@ -590,6 +593,54 @@ typename std::enable_if<isSIMDvectorized<vobj>::value && !isSIMDvectorized<sobj>
extract1(in_vobj, out_ptrs, 0);
}
}
//Copy SIMD-vectorized lattice to array of scalar objects in lexicographic order
template<typename vobj, typename sobj>
typename std::enable_if<isSIMDvectorized<vobj>::value
&& !isSIMDvectorized<sobj>::value, void>::type
vectorizeFromLexOrdArray( std::vector<sobj> &in, Lattice<vobj> &out)
{
typedef typename vobj::vector_type vtype;
GridBase* grid = out._grid;
assert(in.size()==grid->lSites());
int ndim = grid->Nd();
int nsimd = vtype::Nsimd();
std::vector<std::vector<int> > icoor(nsimd);
for(int lane=0; lane < nsimd; lane++){
icoor[lane].resize(ndim);
grid->iCoorFromIindex(icoor[lane],lane);
}
parallel_for(uint64_t oidx = 0; oidx < grid->oSites(); oidx++){ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> ptrs(nsimd);
std::vector<int> ocoor(ndim);
grid->oCoorFromOindex(ocoor, oidx);
std::vector<int> lcoor(grid->Nd());
for(int lane=0; lane < nsimd; lane++){
for(int mu=0;mu<ndim;mu++){
lcoor[mu] = ocoor[mu] + grid->_rdimensions[mu]*icoor[lane][mu];
}
int lex;
Lexicographic::IndexFromCoor(lcoor, lex, grid->_ldimensions);
ptrs[lane] = &in[lex];
}
//pack from those ptrs
vobj vecobj;
merge1(vecobj, ptrs, 0);
out._odata[oidx] = vecobj;
}
}
//Convert a Lattice from one precision to another
template<class VobjOut, class VobjIn>
@ -615,7 +666,7 @@ void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){
std::vector<SobjOut> in_slex_conv(in_grid->lSites());
unvectorizeToLexOrdArray(in_slex_conv, in);
parallel_for(int out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
parallel_for(uint64_t out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
std::vector<int> out_ocoor(ndim);
out_grid->oCoorFromOindex(out_ocoor, out_oidx);

8
lib/lattice/Lattice_unary.h
View File

@ -62,14 +62,20 @@ namespace Grid {
return ret;
}
template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs, ComplexD alpha, Integer Nexp = DEFAULT_MAT_EXP){
template<class obj> Lattice<obj> expMat(const Lattice<obj> &rhs, RealD alpha, Integer Nexp = DEFAULT_MAT_EXP){
Lattice<obj> ret(rhs._grid);
ret.checkerboard = rhs.checkerboard;
conformable(ret,rhs);
parallel_for(int ss=0;ss<rhs._grid->oSites();ss++){
ret._odata[ss]=Exponentiate(rhs._odata[ss],alpha, Nexp);
}
return ret;
}

1
lib/log/Log.cc
View File

@ -30,6 +30,7 @@ directory
*************************************************************************************/
/* END LEGAL */
#include <Grid/GridCore.h>
#include <Grid/util/CompilerCompatible.h>
#include <cxxabi.h>
#include <memory>

1076
lib/parallelIO/BinaryIO.h
View File

File diff suppressed because it is too large Load Diff

716
lib/parallelIO/IldgIO.h Normal file
View File

@ -0,0 +1,716 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/parallelIO/IldgIO.h
Copyright (C) 2015
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_ILDG_IO_H
#define GRID_ILDG_IO_H
#ifdef HAVE_LIME
#include <algorithm>
#include <fstream>
#include <iomanip>
#include <iostream>
#include <map>
#include <pwd.h>
#include <sys/utsname.h>
#include <unistd.h>
//C-Lime is a must have for this functionality
extern "C" {
#include "lime.h"
}
namespace Grid {
namespace QCD {
/////////////////////////////////
// Encode word types as strings
/////////////////////////////////
template<class word> inline std::string ScidacWordMnemonic(void){ return std::string("unknown"); }
template<> inline std::string ScidacWordMnemonic<double> (void){ return std::string("D"); }
template<> inline std::string ScidacWordMnemonic<float> (void){ return std::string("F"); }
template<> inline std::string ScidacWordMnemonic< int32_t>(void){ return std::string("I32_t"); }
template<> inline std::string ScidacWordMnemonic<uint32_t>(void){ return std::string("U32_t"); }
template<> inline std::string ScidacWordMnemonic< int64_t>(void){ return std::string("I64_t"); }
template<> inline std::string ScidacWordMnemonic<uint64_t>(void){ return std::string("U64_t"); }
/////////////////////////////////////////
// Encode a generic tensor as a string
/////////////////////////////////////////
template<class vobj> std::string ScidacRecordTypeString(int &colors, int &spins, int & typesize,int &datacount) {
typedef typename getPrecision<vobj>::real_scalar_type stype;
int _ColourN = indexRank<ColourIndex,vobj>();
int _ColourScalar = isScalar<ColourIndex,vobj>();
int _ColourVector = isVector<ColourIndex,vobj>();
int _ColourMatrix = isMatrix<ColourIndex,vobj>();
int _SpinN = indexRank<SpinIndex,vobj>();
int _SpinScalar = isScalar<SpinIndex,vobj>();
int _SpinVector = isVector<SpinIndex,vobj>();
int _SpinMatrix = isMatrix<SpinIndex,vobj>();
int _LorentzN = indexRank<LorentzIndex,vobj>();
int _LorentzScalar = isScalar<LorentzIndex,vobj>();
int _LorentzVector = isVector<LorentzIndex,vobj>();
int _LorentzMatrix = isMatrix<LorentzIndex,vobj>();
std::stringstream stream;
stream << "GRID_";
stream << ScidacWordMnemonic<stype>();
// std::cout << " Lorentz N/S/V/M : " << _LorentzN<<" "<<_LorentzScalar<<"/"<<_LorentzVector<<"/"<<_LorentzMatrix<<std::endl;
// std::cout << " Spin N/S/V/M : " << _SpinN <<" "<<_SpinScalar <<"/"<<_SpinVector <<"/"<<_SpinMatrix<<std::endl;
// std::cout << " Colour N/S/V/M : " << _ColourN <<" "<<_ColourScalar <<"/"<<_ColourVector <<"/"<<_ColourMatrix<<std::endl;
if ( _LorentzVector ) stream << "_LorentzVector"<<_LorentzN;
if ( _LorentzMatrix ) stream << "_LorentzMatrix"<<_LorentzN;
if ( _SpinVector ) stream << "_SpinVector"<<_SpinN;
if ( _SpinMatrix ) stream << "_SpinMatrix"<<_SpinN;
if ( _ColourVector ) stream << "_ColourVector"<<_ColourN;
if ( _ColourMatrix ) stream << "_ColourMatrix"<<_ColourN;
if ( _ColourScalar && _LorentzScalar && _SpinScalar ) stream << "_Complex";
typesize = sizeof(typename vobj::scalar_type);
if ( _ColourMatrix ) typesize*= _ColourN*_ColourN;
else typesize*= _ColourN;
if ( _SpinMatrix ) typesize*= _SpinN*_SpinN;
else typesize*= _SpinN;
colors = _ColourN;
spins = _SpinN;
datacount = _LorentzN;
return stream.str();
}
template<class vobj> std::string ScidacRecordTypeString(Lattice<vobj> & lat,int &colors, int &spins, int & typesize,int &datacount) {
return ScidacRecordTypeString<vobj>(colors,spins,typesize,datacount);
};
////////////////////////////////////////////////////////////
// Helper to fill out metadata
////////////////////////////////////////////////////////////
template<class vobj> void ScidacMetaData(Lattice<vobj> & field,
FieldMetaData &header,
scidacRecord & _scidacRecord,
scidacFile & _scidacFile)
{
typedef typename getPrecision<vobj>::real_scalar_type stype;
/////////////////////////////////////
// Pull Grid's metadata
/////////////////////////////////////
PrepareMetaData(field,header);
/////////////////////////////////////
// Scidac Private File structure
/////////////////////////////////////
_scidacFile = scidacFile(field._grid);
/////////////////////////////////////
// Scidac Private Record structure
/////////////////////////////////////
scidacRecord sr;
sr.datatype = ScidacRecordTypeString(field,sr.colors,sr.spins,sr.typesize,sr.datacount);
sr.date = header.creation_date;
sr.precision = ScidacWordMnemonic<stype>();
sr.recordtype = GRID_IO_FIELD;
_scidacRecord = sr;
std::cout << GridLogMessage << "Build SciDAC datatype " <<sr.datatype<<std::endl;
}
///////////////////////////////////////////////////////
// Scidac checksum
///////////////////////////////////////////////////////
static int scidacChecksumVerify(scidacChecksum &scidacChecksum_,uint32_t scidac_csuma,uint32_t scidac_csumb)
{
uint32_t scidac_checksuma = stoull(scidacChecksum_.suma,0,16);
uint32_t scidac_checksumb = stoull(scidacChecksum_.sumb,0,16);
if ( scidac_csuma !=scidac_checksuma) return 0;
if ( scidac_csumb !=scidac_checksumb) return 0;
return 1;
}
////////////////////////////////////////////////////////////////////////////////////
// Lime, ILDG and Scidac I/O classes
////////////////////////////////////////////////////////////////////////////////////
class GridLimeReader : public BinaryIO {
public:
///////////////////////////////////////////////////
// FIXME: format for RNG? Now just binary out instead
///////////////////////////////////////////////////
FILE *File;
LimeReader *LimeR;
std::string filename;
/////////////////////////////////////////////
// Open the file
/////////////////////////////////////////////
void open(std::string &_filename)
{
filename= _filename;
File = fopen(filename.c_str(), "r");
LimeR = limeCreateReader(File);
}
/////////////////////////////////////////////
// Close the file
/////////////////////////////////////////////
void close(void){
fclose(File);
// limeDestroyReader(LimeR);
}
////////////////////////////////////////////
// Read a generic lattice field and verify checksum
////////////////////////////////////////////
template<class vobj>
void readLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
{
typedef typename vobj::scalar_object sobj;
scidacChecksum scidacChecksum_;
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
std::string format = getFormatString<vobj>();
while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) {
std::cout << GridLogMessage << limeReaderType(LimeR) <<std::endl;
if ( strncmp(limeReaderType(LimeR), record_name.c_str(),strlen(record_name.c_str()) ) ) {
off_t offset= ftell(File);
BinarySimpleMunger<sobj,sobj> munge;
BinaryIO::readLatticeObject< sobj, sobj >(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
/////////////////////////////////////////////
// Insist checksum is next record
/////////////////////////////////////////////
readLimeObject(scidacChecksum_,std::string("scidacChecksum"),record_name);
/////////////////////////////////////////////
// Verify checksums
/////////////////////////////////////////////
scidacChecksumVerify(scidacChecksum_,scidac_csuma,scidac_csumb);
return;
}
}
}
////////////////////////////////////////////
// Read a generic serialisable object
////////////////////////////////////////////
template<class serialisable_object>
void readLimeObject(serialisable_object &object,std::string object_name,std::string record_name)
{
std::string xmlstring;
// should this be a do while; can we miss a first record??
while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) {
uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
if ( strncmp(limeReaderType(LimeR), record_name.c_str(),strlen(record_name.c_str()) ) ) {
std::vector<char> xmlc(nbytes+1,'\0');
limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);
XmlReader RD(&xmlc[0],"");
read(RD,object_name,object);
return;
}
}
assert(0);
}
};
class GridLimeWriter : public BinaryIO {
public:
///////////////////////////////////////////////////
// FIXME: format for RNG? Now just binary out instead
///////////////////////////////////////////////////
FILE *File;
LimeWriter *LimeW;
std::string filename;
void open(std::string &_filename) {
filename= _filename;
File = fopen(filename.c_str(), "w");
LimeW = limeCreateWriter(File); assert(LimeW != NULL );
}
/////////////////////////////////////////////
// Close the file
/////////////////////////////////////////////
void close(void) {
fclose(File);
// limeDestroyWriter(LimeW);
}
///////////////////////////////////////////////////////
// Lime utility functions
///////////////////////////////////////////////////////
int createLimeRecordHeader(std::string message, int MB, int ME, size_t PayloadSize)
{
LimeRecordHeader *h;
h = limeCreateHeader(MB, ME, const_cast<char *>(message.c_str()), PayloadSize);
assert(limeWriteRecordHeader(h, LimeW) >= 0);
limeDestroyHeader(h);
return LIME_SUCCESS;
}
////////////////////////////////////////////
// Write a generic serialisable object
////////////////////////////////////////////
template<class serialisable_object>
void writeLimeObject(int MB,int ME,serialisable_object &object,std::string object_name,std::string record_name)
{
std::string xmlstring;
{
XmlWriter WR("","");
write(WR,object_name,object);
xmlstring = WR.XmlString();
}
uint64_t nbytes = xmlstring.size();
int err;
LimeRecordHeader *h = limeCreateHeader(MB, ME,(char *)record_name.c_str(), nbytes); assert(h!= NULL);
err=limeWriteRecordHeader(h, LimeW); assert(err>=0);
err=limeWriteRecordData(&xmlstring[0], &nbytes, LimeW); assert(err>=0);
err=limeWriterCloseRecord(LimeW); assert(err>=0);
limeDestroyHeader(h);
}
////////////////////////////////////////////
// Write a generic lattice field and csum
////////////////////////////////////////////
template<class vobj>
void writeLimeLatticeBinaryObject(Lattice<vobj> &field,std::string record_name)
{
////////////////////////////////////////////
// Create record header
////////////////////////////////////////////
typedef typename vobj::scalar_object sobj;
int err;
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
uint64_t PayloadSize = sizeof(sobj) * field._grid->_gsites;
createLimeRecordHeader(record_name, 0, 0, PayloadSize);
////////////////////////////////////////////////////////////////////
// NB: FILE and iostream are jointly writing disjoint sequences in the
// the same file through different file handles (integer units).
//
// These are both buffered, so why I think this code is right is as follows.
//
// i) write record header to FILE *File, telegraphing the size.
// ii) ftell reads the offset from FILE *File .
// iii) iostream / MPI Open independently seek this offset. Write sequence direct to disk.
// Closes iostream and flushes.
// iv) fseek on FILE * to end of this disjoint section.
// v) Continue writing scidac record.
////////////////////////////////////////////////////////////////////
off_t offset = ftell(File);
std::string format = getFormatString<vobj>();
BinarySimpleMunger<sobj,sobj> munge;
BinaryIO::writeLatticeObject<vobj,sobj>(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
err=limeWriterCloseRecord(LimeW); assert(err>=0);
////////////////////////////////////////
// Write checksum element, propagaing forward from the BinaryIO
// Always pair a checksum with a binary object, and close message
////////////////////////////////////////
scidacChecksum checksum;
std::stringstream streama; streama << std::hex << scidac_csuma;
std::stringstream streamb; streamb << std::hex << scidac_csumb;
checksum.suma= streama.str();
checksum.sumb= streamb.str();
std::cout << GridLogMessage<<" writing scidac checksums "<<std::hex<<scidac_csuma<<"/"<<scidac_csumb<<std::dec<<std::endl;
writeLimeObject(0,1,checksum,std::string("scidacChecksum" ),std::string(SCIDAC_CHECKSUM));
}
};
class ScidacWriter : public GridLimeWriter {
public:
template<class SerialisableUserFile>
void writeScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile)
{
scidacFile _scidacFile(grid);
writeLimeObject(1,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
writeLimeObject(0,1,_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
}
////////////////////////////////////////////////
// Write generic lattice field in scidac format
////////////////////////////////////////////////
template <class vobj, class userRecord>
void writeScidacFieldRecord(Lattice<vobj> &field,userRecord _userRecord)
{
typedef typename vobj::scalar_object sobj;
uint64_t nbytes;
GridBase * grid = field._grid;
////////////////////////////////////////
// fill the Grid header
////////////////////////////////////////
FieldMetaData header;
scidacRecord _scidacRecord;
scidacFile _scidacFile;
ScidacMetaData(field,header,_scidacRecord,_scidacFile);
//////////////////////////////////////////////
// Fill the Lime file record by record
//////////////////////////////////////////////
writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message
writeLimeObject(0,0,_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
writeLimeObject(0,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
writeLimeLatticeBinaryObject(field,std::string(ILDG_BINARY_DATA)); // Closes message with checksum
}
};
class IldgWriter : public ScidacWriter {
public:
///////////////////////////////////
// A little helper
///////////////////////////////////
void writeLimeIldgLFN(std::string &LFN)
{
uint64_t PayloadSize = LFN.size();
int err;
createLimeRecordHeader(ILDG_DATA_LFN, 0 , 0, PayloadSize);
err=limeWriteRecordData(const_cast<char*>(LFN.c_str()), &PayloadSize,LimeW); assert(err>=0);
err=limeWriterCloseRecord(LimeW); assert(err>=0);
}
////////////////////////////////////////////////////////////////
// Special ILDG operations ; gauge configs only.
// Don't require scidac records EXCEPT checksum
// Use Grid MetaData object if present.
////////////////////////////////////////////////////////////////
template <class vsimd>
void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,int sequence,std::string LFN,std::string description)
{
GridBase * grid = Umu._grid;
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
typedef iLorentzColourMatrix<vsimd> vobj;
typedef typename vobj::scalar_object sobj;
uint64_t nbytes;
////////////////////////////////////////
// fill the Grid header
////////////////////////////////////////
FieldMetaData header;
scidacRecord _scidacRecord;
scidacFile _scidacFile;
ScidacMetaData(Umu,header,_scidacRecord,_scidacFile);
std::string format = header.floating_point;
header.ensemble_id = description;
header.ensemble_label = description;
header.sequence_number = sequence;
header.ildg_lfn = LFN;
assert ( (format == std::string("IEEE32BIG"))
||(format == std::string("IEEE64BIG")) );
//////////////////////////////////////////////////////
// Fill ILDG header data struct
//////////////////////////////////////////////////////
ildgFormat ildgfmt ;
ildgfmt.field = std::string("su3gauge");
if ( format == std::string("IEEE32BIG") ) {
ildgfmt.precision = 32;
} else {
ildgfmt.precision = 64;
}
ildgfmt.version = 1.0;
ildgfmt.lx = header.dimension[0];
ildgfmt.ly = header.dimension[1];
ildgfmt.lz = header.dimension[2];
ildgfmt.lt = header.dimension[3];
assert(header.nd==4);
assert(header.nd==header.dimension.size());
//////////////////////////////////////////////////////////////////////////////
// Fill the USQCD info field
//////////////////////////////////////////////////////////////////////////////
usqcdInfo info;
info.version=1.0;
info.plaq = header.plaquette;
info.linktr = header.link_trace;
std::cout << GridLogMessage << " Writing config; IldgIO "<<std::endl;
//////////////////////////////////////////////
// Fill the Lime file record by record
//////////////////////////////////////////////
writeLimeObject(1,0,header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message
writeLimeObject(0,0,_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML));
writeLimeObject(0,1,info,info.SerialisableClassName(),std::string(SCIDAC_FILE_XML));
writeLimeObject(1,0,_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML));
writeLimeObject(0,0,info,info.SerialisableClassName(),std::string(SCIDAC_RECORD_XML));
writeLimeObject(0,0,ildgfmt,std::string("ildgFormat") ,std::string(ILDG_FORMAT)); // rec
writeLimeIldgLFN(header.ildg_lfn); // rec
writeLimeLatticeBinaryObject(Umu,std::string(ILDG_BINARY_DATA)); // Closes message with checksum
// limeDestroyWriter(LimeW);
fclose(File);
}
};
class IldgReader : public GridLimeReader {
public:
////////////////////////////////////////////////////////////////
// Read either Grid/SciDAC/ILDG configuration
// Don't require scidac records EXCEPT checksum
// Use Grid MetaData object if present.
// Else use ILDG MetaData object if present.
// Else use SciDAC MetaData object if present.
////////////////////////////////////////////////////////////////
template <class vsimd>
void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu, FieldMetaData &FieldMetaData_) {
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
typedef typename GaugeField::vector_object vobj;
typedef typename vobj::scalar_object sobj;
typedef LorentzColourMatrixF fobj;
typedef LorentzColourMatrixD dobj;
GridBase *grid = Umu._grid;
std::vector<int> dims = Umu._grid->FullDimensions();
assert(dims.size()==4);
// Metadata holders
ildgFormat ildgFormat_ ;
std::string ildgLFN_ ;
scidacChecksum scidacChecksum_;
usqcdInfo usqcdInfo_ ;
// track what we read from file
int found_ildgFormat =0;
int found_ildgLFN =0;
int found_scidacChecksum=0;
int found_usqcdInfo =0;
int found_ildgBinary =0;
int found_FieldMetaData =0;
uint32_t nersc_csum;
uint32_t scidac_csuma;
uint32_t scidac_csumb;
// Binary format
std::string format;
//////////////////////////////////////////////////////////////////////////
// Loop over all records
// -- Order is poorly guaranteed except ILDG header preceeds binary section.
// -- Run like an event loop.
// -- Impose trust hierarchy. Grid takes precedence & look for ILDG, and failing
// that Scidac.
// -- Insist on Scidac checksum record.
//////////////////////////////////////////////////////////////////////////
while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) {
uint64_t nbytes = limeReaderBytes(LimeR);//size of this record (configuration)
//////////////////////////////////////////////////////////////////
// If not BINARY_DATA read a string and parse
//////////////////////////////////////////////////////////////////
if ( strncmp(limeReaderType(LimeR), ILDG_BINARY_DATA,strlen(ILDG_BINARY_DATA) ) ) {
// Copy out the string
std::vector<char> xmlc(nbytes+1,'\0');
limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR);
std::cout << GridLogMessage<< "Non binary record :" <<limeReaderType(LimeR) <<std::endl; //<<"\n"<<(&xmlc[0])<<std::endl;
//////////////////////////////////
// ILDG format record
if ( !strncmp(limeReaderType(LimeR), ILDG_FORMAT,strlen(ILDG_FORMAT)) ) {
XmlReader RD(&xmlc[0],"");
read(RD,"ildgFormat",ildgFormat_);
if ( ildgFormat_.precision == 64 ) format = std::string("IEEE64BIG");
if ( ildgFormat_.precision == 32 ) format = std::string("IEEE32BIG");
assert( ildgFormat_.lx == dims[0]);
assert( ildgFormat_.ly == dims[1]);
assert( ildgFormat_.lz == dims[2]);
assert( ildgFormat_.lt == dims[3]);
found_ildgFormat = 1;
}
if ( !strncmp(limeReaderType(LimeR), ILDG_DATA_LFN,strlen(ILDG_DATA_LFN)) ) {
FieldMetaData_.ildg_lfn = std::string(&xmlc[0]);
found_ildgLFN = 1;
}
if ( !strncmp(limeReaderType(LimeR), GRID_FORMAT,strlen(ILDG_FORMAT)) ) {
XmlReader RD(&xmlc[0],"");
read(RD,"FieldMetaData",FieldMetaData_);
format = FieldMetaData_.floating_point;
assert(FieldMetaData_.dimension[0] == dims[0]);
assert(FieldMetaData_.dimension[1] == dims[1]);
assert(FieldMetaData_.dimension[2] == dims[2]);
assert(FieldMetaData_.dimension[3] == dims[3]);
found_FieldMetaData = 1;
}
if ( !strncmp(limeReaderType(LimeR), SCIDAC_RECORD_XML,strlen(SCIDAC_RECORD_XML)) ) {
std::string xmls(&xmlc[0]);
// is it a USQCD info field
if ( xmls.find(std::string("usqcdInfo")) != std::string::npos ) {
std::cout << GridLogMessage<<"...found a usqcdInfo field"<<std::endl;
XmlReader RD(&xmlc[0],"");
read(RD,"usqcdInfo",usqcdInfo_);
found_usqcdInfo = 1;
}
}
if ( !strncmp(limeReaderType(LimeR), SCIDAC_CHECKSUM,strlen(SCIDAC_CHECKSUM)) ) {
XmlReader RD(&xmlc[0],"");
read(RD,"scidacChecksum",scidacChecksum_);
found_scidacChecksum = 1;
}
} else {
/////////////////////////////////
// Binary data
/////////////////////////////////
std::cout << GridLogMessage << "ILDG Binary record found : " ILDG_BINARY_DATA << std::endl;
off_t offset= ftell(File);
if ( format == std::string("IEEE64BIG") ) {
GaugeSimpleMunger<dobj, sobj> munge;
BinaryIO::readLatticeObject< vobj, dobj >(Umu, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
} else {
GaugeSimpleMunger<fobj, sobj> munge;
BinaryIO::readLatticeObject< vobj, fobj >(Umu, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb);
}
found_ildgBinary = 1;
}
}
//////////////////////////////////////////////////////
// Minimally must find binary segment and checksum
// Since this is an ILDG reader require ILDG format
//////////////////////////////////////////////////////
assert(found_ildgBinary);
assert(found_ildgFormat);
assert(found_scidacChecksum);
// Must find something with the lattice dimensions
assert(found_FieldMetaData||found_ildgFormat);
if ( found_FieldMetaData ) {
std::cout << GridLogMessage<<"Grid MetaData was record found: configuration was probably written by Grid ! Yay ! "<<std::endl;
} else {
assert(found_ildgFormat);
assert ( ildgFormat_.field == std::string("su3gauge") );
///////////////////////////////////////////////////////////////////////////////////////
// Populate our Grid metadata as best we can
///////////////////////////////////////////////////////////////////////////////////////
std::ostringstream vers; vers << ildgFormat_.version;
FieldMetaData_.hdr_version = vers.str();
FieldMetaData_.data_type = std::string("4D_SU3_GAUGE_3X3");
FieldMetaData_.nd=4;
FieldMetaData_.dimension.resize(4);
FieldMetaData_.dimension[0] = ildgFormat_.lx ;
FieldMetaData_.dimension[1] = ildgFormat_.ly ;
FieldMetaData_.dimension[2] = ildgFormat_.lz ;
FieldMetaData_.dimension[3] = ildgFormat_.lt ;
if ( found_usqcdInfo ) {
FieldMetaData_.plaquette = usqcdInfo_.plaq;
FieldMetaData_.link_trace= usqcdInfo_.linktr;
std::cout << GridLogMessage <<"This configuration was probably written by USQCD "<<std::endl;
std::cout << GridLogMessage <<"USQCD xml record Plaquette : "<<FieldMetaData_.plaquette<<std::endl;
std::cout << GridLogMessage <<"USQCD xml record LinkTrace : "<<FieldMetaData_.link_trace<<std::endl;
} else {
FieldMetaData_.plaquette = 0.0;
FieldMetaData_.link_trace= 0.0;
std::cout << GridLogWarning << "This configuration is unsafe with no plaquette records that can verify it !!! "<<std::endl;
}
}
////////////////////////////////////////////////////////////
// Really really want to mandate a scidac checksum
////////////////////////////////////////////////////////////
if ( found_scidacChecksum ) {
FieldMetaData_.scidac_checksuma = stoull(scidacChecksum_.suma,0,16);
FieldMetaData_.scidac_checksumb = stoull(scidacChecksum_.sumb,0,16);
scidacChecksumVerify(scidacChecksum_,scidac_csuma,scidac_csumb);
assert( scidac_csuma ==FieldMetaData_.scidac_checksuma);
assert( scidac_csumb ==FieldMetaData_.scidac_checksumb);
std::cout << GridLogMessage<<"SciDAC checksums match " << std::endl;
} else {
std::cout << GridLogWarning<<"SciDAC checksums not found. This is unsafe. " << std::endl;
assert(0); // Can I insist always checksum ?
}
if ( found_FieldMetaData || found_usqcdInfo ) {
FieldMetaData checker;
GaugeStatistics(Umu,checker);
assert(fabs(checker.plaquette - FieldMetaData_.plaquette )<1.0e-5);
assert(fabs(checker.link_trace - FieldMetaData_.link_trace)<1.0e-5);
std::cout << GridLogMessage<<"Plaquette and link trace match " << std::endl;
}
}
};
}}
//HAVE_LIME
#endif
#endif

231
lib/parallelIO/IldgIOtypes.h Normal file
View File

@ -0,0 +1,231 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/parallelIO/IldgIO.h
Copyright (C) 2015
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_ILDGTYPES_IO_H
#define GRID_ILDGTYPES_IO_H
#ifdef HAVE_LIME
extern "C" { // for linkage
#include "lime.h"
}
namespace Grid {
/////////////////////////////////////////////////////////////////////////////////
// Data representation of records that enter ILDG and SciDac formats
/////////////////////////////////////////////////////////////////////////////////
#define GRID_FORMAT "grid-format"
#define ILDG_FORMAT "ildg-format"
#define ILDG_BINARY_DATA "ildg-binary-data"
#define ILDG_DATA_LFN "ildg-data-lfn"
#define SCIDAC_CHECKSUM "scidac-checksum"
#define SCIDAC_PRIVATE_FILE_XML "scidac-private-file-xml"
#define SCIDAC_FILE_XML "scidac-file-xml"
#define SCIDAC_PRIVATE_RECORD_XML "scidac-private-record-xml"
#define SCIDAC_RECORD_XML "scidac-record-xml"
#define SCIDAC_BINARY_DATA "scidac-binary-data"
// Unused SCIDAC records names; could move to support this functionality
#define SCIDAC_SITELIST "scidac-sitelist"
////////////////////////////////////////////////////////////
const int GRID_IO_SINGLEFILE = 0; // hardcode lift from QIO compat
const int GRID_IO_MULTIFILE = 1; // hardcode lift from QIO compat
const int GRID_IO_FIELD = 0; // hardcode lift from QIO compat
const int GRID_IO_GLOBAL = 1; // hardcode lift from QIO compat
////////////////////////////////////////////////////////////
/////////////////////////////////////////////////////////////////////////////////
// QIO uses mandatory "private" records fixed format
// Private is in principle "opaque" however it can't be changed now because that would break existing
// file compatability, so should be correct to assume the undocumented but defacto file structure.
/////////////////////////////////////////////////////////////////////////////////
////////////////////////
// Scidac private file xml
// <?xml version="1.0" encoding="UTF-8"?><scidacFile><version>1.1</version><spacetime>4</spacetime><dims>16 16 16 32 </dims><volfmt>0</volfmt></scidacFile>
////////////////////////
struct scidacFile : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(scidacFile,
double, version,
int, spacetime,
std::string, dims, // must convert to int
int, volfmt);
std::vector<int> getDimensions(void) {
std::stringstream stream(dims);
std::vector<int> dimensions;
int n;
while(stream >> n){
dimensions.push_back(n);
}
return dimensions;
}
void setDimensions(std::vector<int> dimensions) {
char delimiter = ' ';
std::stringstream stream;
for(int i=0;i<dimensions.size();i++){
stream << dimensions[i];
if ( i != dimensions.size()-1) {
stream << delimiter <<std::endl;
}
}
dims = stream.str();
}
// Constructor provides Grid
scidacFile() =default; // default constructor
scidacFile(GridBase * grid){
version = 1.0;
spacetime = grid->_ndimension;
setDimensions(grid->FullDimensions());
volfmt = GRID_IO_SINGLEFILE;
}
};
///////////////////////////////////////////////////////////////////////
// scidac-private-record-xml : example
// <scidacRecord>
// <version>1.1</version><date>Tue Jul 26 21:14:44 2011 UTC</date><recordtype>0</recordtype>
// <datatype>QDP_D3_ColorMatrix</datatype><precision>D</precision><colors>3</colors><spins>4</spins>
// <typesize>144</typesize><datacount>4</datacount>
// </scidacRecord>
///////////////////////////////////////////////////////////////////////
struct scidacRecord : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(scidacRecord,
double, version,
std::string, date,
int, recordtype,
std::string, datatype,
std::string, precision,
int, colors,
int, spins,
int, typesize,
int, datacount);
scidacRecord() { version =1.0; }
};
////////////////////////
// ILDG format
////////////////////////
struct ildgFormat : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(ildgFormat,
double, version,
std::string, field,
int, precision,
int, lx,
int, ly,
int, lz,
int, lt);
ildgFormat() { version=1.0; };
};
////////////////////////
// USQCD info
////////////////////////
struct usqcdInfo : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdInfo,
double, version,
double, plaq,
double, linktr,
std::string, info);
usqcdInfo() {
version=1.0;
};
};
////////////////////////
// Scidac Checksum
////////////////////////
struct scidacChecksum : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(scidacChecksum,
double, version,
std::string, suma,
std::string, sumb);
scidacChecksum() {
version=1.0;
};
};
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Type: scidac-file-xml <title>MILC ILDG archival gauge configuration</title>
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Type:
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////
// Scidac private file xml
// <?xml version="1.0" encoding="UTF-8"?><scidacFile><version>1.1</version><spacetime>4</spacetime><dims>16 16 16 32 </dims><volfmt>0</volfmt></scidacFile>
////////////////////////
#if 0
////////////////////////////////////////////////////////////////////////////////////////
// From http://www.physics.utah.edu/~detar/scidac/qio_2p3.pdf
////////////////////////////////////////////////////////////////////////////////////////
struct usqcdPropFile : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdPropFile,
double, version,
std::string, type,
std::string, info);
usqcdPropFile() {
version=1.0;
};
};
struct usqcdSourceInfo : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdSourceInfo,
double, version,
std::string, info);
usqcdSourceInfo() {
version=1.0;
};
};
struct usqcdPropInfo : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(usqcdPropInfo,
double, version,
int, spin,
int, color,
std::string, info);
usqcdPropInfo() {
version=1.0;
};
};
#endif
}
#endif
#endif

325
lib/parallelIO/MetaData.h Normal file
View File

@ -0,0 +1,325 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/parallelIO/NerscIO.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 */
#include <algorithm>
#include <iostream>
#include <iomanip>
#include <fstream>
#include <map>
#include <unistd.h>
#include <sys/utsname.h>
#include <pwd.h>
namespace Grid {
///////////////////////////////////////////////////////
// Precision mapping
///////////////////////////////////////////////////////
template<class vobj> static std::string getFormatString (void)
{
std::string format;
typedef typename getPrecision<vobj>::real_scalar_type stype;
if ( sizeof(stype) == sizeof(float) ) {
format = std::string("IEEE32BIG");
}
if ( sizeof(stype) == sizeof(double) ) {
format = std::string("IEEE64BIG");
}
return format;
}
////////////////////////////////////////////////////////////////////////////////
// header specification/interpretation
////////////////////////////////////////////////////////////////////////////////
class FieldMetaData : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(FieldMetaData,
int, nd,
std::vector<int>, dimension,
std::vector<std::string>, boundary,
int, data_start,
std::string, hdr_version,
std::string, storage_format,
double, link_trace,
double, plaquette,
uint32_t, checksum,
uint32_t, scidac_checksuma,
uint32_t, scidac_checksumb,
unsigned int, sequence_number,
std::string, data_type,
std::string, ensemble_id,
std::string, ensemble_label,
std::string, ildg_lfn,
std::string, creator,
std::string, creator_hardware,
std::string, creation_date,
std::string, archive_date,
std::string, floating_point);
FieldMetaData(void) {
nd=4;
dimension.resize(4);
boundary.resize(4);
}
};
namespace QCD {
using namespace Grid;
//////////////////////////////////////////////////////////////////////
// Bit and Physical Checksumming and QA of data
//////////////////////////////////////////////////////////////////////
inline void GridMetaData(GridBase *grid,FieldMetaData &header)
{
int nd = grid->_ndimension;
header.nd = nd;
header.dimension.resize(nd);
header.boundary.resize(nd);
for(int d=0;d<nd;d++) {
header.dimension[d] = grid->_fdimensions[d];
}
for(int d=0;d<nd;d++) {
header.boundary[d] = std::string("PERIODIC");
}
}
inline void MachineCharacteristics(FieldMetaData &header)
{
// Who
struct passwd *pw = getpwuid (getuid());
if (pw) header.creator = std::string(pw->pw_name);
// When
std::time_t t = std::time(nullptr);
std::tm tm_ = *std::localtime(&t);
std::ostringstream oss;
// oss << std::put_time(&tm_, "%c %Z");
header.creation_date = oss.str();
header.archive_date = header.creation_date;
// What
struct utsname name; uname(&name);
header.creator_hardware = std::string(name.nodename)+"-";
header.creator_hardware+= std::string(name.machine)+"-";
header.creator_hardware+= std::string(name.sysname)+"-";
header.creator_hardware+= std::string(name.release);
}
#define dump_meta_data(field, s) \
s << "BEGIN_HEADER" << std::endl; \
s << "HDR_VERSION = " << field.hdr_version << std::endl; \
s << "DATATYPE = " << field.data_type << std::endl; \
s << "STORAGE_FORMAT = " << field.storage_format << std::endl; \
for(int i=0;i<4;i++){ \
s << "DIMENSION_" << i+1 << " = " << field.dimension[i] << std::endl ; \
} \
s << "LINK_TRACE = " << std::setprecision(10) << field.link_trace << std::endl; \
s << "PLAQUETTE = " << std::setprecision(10) << field.plaquette << std::endl; \
for(int i=0;i<4;i++){ \
s << "BOUNDARY_"<<i+1<<" = " << field.boundary[i] << std::endl; \
} \
\
s << "CHECKSUM = "<< std::hex << std::setw(10) << field.checksum << std::dec<<std::endl; \
s << "SCIDAC_CHECKSUMA = "<< std::hex << std::setw(10) << field.scidac_checksuma << std::dec<<std::endl; \
s << "SCIDAC_CHECKSUMB = "<< std::hex << std::setw(10) << field.scidac_checksumb << std::dec<<std::endl; \
s << "ENSEMBLE_ID = " << field.ensemble_id << std::endl; \
s << "ENSEMBLE_LABEL = " << field.ensemble_label << std::endl; \
s << "SEQUENCE_NUMBER = " << field.sequence_number << std::endl; \
s << "CREATOR = " << field.creator << std::endl; \
s << "CREATOR_HARDWARE = "<< field.creator_hardware << std::endl; \
s << "CREATION_DATE = " << field.creation_date << std::endl; \
s << "ARCHIVE_DATE = " << field.archive_date << std::endl; \
s << "FLOATING_POINT = " << field.floating_point << std::endl; \
s << "END_HEADER" << std::endl;
template<class vobj> inline void PrepareMetaData(Lattice<vobj> & field, FieldMetaData &header)
{
GridBase *grid = field._grid;
std::string format = getFormatString<vobj>();
header.floating_point = format;
header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
GridMetaData(grid,header);
MachineCharacteristics(header);
}
inline void GaugeStatistics(Lattice<vLorentzColourMatrixF> & data,FieldMetaData &header)
{
// How to convert data precision etc...
header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplF>::linkTrace(data);
header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplF>::avgPlaquette(data);
}
inline void GaugeStatistics(Lattice<vLorentzColourMatrixD> & data,FieldMetaData &header)
{
// How to convert data precision etc...
header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplD>::linkTrace(data);
header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplD>::avgPlaquette(data);
}
template<> inline void PrepareMetaData<vLorentzColourMatrixF>(Lattice<vLorentzColourMatrixF> & field, FieldMetaData &header)
{
GridBase *grid = field._grid;
std::string format = getFormatString<vLorentzColourMatrixF>();
header.floating_point = format;
header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
GridMetaData(grid,header);
GaugeStatistics(field,header);
MachineCharacteristics(header);
}
template<> inline void PrepareMetaData<vLorentzColourMatrixD>(Lattice<vLorentzColourMatrixD> & field, FieldMetaData &header)
{
GridBase *grid = field._grid;
std::string format = getFormatString<vLorentzColourMatrixD>();
header.floating_point = format;
header.checksum = 0x0; // Nersc checksum unused in ILDG, Scidac
GridMetaData(grid,header);
GaugeStatistics(field,header);
MachineCharacteristics(header);
}
//////////////////////////////////////////////////////////////////////
// Utilities ; these are QCD aware
//////////////////////////////////////////////////////////////////////
inline void reconstruct3(LorentzColourMatrix & cm)
{
const int x=0;
const int y=1;
const int z=2;
for(int mu=0;mu<Nd;mu++){
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
}
}
////////////////////////////////////////////////////////////////////////////////
// Some data types for intermediate storage
////////////////////////////////////////////////////////////////////////////////
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, Nd >;
typedef iLorentzColour2x3<Complex> LorentzColour2x3;
typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
typedef iLorentzColour2x3<ComplexD> LorentzColour2x3D;
/////////////////////////////////////////////////////////////////////////////////
// Simple classes for precision conversion
/////////////////////////////////////////////////////////////////////////////////
template <class fobj, class sobj>
struct BinarySimpleUnmunger {
typedef typename getPrecision<fobj>::real_scalar_type fobj_stype;
typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
void operator()(sobj &in, fobj &out) {
// take word by word and transform accoding to the status
fobj_stype *out_buffer = (fobj_stype *)&out;
sobj_stype *in_buffer = (sobj_stype *)&in;
size_t fobj_words = sizeof(out) / sizeof(fobj_stype);
size_t sobj_words = sizeof(in) / sizeof(sobj_stype);
assert(fobj_words == sobj_words);
for (unsigned int word = 0; word < sobj_words; word++)
out_buffer[word] = in_buffer[word]; // type conversion on the fly
}
};
template <class fobj, class sobj>
struct BinarySimpleMunger {
typedef typename getPrecision<fobj>::real_scalar_type fobj_stype;
typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
void operator()(fobj &in, sobj &out) {
// take word by word and transform accoding to the status
fobj_stype *in_buffer = (fobj_stype *)&in;
sobj_stype *out_buffer = (sobj_stype *)&out;
size_t fobj_words = sizeof(in) / sizeof(fobj_stype);
size_t sobj_words = sizeof(out) / sizeof(sobj_stype);
assert(fobj_words == sobj_words);
for (unsigned int word = 0; word < sobj_words; word++)
out_buffer[word] = in_buffer[word]; // type conversion on the fly
}
};
template<class fobj,class sobj>
struct GaugeSimpleMunger{
void operator()(fobj &in, sobj &out) {
for (int mu = 0; mu < Nd; mu++) {
for (int i = 0; i < Nc; i++) {
for (int j = 0; j < Nc; j++) {
out(mu)()(i, j) = in(mu)()(i, j);
}}
}
};
};
template <class fobj, class sobj>
struct GaugeSimpleUnmunger {
void operator()(sobj &in, fobj &out) {
for (int mu = 0; mu < Nd; mu++) {
for (int i = 0; i < Nc; i++) {
for (int j = 0; j < Nc; j++) {
out(mu)()(i, j) = in(mu)()(i, j);
}}
}
};
};
template<class fobj,class sobj>
struct Gauge3x2munger{
void operator() (fobj &in,sobj &out){
for(int mu=0;mu<Nd;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
out(mu)()(i,j) = in(mu)(i)(j);
}}
}
reconstruct3(out);
}
};
template<class fobj,class sobj>
struct Gauge3x2unmunger{
void operator() (sobj &in,fobj &out){
for(int mu=0;mu<Nd;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
out(mu)(i)(j) = in(mu)()(i,j);
}}
}
}
};
}
}

716
lib/parallelIO/NerscIO.h
View File

@ -1,4 +1,4 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@ -6,9 +6,9 @@
Copyright (C) 2015
Author: Matt Spraggs <matthew.spraggs@gmail.com>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Matt Spraggs <matthew.spraggs@gmail.com>
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
@ -25,253 +25,59 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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 */
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_NERSC_IO_H
#define GRID_NERSC_IO_H
#include <algorithm>
#include <iostream>
#include <iomanip>
#include <fstream>
#include <map>
#include <unistd.h>
#include <sys/utsname.h>
#include <pwd.h>
namespace Grid {
namespace QCD {
namespace QCD {
using namespace Grid;
using namespace Grid;
////////////////////////////////////////////////////////////////////////////////
// Some data types for intermediate storage
////////////////////////////////////////////////////////////////////////////////
template<typename vtype> using iLorentzColour2x3 = iVector<iVector<iVector<vtype, Nc>, 2>, 4 >;
////////////////////////////////////////////////////////////////////////////////
// Write and read from fstream; comput header offset for payload
////////////////////////////////////////////////////////////////////////////////
class NerscIO : public BinaryIO {
public:
typedef iLorentzColour2x3<Complex> LorentzColour2x3;
typedef iLorentzColour2x3<ComplexF> LorentzColour2x3F;
typedef iLorentzColour2x3<ComplexD> LorentzColour2x3D;
////////////////////////////////////////////////////////////////////////////////
// header specification/interpretation
////////////////////////////////////////////////////////////////////////////////
class NerscField {
public:
// header strings (not in order)
int dimension[4];
std::string boundary[4];
int data_start;
std::string hdr_version;
std::string storage_format;
// Checks on data
double link_trace;
double plaquette;
uint32_t checksum;
unsigned int sequence_number;
std::string data_type;
std::string ensemble_id ;
std::string ensemble_label ;
std::string creator ;
std::string creator_hardware ;
std::string creation_date ;
std::string archive_date ;
std::string floating_point;
};
//////////////////////////////////////////////////////////////////////
// Bit and Physical Checksumming and QA of data
//////////////////////////////////////////////////////////////////////
inline void NerscGrid(GridBase *grid,NerscField &header)
{
assert(grid->_ndimension==4);
for(int d=0;d<4;d++) {
header.dimension[d] = grid->_fdimensions[d];
}
for(int d=0;d<4;d++) {
header.boundary[d] = std::string("PERIODIC");
}
}
template<class GaugeField>
inline void NerscStatistics(GaugeField & data,NerscField &header)
{
// How to convert data precision etc...
header.link_trace=Grid::QCD::WilsonLoops<PeriodicGimplR>::linkTrace(data);
header.plaquette =Grid::QCD::WilsonLoops<PeriodicGimplR>::avgPlaquette(data);
}
inline void NerscMachineCharacteristics(NerscField &header)
{
// Who
struct passwd *pw = getpwuid (getuid());
if (pw) header.creator = std::string(pw->pw_name);
// When
std::time_t t = std::time(nullptr);
std::tm tm = *std::localtime(&t);
std::ostringstream oss;
// oss << std::put_time(&tm, "%c %Z");
header.creation_date = oss.str();
header.archive_date = header.creation_date;
// What
struct utsname name; uname(&name);
header.creator_hardware = std::string(name.nodename)+"-";
header.creator_hardware+= std::string(name.machine)+"-";
header.creator_hardware+= std::string(name.sysname)+"-";
header.creator_hardware+= std::string(name.release);
}
//////////////////////////////////////////////////////////////////////
// Utilities ; these are QCD aware
//////////////////////////////////////////////////////////////////////
inline void NerscChecksum(uint32_t *buf,uint32_t buf_size_bytes,uint32_t &csum)
{
BinaryIO::Uint32Checksum(buf,buf_size_bytes,csum);
}
inline void reconstruct3(LorentzColourMatrix & cm)
{
const int x=0;
const int y=1;
const int z=2;
for(int mu=0;mu<4;mu++){
cm(mu)()(2,x) = adj(cm(mu)()(0,y)*cm(mu)()(1,z)-cm(mu)()(0,z)*cm(mu)()(1,y)); //x= yz-zy
cm(mu)()(2,y) = adj(cm(mu)()(0,z)*cm(mu)()(1,x)-cm(mu)()(0,x)*cm(mu)()(1,z)); //y= zx-xz
cm(mu)()(2,z) = adj(cm(mu)()(0,x)*cm(mu)()(1,y)-cm(mu)()(0,y)*cm(mu)()(1,x)); //z= xy-yx
static inline void truncate(std::string file){
std::ofstream fout(file,std::ios::out);
}
static inline unsigned int writeHeader(FieldMetaData &field,std::string file)
{
std::ofstream fout(file,std::ios::out|std::ios::in);
fout.seekp(0,std::ios::beg);
dump_meta_data(field, fout);
field.data_start = fout.tellp();
return field.data_start;
}
template<class fobj,class sobj>
struct NerscSimpleMunger{
// for the header-reader
static inline int readHeader(std::string file,GridBase *grid, FieldMetaData &field)
{
int offset=0;
std::map<std::string,std::string> header;
std::string line;
void operator() (fobj &in,sobj &out,uint32_t &csum){
//////////////////////////////////////////////////
// read the header
//////////////////////////////////////////////////
std::ifstream fin(file);
for(int mu=0;mu<4;mu++){
for(int i=0;i<3;i++){
for(int j=0;j<3;j++){
out(mu)()(i,j) = in(mu)()(i,j);
}}}
NerscChecksum((uint32_t *)&in,sizeof(in),csum);
};
};
getline(fin,line); // read one line and insist is
template<class fobj,class sobj>
struct NerscSimpleUnmunger{
void operator() (sobj &in,fobj &out,uint32_t &csum){
for(int mu=0;mu<Nd;mu++){
for(int i=0;i<Nc;i++){
for(int j=0;j<Nc;j++){
out(mu)()(i,j) = in(mu)()(i,j);
}}}
NerscChecksum((uint32_t *)&out,sizeof(out),csum);
};
};
template<class fobj,class sobj>
struct Nersc3x2munger{
void operator() (fobj &in,sobj &out,uint32_t &csum){
NerscChecksum((uint32_t *)&in,sizeof(in),csum);
removeWhitespace(line);
std::cout << GridLogMessage << "* " << line << std::endl;
for(int mu=0;mu<4;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
out(mu)()(i,j) = in(mu)(i)(j);
}}
}
reconstruct3(out);
}
};
assert(line==std::string("BEGIN_HEADER"));
template<class fobj,class sobj>
struct Nersc3x2unmunger{
void operator() (sobj &in,fobj &out,uint32_t &csum){
for(int mu=0;mu<4;mu++){
for(int i=0;i<2;i++){
for(int j=0;j<3;j++){
out(mu)(i)(j) = in(mu)()(i,j);
}}
}
NerscChecksum((uint32_t *)&out,sizeof(out),csum);
}
};
////////////////////////////////////////////////////////////////////////////////
// Write and read from fstream; comput header offset for payload
////////////////////////////////////////////////////////////////////////////////
class NerscIO : public BinaryIO {
public:
static inline void truncate(std::string file){
std::ofstream fout(file,std::ios::out);
}
#define dump_nersc_header(field, s)\
s << "BEGIN_HEADER" << std::endl;\
s << "HDR_VERSION = " << field.hdr_version << std::endl;\
s << "DATATYPE = " << field.data_type << std::endl;\
s << "STORAGE_FORMAT = " << field.storage_format << std::endl;\
for(int i=0;i<4;i++){\
s << "DIMENSION_" << i+1 << " = " << field.dimension[i] << std::endl ;\
}\
s << "LINK_TRACE = " << std::setprecision(10) << field.link_trace << std::endl;\
s << "PLAQUETTE = " << std::setprecision(10) << field.plaquette << std::endl;\
for(int i=0;i<4;i++){\
s << "BOUNDARY_"<<i+1<<" = " << field.boundary[i] << std::endl;\
}\
\
s << "CHECKSUM = "<< std::hex << std::setw(10) << field.checksum << std::dec<<std::endl;\
s << "ENSEMBLE_ID = " << field.ensemble_id << std::endl;\
s << "ENSEMBLE_LABEL = " << field.ensemble_label << std::endl;\
s << "SEQUENCE_NUMBER = " << field.sequence_number << std::endl;\
s << "CREATOR = " << field.creator << std::endl;\
s << "CREATOR_HARDWARE = "<< field.creator_hardware << std::endl;\
s << "CREATION_DATE = " << field.creation_date << std::endl;\
s << "ARCHIVE_DATE = " << field.archive_date << std::endl;\
s << "FLOATING_POINT = " << field.floating_point << std::endl;\
s << "END_HEADER" << std::endl;
static inline unsigned int writeHeader(NerscField &field,std::string file)
{
std::ofstream fout(file,std::ios::out|std::ios::in);
fout.seekp(0,std::ios::beg);
dump_nersc_header(field, fout);
field.data_start = fout.tellp();
return field.data_start;
}
// for the header-reader
static inline int readHeader(std::string file,GridBase *grid, NerscField &field)
{
int offset=0;
std::map<std::string,std::string> header;
std::string line;
//////////////////////////////////////////////////
// read the header
//////////////////////////////////////////////////
std::ifstream fin(file);
getline(fin,line); // read one line and insist is
removeWhitespace(line);
std::cout << GridLogMessage << "* " << line << std::endl;
assert(line==std::string("BEGIN_HEADER"));
do {
getline(fin,line); // read one line
std::cout << GridLogMessage << "* "<<line<< std::endl;
int eq = line.find("=");
if(eq >0) {
do {
getline(fin,line); // read one line
std::cout << GridLogMessage << "* "<<line<< std::endl;
int eq = line.find("=");
if(eq >0) {
std::string key=line.substr(0,eq);
std::string val=line.substr(eq+1);
removeWhitespace(key);
@ -279,275 +85,269 @@ static inline int readHeader(std::string file,GridBase *grid, NerscField &field
header[key] = val;
}
} while( line.find("END_HEADER") == std::string::npos );
} while( line.find("END_HEADER") == std::string::npos );
field.data_start = fin.tellg();
field.data_start = fin.tellg();
//////////////////////////////////////////////////
// chomp the values
//////////////////////////////////////////////////
field.hdr_version = header["HDR_VERSION"];
field.data_type = header["DATATYPE"];
field.storage_format = header["STORAGE_FORMAT"];
//////////////////////////////////////////////////
// chomp the values
//////////////////////////////////////////////////
field.hdr_version = header["HDR_VERSION"];
field.data_type = header["DATATYPE"];
field.storage_format = header["STORAGE_FORMAT"];
field.dimension[0] = std::stol(header["DIMENSION_1"]);
field.dimension[1] = std::stol(header["DIMENSION_2"]);
field.dimension[2] = std::stol(header["DIMENSION_3"]);
field.dimension[3] = std::stol(header["DIMENSION_4"]);
field.dimension[0] = std::stol(header["DIMENSION_1"]);
field.dimension[1] = std::stol(header["DIMENSION_2"]);
field.dimension[2] = std::stol(header["DIMENSION_3"]);
field.dimension[3] = std::stol(header["DIMENSION_4"]);
assert(grid->_ndimension == 4);
for(int d=0;d<4;d++){
assert(grid->_fdimensions[d]==field.dimension[d]);
}
field.link_trace = std::stod(header["LINK_TRACE"]);
field.plaquette = std::stod(header["PLAQUETTE"]);
field.boundary[0] = header["BOUNDARY_1"];
field.boundary[1] = header["BOUNDARY_2"];
field.boundary[2] = header["BOUNDARY_3"];
field.boundary[3] = header["BOUNDARY_4"];
field.checksum = std::stoul(header["CHECKSUM"],0,16);
field.ensemble_id = header["ENSEMBLE_ID"];
field.ensemble_label = header["ENSEMBLE_LABEL"];
field.sequence_number = std::stol(header["SEQUENCE_NUMBER"]);
field.creator = header["CREATOR"];
field.creator_hardware = header["CREATOR_HARDWARE"];
field.creation_date = header["CREATION_DATE"];
field.archive_date = header["ARCHIVE_DATE"];
field.floating_point = header["FLOATING_POINT"];
return field.data_start;
}
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Now the meat: the object readers
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
#define PARALLEL_READ
#define PARALLEL_WRITE
template<class vsimd>
static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,NerscField& header,std::string file)
{
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
GridBase *grid = Umu._grid;
int offset = readHeader(file,Umu._grid,header);
NerscField clone(header);
std::string format(header.floating_point);
int ieee32big = (format == std::string("IEEE32BIG"));
int ieee32 = (format == std::string("IEEE32"));
int ieee64big = (format == std::string("IEEE64BIG"));
int ieee64 = (format == std::string("IEEE64"));
uint32_t csum;
// depending on datatype, set up munger;
// munger is a function of <floating point, Real, data_type>
if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
if ( ieee32 || ieee32big ) {
#ifdef PARALLEL_READ
csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>, LorentzColour2x3F>
(Umu,file,Nersc3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format);
#else
csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>, LorentzColour2x3F>
(Umu,file,Nersc3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format);
#endif
assert(grid->_ndimension == 4);
for(int d=0;d<4;d++){
assert(grid->_fdimensions[d]==field.dimension[d]);
}
if ( ieee64 || ieee64big ) {
#ifdef PARALLEL_READ
csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>, LorentzColour2x3D>
(Umu,file,Nersc3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format);
#else
csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>, LorentzColour2x3D>
(Umu,file,Nersc3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format);
#endif
field.link_trace = std::stod(header["LINK_TRACE"]);
field.plaquette = std::stod(header["PLAQUETTE"]);
field.boundary[0] = header["BOUNDARY_1"];
field.boundary[1] = header["BOUNDARY_2"];
field.boundary[2] = header["BOUNDARY_3"];
field.boundary[3] = header["BOUNDARY_4"];
field.checksum = std::stoul(header["CHECKSUM"],0,16);
field.ensemble_id = header["ENSEMBLE_ID"];
field.ensemble_label = header["ENSEMBLE_LABEL"];
field.sequence_number = std::stol(header["SEQUENCE_NUMBER"]);
field.creator = header["CREATOR"];
field.creator_hardware = header["CREATOR_HARDWARE"];
field.creation_date = header["CREATION_DATE"];
field.archive_date = header["ARCHIVE_DATE"];
field.floating_point = header["FLOATING_POINT"];
return field.data_start;
}
} else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
if ( ieee32 || ieee32big ) {
#ifdef PARALLEL_READ
csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
(Umu,file,NerscSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format);
#else
csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
(Umu,file,NerscSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format);
#endif
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Now the meat: the object readers
/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vsimd>
static inline void readConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
FieldMetaData& header,
std::string file)
{
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
GridBase *grid = Umu._grid;
int offset = readHeader(file,Umu._grid,header);
FieldMetaData clone(header);
std::string format(header.floating_point);
int ieee32big = (format == std::string("IEEE32BIG"));
int ieee32 = (format == std::string("IEEE32"));
int ieee64big = (format == std::string("IEEE64BIG"));
int ieee64 = (format == std::string("IEEE64"));
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
// depending on datatype, set up munger;
// munger is a function of <floating point, Real, data_type>
if ( header.data_type == std::string("4D_SU3_GAUGE") ) {
if ( ieee32 || ieee32big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3F>
(Umu,file,Gauge3x2munger<LorentzColour2x3F,LorentzColourMatrix>(), offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
if ( ieee64 || ieee64big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>, LorentzColour2x3D>
(Umu,file,Gauge3x2munger<LorentzColour2x3D,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
} else if ( header.data_type == std::string("4D_SU3_GAUGE_3x3") ) {
if ( ieee32 || ieee32big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixF>
(Umu,file,GaugeSimpleMunger<LorentzColourMatrixF,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
if ( ieee64 || ieee64big ) {
BinaryIO::readLatticeObject<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
(Umu,file,GaugeSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format,
nersc_csum,scidac_csuma,scidac_csumb);
}
} else {
assert(0);
}
GaugeStatistics(Umu,clone);
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" checksum "<<std::hex<<nersc_csum<< std::dec
<<" header "<<std::hex<<header.checksum<<std::dec <<std::endl;
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" plaquette "<<clone.plaquette
<<" header "<<header.plaquette<<std::endl;
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" link_trace "<<clone.link_trace
<<" header "<<header.link_trace<<std::endl;
if ( fabs(clone.plaquette -header.plaquette ) >= 1.0e-5 ) {
std::cout << " Plaquette mismatch "<<std::endl;
std::cout << Umu[0]<<std::endl;
std::cout << Umu[1]<<std::endl;
}
if ( nersc_csum != header.checksum ) {
std::cerr << " checksum mismatch " << std::endl;
std::cerr << " plaqs " << clone.plaquette << " " << header.plaquette << std::endl;
std::cerr << " trace " << clone.link_trace<< " " << header.link_trace<< std::endl;
std::cerr << " nersc_csum " <<std::hex<< nersc_csum << " " << header.checksum<< std::dec<< std::endl;
exit(0);
}
assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-5 );
assert(fabs(clone.link_trace-header.link_trace) < 1.0e-6 );
assert(nersc_csum == header.checksum );
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
}
if ( ieee64 || ieee64big ) {
#ifdef PARALLEL_READ
csum=BinaryIO::readObjectParallel<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
(Umu,file,NerscSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format);
#else
csum=BinaryIO::readObjectSerial<iLorentzColourMatrix<vsimd>,LorentzColourMatrixD>
(Umu,file,NerscSimpleMunger<LorentzColourMatrixD,LorentzColourMatrix>(),offset,format);
#endif
}
} else {
assert(0);
}
NerscStatistics<GaugeField>(Umu,clone);
template<class vsimd>
static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,
std::string file,
int two_row,
int bits32)
{
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" checksum "<<std::hex<< csum<< std::dec
<<" header "<<std::hex<<header.checksum<<std::dec <<std::endl;
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" plaquette "<<clone.plaquette
<<" header "<<header.plaquette<<std::endl;
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<<" link_trace "<<clone.link_trace
<<" header "<<header.link_trace<<std::endl;
assert(fabs(clone.plaquette -header.plaquette ) < 1.0e-5 );
assert(fabs(clone.link_trace-header.link_trace) < 1.0e-6 );
assert(csum == header.checksum );
typedef iLorentzColourMatrix<vsimd> vobj;
typedef typename vobj::scalar_object sobj;
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
}
FieldMetaData header;
///////////////////////////////////////////
// Following should become arguments
///////////////////////////////////////////
header.sequence_number = 1;
header.ensemble_id = "UKQCD";
header.ensemble_label = "DWF";
template<class vsimd>
static inline void writeConfiguration(Lattice<iLorentzColourMatrix<vsimd> > &Umu,std::string file, int two_row,int bits32)
{
typedef Lattice<iLorentzColourMatrix<vsimd> > GaugeField;
typedef iLorentzColourMatrix<vsimd> vobj;
typedef typename vobj::scalar_object sobj;
// Following should become arguments
NerscField header;
header.sequence_number = 1;
header.ensemble_id = "UKQCD";
header.ensemble_label = "DWF";
typedef LorentzColourMatrixD fobj3D;
typedef LorentzColour2x3D fobj2D;
typedef LorentzColourMatrixF fobj3f;
typedef LorentzColour2x3F fobj2f;
GridBase *grid = Umu._grid;
NerscGrid(grid,header);
NerscStatistics<GaugeField>(Umu,header);
NerscMachineCharacteristics(header);
uint32_t csum;
int offset;
typedef LorentzColourMatrixD fobj3D;
typedef LorentzColour2x3D fobj2D;
truncate(file);
GridBase *grid = Umu._grid;
if ( two_row ) {
GridMetaData(grid,header);
assert(header.nd==4);
GaugeStatistics(Umu,header);
MachineCharacteristics(header);
header.floating_point = std::string("IEEE64BIG");
header.data_type = std::string("4D_SU3_GAUGE");
Nersc3x2unmunger<fobj2D,sobj> munge;
BinaryIO::Uint32Checksum<vobj,fobj2D>(Umu, munge,header.checksum);
offset = writeHeader(header,file);
#ifdef PARALLEL_WRITE
csum=BinaryIO::writeObjectParallel<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point);
#else
csum=BinaryIO::writeObjectSerial<vobj,fobj2D>(Umu,file,munge,offset,header.floating_point);
#endif
int offset;
truncate(file);
} else {
header.floating_point = std::string("IEEE64BIG");
header.data_type = std::string("4D_SU3_GAUGE_3x3");
NerscSimpleUnmunger<fobj3D,sobj> munge;
BinaryIO::Uint32Checksum<vobj,fobj3D>(Umu, munge,header.checksum);
offset = writeHeader(header,file);
#ifdef PARALLEL_WRITE
csum=BinaryIO::writeObjectParallel<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point);
#else
csum=BinaryIO::writeObjectSerial<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point);
#endif
}
// Sod it -- always write 3x3 double
header.floating_point = std::string("IEEE64BIG");
header.data_type = std::string("4D_SU3_GAUGE_3x3");
GaugeSimpleUnmunger<fobj3D,sobj> munge;
offset = writeHeader(header,file);
std::cout<<GridLogMessage <<"Written NERSC Configuration "<<file<< " checksum "<<std::hex<<csum<< std::dec<<" plaq "<< header.plaquette <<std::endl;
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::writeLatticeObject<vobj,fobj3D>(Umu,file,munge,offset,header.floating_point,
nersc_csum,scidac_csuma,scidac_csumb);
header.checksum = nersc_csum;
writeHeader(header,file);
}
std::cout<<GridLogMessage <<"Written NERSC Configuration on "<< file << " checksum "
<<std::hex<<header.checksum
<<std::dec<<" plaq "<< header.plaquette <<std::endl;
}
///////////////////////////////
// RNG state
///////////////////////////////
static inline void writeRNGState(GridSerialRNG &serial,GridParallelRNG &parallel,std::string file)
{
typedef typename GridParallelRNG::RngStateType RngStateType;
///////////////////////////////
// RNG state
///////////////////////////////
static inline void writeRNGState(GridSerialRNG &serial,GridParallelRNG &parallel,std::string file)
{
typedef typename GridParallelRNG::RngStateType RngStateType;
// Following should become arguments
FieldMetaData header;
header.sequence_number = 1;
header.ensemble_id = "UKQCD";
header.ensemble_label = "DWF";
// Following should become arguments
NerscField header;
header.sequence_number = 1;
header.ensemble_id = "UKQCD";
header.ensemble_label = "DWF";
GridBase *grid = parallel._grid;
GridBase *grid = parallel._grid;
GridMetaData(grid,header);
assert(header.nd==4);
header.link_trace=0.0;
header.plaquette=0.0;
MachineCharacteristics(header);
NerscGrid(grid,header);
header.link_trace=0.0;
header.plaquette=0.0;
NerscMachineCharacteristics(header);
uint32_t csum;
int offset;
int offset;
#ifdef RNG_RANLUX
header.floating_point = std::string("UINT64");
header.data_type = std::string("RANLUX48");
header.floating_point = std::string("UINT64");
header.data_type = std::string("RANLUX48");
#endif
#ifdef RNG_MT19937
header.floating_point = std::string("UINT32");
header.data_type = std::string("MT19937");
header.floating_point = std::string("UINT32");
header.data_type = std::string("MT19937");
#endif
#ifdef RNG_SITMO
header.floating_point = std::string("UINT64");
header.data_type = std::string("SITMO");
header.floating_point = std::string("UINT64");
header.data_type = std::string("SITMO");
#endif
truncate(file);
offset = writeHeader(header,file);
csum=BinaryIO::writeRNGSerial(serial,parallel,file,offset);
header.checksum = csum;
offset = writeHeader(header,file);
truncate(file);
offset = writeHeader(header,file);
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::writeRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
header.checksum = nersc_csum;
offset = writeHeader(header,file);
std::cout<<GridLogMessage <<"Written NERSC RNG STATE "<<file<< " checksum "<<std::hex<<csum<<std::dec<<std::endl;
std::cout<<GridLogMessage
<<"Written NERSC RNG STATE "<<file<< " checksum "
<<std::hex<<header.checksum
<<std::dec<<std::endl;
}
}
static inline void readRNGState(GridSerialRNG &serial,GridParallelRNG & parallel,NerscField& header,std::string file)
{
typedef typename GridParallelRNG::RngStateType RngStateType;
static inline void readRNGState(GridSerialRNG &serial,GridParallelRNG & parallel,FieldMetaData& header,std::string file)
{
typedef typename GridParallelRNG::RngStateType RngStateType;
GridBase *grid = parallel._grid;
GridBase *grid = parallel._grid;
int offset = readHeader(file,grid,header);
int offset = readHeader(file,grid,header);
NerscField clone(header);
FieldMetaData clone(header);
std::string format(header.floating_point);
std::string data_type(header.data_type);
std::string format(header.floating_point);
std::string data_type(header.data_type);
#ifdef RNG_RANLUX
assert(format == std::string("UINT64"));
assert(data_type == std::string("RANLUX48"));
assert(format == std::string("UINT64"));
assert(data_type == std::string("RANLUX48"));
#endif
#ifdef RNG_MT19937
assert(format == std::string("UINT32"));
assert(data_type == std::string("MT19937"));
assert(format == std::string("UINT32"));
assert(data_type == std::string("MT19937"));
#endif
#ifdef RNG_SITMO
assert(format == std::string("UINT64"));
assert(data_type == std::string("SITMO"));
assert(format == std::string("UINT64"));
assert(data_type == std::string("SITMO"));
#endif
// depending on datatype, set up munger;
// munger is a function of <floating point, Real, data_type>
uint32_t csum=BinaryIO::readRNGSerial(serial,parallel,file,offset);
// depending on datatype, set up munger;
// munger is a function of <floating point, Real, data_type>
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::readRNG(serial,parallel,file,offset,nersc_csum,scidac_csuma,scidac_csumb);
assert(csum == header.checksum );
if ( nersc_csum != header.checksum ) {
std::cerr << "checksum mismatch "<<std::hex<< nersc_csum <<" "<<header.checksum<<std::dec<<std::endl;
exit(0);
}
assert(nersc_csum == header.checksum );
std::cout<<GridLogMessage <<"Read NERSC RNG file "<<file<< " format "<< data_type <<std::endl;
}
std::cout<<GridLogMessage <<"Read NERSC RNG file "<<file<< " format "<< data_type <<std::endl;
}
};
};
}}
}}
#endif

2
lib/perfmon/PerfCount.cc
View File

@ -40,7 +40,7 @@ const PerformanceCounter::PerformanceCounterConfig PerformanceCounter::Performan
{ PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES , "CPUCYCLES.........." , INSTRUCTIONS},
{ PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS , "INSTRUCTIONS......." , CPUCYCLES },
// 4
#ifdef AVX512
#ifdef KNL
{ PERF_TYPE_RAW, RawConfig(0x40,0x04), "ALL_LOADS..........", CPUCYCLES },
{ PERF_TYPE_RAW, RawConfig(0x01,0x04), "L1_MISS_LOADS......", L1D_READ_ACCESS },
{ PERF_TYPE_RAW, RawConfig(0x40,0x04), "ALL_LOADS..........", L1D_READ_ACCESS },

5
lib/perfmon/PerfCount.h
View File

@ -205,13 +205,14 @@ public:
void Stop(void) {
count=0;
cycles=0;
size_t ign;
#ifdef __linux__
ssize_t ign;
if ( fd!= -1) {
::ioctl(fd, PERF_EVENT_IOC_DISABLE, 0);
::ioctl(cyclefd, PERF_EVENT_IOC_DISABLE, 0);
ign=::read(fd, &count, sizeof(long long));
ign=::read(cyclefd, &cycles, sizeof(long long));
ign+=::read(cyclefd, &cycles, sizeof(long long));
assert(ign=2*sizeof(long long));
}
elapsed = cyclecount() - begin;
#else

124
lib/qcd/LatticeTheories.h Normal file
View File

@ -0,0 +1,124 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/QCD.h
Copyright (C) 2015
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: neo <cossu@post.kek.jp>
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_LT_H
#define GRID_LT_H
namespace Grid{
// First steps in the complete generalization of the Physics part
// Design not final
namespace LatticeTheories {
template <int Dimensions>
struct LatticeTheory {
static const int Nd = Dimensions;
static const int Nds = Dimensions * 2; // double stored field
template <typename vtype>
using iSinglet = iScalar<iScalar<iScalar<vtype> > >;
};
template <int Dimensions, int Colours>
struct LatticeGaugeTheory : public LatticeTheory<Dimensions> {
static const int Nds = Dimensions * 2;
static const int Nd = Dimensions;
static const int Nc = Colours;
template <typename vtype>
using iColourMatrix = iScalar<iScalar<iMatrix<vtype, Nc> > >;
template <typename vtype>
using iLorentzColourMatrix = iVector<iScalar<iMatrix<vtype, Nc> >, Nd>;
template <typename vtype>
using iDoubleStoredColourMatrix = iVector<iScalar<iMatrix<vtype, Nc> >, Nds>;
template <typename vtype>
using iColourVector = iScalar<iScalar<iVector<vtype, Nc> > >;
};
template <int Dimensions, int Colours, int Spin>
struct FermionicLatticeGaugeTheory
: public LatticeGaugeTheory<Dimensions, Colours> {
static const int Nd = Dimensions;
static const int Nds = Dimensions * 2;
static const int Nc = Colours;
static const int Ns = Spin;
template <typename vtype>
using iSpinMatrix = iScalar<iMatrix<iScalar<vtype>, Ns> >;
template <typename vtype>
using iSpinColourMatrix = iScalar<iMatrix<iMatrix<vtype, Nc>, Ns> >;
template <typename vtype>
using iSpinVector = iScalar<iVector<iScalar<vtype>, Ns> >;
template <typename vtype>
using iSpinColourVector = iScalar<iVector<iVector<vtype, Nc>, Ns> >;
// These 2 only if Spin is a multiple of 2
static const int Nhs = Spin / 2;
template <typename vtype>
using iHalfSpinVector = iScalar<iVector<iScalar<vtype>, Nhs> >;
template <typename vtype>
using iHalfSpinColourVector = iScalar<iVector<iVector<vtype, Nc>, Nhs> >;
//tests
typedef iColourMatrix<Complex> ColourMatrix;
typedef iColourMatrix<ComplexF> ColourMatrixF;
typedef iColourMatrix<ComplexD> ColourMatrixD;
};
// Examples, not complete now.
struct QCD : public FermionicLatticeGaugeTheory<4, 3, 4> {
static const int Xp = 0;
static const int Yp = 1;
static const int Zp = 2;
static const int Tp = 3;
static const int Xm = 4;
static const int Ym = 5;
static const int Zm = 6;
static const int Tm = 7;
typedef FermionicLatticeGaugeTheory FLGT;
typedef FLGT::iSpinMatrix<Complex > SpinMatrix;
typedef FLGT::iSpinMatrix<ComplexF > SpinMatrixF;
typedef FLGT::iSpinMatrix<ComplexD > SpinMatrixD;
};
struct QED : public FermionicLatticeGaugeTheory<4, 1, 4> {//fill
};
template <int Dimensions>
struct Scalar : public LatticeTheory<Dimensions> {};
}; // LatticeTheories
} // Grid
#endif

24
lib/qcd/QCD.h
View File

@ -32,14 +32,14 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#ifndef GRID_QCD_BASE_H
#define GRID_QCD_BASE_H
namespace Grid{
namespace QCD {
static const int Xdir = 0;
static const int Ydir = 1;
static const int Zdir = 2;
static const int Tdir = 3;
static const int Xp = 0;
static const int Yp = 1;
static const int Zp = 2;
@ -358,36 +358,36 @@ namespace QCD {
//////////////////////////////////////////////
template<class vobj>
void pokeColour(Lattice<vobj> &lhs,
const Lattice<decltype(peekIndex<ColourIndex>(lhs._odata[0],0))> & rhs,
int i)
const Lattice<decltype(peekIndex<ColourIndex>(lhs._odata[0],0))> & rhs,
int i)
{
PokeIndex<ColourIndex>(lhs,rhs,i);
}
template<class vobj>
void pokeColour(Lattice<vobj> &lhs,
const Lattice<decltype(peekIndex<ColourIndex>(lhs._odata[0],0,0))> & rhs,
int i,int j)
const Lattice<decltype(peekIndex<ColourIndex>(lhs._odata[0],0,0))> & rhs,
int i,int j)
{
PokeIndex<ColourIndex>(lhs,rhs,i,j);
}
template<class vobj>
void pokeSpin(Lattice<vobj> &lhs,
const Lattice<decltype(peekIndex<SpinIndex>(lhs._odata[0],0))> & rhs,
int i)
const Lattice<decltype(peekIndex<SpinIndex>(lhs._odata[0],0))> & rhs,
int i)
{
PokeIndex<SpinIndex>(lhs,rhs,i);
}
template<class vobj>
void pokeSpin(Lattice<vobj> &lhs,
const Lattice<decltype(peekIndex<SpinIndex>(lhs._odata[0],0,0))> & rhs,
int i,int j)
const Lattice<decltype(peekIndex<SpinIndex>(lhs._odata[0],0,0))> & rhs,
int i,int j)
{
PokeIndex<SpinIndex>(lhs,rhs,i,j);
}
template<class vobj>
void pokeLorentz(Lattice<vobj> &lhs,
const Lattice<decltype(peekIndex<LorentzIndex>(lhs._odata[0],0))> & rhs,
int i)
const Lattice<decltype(peekIndex<LorentzIndex>(lhs._odata[0],0))> & rhs,
int i)
{
PokeIndex<LorentzIndex>(lhs,rhs,i);
}

134
lib/qcd/action/ActionBase.h
View File

@ -4,10 +4,11 @@ Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/ActionBase.h
Copyright (C) 2015
Copyright (C) 2015-2016
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
Author: Guido Cossu <guido.cossu@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
@ -27,128 +28,29 @@ See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef QCD_ACTION_BASE
#define QCD_ACTION_BASE
#ifndef ACTION_BASE_H
#define ACTION_BASE_H
namespace Grid {
namespace QCD {
template <class GaugeField>
class Action {
template <class GaugeField >
class Action
{
public:
bool is_smeared = false;
// Boundary conditions? // Heatbath?
virtual void refresh(const GaugeField& U,
GridParallelRNG& pRNG) = 0; // refresh pseudofermions
virtual RealD S(const GaugeField& U) = 0; // evaluate the action
virtual void deriv(const GaugeField& U,
GaugeField& dSdU) = 0; // evaluate the action derivative
virtual ~Action(){};
// Heatbath?
virtual void refresh(const GaugeField& U, GridParallelRNG& pRNG) = 0; // refresh pseudofermions
virtual RealD S(const GaugeField& U) = 0; // evaluate the action
virtual void deriv(const GaugeField& U, GaugeField& dSdU) = 0; // evaluate the action derivative
virtual std::string action_name() = 0; // return the action name
virtual std::string LogParameters() = 0; // prints action parameters
virtual ~Action(){}
};
// Indexing of tuple types
template <class T, class Tuple>
struct Index;
template <class T, class... Types>
struct Index<T, std::tuple<T, Types...>> {
static const std::size_t value = 0;
};
template <class T, class U, class... Types>
struct Index<T, std::tuple<U, Types...>> {
static const std::size_t value = 1 + Index<T, std::tuple<Types...>>::value;
};
/*
template <class GaugeField>
struct ActionLevel {
public:
typedef Action<GaugeField>*
ActPtr; // now force the same colours as the rest of the code
//Add supported representations here
unsigned int multiplier;
std::vector<ActPtr> actions;
ActionLevel(unsigned int mul = 1) : actions(0), multiplier(mul) {
assert(mul >= 1);
};
void push_back(ActPtr ptr) { actions.push_back(ptr); }
};
*/
template <class GaugeField, class Repr = NoHirep >
struct ActionLevel {
public:
unsigned int multiplier;
// Fundamental repr actions separated because of the smearing
typedef Action<GaugeField>* ActPtr;
// construct a tuple of vectors of the actions for the corresponding higher
// representation fields
typedef typename AccessTypes<Action, Repr>::VectorCollection action_collection;
action_collection actions_hirep;
typedef typename AccessTypes<Action, Repr>::FieldTypeCollection action_hirep_types;
std::vector<ActPtr>& actions;
// Temporary conversion between ActionLevel and ActionLevelHirep
//ActionLevelHirep(ActionLevel<GaugeField>& AL ):actions(AL.actions), multiplier(AL.multiplier){}
ActionLevel(unsigned int mul = 1) : actions(std::get<0>(actions_hirep)), multiplier(mul) {
// initialize the hirep vectors to zero.
//apply(this->resize, actions_hirep, 0); //need a working resize
assert(mul >= 1);
};
//void push_back(ActPtr ptr) { actions.push_back(ptr); }
template < class Field >
void push_back(Action<Field>* ptr) {
// insert only in the correct vector
std::get< Index < Field, action_hirep_types>::value >(actions_hirep).push_back(ptr);
};
template < class ActPtr>
static void resize(ActPtr ap, unsigned int n){
ap->resize(n);
}
//template <std::size_t I>
//auto getRepresentation(Repr& R)->decltype(std::get<I>(R).U) {return std::get<I>(R).U;}
// Loop on tuple for a callable function
template <std::size_t I = 1, typename Callable, typename ...Args>
inline typename std::enable_if<I == std::tuple_size<action_collection>::value, void>::type apply(
Callable, Repr& R,Args&...) const {}
template <std::size_t I = 1, typename Callable, typename ...Args>
inline typename std::enable_if<I < std::tuple_size<action_collection>::value, void>::type apply(
Callable fn, Repr& R, Args&... arguments) const {
fn(std::get<I>(actions_hirep), std::get<I>(R.rep), arguments...);
apply<I + 1>(fn, R, arguments...);
}
};
//template <class GaugeField>
//using ActionSet = std::vector<ActionLevel<GaugeField> >;
template <class GaugeField, class R>
using ActionSet = std::vector<ActionLevel<GaugeField, R> >;
}
}
#endif
#endif // ACTION_BASE_H

16
lib/qcd/action/ActionCore.h
View File

@ -31,15 +31,31 @@ directory
#define QCD_ACTION_CORE
#include <Grid/qcd/action/ActionBase.h>
#include <Grid/qcd/action/ActionSet.h>
#include <Grid/qcd/action/ActionParams.h>
////////////////////////////////////////////
// Gauge Actions
////////////////////////////////////////////
#include <Grid/qcd/action/gauge/Gauge.h>
////////////////////////////////////////////
// Fermion prereqs
////////////////////////////////////////////
#include <Grid/qcd/action/fermion/FermionCore.h>
////////////////////////////////////////////
// Scalar Actions
////////////////////////////////////////////
#include <Grid/qcd/action/scalar/Scalar.h>
////////////////////////////////////////////
// Utility functions
////////////////////////////////////////////
#include <Grid/qcd/utils/Metric.h>
#include <Grid/qcd/utils/CovariantLaplacian.h>
#endif

113
lib/qcd/action/ActionParams.h
View File

@ -1,67 +1,92 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/ActionParams.h
Source file: ./lib/qcd/action/ActionParams.h
Copyright (C) 2015
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@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 free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_QCD_ACTION_PARAMS_H
#define GRID_QCD_ACTION_PARAMS_H
namespace Grid {
namespace QCD {
// These can move into a params header and be given MacroMagic serialisation
struct GparityWilsonImplParams {
bool overlapCommsCompute;
std::vector<int> twists;
GparityWilsonImplParams () : twists(Nd,0), overlapCommsCompute(false) {};
// These can move into a params header and be given MacroMagic serialisation
struct GparityWilsonImplParams {
bool overlapCommsCompute;
std::vector<int> twists;
GparityWilsonImplParams() : twists(Nd, 0), overlapCommsCompute(false){};
};
struct WilsonImplParams {
bool overlapCommsCompute;
std::vector<Complex> boundary_phases;
WilsonImplParams() : overlapCommsCompute(false) {
boundary_phases.resize(Nd, 1.0);
};
WilsonImplParams(const std::vector<Complex> phi)
: boundary_phases(phi), overlapCommsCompute(false) {}
};
struct WilsonImplParams {
bool overlapCommsCompute;
WilsonImplParams() : overlapCommsCompute(false) {};
};
struct StaggeredImplParams {
StaggeredImplParams() {};
};
struct OneFlavourRationalParams : Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(OneFlavourRationalParams,
RealD, lo,
RealD, hi,
int, MaxIter,
RealD, tolerance,
int, degree,
int, precision);
// MaxIter and tolerance, vectors??
// constructor
OneFlavourRationalParams( RealD _lo = 0.0,
RealD _hi = 1.0,
int _maxit = 1000,
RealD tol = 1.0e-8,
int _degree = 10,
int _precision = 64)
: lo(_lo),
hi(_hi),
MaxIter(_maxit),
tolerance(tol),
degree(_degree),
precision(_precision){};
};
}
}
struct StaggeredImplParams {
StaggeredImplParams() {};
};
struct OneFlavourRationalParams {
RealD lo;
RealD hi;
int MaxIter; // Vector?
RealD tolerance; // Vector?
int degree=10;
int precision=64;
OneFlavourRationalParams (RealD _lo,RealD _hi,int _maxit,RealD tol=1.0e-8,int _degree = 10,int _precision=64) :
lo(_lo), hi(_hi), MaxIter(_maxit), tolerance(tol), degree(_degree), precision(_precision)
{};
};
}}
#endif

116
lib/qcd/action/ActionSet.h Normal file
View File

@ -0,0 +1,116 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/ActionSet.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 ACTION_SET_H
#define ACTION_SET_H
namespace Grid {
// Should drop this namespace here
namespace QCD {
//////////////////////////////////
// Indexing of tuple types
//////////////////////////////////
template <class T, class Tuple>
struct Index;
template <class T, class... Types>
struct Index<T, std::tuple<T, Types...>> {
static const std::size_t value = 0;
};
template <class T, class U, class... Types>
struct Index<T, std::tuple<U, Types...>> {
static const std::size_t value = 1 + Index<T, std::tuple<Types...>>::value;
};
////////////////////////////////////////////
// Action Level
// Action collection
// in a integration level
// (for multilevel integration schemes)
////////////////////////////////////////////
template <class Field, class Repr = NoHirep >
struct ActionLevel {
public:
unsigned int multiplier;
// Fundamental repr actions separated because of the smearing
typedef Action<Field>* ActPtr;
// construct a tuple of vectors of the actions for the corresponding higher
// representation fields
typedef typename AccessTypes<Action, Repr>::VectorCollection action_collection;
typedef typename AccessTypes<Action, Repr>::FieldTypeCollection action_hirep_types;
action_collection actions_hirep;
std::vector<ActPtr>& actions;
explicit ActionLevel(unsigned int mul = 1) :
actions(std::get<0>(actions_hirep)), multiplier(mul) {
// initialize the hirep vectors to zero.
// apply(this->resize, actions_hirep, 0); //need a working resize
assert(mul >= 1);
}
template < class GenField >
void push_back(Action<GenField>* ptr) {
// insert only in the correct vector
std::get< Index < GenField, action_hirep_types>::value >(actions_hirep).push_back(ptr);
};
template <class ActPtr>
static void resize(ActPtr ap, unsigned int n) {
ap->resize(n);
}
// Loop on tuple for a callable function
template <std::size_t I = 1, typename Callable, typename ...Args>
inline typename std::enable_if<I == std::tuple_size<action_collection>::value, void>::type apply(Callable, Repr& R,Args&...) const {}
template <std::size_t I = 1, typename Callable, typename ...Args>
inline typename std::enable_if<I < std::tuple_size<action_collection>::value, void>::type apply(Callable fn, Repr& R, Args&... arguments) const {
fn(std::get<I>(actions_hirep), std::get<I>(R.rep), arguments...);
apply<I + 1>(fn, R, arguments...);
}
};
// Define the ActionSet
template <class GaugeField, class R>
using ActionSet = std::vector<ActionLevel<GaugeField, R> >;
} // QCD
} // Grid
#endif // ACTION_SET_H

29
lib/qcd/action/fermion/CayleyFermion5D.cc
View File

@ -29,7 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
*************************************************************************************/
/* END LEGAL */
#include <Grid/Eigen/Dense>
#include <Grid/Grid_Eigen_Dense.h>
#include <Grid/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
@ -320,7 +320,7 @@ void CayleyFermion5D<Impl>::MDeriv (GaugeField &mat,const FermionField &U,const
this->DhopDeriv(mat,U,Din,dag);
} else {
// U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
MeooeDag5D(U,Din);
Meooe5D(U,Din);
this->DhopDeriv(mat,Din,V,dag);
}
};
@ -335,8 +335,8 @@ void CayleyFermion5D<Impl>::MoeDeriv(GaugeField &mat,const FermionField &U,const
this->DhopDerivOE(mat,U,Din,dag);
} else {
// U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
MeooeDag5D(U,Din);
this->DhopDerivOE(mat,Din,V,dag);
Meooe5D(U,Din);
this->DhopDerivOE(mat,Din,V,dag);
}
};
template<class Impl>
@ -350,7 +350,7 @@ void CayleyFermion5D<Impl>::MeoDeriv(GaugeField &mat,const FermionField &U,const
this->DhopDerivEO(mat,U,Din,dag);
} else {
// U d/du [D_w D5]^dag V = U D5^dag d/du DW^dag Y // implicit adj on U in call
MeooeDag5D(U,Din);
Meooe5D(U,Din);
this->DhopDerivEO(mat,Din,V,dag);
}
};
@ -380,6 +380,8 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
///////////////////////////////////////////////////////////
// The Cayley coeffs (unprec)
///////////////////////////////////////////////////////////
assert(gamma.size()==Ls);
omega.resize(Ls);
bs.resize(Ls);
cs.resize(Ls);
@ -412,10 +414,11 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
for(int i=0; i < Ls; i++){
as[i] = 1.0;
omega[i] = gamma[i]*zolo_hi; //NB reciprocal relative to Chroma NEF code
// assert(fabs(omega[i])>0.0);
bs[i] = 0.5*(bpc/omega[i] + bmc);
cs[i] = 0.5*(bpc/omega[i] - bmc);
}
////////////////////////////////////////////////////////
// Constants for the preconditioned matrix Cayley form
////////////////////////////////////////////////////////
@ -425,12 +428,12 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
ceo.resize(Ls);
for(int i=0;i<Ls;i++){
bee[i]=as[i]*(bs[i]*(4.0-this->M5) +1.0);
bee[i]=as[i]*(bs[i]*(4.0-this->M5) +1.0);
// assert(fabs(bee[i])>0.0);
cee[i]=as[i]*(1.0-cs[i]*(4.0-this->M5));
beo[i]=as[i]*bs[i];
ceo[i]=-as[i]*cs[i];
}
aee.resize(Ls);
aeo.resize(Ls);
for(int i=0;i<Ls;i++){
@ -474,14 +477,16 @@ void CayleyFermion5D<Impl>::SetCoefficientsInternal(RealD zolo_hi,std::vector<Co
{
Coeff_t delta_d=mass*cee[Ls-1];
for(int j=0;j<Ls-1;j++) delta_d *= cee[j]/bee[j];
for(int j=0;j<Ls-1;j++) {
// assert(fabs(bee[j])>0.0);
delta_d *= cee[j]/bee[j];
}
dee[Ls-1] += delta_d;
}
int inv=1;
this->MooeeInternalCompute(0,inv,MatpInv,MatmInv);
this->MooeeInternalCompute(1,inv,MatpInvDag,MatmInvDag);
}
@ -495,7 +500,9 @@ void CayleyFermion5D<Impl>::MooeeInternalCompute(int dag, int inv,
GridBase *grid = this->FermionRedBlackGrid();
int LLs = grid->_rdimensions[0];
if ( LLs == Ls ) return; // Not vectorised in 5th direction
if ( LLs == Ls ) {
return; // Not vectorised in 5th direction
}
Eigen::MatrixXcd Pplus = Eigen::MatrixXcd::Zero(Ls,Ls);
Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls);

7
lib/qcd/action/fermion/CayleyFermion5Dcache.cc
View File

@ -237,6 +237,13 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
INSTANTIATE_DPERP(GparityWilsonImplD);
INSTANTIATE_DPERP(ZWilsonImplF);
INSTANTIATE_DPERP(ZWilsonImplD);
INSTANTIATE_DPERP(WilsonImplFH);
INSTANTIATE_DPERP(WilsonImplDF);
INSTANTIATE_DPERP(GparityWilsonImplFH);
INSTANTIATE_DPERP(GparityWilsonImplDF);
INSTANTIATE_DPERP(ZWilsonImplFH);
INSTANTIATE_DPERP(ZWilsonImplDF);
#endif
}}

16
lib/qcd/action/fermion/CayleyFermion5Ddense.cc
View File

@ -29,7 +29,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
*************************************************************************************/
/* END LEGAL */
#include <Grid/Eigen/Dense>
#include <Grid/Grid_Eigen_Dense.h>
#include <Grid/qcd/action/fermion/FermionCore.h>
#include <Grid/qcd/action/fermion/CayleyFermion5D.h>
@ -137,6 +137,20 @@ template void CayleyFermion5D<WilsonImplF>::MooeeInternal(const FermionField &ps
template void CayleyFermion5D<WilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZWilsonImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZWilsonImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
INSTANTIATE_DPERP(GparityWilsonImplFH);
INSTANTIATE_DPERP(GparityWilsonImplDF);
INSTANTIATE_DPERP(WilsonImplFH);
INSTANTIATE_DPERP(WilsonImplDF);
INSTANTIATE_DPERP(ZWilsonImplFH);
INSTANTIATE_DPERP(ZWilsonImplDF);
template void CayleyFermion5D<GparityWilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<GparityWilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<WilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<WilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZWilsonImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZWilsonImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
#endif
}}

8
lib/qcd/action/fermion/CayleyFermion5Dssp.cc
View File

@ -37,7 +37,6 @@ namespace Grid {
namespace QCD {
// FIXME -- make a version of these routines with site loop outermost for cache reuse.
// Pminus fowards
// Pplus backwards
template<class Impl>
@ -152,6 +151,13 @@ void CayleyFermion5D<Impl>::MooeeInvDag (const FermionField &psi, FermionField &
INSTANTIATE_DPERP(GparityWilsonImplD);
INSTANTIATE_DPERP(ZWilsonImplF);
INSTANTIATE_DPERP(ZWilsonImplD);
INSTANTIATE_DPERP(WilsonImplFH);
INSTANTIATE_DPERP(WilsonImplDF);
INSTANTIATE_DPERP(GparityWilsonImplFH);
INSTANTIATE_DPERP(GparityWilsonImplDF);
INSTANTIATE_DPERP(ZWilsonImplFH);
INSTANTIATE_DPERP(ZWilsonImplDF);
#endif
}

11
lib/qcd/action/fermion/CayleyFermion5Dvec.cc
View File

@ -808,10 +808,21 @@ INSTANTIATE_DPERP(DomainWallVec5dImplF);
INSTANTIATE_DPERP(ZDomainWallVec5dImplD);
INSTANTIATE_DPERP(ZDomainWallVec5dImplF);
INSTANTIATE_DPERP(DomainWallVec5dImplDF);
INSTANTIATE_DPERP(DomainWallVec5dImplFH);
INSTANTIATE_DPERP(ZDomainWallVec5dImplDF);
INSTANTIATE_DPERP(ZDomainWallVec5dImplFH);
template void CayleyFermion5D<DomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<DomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZDomainWallVec5dImplF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZDomainWallVec5dImplD>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<DomainWallVec5dImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<DomainWallVec5dImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZDomainWallVec5dImplFH>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
template void CayleyFermion5D<ZDomainWallVec5dImplDF>::MooeeInternal(const FermionField &psi, FermionField &chi,int dag, int inv);
}}

2
lib/qcd/action/fermion/DomainWallFermion.h
View File

@ -68,7 +68,7 @@ namespace Grid {
Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls);// eps is ignored for higham
assert(zdata->n==this->Ls);
// std::cout<<GridLogMessage << "DomainWallFermion with Ls="<<this->Ls<<std::endl;
std::cout<<GridLogMessage << "DomainWallFermion with Ls="<<this->Ls<<std::endl;
// Call base setter
this->SetCoefficientsTanh(zdata,1.0,0.0);

54
lib/qcd/action/fermion/Fermion.h
View File

@ -1,4 +1,4 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@ -91,6 +91,10 @@ typedef WilsonFermion<WilsonImplR> WilsonFermionR;
typedef WilsonFermion<WilsonImplF> WilsonFermionF;
typedef WilsonFermion<WilsonImplD> WilsonFermionD;
typedef WilsonFermion<WilsonImplRL> WilsonFermionRL;
typedef WilsonFermion<WilsonImplFH> WilsonFermionFH;
typedef WilsonFermion<WilsonImplDF> WilsonFermionDF;
typedef WilsonFermion<WilsonAdjImplR> WilsonAdjFermionR;
typedef WilsonFermion<WilsonAdjImplF> WilsonAdjFermionF;
typedef WilsonFermion<WilsonAdjImplD> WilsonAdjFermionD;
@ -113,27 +117,50 @@ typedef DomainWallFermion<WilsonImplR> DomainWallFermionR;
typedef DomainWallFermion<WilsonImplF> DomainWallFermionF;
typedef DomainWallFermion<WilsonImplD> DomainWallFermionD;
typedef DomainWallFermion<WilsonImplRL> DomainWallFermionRL;
typedef DomainWallFermion<WilsonImplFH> DomainWallFermionFH;
typedef DomainWallFermion<WilsonImplDF> DomainWallFermionDF;
typedef MobiusFermion<WilsonImplR> MobiusFermionR;
typedef MobiusFermion<WilsonImplF> MobiusFermionF;
typedef MobiusFermion<WilsonImplD> MobiusFermionD;
typedef MobiusFermion<WilsonImplRL> MobiusFermionRL;
typedef MobiusFermion<WilsonImplFH> MobiusFermionFH;
typedef MobiusFermion<WilsonImplDF> MobiusFermionDF;
typedef ZMobiusFermion<ZWilsonImplR> ZMobiusFermionR;
typedef ZMobiusFermion<ZWilsonImplF> ZMobiusFermionF;
typedef ZMobiusFermion<ZWilsonImplD> ZMobiusFermionD;
typedef ZMobiusFermion<ZWilsonImplRL> ZMobiusFermionRL;
typedef ZMobiusFermion<ZWilsonImplFH> ZMobiusFermionFH;
typedef ZMobiusFermion<ZWilsonImplDF> ZMobiusFermionDF;
// Ls vectorised
typedef DomainWallFermion<DomainWallVec5dImplR> DomainWallFermionVec5dR;
typedef DomainWallFermion<DomainWallVec5dImplF> DomainWallFermionVec5dF;
typedef DomainWallFermion<DomainWallVec5dImplD> DomainWallFermionVec5dD;
typedef DomainWallFermion<DomainWallVec5dImplRL> DomainWallFermionVec5dRL;
typedef DomainWallFermion<DomainWallVec5dImplFH> DomainWallFermionVec5dFH;
typedef DomainWallFermion<DomainWallVec5dImplDF> DomainWallFermionVec5dDF;
typedef MobiusFermion<DomainWallVec5dImplR> MobiusFermionVec5dR;
typedef MobiusFermion<DomainWallVec5dImplF> MobiusFermionVec5dF;
typedef MobiusFermion<DomainWallVec5dImplD> MobiusFermionVec5dD;
typedef MobiusFermion<DomainWallVec5dImplRL> MobiusFermionVec5dRL;
typedef MobiusFermion<DomainWallVec5dImplFH> MobiusFermionVec5dFH;
typedef MobiusFermion<DomainWallVec5dImplDF> MobiusFermionVec5dDF;
typedef ZMobiusFermion<ZDomainWallVec5dImplR> ZMobiusFermionVec5dR;
typedef ZMobiusFermion<ZDomainWallVec5dImplF> ZMobiusFermionVec5dF;
typedef ZMobiusFermion<ZDomainWallVec5dImplD> ZMobiusFermionVec5dD;
typedef ZMobiusFermion<ZDomainWallVec5dImplRL> ZMobiusFermionVec5dRL;
typedef ZMobiusFermion<ZDomainWallVec5dImplFH> ZMobiusFermionVec5dFH;
typedef ZMobiusFermion<ZDomainWallVec5dImplDF> ZMobiusFermionVec5dDF;
typedef ScaledShamirFermion<WilsonImplR> ScaledShamirFermionR;
typedef ScaledShamirFermion<WilsonImplF> ScaledShamirFermionF;
@ -174,17 +201,35 @@ typedef OverlapWilsonPartialFractionZolotarevFermion<WilsonImplD> OverlapWilsonP
typedef WilsonFermion<GparityWilsonImplR> GparityWilsonFermionR;
typedef WilsonFermion<GparityWilsonImplF> GparityWilsonFermionF;
typedef WilsonFermion<GparityWilsonImplD> GparityWilsonFermionD;
typedef WilsonFermion<GparityWilsonImplRL> GparityWilsonFermionRL;
typedef WilsonFermion<GparityWilsonImplFH> GparityWilsonFermionFH;
typedef WilsonFermion<GparityWilsonImplDF> GparityWilsonFermionDF;
typedef DomainWallFermion<GparityWilsonImplR> GparityDomainWallFermionR;
typedef DomainWallFermion<GparityWilsonImplF> GparityDomainWallFermionF;
typedef DomainWallFermion<GparityWilsonImplD> GparityDomainWallFermionD;
typedef DomainWallFermion<GparityWilsonImplRL> GparityDomainWallFermionRL;
typedef DomainWallFermion<GparityWilsonImplFH> GparityDomainWallFermionFH;
typedef DomainWallFermion<GparityWilsonImplDF> GparityDomainWallFermionDF;
typedef WilsonTMFermion<GparityWilsonImplR> GparityWilsonTMFermionR;
typedef WilsonTMFermion<GparityWilsonImplF> GparityWilsonTMFermionF;
typedef WilsonTMFermion<GparityWilsonImplD> GparityWilsonTMFermionD;
typedef WilsonTMFermion<GparityWilsonImplRL> GparityWilsonTMFermionRL;
typedef WilsonTMFermion<GparityWilsonImplFH> GparityWilsonTMFermionFH;
typedef WilsonTMFermion<GparityWilsonImplDF> GparityWilsonTMFermionDF;
typedef MobiusFermion<GparityWilsonImplR> GparityMobiusFermionR;
typedef MobiusFermion<GparityWilsonImplF> GparityMobiusFermionF;
typedef MobiusFermion<GparityWilsonImplD> GparityMobiusFermionD;
typedef MobiusFermion<GparityWilsonImplRL> GparityMobiusFermionRL;
typedef MobiusFermion<GparityWilsonImplFH> GparityMobiusFermionFH;
typedef MobiusFermion<GparityWilsonImplDF> GparityMobiusFermionDF;
typedef ImprovedStaggeredFermion<StaggeredImplR> ImprovedStaggeredFermionR;
typedef ImprovedStaggeredFermion<StaggeredImplF> ImprovedStaggeredFermionF;
typedef ImprovedStaggeredFermion<StaggeredImplD> ImprovedStaggeredFermionD;
@ -200,4 +245,11 @@ typedef ImprovedStaggeredFermion5D<StaggeredVec5dImplD> ImprovedStaggeredFermion
}}
////////////////////
// Scalar QED actions
// TODO: this needs to move to another header after rename to Fermion.h
////////////////////
#include <Grid/qcd/action/scalar/Scalar.h>
#include <Grid/qcd/action/gauge/Photon.h>
#endif

15
lib/qcd/action/fermion/FermionCore.h
View File

@ -55,7 +55,14 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
template class A<ZWilsonImplF>; \
template class A<ZWilsonImplD>; \
template class A<GparityWilsonImplF>; \
template class A<GparityWilsonImplD>;
template class A<GparityWilsonImplD>; \
template class A<WilsonImplFH>; \
template class A<WilsonImplDF>; \
template class A<ZWilsonImplFH>; \
template class A<ZWilsonImplDF>; \
template class A<GparityWilsonImplFH>; \
template class A<GparityWilsonImplDF>;
#define AdjointFermOpTemplateInstantiate(A) \
template class A<WilsonAdjImplF>; \
@ -69,7 +76,11 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
template class A<DomainWallVec5dImplF>; \
template class A<DomainWallVec5dImplD>; \
template class A<ZDomainWallVec5dImplF>; \
template class A<ZDomainWallVec5dImplD>;
template class A<ZDomainWallVec5dImplD>; \
template class A<DomainWallVec5dImplFH>; \
template class A<DomainWallVec5dImplDF>; \
template class A<ZDomainWallVec5dImplFH>; \
template class A<ZDomainWallVec5dImplDF>;
#define FermOpTemplateInstantiate(A) \
FermOp4dVecTemplateInstantiate(A) \

413
lib/qcd/action/fermion/FermionOperatorImpl.h
View File

@ -35,7 +35,6 @@ directory
namespace Grid {
namespace QCD {
//////////////////////////////////////////////
// Template parameter class constructs to package
// externally control Fermion implementations
@ -44,7 +43,7 @@ namespace QCD {
// Ultimately need Impl to always define types where XXX is opaque
//
// typedef typename XXX Simd;
// typedef typename XXX GaugeLinkField;
// typedef typename XXX GaugeLinkField;
// typedef typename XXX GaugeField;
// typedef typename XXX GaugeActField;
// typedef typename XXX FermionField;
@ -89,7 +88,53 @@ namespace QCD {
//
// }
//////////////////////////////////////////////
template <class T> struct SamePrecisionMapper {
typedef T HigherPrecVector ;
typedef T LowerPrecVector ;
};
template <class T> struct LowerPrecisionMapper { };
template <> struct LowerPrecisionMapper<vRealF> {
typedef vRealF HigherPrecVector ;
typedef vRealH LowerPrecVector ;
};
template <> struct LowerPrecisionMapper<vRealD> {
typedef vRealD HigherPrecVector ;
typedef vRealF LowerPrecVector ;
};
template <> struct LowerPrecisionMapper<vComplexF> {
typedef vComplexF HigherPrecVector ;
typedef vComplexH LowerPrecVector ;
};
template <> struct LowerPrecisionMapper<vComplexD> {
typedef vComplexD HigherPrecVector ;
typedef vComplexF LowerPrecVector ;
};
struct CoeffReal {
public:
typedef RealD _Coeff_t;
static const int Nhcs = 2;
template<class Simd> using PrecisionMapper = SamePrecisionMapper<Simd>;
};
struct CoeffRealHalfComms {
public:
typedef RealD _Coeff_t;
static const int Nhcs = 1;
template<class Simd> using PrecisionMapper = LowerPrecisionMapper<Simd>;
};
struct CoeffComplex {
public:
typedef ComplexD _Coeff_t;
static const int Nhcs = 2;
template<class Simd> using PrecisionMapper = SamePrecisionMapper<Simd>;
};
struct CoeffComplexHalfComms {
public:
typedef ComplexD _Coeff_t;
static const int Nhcs = 1;
template<class Simd> using PrecisionMapper = LowerPrecisionMapper<Simd>;
};
////////////////////////////////////////////////////////////////////////
// Implementation dependent fermion types
@ -108,58 +153,63 @@ namespace QCD {
typedef typename Impl::Coeff_t Coeff_t; \
#define INHERIT_IMPL_TYPES(Base) \
INHERIT_GIMPL_TYPES(Base) \
INHERIT_GIMPL_TYPES(Base) \
INHERIT_FIMPL_TYPES(Base)
/////////////////////////////////////////////////////////////////////////////
// Single flavour four spinors with colour index
/////////////////////////////////////////////////////////////////////////////
template <class S, class Representation = FundamentalRepresentation,class _Coeff_t = RealD >
template <class S, class Representation = FundamentalRepresentation,class Options = CoeffReal >
class WilsonImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
public:
static const int Dimension = Representation::Dimension;
static const bool LsVectorised=false;
static const int Nhcs = Options::Nhcs;
typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
//Necessary?
constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
const bool LsVectorised=false;
typedef _Coeff_t Coeff_t;
INHERIT_GIMPL_TYPES(Gimpl);
typedef typename Options::_Coeff_t Coeff_t;
typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
template <typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Dimension>, Ns> >;
template <typename vtype> using iImplPropagator = iScalar<iMatrix<iMatrix<vtype, Dimension>, Ns> >;
template <typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Dimension>, Nhs> >;
template <typename vtype> using iImplHalfCommSpinor = iScalar<iVector<iVector<vtype, Dimension>, Nhcs> >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplHalfCommSpinor<SimdL> SiteHalfCommSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonImplParams ImplParams;
typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
ImplParams Params;
WilsonImpl(const ImplParams &p = ImplParams()) : Params(p){};
WilsonImpl(const ImplParams &p = ImplParams()) : Params(p){
assert(Params.boundary_phases.size() == Nd);
};
bool overlapCommsCompute(void) { return Params.overlapCommsCompute; };
inline void multLink(SiteHalfSpinor &phi,
const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi,
int mu,
StencilEntry *SE,
StencilImpl &St) {
const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi,
int mu,
StencilEntry *SE,
StencilImpl &St) {
mult(&phi(), &U(mu), &chi());
}
@ -169,16 +219,34 @@ namespace QCD {
}
inline void DoubleStore(GridBase *GaugeGrid,
DoubledGaugeField &Uds,
const GaugeField &Umu) {
DoubledGaugeField &Uds,
const GaugeField &Umu)
{
typedef typename Simd::scalar_type scalar_type;
conformable(Uds._grid, GaugeGrid);
conformable(Umu._grid, GaugeGrid);
GaugeLinkField U(GaugeGrid);
GaugeLinkField tmp(GaugeGrid);
Lattice<iScalar<vInteger> > coor(GaugeGrid);
for (int mu = 0; mu < Nd; mu++) {
U = PeekIndex<LorentzIndex>(Umu, mu);
PokeIndex<LorentzIndex>(Uds, U, mu);
U = adj(Cshift(U, mu, -1));
PokeIndex<LorentzIndex>(Uds, U, mu + 4);
auto pha = Params.boundary_phases[mu];
scalar_type phase( real(pha),imag(pha) );
int Lmu = GaugeGrid->GlobalDimensions()[mu] - 1;
LatticeCoordinate(coor, mu);
U = PeekIndex<LorentzIndex>(Umu, mu);
tmp = where(coor == Lmu, phase * U, U);
PokeIndex<LorentzIndex>(Uds, tmp, mu);
U = adj(Cshift(U, mu, -1));
U = where(coor == 0, conjugate(phase) * U, U);
PokeIndex<LorentzIndex>(Uds, U, mu + 4);
}
}
@ -195,11 +263,11 @@ namespace QCD {
tmp = zero;
parallel_for(int sss=0;sss<tmp._grid->oSites();sss++){
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp[sU] = tmp[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde[sF],Atilde[sF])); // ordering here
}
int sU=sss;
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
tmp[sU] = tmp[sU]+ traceIndex<SpinIndex>(outerProduct(Btilde[sF],Atilde[sF])); // ordering here
}
}
PokeIndex<LorentzIndex>(mat,tmp,mu);
@ -209,31 +277,34 @@ namespace QCD {
////////////////////////////////////////////////////////////////////////////////////
// Single flavour four spinors with colour index, 5d redblack
////////////////////////////////////////////////////////////////////////////////////
template<class S,int Nrepresentation=Nc,class _Coeff_t = RealD>
template<class S,int Nrepresentation=Nc, class Options=CoeffReal>
class DomainWallVec5dImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepresentation> > {
public:
static const int Dimension = Nrepresentation;
const bool LsVectorised=true;
typedef _Coeff_t Coeff_t;
typedef PeriodicGaugeImpl<GaugeImplTypes<S, Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
static const int Dimension = Nrepresentation;
static const bool LsVectorised=true;
static const int Nhcs = Options::Nhcs;
typedef typename Options::_Coeff_t Coeff_t;
typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
template <typename vtype> using iImplSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Ns> >;
template <typename vtype> using iImplPropagator = iScalar<iMatrix<iMatrix<vtype, Nrepresentation>, Ns> >;
template <typename vtype> using iImplHalfSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhs> >;
template <typename vtype> using iImplHalfCommSpinor = iScalar<iVector<iVector<vtype, Nrepresentation>, Nhcs> >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>;
template <typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd>;
template <typename vtype> using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplHalfCommSpinor<SimdL> SiteHalfCommSpinor;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
/////////////////////////////////////////////////
// Make the doubled gauge field a *scalar*
@ -241,9 +312,9 @@ class DomainWallVec5dImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
typedef iImplDoubledGaugeField<typename Simd::scalar_type> SiteDoubledGaugeField; // This is a scalar
typedef iImplGaugeField<typename Simd::scalar_type> SiteScalarGaugeField; // scalar
typedef iImplGaugeLink<typename Simd::scalar_type> SiteScalarGaugeLink; // scalar
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonImplParams ImplParams;
typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
@ -259,12 +330,12 @@ class DomainWallVec5dImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
}
inline void multLink(SiteHalfSpinor &phi, const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi, int mu, StencilEntry *SE,
StencilImpl &St) {
const SiteHalfSpinor &chi, int mu, StencilEntry *SE,
StencilImpl &St) {
SiteGaugeLink UU;
for (int i = 0; i < Nrepresentation; i++) {
for (int j = 0; j < Nrepresentation; j++) {
vsplat(UU()()(i, j), U(mu)()(i, j));
vsplat(UU()()(i, j), U(mu)()(i, j));
}
}
mult(&phi(), &UU(), &chi());
@ -301,45 +372,90 @@ class DomainWallVec5dImpl : public PeriodicGaugeImpl< GaugeImplTypes< S,Nrepres
{
assert(0);
}
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde,FermionField &Atilde, int mu)
{
assert(0);
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {
assert(0);
// Following lines to be revised after Peter's addition of half prec
// missing put lane...
/*
typedef decltype(traceIndex<SpinIndex>(outerProduct(Btilde[0], Atilde[0]))) result_type;
unsigned int LLs = Btilde._grid->_rdimensions[0];
conformable(Atilde._grid,Btilde._grid);
GridBase* grid = mat._grid;
GridBase* Bgrid = Btilde._grid;
unsigned int dimU = grid->Nd();
unsigned int dimF = Bgrid->Nd();
GaugeLinkField tmp(grid);
tmp = zero;
// FIXME
// Current implementation works, thread safe, probably suboptimal
// Passing through the local coordinate for grid transformation
// the force grid is in general very different from the Ls vectorized grid
PARALLEL_FOR_LOOP
for (int so = 0; so < grid->oSites(); so++) {
std::vector<typename result_type::scalar_object> vres(Bgrid->Nsimd());
std::vector<int> ocoor; grid->oCoorFromOindex(ocoor,so);
for (int si = 0; si < tmp._grid->iSites(); si++){
typename result_type::scalar_object scalar_object; scalar_object = zero;
std::vector<int> local_coor;
std::vector<int> icoor; grid->iCoorFromIindex(icoor,si);
grid->InOutCoorToLocalCoor(ocoor, icoor, local_coor);
for (int s = 0; s < LLs; s++) {
std::vector<int> slocal_coor(dimF);
slocal_coor[0] = s;
for (int s4d = 1; s4d< dimF; s4d++) slocal_coor[s4d] = local_coor[s4d-1];
int sF = Bgrid->oIndexReduced(slocal_coor);
assert(sF < Bgrid->oSites());
extract(traceIndex<SpinIndex>(outerProduct(Btilde[sF], Atilde[sF])), vres);
// sum across the 5d dimension
for (auto v : vres) scalar_object += v;
}
tmp._odata[so].putlane(scalar_object, si);
}
}
PokeIndex<LorentzIndex>(mat, tmp, mu);
*/
}
};
////////////////////////////////////////////////////////////////////////////////////////
// Flavour doubled spinors; is Gparity the only? what about C*?
////////////////////////////////////////////////////////////////////////////////////////
template <class S, int Nrepresentation,class _Coeff_t = RealD>
template <class S, int Nrepresentation, class Options=CoeffReal>
class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresentation> > {
public:
static const int Dimension = Nrepresentation;
static const int Nhcs = Options::Nhcs;
static const bool LsVectorised=false;
const bool LsVectorised=false;
typedef _Coeff_t Coeff_t;
typedef ConjugateGaugeImpl< GaugeImplTypes<S,Nrepresentation> > Gimpl;
INHERIT_GIMPL_TYPES(Gimpl);
typedef typename Options::_Coeff_t Coeff_t;
typedef typename Options::template PrecisionMapper<Simd>::LowerPrecVector SimdL;
template <typename vtype> using iImplSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Ns>, Ngp>;
template <typename vtype> using iImplPropagator = iVector<iMatrix<iMatrix<vtype, Nrepresentation>, Ns>, Ngp >;
template <typename vtype> using iImplHalfSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>, Ngp>;
template <typename vtype> using iImplSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Ns>, Ngp>;
template <typename vtype> using iImplPropagator = iVector<iMatrix<iMatrix<vtype, Nrepresentation>, Ns>, Ngp>;
template <typename vtype> using iImplHalfSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Nhs>, Ngp>;
template <typename vtype> using iImplHalfCommSpinor = iVector<iVector<iVector<vtype, Nrepresentation>, Nhcs>, Ngp>;
template <typename vtype> using iImplDoubledGaugeField = iVector<iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nds>, Ngp>;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplPropagator<Simd> SitePropagator;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplHalfCommSpinor<SimdL> SiteHalfCommSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SitePropagator> PropagatorField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef WilsonCompressor<SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonCompressor<SiteHalfCommSpinor,SiteHalfSpinor, SiteSpinor> Compressor;
typedef WilsonStencil<SiteSpinor, SiteHalfSpinor> StencilImpl;
typedef GparityWilsonImplParams ImplParams;
@ -353,19 +469,19 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
// provide the multiply by link that is differentiated between Gparity (with
// flavour index) and non-Gparity
inline void multLink(SiteHalfSpinor &phi, const SiteDoubledGaugeField &U,
const SiteHalfSpinor &chi, int mu, StencilEntry *SE,
StencilImpl &St) {
const SiteHalfSpinor &chi, int mu, StencilEntry *SE,
StencilImpl &St) {
typedef SiteHalfSpinor vobj;
typedef typename SiteHalfSpinor::scalar_object sobj;
typedef SiteHalfSpinor vobj;
typedef typename SiteHalfSpinor::scalar_object sobj;
vobj vtmp;
sobj stmp;
GridBase *grid = St._grid;
const int Nsimd = grid->Nsimd();
int direction = St._directions[mu];
int distance = St._distances[mu];
int ptype = St._permute_type[mu];
@ -373,13 +489,13 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
// Fixme X.Y.Z.T hardcode in stencil
int mmu = mu % Nd;
// assert our assumptions
assert((distance == 1) || (distance == -1)); // nearest neighbour stencil hard code
assert((sl == 1) || (sl == 2));
std::vector<int> icoor;
if ( SE->_around_the_world && Params.twists[mmu] ) {
if ( sl == 2 ) {
@ -389,25 +505,25 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
extract(chi,vals);
for(int s=0;s<Nsimd;s++){
grid->iCoorFromIindex(icoor,s);
assert((icoor[direction]==0)||(icoor[direction]==1));
int permute_lane;
if ( distance == 1) {
permute_lane = icoor[direction]?1:0;
} else {
permute_lane = icoor[direction]?0:1;
}
if ( permute_lane ) {
stmp(0) = vals[s](1);
stmp(1) = vals[s](0);
vals[s] = stmp;
}
grid->iCoorFromIindex(icoor,s);
assert((icoor[direction]==0)||(icoor[direction]==1));
int permute_lane;
if ( distance == 1) {
permute_lane = icoor[direction]?1:0;
} else {
permute_lane = icoor[direction]?0:1;
}
if ( permute_lane ) {
stmp(0) = vals[s](1);
stmp(1) = vals[s](0);
vals[s] = stmp;
}
}
merge(vtmp,vals);
} else {
vtmp(0) = chi(1);
vtmp(1) = chi(0);
@ -432,11 +548,11 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
GaugeLinkField Uconj(GaugeGrid);
Lattice<iScalar<vInteger> > coor(GaugeGrid);
for(int mu=0;mu<Nd;mu++){
LatticeCoordinate(coor,mu);
U = PeekIndex<LorentzIndex>(Umu,mu);
Uconj = conjugate(U);
@ -450,7 +566,7 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
Uds[ss](0)(mu) = U[ss]();
Uds[ss](1)(mu) = Uconj[ss]();
}
U = adj(Cshift(U ,mu,-1)); // correct except for spanning the boundary
Uconj = adj(Cshift(Uconj,mu,-1));
@ -458,11 +574,12 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
if ( Params.twists[mu] ) {
Utmp = where(coor==0,Uconj,Utmp);
}
parallel_for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](0)(mu+4) = Utmp[ss]();
}
Utmp = Uconj;
if ( Params.twists[mu] ) {
Utmp = where(coor==0,U,Utmp);
@ -471,11 +588,10 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
parallel_for(auto ss=U.begin();ss<U.end();ss++){
Uds[ss](1)(mu+4) = Utmp[ss]();
}
}
}
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A, int mu) {
// DhopDir provides U or Uconj depending on coor/flavour.
@ -483,7 +599,7 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
// use lorentz for flavour as hack.
auto tmp = TraceIndex<SpinIndex>(outerProduct(Btilde, A));
parallel_for(auto ss = tmp.begin(); ss < tmp.end(); ss++) {
link[ss]() = tmp[ss](0, 0) - conjugate(tmp[ss](1, 1));
link[ss]() = tmp[ss](0, 0) + conjugate(tmp[ss](1, 1));
}
PokeIndex<LorentzIndex>(mat, link, mu);
return;
@ -492,7 +608,7 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde, int mu) {
int Ls = Btilde._grid->_fdimensions[0];
GaugeLinkField tmp(mat._grid);
tmp = zero;
parallel_for(int ss = 0; ss < tmp._grid->oSites(); ss++) {
@ -508,40 +624,36 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
};
/////////////////////////////////////////////////////////////////////////////
// Single flavour one component spinors with colour index
/////////////////////////////////////////////////////////////////////////////
template <class S, class Representation = FundamentalRepresentation >
class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
/////////////////////////////////////////////////////////////////////////////
// Single flavour one component spinors with colour index
/////////////////////////////////////////////////////////////////////////////
template <class S, class Representation = FundamentalRepresentation >
class StaggeredImpl : public PeriodicGaugeImpl<GaugeImplTypes<S, Representation::Dimension > > {
public:
typedef RealD _Coeff_t ;
static const int Dimension = Representation::Dimension;
static const bool LsVectorised=false;
typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
//Necessary?
constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
const bool LsVectorised=false;
typedef _Coeff_t Coeff_t;
INHERIT_GIMPL_TYPES(Gimpl);
template <typename vtype> using iImplScalar = iScalar<iScalar<iScalar<vtype> > >;
template <typename vtype> using iImplSpinor = iScalar<iScalar<iVector<vtype, Dimension> > >;
template <typename vtype> using iImplHalfSpinor = iScalar<iScalar<iVector<vtype, Dimension> > >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
template <typename vtype> using iImplPropagator = iScalar<iScalar<iMatrix<vtype, Dimension> > >;
typedef iImplScalar<Simd> SiteComplex;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef iImplDoubledGaugeField<Simd> SiteDoubledGaugeField;
typedef iImplPropagator<Simd> SitePropagator;
typedef Lattice<SiteComplex> ComplexField;
typedef Lattice<SiteSpinor> FermionField;
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef Lattice<SitePropagator> PropagatorField;
@ -641,8 +753,6 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
}
};
/////////////////////////////////////////////////////////////////////////////
// Single flavour one component spinors with colour index. 5d vec
/////////////////////////////////////////////////////////////////////////////
@ -651,20 +761,17 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
public:
typedef RealD _Coeff_t ;
static const int Dimension = Representation::Dimension;
static const bool LsVectorised=true;
typedef RealD Coeff_t ;
typedef PeriodicGaugeImpl<GaugeImplTypes<S, Dimension > > Gimpl;
//Necessary?
constexpr bool is_fundamental() const{return Dimension == Nc ? 1 : 0;}
const bool LsVectorised=true;
typedef _Coeff_t Coeff_t;
INHERIT_GIMPL_TYPES(Gimpl);
template <typename vtype> using iImplScalar = iScalar<iScalar<iScalar<vtype> > >;
template <typename vtype> using iImplSpinor = iScalar<iScalar<iVector<vtype, Dimension> > >;
template <typename vtype> using iImplHalfSpinor = iScalar<iScalar<iVector<vtype, Dimension> > >;
template <typename vtype> using iImplDoubledGaugeField = iVector<iScalar<iMatrix<vtype, Dimension> >, Nds>;
@ -681,12 +788,10 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
typedef Lattice<SiteDoubledGaugeField> DoubledGaugeField;
typedef Lattice<SitePropagator> PropagatorField;
typedef iImplScalar<Simd> SiteComplex;
typedef iImplSpinor<Simd> SiteSpinor;
typedef iImplHalfSpinor<Simd> SiteHalfSpinor;
typedef Lattice<SiteComplex> ComplexField;
typedef Lattice<SiteSpinor> FermionField;
typedef SimpleCompressor<SiteSpinor> Compressor;
@ -823,43 +928,61 @@ class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Nrepresent
}
};
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffReal > WilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffReal > WilsonImplF; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffReal > WilsonImplD; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplRL; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplFH; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffRealHalfComms > WilsonImplDF; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation > WilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation > WilsonImplF; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation > WilsonImplD; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffComplex > ZWilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplex > ZWilsonImplF; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplex > ZWilsonImplD; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation, ComplexD > ZWilsonImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation, ComplexD > ZWilsonImplF; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation, ComplexD > ZWilsonImplD; // Double
typedef WilsonImpl<vComplex, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplRL; // Real.. whichever prec
typedef WilsonImpl<vComplexF, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplFH; // Float
typedef WilsonImpl<vComplexD, FundamentalRepresentation, CoeffComplexHalfComms > ZWilsonImplDF; // Double
typedef WilsonImpl<vComplex, AdjointRepresentation > WilsonAdjImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, AdjointRepresentation > WilsonAdjImplF; // Float
typedef WilsonImpl<vComplexD, AdjointRepresentation > WilsonAdjImplD; // Double
typedef WilsonImpl<vComplex, AdjointRepresentation, CoeffReal > WilsonAdjImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, AdjointRepresentation, CoeffReal > WilsonAdjImplF; // Float
typedef WilsonImpl<vComplexD, AdjointRepresentation, CoeffReal > WilsonAdjImplD; // Double
typedef WilsonImpl<vComplex, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplF; // Float
typedef WilsonImpl<vComplexD, TwoIndexSymmetricRepresentation > WilsonTwoIndexSymmetricImplD; // Double
typedef WilsonImpl<vComplex, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplR; // Real.. whichever prec
typedef WilsonImpl<vComplexF, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplF; // Float
typedef WilsonImpl<vComplexD, TwoIndexSymmetricRepresentation, CoeffReal > WilsonTwoIndexSymmetricImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc> DomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc> DomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc> DomainWallVec5dImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc, CoeffReal> DomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc, CoeffReal> DomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc, CoeffReal> DomainWallVec5dImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc,ComplexD> ZDomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc,ComplexD> ZDomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc,ComplexD> ZDomainWallVec5dImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc, CoeffRealHalfComms> DomainWallVec5dImplRL; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc, CoeffRealHalfComms> DomainWallVec5dImplFH; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc, CoeffRealHalfComms> DomainWallVec5dImplDF; // Double
typedef GparityWilsonImpl<vComplex , Nc> GparityWilsonImplR; // Real.. whichever prec
typedef GparityWilsonImpl<vComplexF, Nc> GparityWilsonImplF; // Float
typedef GparityWilsonImpl<vComplexD, Nc> GparityWilsonImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc,CoeffComplex> ZDomainWallVec5dImplR; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc,CoeffComplex> ZDomainWallVec5dImplF; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc,CoeffComplex> ZDomainWallVec5dImplD; // Double
typedef DomainWallVec5dImpl<vComplex ,Nc,CoeffComplexHalfComms> ZDomainWallVec5dImplRL; // Real.. whichever prec
typedef DomainWallVec5dImpl<vComplexF,Nc,CoeffComplexHalfComms> ZDomainWallVec5dImplFH; // Float
typedef DomainWallVec5dImpl<vComplexD,Nc,CoeffComplexHalfComms> ZDomainWallVec5dImplDF; // Double
typedef GparityWilsonImpl<vComplex , Nc,CoeffReal> GparityWilsonImplR; // Real.. whichever prec
typedef GparityWilsonImpl<vComplexF, Nc,CoeffReal> GparityWilsonImplF; // Float
typedef GparityWilsonImpl<vComplexD, Nc,CoeffReal> GparityWilsonImplD; // Double
typedef GparityWilsonImpl<vComplex , Nc,CoeffRealHalfComms> GparityWilsonImplRL; // Real.. whichever prec
typedef GparityWilsonImpl<vComplexF, Nc,CoeffRealHalfComms> GparityWilsonImplFH; // Float
typedef GparityWilsonImpl<vComplexD, Nc,CoeffRealHalfComms> GparityWilsonImplDF; // Double
typedef StaggeredImpl<vComplex, FundamentalRepresentation > StaggeredImplR; // Real.. whichever prec
typedef StaggeredImpl<vComplexF, FundamentalRepresentation > StaggeredImplF; // Float
typedef StaggeredImpl<vComplexD, FundamentalRepresentation > StaggeredImplD; // Double
typedef StaggeredImpl<vComplex, FundamentalRepresentation > StaggeredImplR; // Real.. whichever prec
typedef StaggeredImpl<vComplexF, FundamentalRepresentation > StaggeredImplF; // Float
typedef StaggeredImpl<vComplexD, FundamentalRepresentation > StaggeredImplD; // Double
typedef StaggeredVec5dImpl<vComplex, FundamentalRepresentation > StaggeredVec5dImplR; // Real.. whichever prec
typedef StaggeredVec5dImpl<vComplexF, FundamentalRepresentation > StaggeredVec5dImplF; // Float
typedef StaggeredVec5dImpl<vComplexD, FundamentalRepresentation > StaggeredVec5dImplD; // Double
typedef StaggeredVec5dImpl<vComplex, FundamentalRepresentation > StaggeredVec5dImplR; // Real.. whichever prec
typedef StaggeredVec5dImpl<vComplexF, FundamentalRepresentation > StaggeredVec5dImplF; // Float
typedef StaggeredVec5dImpl<vComplexD, FundamentalRepresentation > StaggeredVec5dImplD; // Double
}}

511
lib/qcd/action/fermion/WilsonCompressor.h
View File

@ -33,228 +33,321 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
namespace Grid {
namespace QCD {
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonCompressor {
public:
int mu;
int dag;
/////////////////////////////////////////////////////////////////////////////////////////////
// optimised versions supporting half precision too
/////////////////////////////////////////////////////////////////////////////////////////////
WilsonCompressor(int _dag){
mu=0;
dag=_dag;
assert((dag==0)||(dag==1));
}
void Point(int p) {
mu=p;
};
template<class _HCspinor,class _Hspinor,class _Spinor, class projector,typename SFINAE = void >
class WilsonCompressorTemplate;
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
int mudag=mu;
if (!dag) {
mudag=(mu+Nd)%(2*Nd);
}
switch(mudag) {
case Xp:
spProjXp(ret,in);
break;
case Yp:
spProjYp(ret,in);
break;
case Zp:
spProjZp(ret,in);
break;
case Tp:
spProjTp(ret,in);
break;
case Xm:
spProjXm(ret,in);
break;
case Ym:
spProjYm(ret,in);
break;
case Zm:
spProjZm(ret,in);
break;
case Tm:
spProjTm(ret,in);
break;
default:
assert(0);
break;
}
return ret;
}
};
/////////////////////////
// optimised versions
/////////////////////////
template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
typename std::enable_if<std::is_same<_HCspinor,_Hspinor>::value>::type >
{
public:
int mu,dag;
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonXpCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjXp(ret,in);
return ret;
}
};
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonYpCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjYp(ret,in);
return ret;
}
};
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonZpCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjZp(ret,in);
return ret;
}
};
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonTpCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjTp(ret,in);
return ret;
}
};
void Point(int p) { mu=p; };
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonXmCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjXm(ret,in);
return ret;
}
};
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonYmCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjYm(ret,in);
return ret;
}
};
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonZmCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjZm(ret,in);
return ret;
}
};
template<class SiteHalfSpinor,class SiteSpinor>
class WilsonTmCompressor {
public:
inline SiteHalfSpinor operator () (const SiteSpinor &in) {
SiteHalfSpinor ret;
spProjTm(ret,in);
return ret;
}
};
WilsonCompressorTemplate(int _dag=0){
dag = _dag;
}
// Fast comms buffer manipulation which should inline right through (avoid direction
// dependent logic that prevents inlining
template<class vobj,class cobj>
class WilsonStencil : public CartesianStencil<vobj,cobj> {
public:
typedef _Spinor SiteSpinor;
typedef _Hspinor SiteHalfSpinor;
typedef _HCspinor SiteHalfCommSpinor;
typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
typedef typename SiteHalfSpinor::vector_type vComplexHigh;
constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
typedef CartesianCommunicator::CommsRequest_t CommsRequest_t;
inline int CommDatumSize(void) {
return sizeof(SiteHalfCommSpinor);
}
WilsonStencil(GridBase *grid,
/*****************************************************/
/* Compress includes precision change if mpi data is not same */
/*****************************************************/
inline void Compress(SiteHalfSpinor *buf,Integer o,const SiteSpinor &in) {
projector::Proj(buf[o],in,mu,dag);
}
/*****************************************************/
/* Exchange includes precision change if mpi data is not same */
/*****************************************************/
inline void Exchange(SiteHalfSpinor *mp,
SiteHalfSpinor *vp0,
SiteHalfSpinor *vp1,
Integer type,Integer o){
exchange(mp[2*o],mp[2*o+1],vp0[o],vp1[o],type);
}
/*****************************************************/
/* Have a decompression step if mpi data is not same */
/*****************************************************/
inline void Decompress(SiteHalfSpinor *out,
SiteHalfSpinor *in, Integer o) {
assert(0);
}
/*****************************************************/
/* Compress Exchange */
/*****************************************************/
inline void CompressExchange(SiteHalfSpinor *out0,
SiteHalfSpinor *out1,
const SiteSpinor *in,
Integer j,Integer k, Integer m,Integer type){
SiteHalfSpinor temp1, temp2,temp3,temp4;
projector::Proj(temp1,in[k],mu,dag);
projector::Proj(temp2,in[m],mu,dag);
exchange(out0[j],out1[j],temp1,temp2,type);
}
/*****************************************************/
/* Pass the info to the stencil */
/*****************************************************/
inline bool DecompressionStep(void) { return false; }
};
template<class _HCspinor,class _Hspinor,class _Spinor, class projector>
class WilsonCompressorTemplate< _HCspinor, _Hspinor, _Spinor, projector,
typename std::enable_if<!std::is_same<_HCspinor,_Hspinor>::value>::type >
{
public:
int mu,dag;
void Point(int p) { mu=p; };
WilsonCompressorTemplate(int _dag=0){
dag = _dag;
}
typedef _Spinor SiteSpinor;
typedef _Hspinor SiteHalfSpinor;
typedef _HCspinor SiteHalfCommSpinor;
typedef typename SiteHalfCommSpinor::vector_type vComplexLow;
typedef typename SiteHalfSpinor::vector_type vComplexHigh;
constexpr static int Nw=sizeof(SiteHalfSpinor)/sizeof(vComplexHigh);
inline int CommDatumSize(void) {
return sizeof(SiteHalfCommSpinor);
}
/*****************************************************/
/* Compress includes precision change if mpi data is not same */
/*****************************************************/
inline void Compress(SiteHalfSpinor *buf,Integer o,const SiteSpinor &in) {
SiteHalfSpinor hsp;
SiteHalfCommSpinor *hbuf = (SiteHalfCommSpinor *)buf;
projector::Proj(hsp,in,mu,dag);
precisionChange((vComplexLow *)&hbuf[o],(vComplexHigh *)&hsp,Nw);
}
/*****************************************************/
/* Exchange includes precision change if mpi data is not same */
/*****************************************************/
inline void Exchange(SiteHalfSpinor *mp,
SiteHalfSpinor *vp0,
SiteHalfSpinor *vp1,
Integer type,Integer o){
SiteHalfSpinor vt0,vt1;
SiteHalfCommSpinor *vpp0 = (SiteHalfCommSpinor *)vp0;
SiteHalfCommSpinor *vpp1 = (SiteHalfCommSpinor *)vp1;
precisionChange((vComplexHigh *)&vt0,(vComplexLow *)&vpp0[o],Nw);
precisionChange((vComplexHigh *)&vt1,(vComplexLow *)&vpp1[o],Nw);
exchange(mp[2*o],mp[2*o+1],vt0,vt1,type);
}
/*****************************************************/
/* Have a decompression step if mpi data is not same */
/*****************************************************/
inline void Decompress(SiteHalfSpinor *out,
SiteHalfSpinor *in, Integer o){
SiteHalfCommSpinor *hin=(SiteHalfCommSpinor *)in;
precisionChange((vComplexHigh *)&out[o],(vComplexLow *)&hin[o],Nw);
}
/*****************************************************/
/* Compress Exchange */
/*****************************************************/
inline void CompressExchange(SiteHalfSpinor *out0,
SiteHalfSpinor *out1,
const SiteSpinor *in,
Integer j,Integer k, Integer m,Integer type){
SiteHalfSpinor temp1, temp2,temp3,temp4;
SiteHalfCommSpinor *hout0 = (SiteHalfCommSpinor *)out0;
SiteHalfCommSpinor *hout1 = (SiteHalfCommSpinor *)out1;
projector::Proj(temp1,in[k],mu,dag);
projector::Proj(temp2,in[m],mu,dag);
exchange(temp3,temp4,temp1,temp2,type);
precisionChange((vComplexLow *)&hout0[j],(vComplexHigh *)&temp3,Nw);
precisionChange((vComplexLow *)&hout1[j],(vComplexHigh *)&temp4,Nw);
}
/*****************************************************/
/* Pass the info to the stencil */
/*****************************************************/
inline bool DecompressionStep(void) { return true; }
};
#define DECLARE_PROJ(Projector,Compressor,spProj) \
class Projector { \
public: \
template<class hsp,class fsp> \
static void Proj(hsp &result,const fsp &in,int mu,int dag){ \
spProj(result,in); \
} \
}; \
template<typename HCS,typename HS,typename S> using Compressor = WilsonCompressorTemplate<HCS,HS,S,Projector>;
DECLARE_PROJ(WilsonXpProjector,WilsonXpCompressor,spProjXp);
DECLARE_PROJ(WilsonYpProjector,WilsonYpCompressor,spProjYp);
DECLARE_PROJ(WilsonZpProjector,WilsonZpCompressor,spProjZp);
DECLARE_PROJ(WilsonTpProjector,WilsonTpCompressor,spProjTp);
DECLARE_PROJ(WilsonXmProjector,WilsonXmCompressor,spProjXm);
DECLARE_PROJ(WilsonYmProjector,WilsonYmCompressor,spProjYm);
DECLARE_PROJ(WilsonZmProjector,WilsonZmCompressor,spProjZm);
DECLARE_PROJ(WilsonTmProjector,WilsonTmCompressor,spProjTm);
class WilsonProjector {
public:
template<class hsp,class fsp>
static void Proj(hsp &result,const fsp &in,int mu,int dag){
int mudag=dag? mu : (mu+Nd)%(2*Nd);
switch(mudag) {
case Xp: spProjXp(result,in); break;
case Yp: spProjYp(result,in); break;
case Zp: spProjZp(result,in); break;
case Tp: spProjTp(result,in); break;
case Xm: spProjXm(result,in); break;
case Ym: spProjYm(result,in); break;
case Zm: spProjZm(result,in); break;
case Tm: spProjTm(result,in); break;
default: assert(0); break;
}
}
};
template<typename HCS,typename HS,typename S> using WilsonCompressor = WilsonCompressorTemplate<HCS,HS,S,WilsonProjector>;
// Fast comms buffer manipulation which should inline right through (avoid direction
// dependent logic that prevents inlining
template<class vobj,class cobj>
class WilsonStencil : public CartesianStencil<vobj,cobj> {
public:
typedef CartesianCommunicator::CommsRequest_t CommsRequest_t;
std::vector<int> same_node;
std::vector<int> surface_list;
WilsonStencil(GridBase *grid,
int npoints,
int checkerboard,
const std::vector<int> &directions,
const std::vector<int> &distances) : CartesianStencil<vobj,cobj> (grid,npoints,checkerboard,directions,distances)
{ };
template < class compressor>
void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress)
{
std::vector<std::vector<CommsRequest_t> > reqs;
HaloExchangeOptGather(source,compress);
this->CommunicateBegin(reqs);
this->calls++;
this->CommunicateComplete(reqs);
this->CommsMerge();
}
template < class compressor>
void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress)
{
this->calls++;
this->Mergers.resize(0);
this->Packets.resize(0);
this->HaloGatherOpt(source,compress);
}
template < class compressor>
void HaloGatherOpt(const Lattice<vobj> &source,compressor &compress)
{
this->_grid->StencilBarrier();
// conformable(source._grid,_grid);
assert(source._grid==this->_grid);
this->halogtime-=usecond();
this->u_comm_offset=0;
int dag = compress.dag;
WilsonXpCompressor<cobj,vobj> XpCompress;
WilsonYpCompressor<cobj,vobj> YpCompress;
WilsonZpCompressor<cobj,vobj> ZpCompress;
WilsonTpCompressor<cobj,vobj> TpCompress;
WilsonXmCompressor<cobj,vobj> XmCompress;
WilsonYmCompressor<cobj,vobj> YmCompress;
WilsonZmCompressor<cobj,vobj> ZmCompress;
WilsonTmCompressor<cobj,vobj> TmCompress;
// Gather all comms buffers
// for(int point = 0 ; point < _npoints; point++) {
// compress.Point(point);
// HaloGatherDir(source,compress,point,face_idx);
// }
int face_idx=0;
if ( dag ) {
// std::cout << " Optimised Dagger compress " <<std::endl;
this->HaloGatherDir(source,XpCompress,Xp,face_idx);
this->HaloGatherDir(source,YpCompress,Yp,face_idx);
this->HaloGatherDir(source,ZpCompress,Zp,face_idx);
this->HaloGatherDir(source,TpCompress,Tp,face_idx);
this->HaloGatherDir(source,XmCompress,Xm,face_idx);
this->HaloGatherDir(source,YmCompress,Ym,face_idx);
this->HaloGatherDir(source,ZmCompress,Zm,face_idx);
this->HaloGatherDir(source,TmCompress,Tm,face_idx);
} else {
this->HaloGatherDir(source,XmCompress,Xp,face_idx);
this->HaloGatherDir(source,YmCompress,Yp,face_idx);
this->HaloGatherDir(source,ZmCompress,Zp,face_idx);
this->HaloGatherDir(source,TmCompress,Tp,face_idx);
this->HaloGatherDir(source,XpCompress,Xm,face_idx);
this->HaloGatherDir(source,YpCompress,Ym,face_idx);
this->HaloGatherDir(source,ZpCompress,Zm,face_idx);
this->HaloGatherDir(source,TpCompress,Tm,face_idx);
}
this->face_table_computed=1;
assert(this->u_comm_offset==this->_unified_buffer_size);
this->halogtime+=usecond();
}
const std::vector<int> &distances)
: CartesianStencil<vobj,cobj> (grid,npoints,checkerboard,directions,distances) ,
same_node(npoints)
{
surface_list.resize(0);
};
void BuildSurfaceList(int Ls,int vol4){
// find same node for SHM
// Here we know the distance is 1 for WilsonStencil
for(int point=0;point<this->_npoints;point++){
same_node[point] = this->SameNode(point);
// std::cout << " dir " <<point<<" same_node " <<same_node[point]<<std::endl;
}
for(int site = 0 ;site< vol4;site++){
int local = 1;
for(int point=0;point<this->_npoints;point++){
if( (!this->GetNodeLocal(site*Ls,point)) && (!same_node[point]) ){
local = 0;
}
}
if(local == 0) {
surface_list.push_back(site);
}
}
}
template < class compressor>
void HaloExchangeOpt(const Lattice<vobj> &source,compressor &compress)
{
std::vector<std::vector<CommsRequest_t> > reqs;
this->HaloExchangeOptGather(source,compress);
this->CommunicateBegin(reqs);
this->CommunicateComplete(reqs);
this->CommsMerge(compress);
this->CommsMergeSHM(compress);
}
template <class compressor>
void HaloExchangeOptGather(const Lattice<vobj> &source,compressor &compress)
{
this->Prepare();
this->HaloGatherOpt(source,compress);
}
template <class compressor>
void HaloGatherOpt(const Lattice<vobj> &source,compressor &compress)
{
// Strategy. Inherit types from Compressor.
// Use types to select the write direction by directon compressor
typedef typename compressor::SiteSpinor SiteSpinor;
typedef typename compressor::SiteHalfSpinor SiteHalfSpinor;
typedef typename compressor::SiteHalfCommSpinor SiteHalfCommSpinor;
this->_grid->StencilBarrier();
assert(source._grid==this->_grid);
this->halogtime-=usecond();
this->u_comm_offset=0;
WilsonXpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> XpCompress;
WilsonYpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> YpCompress;
WilsonZpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> ZpCompress;
WilsonTpCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> TpCompress;
WilsonXmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> XmCompress;
WilsonYmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> YmCompress;
WilsonZmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> ZmCompress;
WilsonTmCompressor<SiteHalfCommSpinor,SiteHalfSpinor,SiteSpinor> TmCompress;
int dag = compress.dag;
int face_idx=0;
if ( dag ) {
// std::cout << " Optimised Dagger compress " <<std::endl;
assert(same_node[Xp]==this->HaloGatherDir(source,XpCompress,Xp,face_idx));
assert(same_node[Yp]==this->HaloGatherDir(source,YpCompress,Yp,face_idx));
assert(same_node[Zp]==this->HaloGatherDir(source,ZpCompress,Zp,face_idx));
assert(same_node[Tp]==this->HaloGatherDir(source,TpCompress,Tp,face_idx));
assert(same_node[Xm]==this->HaloGatherDir(source,XmCompress,Xm,face_idx));
assert(same_node[Ym]==this->HaloGatherDir(source,YmCompress,Ym,face_idx));
assert(same_node[Zm]==this->HaloGatherDir(source,ZmCompress,Zm,face_idx));
assert(same_node[Tm]==this->HaloGatherDir(source,TmCompress,Tm,face_idx));
} else {
assert(same_node[Xp]==this->HaloGatherDir(source,XmCompress,Xp,face_idx));
assert(same_node[Yp]==this->HaloGatherDir(source,YmCompress,Yp,face_idx));
assert(same_node[Zp]==this->HaloGatherDir(source,ZmCompress,Zp,face_idx));
assert(same_node[Tp]==this->HaloGatherDir(source,TmCompress,Tp,face_idx));
assert(same_node[Xm]==this->HaloGatherDir(source,XpCompress,Xm,face_idx));
assert(same_node[Ym]==this->HaloGatherDir(source,YpCompress,Ym,face_idx));
assert(same_node[Zm]==this->HaloGatherDir(source,ZpCompress,Zm,face_idx));
assert(same_node[Tm]==this->HaloGatherDir(source,TpCompress,Tm,face_idx));
}
this->face_table_computed=1;
assert(this->u_comm_offset==this->_unified_buffer_size);
this->halogtime+=usecond();
}
};
}} // namespace close
#endif

16
lib/qcd/action/fermion/WilsonFermion.cc
View File

@ -230,8 +230,7 @@ void WilsonFermion<Impl>::DerivInternal(StencilImpl &st, DoubledGaugeField &U,
}
template <class Impl>
void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag) {
void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
conformable(U._grid, _grid);
conformable(U._grid, V._grid);
conformable(U._grid, mat._grid);
@ -242,12 +241,12 @@ void WilsonFermion<Impl>::DhopDeriv(GaugeField &mat, const FermionField &U,
}
template <class Impl>
void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag) {
void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
conformable(U._grid, _cbgrid);
conformable(U._grid, V._grid);
conformable(U._grid, mat._grid);
//conformable(U._grid, mat._grid); not general, leaving as a comment (Guido)
// Motivation: look at the SchurDiff operator
assert(V.checkerboard == Even);
assert(U.checkerboard == Odd);
mat.checkerboard = Odd;
@ -256,11 +255,10 @@ void WilsonFermion<Impl>::DhopDerivOE(GaugeField &mat, const FermionField &U,
}
template <class Impl>
void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U,
const FermionField &V, int dag) {
void WilsonFermion<Impl>::DhopDerivEO(GaugeField &mat, const FermionField &U, const FermionField &V, int dag) {
conformable(U._grid, _cbgrid);
conformable(U._grid, V._grid);
conformable(U._grid, mat._grid);
//conformable(U._grid, mat._grid);
assert(V.checkerboard == Odd);
assert(U.checkerboard == Even);

143
lib/qcd/action/fermion/WilsonFermion5D.cc
View File

@ -11,6 +11,7 @@ Author: Peter Boyle <pabobyle@ph.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Peter Boyle <peterboyle@Peters-MacBook-Pro-2.local>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@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
@ -117,49 +118,19 @@ WilsonFermion5D<Impl>::WilsonFermion5D(GaugeField &_Umu,
// Allocate the required comms buffer
ImportGauge(_Umu);
// Build lists of exterior only nodes
int LLs = FiveDimGrid._rdimensions[0];
int vol4;
vol4=FourDimGrid.oSites();
Stencil.BuildSurfaceList(LLs,vol4);
vol4=FourDimRedBlackGrid.oSites();
StencilEven.BuildSurfaceList(LLs,vol4);
StencilOdd.BuildSurfaceList(LLs,vol4);
std::cout << GridLogMessage << " SurfaceLists "<< Stencil.surface_list.size()
<<" " << StencilEven.surface_list.size()<<std::endl;
}
/*
template<class Impl>
WilsonFermion5D<Impl>::WilsonFermion5D(int simd,GaugeField &_Umu,
GridCartesian &FiveDimGrid,
GridRedBlackCartesian &FiveDimRedBlackGrid,
GridCartesian &FourDimGrid,
RealD _M5,const ImplParams &p) :
{
int nsimd = Simd::Nsimd();
// some assertions
assert(FiveDimGrid._ndimension==5);
assert(FiveDimRedBlackGrid._ndimension==5);
assert(FiveDimRedBlackGrid._checker_dim==0); // Checkerboard the s-direction
assert(FourDimGrid._ndimension==4);
// Dimension zero of the five-d is the Ls direction
Ls=FiveDimGrid._fdimensions[0];
assert(FiveDimGrid._processors[0] ==1);
assert(FiveDimGrid._simd_layout[0] ==nsimd);
assert(FiveDimRedBlackGrid._fdimensions[0]==Ls);
assert(FiveDimRedBlackGrid._processors[0] ==1);
assert(FiveDimRedBlackGrid._simd_layout[0]==nsimd);
// Other dimensions must match the decomposition of the four-D fields
for(int d=0;d<4;d++){
assert(FiveDimRedBlackGrid._fdimensions[d+1]==FourDimGrid._fdimensions[d]);
assert(FiveDimRedBlackGrid._processors[d+1] ==FourDimGrid._processors[d]);
assert(FourDimGrid._simd_layout[d]=1);
assert(FiveDimRedBlackGrid._simd_layout[d+1]==1);
assert(FiveDimGrid._fdimensions[d+1] ==FourDimGrid._fdimensions[d]);
assert(FiveDimGrid._processors[d+1] ==FourDimGrid._processors[d]);
assert(FiveDimGrid._simd_layout[d+1] ==FourDimGrid._simd_layout[d]);
}
{
}
}
*/
template<class Impl>
void WilsonFermion5D<Impl>::Report(void)
@ -296,6 +267,8 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
DerivCommTime+=usecond();
Atilde=A;
int LLs = B._grid->_rdimensions[0];
DerivComputeTime-=usecond();
for (int mu = 0; mu < Nd; mu++) {
@ -325,6 +298,9 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
////////////////////////////
}
}
////////////////////////////
// spin trace outer product
////////////////////////////
DerivDhopComputeTime += usecond();
Impl::InsertForce5D(mat, Btilde, Atilde, mu);
}
@ -333,13 +309,14 @@ void WilsonFermion5D<Impl>::DerivInternal(StencilImpl & st,
template<class Impl>
void WilsonFermion5D<Impl>::DhopDeriv(GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
const FermionField &A,
const FermionField &B,
int dag)
{
conformable(A._grid,FermionGrid());
conformable(A._grid,B._grid);
conformable(GaugeGrid(),mat._grid);
//conformable(GaugeGrid(),mat._grid);// this is not general! leaving as a comment
mat.checkerboard = A.checkerboard;
@ -348,12 +325,11 @@ void WilsonFermion5D<Impl>::DhopDeriv(GaugeField &mat,
template<class Impl>
void WilsonFermion5D<Impl>::DhopDerivEO(GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
const FermionField &A,
const FermionField &B,
int dag)
{
conformable(A._grid,FermionRedBlackGrid());
conformable(GaugeRedBlackGrid(),mat._grid);
conformable(A._grid,B._grid);
assert(B.checkerboard==Odd);
@ -366,12 +342,11 @@ void WilsonFermion5D<Impl>::DhopDerivEO(GaugeField &mat,
template<class Impl>
void WilsonFermion5D<Impl>::DhopDerivOE(GaugeField &mat,
const FermionField &A,
const FermionField &B,
int dag)
const FermionField &A,
const FermionField &B,
int dag)
{
conformable(A._grid,FermionRedBlackGrid());
conformable(GaugeRedBlackGrid(),mat._grid);
conformable(A._grid,B._grid);
assert(B.checkerboard==Even);
@ -383,8 +358,8 @@ void WilsonFermion5D<Impl>::DhopDerivOE(GaugeField &mat,
template<class Impl>
void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
DoubledGaugeField & U,
const FermionField &in, FermionField &out,int dag)
{
DhopTotalTime-=usecond();
#ifdef GRID_OMP
@ -396,6 +371,7 @@ void WilsonFermion5D<Impl>::DhopInternal(StencilImpl & st, LebesgueOrder &lo,
DhopTotalTime+=usecond();
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
@ -409,12 +385,21 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
int LLs = in._grid->_rdimensions[0];
int len = U._grid->oSites();
DhopFaceTime-=usecond();
st.HaloExchangeOptGather(in,compressor);
DhopFaceTime+=usecond();
std::vector<std::vector<CommsRequest_t> > reqs;
// Rely on async comms; start comms before merge of local data
DhopCommTime-=usecond();
st.CommunicateBegin(reqs);
DhopFaceTime-=usecond();
st.CommsMergeSHM(compressor);
DhopFaceTime+=usecond();
// Perhaps use omp task and region
#pragma omp parallel
{
int nthreads = omp_get_num_threads();
@ -425,8 +410,6 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
int sF = LLs * myoff;
if ( me == 0 ) {
DhopCommTime-=usecond();
st.CommunicateBegin(reqs);
st.CommunicateComplete(reqs);
DhopCommTime+=usecond();
} else {
@ -439,28 +422,37 @@ void WilsonFermion5D<Impl>::DhopInternalOverlappedComms(StencilImpl & st, Lebesg
}
DhopFaceTime-=usecond();
st.CommsMerge();
st.CommsMerge(compressor);
DhopFaceTime+=usecond();
#pragma omp parallel
{
int nthreads = omp_get_num_threads();
int me = omp_get_thread_num();
int myoff, mywork;
GridThread::GetWork(len,me,mywork,myoff,nthreads);
int sF = LLs * myoff;
// Exterior links in stencil
if ( me==0 ) DhopComputeTime2-=usecond();
if (dag == DaggerYes) Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,myoff,LLs,mywork,in,out,0,1);
else Kernels::DhopSite (st,lo,U,st.CommBuf(),sF,myoff,LLs,mywork,in,out,0,1);
if ( me==0 ) DhopComputeTime2+=usecond();
}// end parallel region
// Load imbalance alert. Should use dynamic schedule OMP for loop
// Perhaps create a list of only those sites with face work, and
// load balance process the list.
DhopComputeTime2-=usecond();
if (dag == DaggerYes) {
int sz=st.surface_list.size();
parallel_for (int ss = 0; ss < sz; ss++) {
int sU = st.surface_list[ss];
int sF = LLs * sU;
Kernels::DhopSiteDag(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
}
} else {
int sz=st.surface_list.size();
parallel_for (int ss = 0; ss < sz; ss++) {
int sU = st.surface_list[ss];
int sF = LLs * sU;
Kernels::DhopSite(st,lo,U,st.CommBuf(),sF,sU,LLs,1,in,out,0,1);
}
}
DhopComputeTime2+=usecond();
#else
assert(0);
#endif
}
template<class Impl>
void WilsonFermion5D<Impl>::DhopInternalSerialComms(StencilImpl & st, LebesgueOrder &lo,
DoubledGaugeField & U,
@ -679,7 +671,6 @@ void WilsonFermion5D<Impl>::MomentumSpacePropagatorHw(FermionField &out,const Fe
}
FermOpTemplateInstantiate(WilsonFermion5D);
GparityFermOpTemplateInstantiate(WilsonFermion5D);

684
lib/qcd/action/fermion/WilsonKernels.cc
View File

@ -33,52 +33,8 @@ directory
namespace Grid {
namespace QCD {
int WilsonKernelsStatic::Opt = WilsonKernelsStatic::OptGeneric;
int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
#ifdef QPX
#include <spi/include/kernel/location.h>
#include <spi/include/l1p/types.h>
#include <hwi/include/bqc/l1p_mmio.h>
#include <hwi/include/bqc/A2_inlines.h>
#endif
void bgq_l1p_optimisation(int mode)
{
#ifdef QPX
#undef L1P_CFG_PF_USR
#define L1P_CFG_PF_USR (0x3fde8000108ll) /* (64 bit reg, 23 bits wide, user/unpriv) */
uint64_t cfg_pf_usr;
if ( mode ) {
cfg_pf_usr =
L1P_CFG_PF_USR_ifetch_depth(0)
| L1P_CFG_PF_USR_ifetch_max_footprint(1)
| L1P_CFG_PF_USR_pf_stream_est_on_dcbt
| L1P_CFG_PF_USR_pf_stream_establish_enable
| L1P_CFG_PF_USR_pf_stream_optimistic
| L1P_CFG_PF_USR_pf_adaptive_throttle(0xF) ;
// if ( sizeof(Float) == sizeof(double) ) {
cfg_pf_usr |= L1P_CFG_PF_USR_dfetch_depth(2)| L1P_CFG_PF_USR_dfetch_max_footprint(3) ;
// } else {
// cfg_pf_usr |= L1P_CFG_PF_USR_dfetch_depth(1)| L1P_CFG_PF_USR_dfetch_max_footprint(2) ;
// }
} else {
cfg_pf_usr = L1P_CFG_PF_USR_dfetch_depth(1)
| L1P_CFG_PF_USR_dfetch_max_footprint(2)
| L1P_CFG_PF_USR_ifetch_depth(0)
| L1P_CFG_PF_USR_ifetch_max_footprint(1)
| L1P_CFG_PF_USR_pf_stream_est_on_dcbt
| L1P_CFG_PF_USR_pf_stream_establish_enable
| L1P_CFG_PF_USR_pf_stream_optimistic
| L1P_CFG_PF_USR_pf_stream_prefetch_enable;
}
*((uint64_t *)L1P_CFG_PF_USR) = cfg_pf_usr;
#endif
}
int WilsonKernelsStatic::Opt = WilsonKernelsStatic::OptGeneric;
int WilsonKernelsStatic::Comms = WilsonKernelsStatic::CommsAndCompute;
template <class Impl>
WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
@ -86,12 +42,72 @@ WilsonKernels<Impl>::WilsonKernels(const ImplParams &p) : Base(p){};
////////////////////////////////////////////
// Generic implementation; move to different file?
////////////////////////////////////////////
#define GENERIC_STENCIL_LEG(Dir,spProj,Recon) \
SE = st.GetEntry(ptype, Dir, sF); \
if (SE->_is_local) { \
chi_p = &chi; \
if (SE->_permute) { \
spProj(tmp, in._odata[SE->_offset]); \
permute(chi, tmp, ptype); \
} else { \
spProj(chi, in._odata[SE->_offset]); \
} \
} else { \
chi_p = &buf[SE->_offset]; \
} \
Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st); \
Recon(result, Uchi);
#define GENERIC_STENCIL_LEG_INT(Dir,spProj,Recon) \
SE = st.GetEntry(ptype, Dir, sF); \
if (SE->_is_local) { \
chi_p = &chi; \
if (SE->_permute) { \
spProj(tmp, in._odata[SE->_offset]); \
permute(chi, tmp, ptype); \
} else { \
spProj(chi, in._odata[SE->_offset]); \
} \
} else if ( st.same_node[Dir] ) { \
chi_p = &buf[SE->_offset]; \
} \
if (SE->_is_local || st.same_node[Dir] ) { \
Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st); \
Recon(result, Uchi); \
}
#define GENERIC_STENCIL_LEG_EXT(Dir,spProj,Recon) \
SE = st.GetEntry(ptype, Dir, sF); \
if ((!SE->_is_local) && (!st.same_node[Dir]) ) { \
chi_p = &buf[SE->_offset]; \
Impl::multLink(Uchi, U._odata[sU], *chi_p, Dir, SE, st); \
Recon(result, Uchi); \
nmu++; \
}
#define GENERIC_DHOPDIR_LEG(Dir,spProj,Recon) \
if (gamma == Dir) { \
if (SE->_is_local && SE->_permute) { \
spProj(tmp, in._odata[SE->_offset]); \
permute(chi, tmp, ptype); \
} else if (SE->_is_local) { \
spProj(chi, in._odata[SE->_offset]); \
} else { \
chi = buf[SE->_offset]; \
} \
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st); \
Recon(result, Uchi); \
}
////////////////////////////////////////////////////////////////////
// All legs kernels ; comms then compute
////////////////////////////////////////////////////////////////////
template <class Impl>
void WilsonKernels<Impl>::GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out,
int interior,int exterior) {
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
@ -100,174 +116,22 @@ void WilsonKernels<Impl>::GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo,
StencilEntry *SE;
int ptype;
///////////////////////////
// Xp
///////////////////////////
SE = st.GetEntry(ptype, Xp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjXp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xp, SE, st);
spReconXp(result, Uchi);
///////////////////////////
// Yp
///////////////////////////
SE = st.GetEntry(ptype, Yp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjYp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Yp, SE, st);
accumReconYp(result, Uchi);
///////////////////////////
// Zp
///////////////////////////
SE = st.GetEntry(ptype, Zp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjZp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zp, SE, st);
accumReconZp(result, Uchi);
///////////////////////////
// Tp
///////////////////////////
SE = st.GetEntry(ptype, Tp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjTp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tp, SE, st);
accumReconTp(result, Uchi);
///////////////////////////
// Xm
///////////////////////////
SE = st.GetEntry(ptype, Xm, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjXm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xm, SE, st);
accumReconXm(result, Uchi);
///////////////////////////
// Ym
///////////////////////////
SE = st.GetEntry(ptype, Ym, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjYm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Ym, SE, st);
accumReconYm(result, Uchi);
///////////////////////////
// Zm
///////////////////////////
SE = st.GetEntry(ptype, Zm, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjZm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zm, SE, st);
accumReconZm(result, Uchi);
///////////////////////////
// Tm
///////////////////////////
SE = st.GetEntry(ptype, Tm, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjTm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tm, SE, st);
accumReconTm(result, Uchi);
GENERIC_STENCIL_LEG(Xp,spProjXp,spReconXp);
GENERIC_STENCIL_LEG(Yp,spProjYp,accumReconYp);
GENERIC_STENCIL_LEG(Zp,spProjZp,accumReconZp);
GENERIC_STENCIL_LEG(Tp,spProjTp,accumReconTp);
GENERIC_STENCIL_LEG(Xm,spProjXm,accumReconXm);
GENERIC_STENCIL_LEG(Ym,spProjYm,accumReconYm);
GENERIC_STENCIL_LEG(Zm,spProjZm,accumReconZm);
GENERIC_STENCIL_LEG(Tm,spProjTm,accumReconTm);
vstream(out._odata[sF], result);
};
// Need controls to do interior, exterior, or both
template <class Impl>
void WilsonKernels<Impl>::GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out,int interior,int exterior) {
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
@ -276,168 +140,123 @@ void WilsonKernels<Impl>::GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, Do
StencilEntry *SE;
int ptype;
///////////////////////////
// Xp
///////////////////////////
SE = st.GetEntry(ptype, Xm, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjXp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xm, SE, st);
spReconXp(result, Uchi);
///////////////////////////
// Yp
///////////////////////////
SE = st.GetEntry(ptype, Ym, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjYp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Ym, SE, st);
accumReconYp(result, Uchi);
///////////////////////////
// Zp
///////////////////////////
SE = st.GetEntry(ptype, Zm, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjZp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zm, SE, st);
accumReconZp(result, Uchi);
///////////////////////////
// Tp
///////////////////////////
SE = st.GetEntry(ptype, Tm, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjTp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTp(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tm, SE, st);
accumReconTp(result, Uchi);
///////////////////////////
// Xm
///////////////////////////
SE = st.GetEntry(ptype, Xp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjXm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjXm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Xp, SE, st);
accumReconXm(result, Uchi);
///////////////////////////
// Ym
///////////////////////////
SE = st.GetEntry(ptype, Yp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjYm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjYm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Yp, SE, st);
accumReconYm(result, Uchi);
///////////////////////////
// Zm
///////////////////////////
SE = st.GetEntry(ptype, Zp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjZm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjZm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Zp, SE, st);
accumReconZm(result, Uchi);
///////////////////////////
// Tm
///////////////////////////
SE = st.GetEntry(ptype, Tp, sF);
if (SE->_is_local) {
chi_p = &chi;
if (SE->_permute) {
spProjTm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else {
spProjTm(chi, in._odata[SE->_offset]);
}
} else {
chi_p = &buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], *chi_p, Tp, SE, st);
accumReconTm(result, Uchi);
GENERIC_STENCIL_LEG(Xm,spProjXp,spReconXp);
GENERIC_STENCIL_LEG(Ym,spProjYp,accumReconYp);
GENERIC_STENCIL_LEG(Zm,spProjZp,accumReconZp);
GENERIC_STENCIL_LEG(Tm,spProjTp,accumReconTp);
GENERIC_STENCIL_LEG(Xp,spProjXm,accumReconXm);
GENERIC_STENCIL_LEG(Yp,spProjYm,accumReconYm);
GENERIC_STENCIL_LEG(Zp,spProjZm,accumReconZm);
GENERIC_STENCIL_LEG(Tp,spProjTm,accumReconTm);
vstream(out._odata[sF], result);
};
////////////////////////////////////////////////////////////////////
// Interior kernels
////////////////////////////////////////////////////////////////////
template <class Impl>
void WilsonKernels<Impl>::GenericDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
SiteHalfSpinor Uchi;
SiteSpinor result;
StencilEntry *SE;
int ptype;
result=zero;
GENERIC_STENCIL_LEG_INT(Xp,spProjXp,accumReconXp);
GENERIC_STENCIL_LEG_INT(Yp,spProjYp,accumReconYp);
GENERIC_STENCIL_LEG_INT(Zp,spProjZp,accumReconZp);
GENERIC_STENCIL_LEG_INT(Tp,spProjTp,accumReconTp);
GENERIC_STENCIL_LEG_INT(Xm,spProjXm,accumReconXm);
GENERIC_STENCIL_LEG_INT(Ym,spProjYm,accumReconYm);
GENERIC_STENCIL_LEG_INT(Zm,spProjZm,accumReconZm);
GENERIC_STENCIL_LEG_INT(Tm,spProjTm,accumReconTm);
vstream(out._odata[sF], result);
};
template <class Impl>
void WilsonKernels<Impl>::GenericDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
SiteHalfSpinor Uchi;
SiteSpinor result;
StencilEntry *SE;
int ptype;
result=zero;
GENERIC_STENCIL_LEG_INT(Xm,spProjXp,accumReconXp);
GENERIC_STENCIL_LEG_INT(Ym,spProjYp,accumReconYp);
GENERIC_STENCIL_LEG_INT(Zm,spProjZp,accumReconZp);
GENERIC_STENCIL_LEG_INT(Tm,spProjTp,accumReconTp);
GENERIC_STENCIL_LEG_INT(Xp,spProjXm,accumReconXm);
GENERIC_STENCIL_LEG_INT(Yp,spProjYm,accumReconYm);
GENERIC_STENCIL_LEG_INT(Zp,spProjZm,accumReconZm);
GENERIC_STENCIL_LEG_INT(Tp,spProjTm,accumReconTm);
vstream(out._odata[sF], result);
};
////////////////////////////////////////////////////////////////////
// Exterior kernels
////////////////////////////////////////////////////////////////////
template <class Impl>
void WilsonKernels<Impl>::GenericDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
SiteHalfSpinor Uchi;
SiteSpinor result;
StencilEntry *SE;
int ptype;
int nmu=0;
result=zero;
GENERIC_STENCIL_LEG_EXT(Xp,spProjXp,accumReconXp);
GENERIC_STENCIL_LEG_EXT(Yp,spProjYp,accumReconYp);
GENERIC_STENCIL_LEG_EXT(Zp,spProjZp,accumReconZp);
GENERIC_STENCIL_LEG_EXT(Tp,spProjTp,accumReconTp);
GENERIC_STENCIL_LEG_EXT(Xm,spProjXm,accumReconXm);
GENERIC_STENCIL_LEG_EXT(Ym,spProjYm,accumReconYm);
GENERIC_STENCIL_LEG_EXT(Zm,spProjZm,accumReconZm);
GENERIC_STENCIL_LEG_EXT(Tm,spProjTm,accumReconTm);
if ( nmu ) {
out._odata[sF] = out._odata[sF] + result;
}
};
template <class Impl>
void WilsonKernels<Impl>::GenericDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U,
SiteHalfSpinor *buf, int sF,
int sU, const FermionField &in, FermionField &out)
{
SiteHalfSpinor tmp;
SiteHalfSpinor chi;
SiteHalfSpinor *chi_p;
SiteHalfSpinor Uchi;
SiteSpinor result;
StencilEntry *SE;
int ptype;
int nmu=0;
result=zero;
GENERIC_STENCIL_LEG_EXT(Xm,spProjXp,accumReconXp);
GENERIC_STENCIL_LEG_EXT(Ym,spProjYp,accumReconYp);
GENERIC_STENCIL_LEG_EXT(Zm,spProjZp,accumReconZp);
GENERIC_STENCIL_LEG_EXT(Tm,spProjTp,accumReconTp);
GENERIC_STENCIL_LEG_EXT(Xp,spProjXm,accumReconXm);
GENERIC_STENCIL_LEG_EXT(Yp,spProjYm,accumReconYm);
GENERIC_STENCIL_LEG_EXT(Zp,spProjZm,accumReconZm);
GENERIC_STENCIL_LEG_EXT(Tp,spProjTm,accumReconTm);
if ( nmu ) {
out._odata[sF] = out._odata[sF] + result;
}
};
template <class Impl>
void WilsonKernels<Impl>::DhopDir( StencilImpl &st, DoubledGaugeField &U,SiteHalfSpinor *buf, int sF,
@ -451,119 +270,14 @@ void WilsonKernels<Impl>::DhopDir( StencilImpl &st, DoubledGaugeField &U,SiteHal
int ptype;
SE = st.GetEntry(ptype, dir, sF);
// Xp
if (gamma == Xp) {
if (SE->_is_local && SE->_permute) {
spProjXp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjXp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconXp(result, Uchi);
}
// Yp
if (gamma == Yp) {
if (SE->_is_local && SE->_permute) {
spProjYp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjYp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconYp(result, Uchi);
}
// Zp
if (gamma == Zp) {
if (SE->_is_local && SE->_permute) {
spProjZp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjZp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconZp(result, Uchi);
}
// Tp
if (gamma == Tp) {
if (SE->_is_local && SE->_permute) {
spProjTp(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjTp(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconTp(result, Uchi);
}
// Xm
if (gamma == Xm) {
if (SE->_is_local && SE->_permute) {
spProjXm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjXm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconXm(result, Uchi);
}
// Ym
if (gamma == Ym) {
if (SE->_is_local && SE->_permute) {
spProjYm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjYm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconYm(result, Uchi);
}
// Zm
if (gamma == Zm) {
if (SE->_is_local && SE->_permute) {
spProjZm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjZm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconZm(result, Uchi);
}
// Tm
if (gamma == Tm) {
if (SE->_is_local && SE->_permute) {
spProjTm(tmp, in._odata[SE->_offset]);
permute(chi, tmp, ptype);
} else if (SE->_is_local) {
spProjTm(chi, in._odata[SE->_offset]);
} else {
chi = buf[SE->_offset];
}
Impl::multLink(Uchi, U._odata[sU], chi, dir, SE, st);
spReconTm(result, Uchi);
}
GENERIC_DHOPDIR_LEG(Xp,spProjXp,spReconXp);
GENERIC_DHOPDIR_LEG(Yp,spProjYp,spReconYp);
GENERIC_DHOPDIR_LEG(Zp,spProjZp,spReconZp);
GENERIC_DHOPDIR_LEG(Tp,spProjTp,spReconTp);
GENERIC_DHOPDIR_LEG(Xm,spProjXm,spReconXm);
GENERIC_DHOPDIR_LEG(Ym,spProjYm,spReconYm);
GENERIC_DHOPDIR_LEG(Zm,spProjZm,spReconZm);
GENERIC_DHOPDIR_LEG(Tm,spProjTm,spReconTm);
vstream(out._odata[sF], result);
}

90
lib/qcd/action/fermion/WilsonKernels.h
View File

@ -34,8 +34,6 @@ directory
namespace Grid {
namespace QCD {
void bgq_l1p_optimisation(int mode);
////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Helper routines that implement Wilson stencil for a single site.
// Common to both the WilsonFermion and WilsonFermion5D
@ -44,9 +42,8 @@ class WilsonKernelsStatic {
public:
enum { OptGeneric, OptHandUnroll, OptInlineAsm };
enum { CommsAndCompute, CommsThenCompute };
// S-direction is INNERMOST and takes no part in the parity.
static int Opt; // these are a temporary hack
static int Comms; // these are a temporary hack
static int Opt;
static int Comms;
};
template<class Impl> class WilsonKernels : public FermionOperator<Impl> , public WilsonKernelsStatic {
@ -66,7 +63,7 @@ public:
switch(Opt) {
#if defined(AVX512) || defined (QPX)
case OptInlineAsm:
if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSite (st,lo,U,buf,sF,sU,Ls,Ns,in,out);
else if (interior) WilsonKernels<Impl>::AsmDhopSiteInt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
else if (exterior) WilsonKernels<Impl>::AsmDhopSiteExt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
else assert(0);
@ -75,7 +72,9 @@ public:
case OptHandUnroll:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if( exterior) WilsonKernels<Impl>::HandDhopSite(st,lo,U,buf,sF,sU,in,out,interior,exterior);
if(interior&&exterior) WilsonKernels<Impl>::HandDhopSite(st,lo,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::HandDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::HandDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
sF++;
}
sU++;
@ -84,7 +83,10 @@ public:
case OptGeneric:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if( exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out,interior,exterior);
if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::GenericDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::GenericDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
@ -99,11 +101,14 @@ public:
template <bool EnableBool = true>
typename std::enable_if<(Impl::Dimension != 3 || (Impl::Dimension == 3 && Nc != 3)) && EnableBool, void>::type
DhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1 ) {
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1 ) {
// no kernel choice
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if( exterior) WilsonKernels<Impl>::GenericDhopSite(st, lo, U, buf, sF, sU, in, out,interior,exterior);
if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSite(st,lo,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::GenericDhopSiteInt(st,lo,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::GenericDhopSiteExt(st,lo,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
@ -113,13 +118,13 @@ public:
template <bool EnableBool = true>
typename std::enable_if<Impl::Dimension == 3 && Nc == 3 && EnableBool,void>::type
DhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1) {
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out,int interior=1,int exterior=1)
{
bgq_l1p_optimisation(1);
switch(Opt) {
#if defined(AVX512) || defined (QPX)
case OptInlineAsm:
if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSiteDag(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
if(interior&&exterior) WilsonKernels<Impl>::AsmDhopSiteDag (st,lo,U,buf,sF,sU,Ls,Ns,in,out);
else if (interior) WilsonKernels<Impl>::AsmDhopSiteDagInt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
else if (exterior) WilsonKernels<Impl>::AsmDhopSiteDagExt(st,lo,U,buf,sF,sU,Ls,Ns,in,out);
else assert(0);
@ -128,7 +133,10 @@ public:
case OptHandUnroll:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if( exterior) WilsonKernels<Impl>::HandDhopSiteDag(st,lo,U,buf,sF,sU,in,out,interior,exterior);
if(interior&&exterior) WilsonKernels<Impl>::HandDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::HandDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::HandDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
@ -137,7 +145,10 @@ public:
case OptGeneric:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if( exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out,interior,exterior);
if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::GenericDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::GenericDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
@ -156,7 +167,10 @@ public:
for (int site = 0; site < Ns; site++) {
for (int s = 0; s < Ls; s++) {
if( exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out,interior,exterior);
if(interior&&exterior) WilsonKernels<Impl>::GenericDhopSiteDag(st,lo,U,buf,sF,sU,in,out);
else if (interior) WilsonKernels<Impl>::GenericDhopSiteDagInt(st,lo,U,buf,sF,sU,in,out);
else if (exterior) WilsonKernels<Impl>::GenericDhopSiteDagExt(st,lo,U,buf,sF,sU,in,out);
else assert(0);
sF++;
}
sU++;
@ -169,36 +183,60 @@ public:
private:
// Specialised variants
void GenericDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
int sF, int sU, const FermionField &in, FermionField &out);
void GenericDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
int sF, int sU, const FermionField &in, FermionField &out);
void GenericDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void GenericDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void GenericDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void GenericDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void AsmDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
void AsmDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
void AsmDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
void AsmDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
void AsmDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
int sF, int sU, int Ls, int Ns, const FermionField &in,FermionField &out);
void AsmDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
int sF, int sU, int Ls, int Ns, const FermionField &in, FermionField &out);
void HandDhopSite(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
int sF, int sU, const FermionField &in, FermionField &out);
void HandDhopSiteDag(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out,int interior,int exterior);
int sF, int sU, const FermionField &in, FermionField &out);
void HandDhopSiteInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void HandDhopSiteDagInt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void HandDhopSiteExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
void HandDhopSiteDagExt(StencilImpl &st, LebesgueOrder &lo, DoubledGaugeField &U, SiteHalfSpinor * buf,
int sF, int sU, const FermionField &in, FermionField &out);
public:
WilsonKernels(const ImplParams &p = ImplParams());

11
lib/qcd/action/fermion/WilsonKernelsAsm.cc
View File

@ -112,5 +112,16 @@ INSTANTIATE_ASM(DomainWallVec5dImplD);
INSTANTIATE_ASM(ZDomainWallVec5dImplF);
INSTANTIATE_ASM(ZDomainWallVec5dImplD);
INSTANTIATE_ASM(WilsonImplFH);
INSTANTIATE_ASM(WilsonImplDF);
INSTANTIATE_ASM(ZWilsonImplFH);
INSTANTIATE_ASM(ZWilsonImplDF);
INSTANTIATE_ASM(GparityWilsonImplFH);
INSTANTIATE_ASM(GparityWilsonImplDF);
INSTANTIATE_ASM(DomainWallVec5dImplFH);
INSTANTIATE_ASM(DomainWallVec5dImplDF);
INSTANTIATE_ASM(ZDomainWallVec5dImplFH);
INSTANTIATE_ASM(ZDomainWallVec5dImplDF);
}}

223
lib/qcd/action/fermion/WilsonKernelsAsmAvx512.h
View File

@ -71,6 +71,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,Doub
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -84,6 +94,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,D
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
@ -97,6 +117,16 @@ template<> void
WilsonKernels<ZWilsonImplF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
/////////////////////////////////////////////////////////////////
// XYZT vectorised, dag Kernel, single
@ -115,6 +145,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,D
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -128,6 +168,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & l
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -141,6 +191,16 @@ WilsonKernels<ZWilsonImplF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & l
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef MAYBEPERM
#undef MULT_2SPIN
#define MAYBEPERM(A,B)
@ -162,6 +222,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSite(StencilImpl &st,LebesgueOrder
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -174,6 +243,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrd
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -189,6 +267,16 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrd
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
/////////////////////////////////////////////////////////////////
// Ls vectorised, dag Kernel, single
/////////////////////////////////////////////////////////////////
@ -205,6 +293,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrd
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -217,6 +314,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagInt(StencilImpl &st,Lebesgue
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -229,6 +335,15 @@ WilsonKernels<ZDomainWallVec5dImplF>::AsmDhopSiteDagExt(StencilImpl &st,Lebesgue
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplFH>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef COMPLEX_SIGNS
#undef MAYBEPERM
#undef MULT_2SPIN
@ -269,6 +384,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,Doub
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -281,6 +405,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,D
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -293,6 +426,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,D
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
/////////////////////////////////////////////////////////////////
// XYZT vectorised, dag Kernel, single
/////////////////////////////////////////////////////////////////
@ -309,6 +451,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,D
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -321,6 +472,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & l
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -333,6 +493,15 @@ WilsonKernels<ZWilsonImplD>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & l
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<WilsonImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZWilsonImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef MAYBEPERM
#undef MULT_2SPIN
#define MAYBEPERM(A,B)
@ -354,6 +523,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSite(StencilImpl &st,LebesgueOrder
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSite(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -366,6 +544,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrd
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -380,6 +567,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrd
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U, SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
/////////////////////////////////////////////////////////////////
// Ls vectorised, dag Kernel, single
/////////////////////////////////////////////////////////////////
@ -396,6 +592,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrd
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDag(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#define INTERIOR
#undef EXTERIOR
@ -408,6 +613,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagInt(StencilImpl &st,Lebesgue
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagInt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef INTERIOR_AND_EXTERIOR
#undef INTERIOR
#define EXTERIOR
@ -420,6 +634,15 @@ WilsonKernels<ZDomainWallVec5dImplD>::AsmDhopSiteDagExt(StencilImpl &st,Lebesgue
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<DomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
template<> void
WilsonKernels<ZDomainWallVec5dImplDF>::AsmDhopSiteDagExt(StencilImpl &st,LebesgueOrder & lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int ssU,int Ls,int Ns,const FermionField &in, FermionField &out)
#include <qcd/action/fermion/WilsonKernelsAsmBody.h>
#undef COMPLEX_SIGNS
#undef MAYBEPERM
#undef MULT_2SPIN

231
lib/qcd/action/fermion/WilsonKernelsAsmBody.h
View File

@ -39,24 +39,26 @@
////////////////////////////////////////////////////////////////////////////////
#ifdef INTERIOR_AND_EXTERIOR
#define ZERO_NMU(A)
#define INTERIOR_BLOCK_XP(a,b,PERMUTE_DIR,PROJMEM,RECON) INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)
#define EXTERIOR_BLOCK_XP(a,b,RECON) EXTERIOR_BLOCK(a,b,RECON)
#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
basep = st.GetPFInfo(nent,plocal); nent++; \
if ( local ) { \
LOAD64(%r10,isigns); \
PROJ(base); \
MAYBEPERM(PERMUTE_DIR,perm); \
} else { \
LOAD_CHI(base); \
} \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
MULT_2SPIN_DIR_PF(Dir,basep); \
LOAD64(%r10,isigns); \
RECON; \
#define INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON) \
LOAD64(%r10,isigns); \
PROJMEM(base); \
MAYBEPERM(PERMUTE_DIR,perm);
#define EXTERIOR_BLOCK(a,b,RECON) \
LOAD_CHI(base);
#define COMMON_BLOCK(a,b,RECON) \
base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
MULT_2SPIN_DIR_PF(a,basep); \
LOAD64(%r10,isigns); \
RECON;
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
PF_GAUGE(Xp); \
PREFETCH1_CHIMU(base); \
ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
#define RESULT(base,basep) SAVE_RESULT(base,basep);
@ -67,62 +69,62 @@
////////////////////////////////////////////////////////////////////////////////
#ifdef INTERIOR
#define COMMON_BLOCK(a,b,RECON)
#define ZERO_NMU(A)
#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
basep = st.GetPFInfo(nent,plocal); nent++; \
if ( local ) { \
LOAD64(%r10,isigns); \
PROJ(base); \
MAYBEPERM(PERMUTE_DIR,perm); \
}else if ( st.same_node[Dir] ) {LOAD_CHI(base);} \
if ( local || st.same_node[Dir] ) { \
MULT_2SPIN_DIR_PF(Dir,basep); \
LOAD64(%r10,isigns); \
RECON; \
} \
base = st.GetInfo(ptype,local,perm,NxtDir,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
// No accumulate for DIR0
#define EXTERIOR_BLOCK_XP(a,b,RECON) \
ZERO_PSI; \
base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;
#define EXTERIOR_BLOCK(a,b,RECON) \
base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;
#define INTERIOR_BLOCK_XP(a,b,PERMUTE_DIR,PROJMEM,RECON) INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON)
#define INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON) \
LOAD64(%r10,isigns); \
PROJMEM(base); \
MAYBEPERM(PERMUTE_DIR,perm); \
base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++; \
PREFETCH_CHIMU(base); \
MULT_2SPIN_DIR_PF(a,basep); \
LOAD64(%r10,isigns); \
RECON;
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
PF_GAUGE(Xp); \
PREFETCH1_CHIMU(base); \
{ ZERO_PSI; } \
ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON)
#define RESULT(base,basep) SAVE_RESULT(base,basep);
#endif
////////////////////////////////////////////////////////////////////////////////
// Post comms kernel
////////////////////////////////////////////////////////////////////////////////
#ifdef EXTERIOR
#define ZERO_NMU(A) nmu=0;
#define INTERIOR_BLOCK_XP(a,b,PERMUTE_DIR,PROJMEM,RECON) \
ZERO_PSI; base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;
#define ASM_LEG(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \
MULT_2SPIN_DIR_PF(Dir,base); \
LOAD64(%r10,isigns); \
RECON; \
nmu++; \
}
#define EXTERIOR_BLOCK_XP(a,b,RECON) EXTERIOR_BLOCK(a,b,RECON)
#define ASM_LEG_XP(Dir,NxtDir,PERMUTE_DIR,PROJ,RECON) \
nmu=0; \
{ ZERO_PSI;} \
base = st.GetInfo(ptype,local,perm,Dir,ent,plocal); ent++; \
if((!local)&&(!st.same_node[Dir]) ) { \
LOAD_CHI(base); \
MULT_2SPIN_DIR_PF(Dir,base); \
LOAD64(%r10,isigns); \
RECON; \
nmu++; \
}
#define INTERIOR_BLOCK(a,b,PERMUTE_DIR,PROJMEM,RECON) \
base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++;
#define EXTERIOR_BLOCK(a,b,RECON) \
nmu++; \
LOAD_CHI(base); \
MULT_2SPIN_DIR_PF(a,base); \
base = st.GetInfo(ptype,local,perm,b,ent,plocal); ent++; \
LOAD64(%r10,isigns); \
RECON;
#define COMMON_BLOCK(a,b,RECON)
#define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);}
#define RESULT(base,basep) if (nmu){ ADD_RESULT(base,base);}
#endif
{
int nmu;
int local,perm, ptype;
@ -134,11 +136,15 @@
MASK_REGS;
int nmax=U._grid->oSites();
for(int site=0;site<Ns;site++) {
#ifndef EXTERIOR
int sU =lo.Reorder(ssU);
int ssn=ssU+1; if(ssn>=nmax) ssn=0;
int sUn=lo.Reorder(ssn);
#ifndef EXTERIOR
LOCK_GAUGE(0);
#else
int sU =ssU;
int ssn=ssU+1; if(ssn>=nmax) ssn=0;
int sUn=ssn;
#endif
for(int s=0;s<Ls;s++) {
ss =sU*Ls+s;
@ -146,93 +152,20 @@
int ent=ss*8;// 2*Ndim
int nent=ssn*8;
ZERO_NMU(0);
base = st.GetInfo(ptype,local,perm,Xp,ent,plocal); ent++;
#ifndef EXTERIOR
PF_GAUGE(Xp);
PREFETCH1_CHIMU(base);
#endif
////////////////////////////////
// Xp
////////////////////////////////
basep = st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJMEM,DIR0_RECON);
} else {
EXTERIOR_BLOCK_XP(Xp,Yp,DIR0_RECON);
}
COMMON_BLOCK(Xp,Yp,DIR0_RECON);
////////////////////////////////
// Yp
////////////////////////////////
basep = st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Yp,Zp,PERMUTE_DIR2,DIR1_PROJMEM,DIR1_RECON);
} else {
EXTERIOR_BLOCK(Yp,Zp,DIR1_RECON);
}
COMMON_BLOCK(Yp,Zp,DIR1_RECON);
////////////////////////////////
// Zp
////////////////////////////////
basep = st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Zp,Tp,PERMUTE_DIR1,DIR2_PROJMEM,DIR2_RECON);
} else {
EXTERIOR_BLOCK(Zp,Tp,DIR2_RECON);
}
COMMON_BLOCK(Zp,Tp,DIR2_RECON);
////////////////////////////////
// Tp
////////////////////////////////
basep = st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Tp,Xm,PERMUTE_DIR0,DIR3_PROJMEM,DIR3_RECON);
} else {
EXTERIOR_BLOCK(Tp,Xm,DIR3_RECON);
}
COMMON_BLOCK(Tp,Xm,DIR3_RECON);
////////////////////////////////
// Xm
////////////////////////////////
// basep= st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Xm,Ym,PERMUTE_DIR3,DIR4_PROJMEM,DIR4_RECON);
} else {
EXTERIOR_BLOCK(Xm,Ym,DIR4_RECON);
}
COMMON_BLOCK(Xm,Ym,DIR4_RECON);
////////////////////////////////
// Ym
////////////////////////////////
basep= st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Ym,Zm,PERMUTE_DIR2,DIR5_PROJMEM,DIR5_RECON);
} else {
EXTERIOR_BLOCK(Ym,Zm,DIR5_RECON);
}
COMMON_BLOCK(Ym,Zm,DIR5_RECON);
////////////////////////////////
// Zm
////////////////////////////////
basep= st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Zm,Tm,PERMUTE_DIR1,DIR6_PROJMEM,DIR6_RECON);
} else {
EXTERIOR_BLOCK(Zm,Tm,DIR6_RECON);
}
COMMON_BLOCK(Zm,Tm,DIR6_RECON);
////////////////////////////////
// Tm
////////////////////////////////
basep= st.GetPFInfo(nent,plocal); nent++;
if ( local ) {
INTERIOR_BLOCK(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON);
} else {
EXTERIOR_BLOCK(Tm,Xp,DIR7_RECON);
}
COMMON_BLOCK(Tm,Xp,DIR7_RECON);
ASM_LEG_XP(Xp,Yp,PERMUTE_DIR3,DIR0_PROJMEM,DIR0_RECON);
ASM_LEG(Yp,Zp,PERMUTE_DIR2,DIR1_PROJMEM,DIR1_RECON);
ASM_LEG(Zp,Tp,PERMUTE_DIR1,DIR2_PROJMEM,DIR2_RECON);
ASM_LEG(Tp,Xm,PERMUTE_DIR0,DIR3_PROJMEM,DIR3_RECON);
ASM_LEG(Xm,Ym,PERMUTE_DIR3,DIR4_PROJMEM,DIR4_RECON);
ASM_LEG(Ym,Zm,PERMUTE_DIR2,DIR5_PROJMEM,DIR5_RECON);
ASM_LEG(Zm,Tm,PERMUTE_DIR1,DIR6_PROJMEM,DIR6_RECON);
ASM_LEG(Tm,Xp,PERMUTE_DIR0,DIR7_PROJMEM,DIR7_RECON);
#ifdef EXTERIOR
if (nmu==0) break;
// if (nmu!=0) std::cout << "EXT "<<sU<<std::endl;
#endif
base = (uint64_t) &out._odata[ss];
basep= st.GetPFInfo(nent,plocal); nent++;
RESULT(base,basep);
@ -258,10 +191,6 @@
#undef DIR5_RECON
#undef DIR6_RECON
#undef DIR7_RECON
#undef EXTERIOR_BLOCK
#undef INTERIOR_BLOCK
#undef EXTERIOR_BLOCK_XP
#undef INTERIOR_BLOCK_XP
#undef COMMON_BLOCK
#undef ZERO_NMU
#undef ASM_LEG
#undef ASM_LEG_XP
#undef RESULT

805
lib/qcd/action/fermion/WilsonKernelsHand.cc
View File

@ -31,7 +31,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#define REGISTER
#define LOAD_CHIMU \
const SiteSpinor & ref (in._odata[offset]); \
{const SiteSpinor & ref (in._odata[offset]); \
Chimu_00=ref()(0)(0);\
Chimu_01=ref()(0)(1);\
Chimu_02=ref()(0)(2);\
@ -43,20 +43,20 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
Chimu_22=ref()(2)(2);\
Chimu_30=ref()(3)(0);\
Chimu_31=ref()(3)(1);\
Chimu_32=ref()(3)(2);
Chimu_32=ref()(3)(2);}
#define LOAD_CHI\
const SiteHalfSpinor &ref(buf[offset]); \
{const SiteHalfSpinor &ref(buf[offset]); \
Chi_00 = ref()(0)(0);\
Chi_01 = ref()(0)(1);\
Chi_02 = ref()(0)(2);\
Chi_10 = ref()(1)(0);\
Chi_11 = ref()(1)(1);\
Chi_12 = ref()(1)(2);
Chi_12 = ref()(1)(2);}
// To splat or not to splat depends on the implementation
#define MULT_2SPIN(A)\
auto & ref(U._odata[sU](A)); \
{auto & ref(U._odata[sU](A)); \
Impl::loadLinkElement(U_00,ref()(0,0)); \
Impl::loadLinkElement(U_10,ref()(1,0)); \
Impl::loadLinkElement(U_20,ref()(2,0)); \
@ -83,7 +83,7 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
UChi_01+= U_10*Chi_02;\
UChi_11+= U_10*Chi_12;\
UChi_02+= U_20*Chi_02;\
UChi_12+= U_20*Chi_12;
UChi_12+= U_20*Chi_12;}
#define PERMUTE_DIR(dir) \
@ -307,55 +307,132 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
result_31-= UChi_11; \
result_32-= UChi_12;
namespace Grid {
namespace QCD {
#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if ( local ) { \
LOAD_CHIMU; \
PROJ; \
if ( perm) { \
PERMUTE_DIR(PERM); \
} \
} else { \
LOAD_CHI; \
} \
MULT_2SPIN(DIR); \
RECON;
#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
local = SE->_is_local; \
perm = SE->_permute; \
if ( local ) { \
LOAD_CHIMU; \
PROJ; \
if ( perm) { \
PERMUTE_DIR(PERM); \
} \
} else if ( st.same_node[DIR] ) { \
LOAD_CHI; \
} \
if (local || st.same_node[DIR] ) { \
MULT_2SPIN(DIR); \
RECON; \
}
#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON) \
SE=st.GetEntry(ptype,DIR,ss); \
offset = SE->_offset; \
if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \
LOAD_CHI; \
MULT_2SPIN(DIR); \
RECON; \
nmu++; \
}
#define HAND_RESULT(ss) \
{ \
SiteSpinor & ref (out._odata[ss]); \
vstream(ref()(0)(0),result_00); \
vstream(ref()(0)(1),result_01); \
vstream(ref()(0)(2),result_02); \
vstream(ref()(1)(0),result_10); \
vstream(ref()(1)(1),result_11); \
vstream(ref()(1)(2),result_12); \
vstream(ref()(2)(0),result_20); \
vstream(ref()(2)(1),result_21); \
vstream(ref()(2)(2),result_22); \
vstream(ref()(3)(0),result_30); \
vstream(ref()(3)(1),result_31); \
vstream(ref()(3)(2),result_32); \
}
#define HAND_RESULT_EXT(ss) \
if (nmu){ \
SiteSpinor & ref (out._odata[ss]); \
ref()(0)(0)+=result_00; \
ref()(0)(1)+=result_01; \
ref()(0)(2)+=result_02; \
ref()(1)(0)+=result_10; \
ref()(1)(1)+=result_11; \
ref()(1)(2)+=result_12; \
ref()(2)(0)+=result_20; \
ref()(2)(1)+=result_21; \
ref()(2)(2)+=result_22; \
ref()(3)(0)+=result_30; \
ref()(3)(1)+=result_31; \
ref()(3)(2)+=result_32; \
}
template<class Impl> void
WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
#define HAND_DECLARATIONS(a) \
Simd result_00; \
Simd result_01; \
Simd result_02; \
Simd result_10; \
Simd result_11; \
Simd result_12; \
Simd result_20; \
Simd result_21; \
Simd result_22; \
Simd result_30; \
Simd result_31; \
Simd result_32; \
Simd Chi_00; \
Simd Chi_01; \
Simd Chi_02; \
Simd Chi_10; \
Simd Chi_11; \
Simd Chi_12; \
Simd UChi_00; \
Simd UChi_01; \
Simd UChi_02; \
Simd UChi_10; \
Simd UChi_11; \
Simd UChi_12; \
Simd U_00; \
Simd U_10; \
Simd U_20; \
Simd U_01; \
Simd U_11; \
Simd U_21;
REGISTER Simd result_00; // 12 regs on knc
REGISTER Simd result_01;
REGISTER Simd result_02;
REGISTER Simd result_10;
REGISTER Simd result_11;
REGISTER Simd result_12;
REGISTER Simd result_20;
REGISTER Simd result_21;
REGISTER Simd result_22;
REGISTER Simd result_30;
REGISTER Simd result_31;
REGISTER Simd result_32; // 20 left
REGISTER Simd Chi_00; // two spinor; 6 regs
REGISTER Simd Chi_01;
REGISTER Simd Chi_02;
REGISTER Simd Chi_10;
REGISTER Simd Chi_11;
REGISTER Simd Chi_12; // 14 left
REGISTER Simd UChi_00; // two spinor; 6 regs
REGISTER Simd UChi_01;
REGISTER Simd UChi_02;
REGISTER Simd UChi_10;
REGISTER Simd UChi_11;
REGISTER Simd UChi_12; // 8 left
REGISTER Simd U_00; // two rows of U matrix
REGISTER Simd U_10;
REGISTER Simd U_20;
REGISTER Simd U_01;
REGISTER Simd U_11;
REGISTER Simd U_21; // 2 reg left.
#define ZERO_RESULT \
result_00=zero; \
result_01=zero; \
result_02=zero; \
result_10=zero; \
result_11=zero; \
result_12=zero; \
result_20=zero; \
result_21=zero; \
result_22=zero; \
result_30=zero; \
result_31=zero; \
result_32=zero;
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
@ -370,475 +447,225 @@ WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGauge
#define Chimu_31 UChi_11
#define Chimu_32 UChi_12
namespace Grid {
namespace QCD {
template<class Impl> void
WilsonKernels<Impl>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
int offset,local,perm, ptype;
StencilEntry *SE;
// Xp
SE=st.GetEntry(ptype,Xp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
XM_PROJ;
if ( perm) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Xp);
}
XM_RECON;
// Yp
SE=st.GetEntry(ptype,Yp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
YM_PROJ;
if ( perm) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Yp);
}
YM_RECON_ACCUM;
// Zp
SE=st.GetEntry(ptype,Zp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
ZM_PROJ;
if ( perm) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Zp);
}
ZM_RECON_ACCUM;
// Tp
SE=st.GetEntry(ptype,Tp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
TM_PROJ;
if ( perm) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Tp);
}
TM_RECON_ACCUM;
// Xm
SE=st.GetEntry(ptype,Xm,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
XP_PROJ;
if ( perm) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Xm);
}
XP_RECON_ACCUM;
// Ym
SE=st.GetEntry(ptype,Ym,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
YP_PROJ;
if ( perm) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Ym);
}
YP_RECON_ACCUM;
// Zm
SE=st.GetEntry(ptype,Zm,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
ZP_PROJ;
if ( perm) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Zm);
}
ZP_RECON_ACCUM;
// Tm
SE=st.GetEntry(ptype,Tm,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
TP_PROJ;
if ( perm) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Tm);
}
TP_RECON_ACCUM;
{
SiteSpinor & ref (out._odata[ss]);
vstream(ref()(0)(0),result_00);
vstream(ref()(0)(1),result_01);
vstream(ref()(0)(2),result_02);
vstream(ref()(1)(0),result_10);
vstream(ref()(1)(1),result_11);
vstream(ref()(1)(2),result_12);
vstream(ref()(2)(0),result_20);
vstream(ref()(2)(1),result_21);
vstream(ref()(2)(2),result_22);
vstream(ref()(3)(0),result_30);
vstream(ref()(3)(1),result_31);
vstream(ref()(3)(2),result_32);
}
HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON);
HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_RESULT(ss);
}
template<class Impl>
void WilsonKernels<Impl>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior)
int ss,int sU,const FermionField &in, FermionField &out)
{
// std::cout << "Hand op Dhop "<<std::endl;
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
REGISTER Simd result_00; // 12 regs on knc
REGISTER Simd result_01;
REGISTER Simd result_02;
REGISTER Simd result_10;
REGISTER Simd result_11;
REGISTER Simd result_12;
REGISTER Simd result_20;
REGISTER Simd result_21;
REGISTER Simd result_22;
REGISTER Simd result_30;
REGISTER Simd result_31;
REGISTER Simd result_32; // 20 left
REGISTER Simd Chi_00; // two spinor; 6 regs
REGISTER Simd Chi_01;
REGISTER Simd Chi_02;
REGISTER Simd Chi_10;
REGISTER Simd Chi_11;
REGISTER Simd Chi_12; // 14 left
REGISTER Simd UChi_00; // two spinor; 6 regs
REGISTER Simd UChi_01;
REGISTER Simd UChi_02;
REGISTER Simd UChi_10;
REGISTER Simd UChi_11;
REGISTER Simd UChi_12; // 8 left
REGISTER Simd U_00; // two rows of U matrix
REGISTER Simd U_10;
REGISTER Simd U_20;
REGISTER Simd U_01;
REGISTER Simd U_11;
REGISTER Simd U_21; // 2 reg left.
#define Chimu_00 Chi_00
#define Chimu_01 Chi_01
#define Chimu_02 Chi_02
#define Chimu_10 Chi_10
#define Chimu_11 Chi_11
#define Chimu_12 Chi_12
#define Chimu_20 UChi_00
#define Chimu_21 UChi_01
#define Chimu_22 UChi_02
#define Chimu_30 UChi_10
#define Chimu_31 UChi_11
#define Chimu_32 UChi_12
HAND_DECLARATIONS(ignore);
StencilEntry *SE;
int offset,local,perm, ptype;
// Xp
SE=st.GetEntry(ptype,Xp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
XP_PROJ;
if ( perm) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON);
HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_RESULT(ss);
}
{
MULT_2SPIN(Xp);
}
XP_RECON;
template<class Impl> void
WilsonKernels<Impl>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
// Yp
SE=st.GetEntry(ptype,Yp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
YP_PROJ;
if ( perm) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Yp);
}
YP_RECON_ACCUM;
HAND_DECLARATIONS(ignore);
int offset,local,perm, ptype;
StencilEntry *SE;
ZERO_RESULT;
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_RESULT(ss);
}
// Zp
SE=st.GetEntry(ptype,Zp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
ZP_PROJ;
if ( perm) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Zp);
}
ZP_RECON_ACCUM;
template<class Impl>
void WilsonKernels<Impl>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
// Tp
SE=st.GetEntry(ptype,Tp,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
TP_PROJ;
if ( perm) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Tp);
}
TP_RECON_ACCUM;
// Xm
SE=st.GetEntry(ptype,Xm,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
XM_PROJ;
if ( perm) {
PERMUTE_DIR(3); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Xm);
}
XM_RECON_ACCUM;
// Ym
SE=st.GetEntry(ptype,Ym,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
if ( local ) {
LOAD_CHIMU;
YM_PROJ;
if ( perm) {
PERMUTE_DIR(2); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Ym);
}
YM_RECON_ACCUM;
HAND_DECLARATIONS(ignore);
// Zm
SE=st.GetEntry(ptype,Zm,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
StencilEntry *SE;
int offset,local,perm, ptype;
ZERO_RESULT;
HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_RESULT(ss);
}
if ( local ) {
LOAD_CHIMU;
ZM_PROJ;
if ( perm) {
PERMUTE_DIR(1); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Zm);
}
ZM_RECON_ACCUM;
template<class Impl> void
WilsonKernels<Impl>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
// T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
// Tm
SE=st.GetEntry(ptype,Tm,ss);
offset = SE->_offset;
local = SE->_is_local;
perm = SE->_permute;
HAND_DECLARATIONS(ignore);
if ( local ) {
LOAD_CHIMU;
TM_PROJ;
if ( perm) {
PERMUTE_DIR(0); // T==0, Z==1, Y==2, Z==3 expect 1,2,2,2 simd layout etc...
}
} else {
LOAD_CHI;
}
{
MULT_2SPIN(Tm);
}
TM_RECON_ACCUM;
int offset,local,perm, ptype;
StencilEntry *SE;
int nmu=0;
ZERO_RESULT;
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM);
HAND_RESULT_EXT(ss);
}
{
SiteSpinor & ref (out._odata[ss]);
vstream(ref()(0)(0),result_00);
vstream(ref()(0)(1),result_01);
vstream(ref()(0)(2),result_02);
vstream(ref()(1)(0),result_10);
vstream(ref()(1)(1),result_11);
vstream(ref()(1)(2),result_12);
vstream(ref()(2)(0),result_20);
vstream(ref()(2)(1),result_21);
vstream(ref()(2)(2),result_22);
vstream(ref()(3)(0),result_30);
vstream(ref()(3)(1),result_31);
vstream(ref()(3)(2),result_32);
}
template<class Impl>
void WilsonKernels<Impl>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int ss,int sU,const FermionField &in, FermionField &out)
{
typedef typename Simd::scalar_type S;
typedef typename Simd::vector_type V;
HAND_DECLARATIONS(ignore);
StencilEntry *SE;
int offset,local,perm, ptype;
int nmu=0;
ZERO_RESULT;
HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM);
HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM);
HAND_RESULT_EXT(ss);
}
////////////////////////////////////////////////
// Specialise Gparity to simple implementation
////////////////////////////////////////////////
template<> void
WilsonKernels<GparityWilsonImplF>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
SiteHalfSpinor *buf,
int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
{
assert(0);
}
template<> void
WilsonKernels<GparityWilsonImplF>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,
SiteHalfSpinor *buf,
int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
{
assert(0);
}
template<> void
WilsonKernels<GparityWilsonImplD>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
{
assert(0);
}
template<> void
WilsonKernels<GparityWilsonImplD>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,
int sF,int sU,const FermionField &in, FermionField &out,int internal,int external)
{
assert(0);
}
#define HAND_SPECIALISE_EMPTY(IMPL) \
template<> void \
WilsonKernels<IMPL>::HandDhopSite(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteDag(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteInt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteExt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteDagInt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
template<> void \
WilsonKernels<IMPL>::HandDhopSiteDagExt(StencilImpl &st, \
LebesgueOrder &lo, \
DoubledGaugeField &U, \
SiteHalfSpinor *buf, \
int sF,int sU, \
const FermionField &in, \
FermionField &out){ assert(0); } \
HAND_SPECIALISE_EMPTY(GparityWilsonImplF);
HAND_SPECIALISE_EMPTY(GparityWilsonImplD);
HAND_SPECIALISE_EMPTY(GparityWilsonImplFH);
HAND_SPECIALISE_EMPTY(GparityWilsonImplDF);
////////////// Wilson ; uses this implementation /////////////////////
// Need Nc=3 though //
#define INSTANTIATE_THEM(A) \
template void WilsonKernels<A>::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior); \
template void WilsonKernels<A>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out,int interior,int exterior);
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out);\
template void WilsonKernels<A>::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf,\
int ss,int sU,const FermionField &in, FermionField &out); \
template void WilsonKernels<A>::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \
int ss,int sU,const FermionField &in, FermionField &out);
INSTANTIATE_THEM(WilsonImplF);
INSTANTIATE_THEM(WilsonImplD);
@ -850,5 +677,15 @@ INSTANTIATE_THEM(DomainWallVec5dImplF);
INSTANTIATE_THEM(DomainWallVec5dImplD);
INSTANTIATE_THEM(ZDomainWallVec5dImplF);
INSTANTIATE_THEM(ZDomainWallVec5dImplD);
INSTANTIATE_THEM(WilsonImplFH);
INSTANTIATE_THEM(WilsonImplDF);
INSTANTIATE_THEM(ZWilsonImplFH);
INSTANTIATE_THEM(ZWilsonImplDF);
INSTANTIATE_THEM(GparityWilsonImplFH);
INSTANTIATE_THEM(GparityWilsonImplDF);
INSTANTIATE_THEM(DomainWallVec5dImplFH);
INSTANTIATE_THEM(DomainWallVec5dImplDF);
INSTANTIATE_THEM(ZDomainWallVec5dImplFH);
INSTANTIATE_THEM(ZDomainWallVec5dImplDF);
}}

2
lib/qcd/action/gauge/Gauge.h
View File

@ -29,7 +29,7 @@ directory
#ifndef GRID_QCD_GAUGE_H
#define GRID_QCD_GAUGE_H
#include <Grid/qcd/action/gauge/GaugeImpl.h>
#include <Grid/qcd/action/gauge/GaugeImplementations.h>
#include <Grid/qcd/utils/WilsonLoops.h>
#include <Grid/qcd/action/gauge/WilsonGaugeAction.h>
#include <Grid/qcd/action/gauge/PlaqPlusRectangleAction.h>

153
lib/qcd/action/gauge/GaugeImplTypes.h Normal file
View File

@ -0,0 +1,153 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/GaugeImpl.h
Copyright (C) 2015
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_GAUGE_IMPL_TYPES_H
#define GRID_GAUGE_IMPL_TYPES_H
namespace Grid {
namespace QCD {
////////////////////////////////////////////////////////////////////////
// Implementation dependent gauge types
////////////////////////////////////////////////////////////////////////
#define INHERIT_GIMPL_TYPES(GImpl) \
typedef typename GImpl::Simd Simd; \
typedef typename GImpl::LinkField GaugeLinkField; \
typedef typename GImpl::Field GaugeField; \
typedef typename GImpl::ComplexField ComplexField;\
typedef typename GImpl::SiteField SiteGaugeField; \
typedef typename GImpl::SiteComplex SiteComplex; \
typedef typename GImpl::SiteLink SiteGaugeLink;
#define INHERIT_FIELD_TYPES(Impl) \
typedef typename Impl::Simd Simd; \
typedef typename Impl::ComplexField ComplexField; \
typedef typename Impl::SiteField SiteField; \
typedef typename Impl::Field Field;
// hardcodes the exponential approximation in the template
template <class S, int Nrepresentation = Nc, int Nexp = 12 > class GaugeImplTypes {
public:
typedef S Simd;
template <typename vtype> using iImplScalar = iScalar<iScalar<iScalar<vtype> > >;
template <typename vtype> using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation> > >;
template <typename vtype> using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation> >, Nd>;
typedef iImplScalar<Simd> SiteComplex;
typedef iImplGaugeLink<Simd> SiteLink;
typedef iImplGaugeField<Simd> SiteField;
typedef Lattice<SiteComplex> ComplexField;
typedef Lattice<SiteLink> LinkField;
typedef Lattice<SiteField> Field;
// Guido: we can probably separate the types from the HMC functions
// this will create 2 kind of implementations
// probably confusing the users
// Now keeping only one class
// Move this elsewhere? FIXME
static inline void AddLink(Field &U, LinkField &W,
int mu) { // U[mu] += W
PARALLEL_FOR_LOOP
for (auto ss = 0; ss < U._grid->oSites(); ss++) {
U._odata[ss]._internal[mu] =
U._odata[ss]._internal[mu] + W._odata[ss]._internal;
}
}
///////////////////////////////////////////////////////////
// Move these to another class
// HMC auxiliary functions
static inline void generate_momenta(Field &P, GridParallelRNG &pRNG) {
// specific for SU gauge fields
LinkField Pmu(P._grid);
Pmu = zero;
for (int mu = 0; mu < Nd; mu++) {
SU<Nrepresentation>::GaussianFundamentalLieAlgebraMatrix(pRNG, Pmu);
PokeIndex<LorentzIndex>(P, Pmu, mu);
}
}
static inline Field projectForce(Field &P) { return Ta(P); }
static inline void update_field(Field& P, Field& U, double ep){
//static std::chrono::duration<double> diff;
//auto start = std::chrono::high_resolution_clock::now();
parallel_for(int ss=0;ss<P._grid->oSites();ss++){
for (int mu = 0; mu < Nd; mu++)
U[ss]._internal[mu] = ProjectOnGroup(Exponentiate(P[ss]._internal[mu], ep, Nexp) * U[ss]._internal[mu]);
}
//auto end = std::chrono::high_resolution_clock::now();
// diff += end - start;
// std::cout << "Time to exponentiate matrix " << diff.count() << " s\n";
}
static inline RealD FieldSquareNorm(Field& U){
LatticeComplex Hloc(U._grid);
Hloc = zero;
for (int mu = 0; mu < Nd; mu++) {
auto Umu = PeekIndex<LorentzIndex>(U, mu);
Hloc += trace(Umu * Umu);
}
Complex Hsum = sum(Hloc);
return Hsum.real();
}
static inline void HotConfiguration(GridParallelRNG &pRNG, Field &U) {
SU<Nc>::HotConfiguration(pRNG, U);
}
static inline void TepidConfiguration(GridParallelRNG &pRNG, Field &U) {
SU<Nc>::TepidConfiguration(pRNG, U);
}
static inline void ColdConfiguration(GridParallelRNG &pRNG, Field &U) {
SU<Nc>::ColdConfiguration(pRNG, U);
}
};
typedef GaugeImplTypes<vComplex, Nc> GimplTypesR;
typedef GaugeImplTypes<vComplexF, Nc> GimplTypesF;
typedef GaugeImplTypes<vComplexD, Nc> GimplTypesD;
typedef GaugeImplTypes<vComplex, SU<Nc>::AdjointDimension> GimplAdjointTypesR;
typedef GaugeImplTypes<vComplexF, SU<Nc>::AdjointDimension> GimplAdjointTypesF;
typedef GaugeImplTypes<vComplexD, SU<Nc>::AdjointDimension> GimplAdjointTypesD;
} // QCD
} // Grid
#endif // GRID_GAUGE_IMPL_TYPES_H

59
lib/qcd/action/gauge/GaugeImpl.h → lib/qcd/action/gauge/GaugeImplementations.h
View File

@ -2,7 +2,7 @@
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/GaugeImpl.h
Source file: ./lib/qcd/action/gauge/GaugeImplementations.h
Copyright (C) 2015
@ -26,53 +26,14 @@ See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_QCD_GAUGE_IMPL_H
#define GRID_QCD_GAUGE_IMPL_H
#ifndef GRID_QCD_GAUGE_IMPLEMENTATIONS_H
#define GRID_QCD_GAUGE_IMPLEMENTATIONS_H
#include "GaugeImplTypes.h"
namespace Grid {
namespace QCD {
////////////////////////////////////////////////////////////////////////
// Implementation dependent gauge types
////////////////////////////////////////////////////////////////////////
template <class Gimpl> class WilsonLoops;
#define INHERIT_GIMPL_TYPES(GImpl) \
typedef typename GImpl::Simd Simd; \
typedef typename GImpl::GaugeLinkField GaugeLinkField; \
typedef typename GImpl::GaugeField GaugeField; \
typedef typename GImpl::SiteGaugeField SiteGaugeField; \
typedef typename GImpl::SiteGaugeLink SiteGaugeLink;
//
template <class S, int Nrepresentation = Nc> class GaugeImplTypes {
public:
typedef S Simd;
template <typename vtype>
using iImplGaugeLink = iScalar<iScalar<iMatrix<vtype, Nrepresentation>>>;
template <typename vtype>
using iImplGaugeField = iVector<iScalar<iMatrix<vtype, Nrepresentation>>, Nd>;
typedef iImplGaugeLink<Simd> SiteGaugeLink;
typedef iImplGaugeField<Simd> SiteGaugeField;
typedef Lattice<SiteGaugeLink> GaugeLinkField; // bit ugly naming; polarised
// gauge field, lorentz... all
// ugly
typedef Lattice<SiteGaugeField> GaugeField;
// Move this elsewhere? FIXME
static inline void AddGaugeLink(GaugeField &U, GaugeLinkField &W,
int mu) { // U[mu] += W
parallel_for (auto ss = 0; ss < U._grid->oSites(); ss++) {
U._odata[ss]._internal[mu] =
U._odata[ss]._internal[mu] + W._odata[ss]._internal;
}
}
};
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
template <class GimplTypes> class PeriodicGaugeImpl : public GimplTypes {
@ -168,14 +129,6 @@ public:
static inline bool isPeriodicGaugeField(void) { return false; }
};
typedef GaugeImplTypes<vComplex, Nc> GimplTypesR;
typedef GaugeImplTypes<vComplexF, Nc> GimplTypesF;
typedef GaugeImplTypes<vComplexD, Nc> GimplTypesD;
typedef GaugeImplTypes<vComplex, SU<Nc>::AdjointDimension> GimplAdjointTypesR;
typedef GaugeImplTypes<vComplexF, SU<Nc>::AdjointDimension> GimplAdjointTypesF;
typedef GaugeImplTypes<vComplexD, SU<Nc>::AdjointDimension> GimplAdjointTypesD;
typedef PeriodicGaugeImpl<GimplTypesR> PeriodicGimplR; // Real.. whichever prec
typedef PeriodicGaugeImpl<GimplTypesF> PeriodicGimplF; // Float
typedef PeriodicGaugeImpl<GimplTypesD> PeriodicGimplD; // Double
@ -187,6 +140,8 @@ typedef PeriodicGaugeImpl<GimplAdjointTypesD> PeriodicGimplAdjD; // Double
typedef ConjugateGaugeImpl<GimplTypesR> ConjugateGimplR; // Real.. whichever prec
typedef ConjugateGaugeImpl<GimplTypesF> ConjugateGimplF; // Float
typedef ConjugateGaugeImpl<GimplTypesD> ConjugateGimplD; // Double
}
}

286
lib/qcd/action/gauge/Photon.h Normal file
View File

@ -0,0 +1,286 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/Photon.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 QCD_PHOTON_ACTION_H
#define QCD_PHOTON_ACTION_H
namespace Grid{
namespace QCD{
template <class S>
class QedGimpl
{
public:
typedef S Simd;
template <typename vtype>
using iImplGaugeLink = iScalar<iScalar<iScalar<vtype>>>;
template <typename vtype>
using iImplGaugeField = iVector<iScalar<iScalar<vtype>>, Nd>;
typedef iImplGaugeLink<Simd> SiteLink;
typedef iImplGaugeField<Simd> SiteField;
typedef SiteField SiteComplex;
typedef Lattice<SiteLink> LinkField;
typedef Lattice<SiteField> Field;
typedef Field ComplexField;
};
typedef QedGimpl<vComplex> QedGimplR;
template<class Gimpl>
class Photon
{
public:
INHERIT_GIMPL_TYPES(Gimpl);
GRID_SERIALIZABLE_ENUM(Gauge, undef, feynman, 1, coulomb, 2, landau, 3);
GRID_SERIALIZABLE_ENUM(ZmScheme, undef, qedL, 1, qedTL, 2);
public:
Photon(Gauge gauge, ZmScheme zmScheme);
virtual ~Photon(void) = default;
void FreePropagator(const GaugeField &in, GaugeField &out);
void MomentumSpacePropagator(const GaugeField &in, GaugeField &out);
void StochasticWeight(GaugeLinkField &weight);
void StochasticField(GaugeField &out, GridParallelRNG &rng);
void StochasticField(GaugeField &out, GridParallelRNG &rng,
const GaugeLinkField &weight);
private:
void invKHatSquared(GaugeLinkField &out);
void zmSub(GaugeLinkField &out);
private:
Gauge gauge_;
ZmScheme zmScheme_;
};
typedef Photon<QedGimplR> PhotonR;
template<class Gimpl>
Photon<Gimpl>::Photon(Gauge gauge, ZmScheme zmScheme)
: gauge_(gauge), zmScheme_(zmScheme)
{}
template<class Gimpl>
void Photon<Gimpl>::FreePropagator (const GaugeField &in,GaugeField &out)
{
FFT theFFT(in._grid);
GaugeField in_k(in._grid);
GaugeField prop_k(in._grid);
theFFT.FFT_all_dim(in_k,in,FFT::forward);
MomentumSpacePropagator(prop_k,in_k);
theFFT.FFT_all_dim(out,prop_k,FFT::backward);
}
template<class Gimpl>
void Photon<Gimpl>::invKHatSquared(GaugeLinkField &out)
{
GridBase *grid = out._grid;
GaugeLinkField kmu(grid), one(grid);
const unsigned int nd = grid->_ndimension;
std::vector<int> &l = grid->_fdimensions;
std::vector<int> zm(nd,0);
TComplex Tone = Complex(1.0,0.0);
TComplex Tzero= Complex(0.0,0.0);
one = Complex(1.0,0.0);
out = zero;
for(int mu = 0; mu < nd; mu++)
{
Real twoPiL = M_PI*2./l[mu];
LatticeCoordinate(kmu,mu);
kmu = 2.*sin(.5*twoPiL*kmu);
out = out + kmu*kmu;
}
pokeSite(Tone, out, zm);
out = one/out;
pokeSite(Tzero, out, zm);
}
template<class Gimpl>
void Photon<Gimpl>::zmSub(GaugeLinkField &out)
{
GridBase *grid = out._grid;
const unsigned int nd = grid->_ndimension;
switch (zmScheme_)
{
case ZmScheme::qedTL:
{
std::vector<int> zm(nd,0);
TComplex Tzero = Complex(0.0,0.0);
pokeSite(Tzero, out, zm);
break;
}
case ZmScheme::qedL:
{
LatticeInteger spNrm(grid), coor(grid);
GaugeLinkField z(grid);
spNrm = zero;
for(int d = 0; d < grid->_ndimension - 1; d++)
{
LatticeCoordinate(coor,d);
spNrm = spNrm + coor*coor;
}
out = where(spNrm == Integer(0), 0.*out, out);
break;
}
default:
break;
}
}
template<class Gimpl>
void Photon<Gimpl>::MomentumSpacePropagator(const GaugeField &in,
GaugeField &out)
{
GridBase *grid = out._grid;
LatticeComplex k2Inv(grid);
invKHatSquared(k2Inv);
zmSub(k2Inv);
out = in*k2Inv;
}
template<class Gimpl>
void Photon<Gimpl>::StochasticWeight(GaugeLinkField &weight)
{
auto *grid = dynamic_cast<GridCartesian *>(weight._grid);
const unsigned int nd = grid->_ndimension;
std::vector<int> latt_size = grid->_fdimensions;
Integer vol = 1;
for(int d = 0; d < nd; d++)
{
vol = vol * latt_size[d];
}
invKHatSquared(weight);
weight = sqrt(vol*real(weight));
zmSub(weight);
}
template<class Gimpl>
void Photon<Gimpl>::StochasticField(GaugeField &out, GridParallelRNG &rng)
{
auto *grid = dynamic_cast<GridCartesian *>(out._grid);
GaugeLinkField weight(grid);
StochasticWeight(weight);
StochasticField(out, rng, weight);
}
template<class Gimpl>
void Photon<Gimpl>::StochasticField(GaugeField &out, GridParallelRNG &rng,
const GaugeLinkField &weight)
{
auto *grid = dynamic_cast<GridCartesian *>(out._grid);
const unsigned int nd = grid->_ndimension;
GaugeLinkField r(grid);
GaugeField aTilde(grid);
FFT fft(grid);
for(int mu = 0; mu < nd; mu++)
{
gaussian(rng, r);
r = weight*r;
pokeLorentz(aTilde, r, mu);
}
fft.FFT_all_dim(out, aTilde, FFT::backward);
out = real(out);
}
// template<class Gimpl>
// void Photon<Gimpl>::FeynmanGaugeMomentumSpacePropagator_L(GaugeField &out,
// const GaugeField &in)
// {
//
// FeynmanGaugeMomentumSpacePropagator_TL(out,in);
//
// GridBase *grid = out._grid;
// LatticeInteger coor(grid);
// GaugeField zz(grid); zz=zero;
//
// // xyzt
// for(int d = 0; d < grid->_ndimension-1;d++){
// LatticeCoordinate(coor,d);
// out = where(coor==Integer(0),zz,out);
// }
// }
//
// template<class Gimpl>
// void Photon<Gimpl>::FeynmanGaugeMomentumSpacePropagator_TL(GaugeField &out,
// const GaugeField &in)
// {
//
// // what type LatticeComplex
// GridBase *grid = out._grid;
// int nd = grid->_ndimension;
//
// typedef typename GaugeField::vector_type vector_type;
// typedef typename GaugeField::scalar_type ScalComplex;
// typedef Lattice<iSinglet<vector_type> > LatComplex;
//
// std::vector<int> latt_size = grid->_fdimensions;
//
// LatComplex denom(grid); denom= zero;
// LatComplex one(grid); one = ScalComplex(1.0,0.0);
// LatComplex kmu(grid);
//
// ScalComplex ci(0.0,1.0);
// // momphase = n * 2pi / L
// for(int mu=0;mu<Nd;mu++) {
//
// LatticeCoordinate(kmu,mu);
//
// RealD TwoPiL = M_PI * 2.0/ latt_size[mu];
//
// kmu = TwoPiL * kmu ;
//
// denom = denom + 4.0*sin(kmu*0.5)*sin(kmu*0.5); // Wilson term
// }
// std::vector<int> zero_mode(nd,0);
// TComplexD Tone = ComplexD(1.0,0.0);
// TComplexD Tzero= ComplexD(0.0,0.0);
//
// pokeSite(Tone,denom,zero_mode);
//
// denom= one/denom;
//
// pokeSite(Tzero,denom,zero_mode);
//
// out = zero;
// out = in*denom;
// };
}}
#endif

26
lib/qcd/action/gauge/PlaqPlusRectangleAction.h
View File

@ -47,9 +47,19 @@ namespace Grid{
public:
PlaqPlusRectangleAction(RealD b,RealD c): c_plaq(b),c_rect(c){};
virtual std::string action_name(){return "PlaqPlusRectangleAction";}
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {}; // noop as no pseudoferms
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name() <<"] c_plaq: " << c_plaq << std::endl;
sstream << GridLogMessage << "["<<action_name() <<"] c_rect: " << c_rect << std::endl;
return sstream.str();
}
virtual RealD S(const GaugeField &U) {
RealD vol = U._grid->gSites();
@ -108,32 +118,32 @@ namespace Grid{
class RBCGaugeAction : public PlaqPlusRectangleAction<Gimpl> {
public:
INHERIT_GIMPL_TYPES(Gimpl);
RBCGaugeAction(RealD beta,RealD c1) : PlaqPlusRectangleAction<Gimpl>(beta*(1.0-8.0*c1), beta*c1) {
};
RBCGaugeAction(RealD beta,RealD c1) : PlaqPlusRectangleAction<Gimpl>(beta*(1.0-8.0*c1), beta*c1) {};
virtual std::string action_name(){return "RBCGaugeAction";}
};
template<class Gimpl>
class IwasakiGaugeAction : public RBCGaugeAction<Gimpl> {
public:
INHERIT_GIMPL_TYPES(Gimpl);
IwasakiGaugeAction(RealD beta) : RBCGaugeAction<Gimpl>(beta,-0.331) {
};
IwasakiGaugeAction(RealD beta) : RBCGaugeAction<Gimpl>(beta,-0.331) {};
virtual std::string action_name(){return "IwasakiGaugeAction";}
};
template<class Gimpl>
class SymanzikGaugeAction : public RBCGaugeAction<Gimpl> {
public:
INHERIT_GIMPL_TYPES(Gimpl);
SymanzikGaugeAction(RealD beta) : RBCGaugeAction<Gimpl>(beta,-1.0/12.0) {
};
SymanzikGaugeAction(RealD beta) : RBCGaugeAction<Gimpl>(beta,-1.0/12.0) {};
virtual std::string action_name(){return "SymanzikGaugeAction";}
};
template<class Gimpl>
class DBW2GaugeAction : public RBCGaugeAction<Gimpl> {
public:
INHERIT_GIMPL_TYPES(Gimpl);
DBW2GaugeAction(RealD beta) : RBCGaugeAction<Gimpl>(beta,-1.4067) {
};
DBW2GaugeAction(RealD beta) : RBCGaugeAction<Gimpl>(beta,-1.4067) {};
virtual std::string action_name(){return "DBW2GaugeAction";}
};
}

139
lib/qcd/action/gauge/WilsonGaugeAction.h
View File

@ -1,86 +1,99 @@
/*************************************************************************************
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/WilsonGaugeAction.h
Source file: ./lib/qcd/action/gauge/WilsonGaugeAction.h
Copyright (C) 2015
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: Guido Cossu <guido.cossu@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 free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef QCD_WILSON_GAUGE_ACTION_H
#define QCD_WILSON_GAUGE_ACTION_H
namespace Grid{
namespace QCD{
////////////////////////////////////////////////////////////////////////
// Wilson Gauge Action .. should I template the Nc etc..
////////////////////////////////////////////////////////////////////////
template<class Gimpl>
class WilsonGaugeAction : public Action<typename Gimpl::GaugeField> {
public:
namespace Grid {
namespace QCD {
INHERIT_GIMPL_TYPES(Gimpl);
////////////////////////////////////////////////////////////////////////
// Wilson Gauge Action .. should I template the Nc etc..
////////////////////////////////////////////////////////////////////////
template <class Gimpl>
class WilsonGaugeAction : public Action<typename Gimpl::GaugeField> {
public:
INHERIT_GIMPL_TYPES(Gimpl);
// typedef LorentzScalar<GaugeField> GaugeLinkField;
/////////////////////////// constructors
explicit WilsonGaugeAction(RealD beta_):beta(beta_){};
private:
RealD beta;
public:
WilsonGaugeAction(RealD b):beta(b){};
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {}; // noop as no pseudoferms
virtual RealD S(const GaugeField &U) {
RealD plaq = WilsonLoops<Gimpl>::avgPlaquette(U);
RealD vol = U._grid->gSites();
RealD action=beta*(1.0 -plaq)*(Nd*(Nd-1.0))*vol*0.5;
return action;
};
virtual std::string action_name() {return "WilsonGaugeAction";}
virtual void deriv(const GaugeField &U,GaugeField & dSdU) {
//not optimal implementation FIXME
//extend Ta to include Lorentz indexes
//RealD factor = 0.5*beta/RealD(Nc);
RealD factor = 0.5*beta/RealD(Nc);
GaugeLinkField Umu(U._grid);
GaugeLinkField dSdU_mu(U._grid);
for (int mu=0; mu < Nd; mu++){
Umu = PeekIndex<LorentzIndex>(U,mu);
// Staple in direction mu
WilsonLoops<Gimpl>::Staple(dSdU_mu,U,mu);
dSdU_mu = Ta(Umu*dSdU_mu)*factor;
PokeIndex<LorentzIndex>(dSdU, dSdU_mu, mu);
}
};
};
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "[WilsonGaugeAction] Beta: " << beta << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U,
GridParallelRNG &pRNG){}; // noop as no pseudoferms
virtual RealD S(const GaugeField &U) {
RealD plaq = WilsonLoops<Gimpl>::avgPlaquette(U);
RealD vol = U._grid->gSites();
RealD action = beta * (1.0 - plaq) * (Nd * (Nd - 1.0)) * vol * 0.5;
return action;
};
virtual void deriv(const GaugeField &U, GaugeField &dSdU) {
// not optimal implementation FIXME
// extend Ta to include Lorentz indexes
RealD factor = 0.5 * beta / RealD(Nc);
//GaugeLinkField Umu(U._grid);
GaugeLinkField dSdU_mu(U._grid);
for (int mu = 0; mu < Nd; mu++) {
//Umu = PeekIndex<LorentzIndex>(U, mu);
// Staple in direction mu
//WilsonLoops<Gimpl>::Staple(dSdU_mu, U, mu);
//dSdU_mu = Ta(Umu * dSdU_mu) * factor;
WilsonLoops<Gimpl>::StapleMult(dSdU_mu, U, mu);
dSdU_mu = Ta(dSdU_mu) * factor;
PokeIndex<LorentzIndex>(dSdU, dSdU_mu, mu);
}
}
private:
RealD beta;
};
}
}
#endif

138
lib/qcd/action/pseudofermion/EvenOddSchurDifferentiable.h
View File

@ -7,6 +7,7 @@
Copyright (C) 2015
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@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
@ -45,92 +46,97 @@ namespace Grid{
public:
INHERIT_IMPL_TYPES(Impl);
typedef FermionOperator<Impl> Matrix;
typedef FermionOperator<Impl> Matrix;
SchurDifferentiableOperator (Matrix &Mat) : SchurDiagMooeeOperator<Matrix,FermionField>(Mat) {};
SchurDifferentiableOperator (Matrix &Mat) : SchurDiagMooeeOperator<Matrix,FermionField>(Mat) {};
void MpcDeriv(GaugeField &Force,const FermionField &U,const FermionField &V) {
GridBase *fgrid = this->_Mat.FermionGrid();
GridBase *fcbgrid = this->_Mat.FermionRedBlackGrid();
GridBase *ugrid = this->_Mat.GaugeGrid();
GridBase *ucbgrid = this->_Mat.GaugeRedBlackGrid();
void MpcDeriv(GaugeField &Force,const FermionField &U,const FermionField &V) {
GridBase *fgrid = this->_Mat.FermionGrid();
GridBase *fcbgrid = this->_Mat.FermionRedBlackGrid();
Real coeff = 1.0;
FermionField tmp1(fcbgrid);
FermionField tmp2(fcbgrid);
FermionField tmp1(fcbgrid);
FermionField tmp2(fcbgrid);
conformable(fcbgrid,U._grid);
conformable(fcbgrid,V._grid);
conformable(fcbgrid,U._grid);
conformable(fcbgrid,V._grid);
// Assert the checkerboard?? or code for either
assert(U.checkerboard==Odd);
assert(V.checkerboard==U.checkerboard);
// Assert the checkerboard?? or code for either
assert(U.checkerboard==Odd);
assert(V.checkerboard==U.checkerboard);
// NOTE Guido: WE DO NOT WANT TO USE THE ucbgrid GRID FOR THE FORCE
// it is not conformable with the HMC force field
// Case: Ls vectorised fields
// INHERIT FROM THE Force field instead
GridRedBlackCartesian* forcecb = new GridRedBlackCartesian(Force._grid);
GaugeField ForceO(forcecb);
GaugeField ForceE(forcecb);
GaugeField ForceO(ucbgrid);
GaugeField ForceE(ucbgrid);
// X^dag Der_oe MeeInv Meo Y
// Use Mooee as nontrivial but gauge field indept
this->_Mat.Meooe (V,tmp1); // odd->even -- implicit -0.5 factor to be applied
// X^dag Der_oe MeeInv Meo Y
// Use Mooee as nontrivial but gauge field indept
this->_Mat.Meooe (V,tmp1); // odd->even -- implicit -0.5 factor to be applied
this->_Mat.MooeeInv(tmp1,tmp2); // even->even
this->_Mat.MoeDeriv(ForceO,U,tmp2,DaggerNo);
// Accumulate X^dag M_oe MeeInv Der_eo Y
this->_Mat.MeooeDag (U,tmp1); // even->odd -- implicit -0.5 factor to be applied
this->_Mat.MooeeInvDag(tmp1,tmp2); // even->even
this->_Mat.MeoDeriv(ForceE,tmp2,V,DaggerNo);
assert(ForceE.checkerboard==Even);
assert(ForceO.checkerboard==Odd);
this->_Mat.MoeDeriv(ForceO,U,tmp2,DaggerNo);
// Accumulate X^dag M_oe MeeInv Der_eo Y
this->_Mat.MeooeDag (U,tmp1); // even->odd -- implicit -0.5 factor to be applied
this->_Mat.MooeeInvDag(tmp1,tmp2); // even->even
this->_Mat.MeoDeriv(ForceE,tmp2,V,DaggerNo);
assert(ForceE.checkerboard==Even);
assert(ForceO.checkerboard==Odd);
setCheckerboard(Force,ForceE);
setCheckerboard(Force,ForceO);
Force=-Force;
}
setCheckerboard(Force,ForceE);
setCheckerboard(Force,ForceO);
Force=-Force;
delete forcecb;
}
void MpcDagDeriv(GaugeField &Force,const FermionField &U,const FermionField &V) {
GridBase *fgrid = this->_Mat.FermionGrid();
GridBase *fcbgrid = this->_Mat.FermionRedBlackGrid();
GridBase *ugrid = this->_Mat.GaugeGrid();
GridBase *ucbgrid = this->_Mat.GaugeRedBlackGrid();
void MpcDagDeriv(GaugeField &Force,const FermionField &U,const FermionField &V) {
GridBase *fgrid = this->_Mat.FermionGrid();
GridBase *fcbgrid = this->_Mat.FermionRedBlackGrid();
Real coeff = 1.0;
FermionField tmp1(fcbgrid);
FermionField tmp2(fcbgrid);
FermionField tmp1(fcbgrid);
FermionField tmp2(fcbgrid);
conformable(fcbgrid,U._grid);
conformable(fcbgrid,V._grid);
conformable(fcbgrid,U._grid);
conformable(fcbgrid,V._grid);
// Assert the checkerboard?? or code for either
assert(V.checkerboard==Odd);
assert(V.checkerboard==V.checkerboard);
// Assert the checkerboard?? or code for either
assert(V.checkerboard==Odd);
assert(V.checkerboard==V.checkerboard);
// NOTE Guido: WE DO NOT WANT TO USE THE ucbgrid GRID FOR THE FORCE
// it is not conformable with the HMC force field
// INHERIT FROM THE Force field instead
GridRedBlackCartesian* forcecb = new GridRedBlackCartesian(Force._grid);
GaugeField ForceO(forcecb);
GaugeField ForceE(forcecb);
GaugeField ForceO(ucbgrid);
GaugeField ForceE(ucbgrid);
// X^dag Der_oe MeeInv Meo Y
// Use Mooee as nontrivial but gauge field indept
this->_Mat.MeooeDag (V,tmp1); // odd->even -- implicit -0.5 factor to be applied
this->_Mat.MooeeInvDag(tmp1,tmp2); // even->even
this->_Mat.MoeDeriv(ForceO,U,tmp2,DaggerYes);
// Accumulate X^dag M_oe MeeInv Der_eo Y
this->_Mat.Meooe (U,tmp1); // even->odd -- implicit -0.5 factor to be applied
this->_Mat.MooeeInv(tmp1,tmp2); // even->even
this->_Mat.MeoDeriv(ForceE,tmp2,V,DaggerYes);
// X^dag Der_oe MeeInv Meo Y
// Use Mooee as nontrivial but gauge field indept
this->_Mat.MeooeDag (V,tmp1); // odd->even -- implicit -0.5 factor to be applied
this->_Mat.MooeeInvDag(tmp1,tmp2); // even->even
this->_Mat.MoeDeriv(ForceO,U,tmp2,DaggerYes);
// Accumulate X^dag M_oe MeeInv Der_eo Y
this->_Mat.Meooe (U,tmp1); // even->odd -- implicit -0.5 factor to be applied
this->_Mat.MooeeInv(tmp1,tmp2); // even->even
this->_Mat.MeoDeriv(ForceE,tmp2,V,DaggerYes);
assert(ForceE.checkerboard==Even);
assert(ForceO.checkerboard==Odd);
assert(ForceE.checkerboard==Even);
assert(ForceO.checkerboard==Odd);
setCheckerboard(Force,ForceE);
setCheckerboard(Force,ForceO);
Force=-Force;
setCheckerboard(Force,ForceE);
setCheckerboard(Force,ForceO);
Force=-Force;
}
delete forcecb;
}
};

14
lib/qcd/action/pseudofermion/OneFlavourEvenOddRational.h
View File

@ -1,3 +1,4 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
@ -90,6 +91,19 @@ class OneFlavourEvenOddRationalPseudoFermionAction
PowerNegQuarter.Init(remez, param.tolerance, true);
};
virtual std::string action_name(){return "OneFlavourEvenOddRationalPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] Low :" << param.lo << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] High :" << param.hi << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Max iterations :" << param.MaxIter << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Tolerance :" << param.tolerance << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Degree :" << param.degree << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Precision :" << param.precision << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U, GridParallelRNG &pRNG) {
// P(phi) = e^{- phi^dag (MpcdagMpc)^-1/2 phi}
// = e^{- phi^dag (MpcdagMpc)^-1/4 (MpcdagMpc)^-1/4 phi}

14
lib/qcd/action/pseudofermion/OneFlavourEvenOddRationalRatio.h
View File

@ -87,6 +87,20 @@ namespace Grid{
PowerQuarter.Init(remez,param.tolerance,false);
PowerNegQuarter.Init(remez,param.tolerance,true);
};
virtual std::string action_name(){return "OneFlavourEvenOddRatioRationalPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] Low :" << param.lo << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] High :" << param.hi << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Max iterations :" << param.MaxIter << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Tolerance :" << param.tolerance << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Degree :" << param.degree << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Precision :" << param.precision << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {

16
lib/qcd/action/pseudofermion/OneFlavourRational.h
View File

@ -83,9 +83,25 @@ namespace Grid{
PowerQuarter.Init(remez,param.tolerance,false);
PowerNegQuarter.Init(remez,param.tolerance,true);
};
virtual std::string action_name(){return "OneFlavourRationalPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] Low :" << param.lo << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] High :" << param.hi << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Max iterations :" << param.MaxIter << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Tolerance :" << param.tolerance << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Degree :" << param.degree << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Precision :" << param.precision << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {
// P(phi) = e^{- phi^dag (MdagM)^-1/2 phi}
// = e^{- phi^dag (MdagM)^-1/4 (MdagM)^-1/4 phi}
// Phi = Mdag^{1/4} eta

14
lib/qcd/action/pseudofermion/OneFlavourRationalRatio.h
View File

@ -81,7 +81,21 @@ namespace Grid{
PowerQuarter.Init(remez,param.tolerance,false);
PowerNegQuarter.Init(remez,param.tolerance,true);
};
virtual std::string action_name(){return "OneFlavourRatioRationalPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] Low :" << param.lo << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] High :" << param.hi << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Max iterations :" << param.MaxIter << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Tolerance :" << param.tolerance << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Degree :" << param.degree << std::endl;
sstream << GridLogMessage << "["<<action_name()<<"] Precision :" << param.precision << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {
// S_f = chi^dag* P(V^dag*V)/Q(V^dag*V)* N(M^dag*M)/D(M^dag*M)* P(V^dag*V)/Q(V^dag*V)* chi

11
lib/qcd/action/pseudofermion/TwoFlavour.h
View File

@ -62,6 +62,15 @@ class TwoFlavourPseudoFermionAction : public Action<typename Impl::GaugeField> {
ActionSolver(AS),
Phi(Op.FermionGrid()){};
virtual std::string action_name(){return "TwoFlavourPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] has no parameters" << std::endl;
return sstream.str();
}
//////////////////////////////////////////////////////////////////////////////////////
// Push the gauge field in to the dops. Assume any BC's and smearing already applied
//////////////////////////////////////////////////////////////////////////////////////
@ -80,7 +89,9 @@ class TwoFlavourPseudoFermionAction : public Action<typename Impl::GaugeField> {
// in the Phi integral, and thus is only an irrelevant prefactor for
// the partition function.
//
RealD scale = std::sqrt(0.5);
FermionField eta(FermOp.FermionGrid());
gaussian(pRNG, eta);

102
lib/qcd/action/pseudofermion/TwoFlavourEvenOdd.h
View File

@ -31,80 +31,89 @@ directory
#define QCD_PSEUDOFERMION_TWO_FLAVOUR_EVEN_ODD_H
namespace Grid {
namespace QCD {
namespace QCD {
////////////////////////////////////////////////////////////////////////
// Two flavour pseudofermion action for any EO prec dop
////////////////////////////////////////////////////////////////////////
template <class Impl>
class TwoFlavourEvenOddPseudoFermionAction
: public Action<typename Impl::GaugeField> {
public:
INHERIT_IMPL_TYPES(Impl);
////////////////////////////////////////////////////////////////////////
// Two flavour pseudofermion action for any EO prec dop
////////////////////////////////////////////////////////////////////////
template <class Impl>
class TwoFlavourEvenOddPseudoFermionAction
: public Action<typename Impl::GaugeField> {
public:
INHERIT_IMPL_TYPES(Impl);
private:
FermionOperator<Impl> &FermOp; // the basic operator
private:
FermionOperator<Impl> &FermOp; // the basic operator
OperatorFunction<FermionField> &DerivativeSolver;
OperatorFunction<FermionField> &ActionSolver;
OperatorFunction<FermionField> &DerivativeSolver;
OperatorFunction<FermionField> &ActionSolver;
FermionField PhiOdd; // the pseudo fermion field for this trajectory
FermionField PhiEven; // the pseudo fermion field for this trajectory
FermionField PhiOdd; // the pseudo fermion field for this trajectory
FermionField PhiEven; // the pseudo fermion field for this trajectory
public:
/////////////////////////////////////////////////
// Pass in required objects.
/////////////////////////////////////////////////
TwoFlavourEvenOddPseudoFermionAction(FermionOperator<Impl> &Op,
OperatorFunction<FermionField> &DS,
OperatorFunction<FermionField> &AS)
: FermOp(Op),
DerivativeSolver(DS),
ActionSolver(AS),
PhiEven(Op.FermionRedBlackGrid()),
PhiOdd(Op.FermionRedBlackGrid())
{};
public:
/////////////////////////////////////////////////
// Pass in required objects.
/////////////////////////////////////////////////
TwoFlavourEvenOddPseudoFermionAction(FermionOperator<Impl> &Op,
OperatorFunction<FermionField> &DS,
OperatorFunction<FermionField> &AS)
: FermOp(Op),
DerivativeSolver(DS),
ActionSolver(AS),
PhiEven(Op.FermionRedBlackGrid()),
PhiOdd(Op.FermionRedBlackGrid())
{};
virtual std::string action_name(){return "TwoFlavourEvenOddPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] has no parameters" << std::endl;
return sstream.str();
}
//////////////////////////////////////////////////////////////////////////////////////
// Push the gauge field in to the dops. Assume any BC's and smearing already applied
//////////////////////////////////////////////////////////////////////////////////////
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {
// P(phi) = e^{- phi^dag (MpcdagMpc)^-1 phi}
// Phi = McpDag eta
// P(eta) = e^{- eta^dag eta}
//
// e^{x^2/2 sig^2} => sig^2 = 0.5.
RealD scale = std::sqrt(0.5);
FermionField eta (FermOp.FermionGrid());
FermionField etaOdd (FermOp.FermionRedBlackGrid());
FermionField etaEven(FermOp.FermionRedBlackGrid());
gaussian(pRNG,eta);
pickCheckerboard(Even,etaEven,eta);
pickCheckerboard(Odd,etaOdd,eta);
FermOp.ImportGauge(U);
SchurDifferentiableOperator<Impl> PCop(FermOp);
PCop.MpcDag(etaOdd,PhiOdd);
FermOp.MooeeDag(etaEven,PhiEven);
PhiOdd =PhiOdd*scale;
PhiEven=PhiEven*scale;
};
//////////////////////////////////////////////////////
// S = phi^dag (Mdag M)^-1 phi (odd)
// + phi^dag (Mdag M)^-1 phi (even)
//////////////////////////////////////////////////////
virtual RealD S(const GaugeField &U) {
FermOp.ImportGauge(U);
FermionField X(FermOp.FermionRedBlackGrid());
@ -135,7 +144,6 @@ class TwoFlavourEvenOddPseudoFermionAction
//
//////////////////////////////////////////////////////
virtual void deriv(const GaugeField &U,GaugeField & dSdU) {
FermOp.ImportGauge(U);
FermionField X(FermOp.FermionRedBlackGrid());
@ -150,8 +158,8 @@ class TwoFlavourEvenOddPseudoFermionAction
X=zero;
DerivativeSolver(Mpc,PhiOdd,X);
Mpc.Mpc(X,Y);
Mpc.MpcDeriv(tmp , Y, X ); dSdU=tmp;
Mpc.MpcDagDeriv(tmp , X, Y); dSdU=dSdU+tmp;
Mpc.MpcDeriv(tmp , Y, X ); dSdU=tmp;
Mpc.MpcDagDeriv(tmp , X, Y); dSdU=dSdU+tmp;
// Treat the EE case. (MdagM)^-1 = Minv Minvdag
// Deriv defaults to zero.
@ -163,10 +171,10 @@ class TwoFlavourEvenOddPseudoFermionAction
assert(FermOp.ConstEE() == 1);
/*
FermOp.MooeeInvDag(PhiOdd,Y);
FermOp.MooeeInv(Y,X);
FermOp.MeeDeriv(tmp , Y, X,DaggerNo ); dSdU=tmp;
FermOp.MeeDeriv(tmp , X, Y,DaggerYes); dSdU=dSdU+tmp;
FermOp.MooeeInvDag(PhiOdd,Y);
FermOp.MooeeInv(Y,X);
FermOp.MeeDeriv(tmp , Y, X,DaggerNo ); dSdU=tmp;
FermOp.MeeDeriv(tmp , X, Y,DaggerYes); dSdU=dSdU+tmp;
*/
//dSdU = Ta(dSdU);

195
lib/qcd/action/pseudofermion/TwoFlavourEvenOddRatio.h
View File

@ -52,66 +52,75 @@ namespace Grid{
public:
TwoFlavourEvenOddRatioPseudoFermionAction(FermionOperator<Impl> &_NumOp,
FermionOperator<Impl> &_DenOp,
OperatorFunction<FermionField> & DS,
OperatorFunction<FermionField> & AS) :
FermionOperator<Impl> &_DenOp,
OperatorFunction<FermionField> & DS,
OperatorFunction<FermionField> & AS) :
NumOp(_NumOp),
DenOp(_DenOp),
DerivativeSolver(DS),
ActionSolver(AS),
PhiEven(_NumOp.FermionRedBlackGrid()),
PhiOdd(_NumOp.FermionRedBlackGrid())
{
conformable(_NumOp.FermionGrid(), _DenOp.FermionGrid());
conformable(_NumOp.FermionRedBlackGrid(), _DenOp.FermionRedBlackGrid());
conformable(_NumOp.GaugeGrid(), _DenOp.GaugeGrid());
conformable(_NumOp.GaugeRedBlackGrid(), _DenOp.GaugeRedBlackGrid());
};
{
conformable(_NumOp.FermionGrid(), _DenOp.FermionGrid());
conformable(_NumOp.FermionRedBlackGrid(), _DenOp.FermionRedBlackGrid());
conformable(_NumOp.GaugeGrid(), _DenOp.GaugeGrid());
conformable(_NumOp.GaugeRedBlackGrid(), _DenOp.GaugeRedBlackGrid());
};
virtual std::string action_name(){return "TwoFlavourEvenOddRatioPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] has no parameters" << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {
// P(phi) = e^{- phi^dag Vpc (MpcdagMpc)^-1 Vpcdag phi}
//
// NumOp == V
// DenOp == M
//
// Take phi_o = Vpcdag^{-1} Mpcdag eta_o ; eta_o = Mpcdag^{-1} Vpcdag Phi
//
// P(eta_o) = e^{- eta_o^dag eta_o}
//
// e^{x^2/2 sig^2} => sig^2 = 0.5.
//
RealD scale = std::sqrt(0.5);
// P(phi) = e^{- phi^dag Vpc (MpcdagMpc)^-1 Vpcdag phi}
//
// NumOp == V
// DenOp == M
//
// Take phi_o = Vpcdag^{-1} Mpcdag eta_o ; eta_o = Mpcdag^{-1} Vpcdag Phi
//
// P(eta_o) = e^{- eta_o^dag eta_o}
//
// e^{x^2/2 sig^2} => sig^2 = 0.5.
//
RealD scale = std::sqrt(0.5);
FermionField eta (NumOp.FermionGrid());
FermionField etaOdd (NumOp.FermionRedBlackGrid());
FermionField etaEven(NumOp.FermionRedBlackGrid());
FermionField tmp (NumOp.FermionRedBlackGrid());
FermionField eta (NumOp.FermionGrid());
FermionField etaOdd (NumOp.FermionRedBlackGrid());
FermionField etaEven(NumOp.FermionRedBlackGrid());
FermionField tmp (NumOp.FermionRedBlackGrid());
gaussian(pRNG,eta);
gaussian(pRNG,eta);
pickCheckerboard(Even,etaEven,eta);
pickCheckerboard(Odd,etaOdd,eta);
pickCheckerboard(Even,etaEven,eta);
pickCheckerboard(Odd,etaOdd,eta);
NumOp.ImportGauge(U);
DenOp.ImportGauge(U);
NumOp.ImportGauge(U);
DenOp.ImportGauge(U);
SchurDifferentiableOperator<Impl> Mpc(DenOp);
SchurDifferentiableOperator<Impl> Vpc(NumOp);
SchurDifferentiableOperator<Impl> Mpc(DenOp);
SchurDifferentiableOperator<Impl> Vpc(NumOp);
// Odd det factors
Mpc.MpcDag(etaOdd,PhiOdd);
tmp=zero;
ActionSolver(Vpc,PhiOdd,tmp);
Vpc.Mpc(tmp,PhiOdd);
// Odd det factors
Mpc.MpcDag(etaOdd,PhiOdd);
tmp=zero;
ActionSolver(Vpc,PhiOdd,tmp);
Vpc.Mpc(tmp,PhiOdd);
// Even det factors
DenOp.MooeeDag(etaEven,tmp);
NumOp.MooeeInvDag(tmp,PhiEven);
// Even det factors
DenOp.MooeeDag(etaEven,tmp);
NumOp.MooeeInvDag(tmp,PhiEven);
PhiOdd =PhiOdd*scale;
PhiEven=PhiEven*scale;
PhiOdd =PhiOdd*scale;
PhiEven=PhiEven*scale;
};
//////////////////////////////////////////////////////
@ -119,33 +128,33 @@ namespace Grid{
//////////////////////////////////////////////////////
virtual RealD S(const GaugeField &U) {
NumOp.ImportGauge(U);
DenOp.ImportGauge(U);
NumOp.ImportGauge(U);
DenOp.ImportGauge(U);
SchurDifferentiableOperator<Impl> Mpc(DenOp);
SchurDifferentiableOperator<Impl> Vpc(NumOp);
SchurDifferentiableOperator<Impl> Mpc(DenOp);
SchurDifferentiableOperator<Impl> Vpc(NumOp);
FermionField X(NumOp.FermionRedBlackGrid());
FermionField Y(NumOp.FermionRedBlackGrid());
FermionField X(NumOp.FermionRedBlackGrid());
FermionField Y(NumOp.FermionRedBlackGrid());
Vpc.MpcDag(PhiOdd,Y); // Y= Vdag phi
X=zero;
ActionSolver(Mpc,Y,X); // X= (MdagM)^-1 Vdag phi
//Mpc.Mpc(X,Y); // Y= Mdag^-1 Vdag phi
// Multiply by Ydag
RealD action = real(innerProduct(Y,X));
Vpc.MpcDag(PhiOdd,Y); // Y= Vdag phi
X=zero;
ActionSolver(Mpc,Y,X); // X= (MdagM)^-1 Vdag phi
//Mpc.Mpc(X,Y); // Y= Mdag^-1 Vdag phi
// Multiply by Ydag
RealD action = real(innerProduct(Y,X));
//RealD action = norm2(Y);
//RealD action = norm2(Y);
// The EE factorised block; normally can replace with zero if det is constant (gauge field indept)
// Only really clover term that creates this. Leave the EE portion as a future to do to make most
// rapid progresss on DWF for now.
//
NumOp.MooeeDag(PhiEven,X);
DenOp.MooeeInvDag(X,Y);
action = action + norm2(Y);
// The EE factorised block; normally can replace with zero if det is constant (gauge field indept)
// Only really clover term that creates this. Leave the EE portion as a future to do to make most
// rapid progresss on DWF for now.
//
NumOp.MooeeDag(PhiEven,X);
DenOp.MooeeInvDag(X,Y);
action = action + norm2(Y);
return action;
return action;
};
//////////////////////////////////////////////////////
@ -155,44 +164,44 @@ namespace Grid{
//////////////////////////////////////////////////////
virtual void deriv(const GaugeField &U,GaugeField & dSdU) {
NumOp.ImportGauge(U);
DenOp.ImportGauge(U);
NumOp.ImportGauge(U);
DenOp.ImportGauge(U);
SchurDifferentiableOperator<Impl> Mpc(DenOp);
SchurDifferentiableOperator<Impl> Vpc(NumOp);
SchurDifferentiableOperator<Impl> Mpc(DenOp);
SchurDifferentiableOperator<Impl> Vpc(NumOp);
FermionField X(NumOp.FermionRedBlackGrid());
FermionField Y(NumOp.FermionRedBlackGrid());
FermionField X(NumOp.FermionRedBlackGrid());
FermionField Y(NumOp.FermionRedBlackGrid());
GaugeField force(NumOp.GaugeGrid());
// This assignment is necessary to be compliant with the HMC grids
GaugeField force(dSdU._grid);
//Y=Vdag phi
//X = (Mdag M)^-1 V^dag phi
//Y = (Mdag)^-1 V^dag phi
Vpc.MpcDag(PhiOdd,Y); // Y= Vdag phi
X=zero;
DerivativeSolver(Mpc,Y,X); // X= (MdagM)^-1 Vdag phi
Mpc.Mpc(X,Y); // Y= Mdag^-1 Vdag phi
//Y=Vdag phi
//X = (Mdag M)^-1 V^dag phi
//Y = (Mdag)^-1 V^dag phi
Vpc.MpcDag(PhiOdd,Y); // Y= Vdag phi
X=zero;
DerivativeSolver(Mpc,Y,X); // X= (MdagM)^-1 Vdag phi
Mpc.Mpc(X,Y); // Y= Mdag^-1 Vdag phi
// phi^dag V (Mdag M)^-1 dV^dag phi
Vpc.MpcDagDeriv(force , X, PhiOdd ); dSdU=force;
// phi^dag V (Mdag M)^-1 dV^dag phi
Vpc.MpcDagDeriv(force , X, PhiOdd ); dSdU = force;
// phi^dag dV (Mdag M)^-1 V^dag phi
Vpc.MpcDeriv(force , PhiOdd, X ); dSdU=dSdU+force;
// phi^dag dV (Mdag M)^-1 V^dag phi
Vpc.MpcDeriv(force , PhiOdd, X ); dSdU = dSdU+force;
// - phi^dag V (Mdag M)^-1 Mdag dM (Mdag M)^-1 V^dag phi
// - phi^dag V (Mdag M)^-1 dMdag M (Mdag M)^-1 V^dag phi
Mpc.MpcDeriv(force,Y,X); dSdU=dSdU-force;
Mpc.MpcDagDeriv(force,X,Y); dSdU=dSdU-force;
// - phi^dag V (Mdag M)^-1 Mdag dM (Mdag M)^-1 V^dag phi
// - phi^dag V (Mdag M)^-1 dMdag M (Mdag M)^-1 V^dag phi
Mpc.MpcDeriv(force,Y,X); dSdU = dSdU-force;
Mpc.MpcDagDeriv(force,X,Y); dSdU = dSdU-force;
// FIXME No force contribution from EvenEven assumed here
// Needs a fix for clover.
assert(NumOp.ConstEE() == 1);
assert(DenOp.ConstEE() == 1);
// FIXME No force contribution from EvenEven assumed here
// Needs a fix for clover.
assert(NumOp.ConstEE() == 1);
assert(DenOp.ConstEE() == 1);
//dSdU = -Ta(dSdU);
dSdU = -dSdU;
dSdU = -dSdU;
};
};
}

8
lib/qcd/action/pseudofermion/TwoFlavourRatio.h
View File

@ -57,6 +57,14 @@ namespace Grid{
OperatorFunction<FermionField> & AS
) : NumOp(_NumOp), DenOp(_DenOp), DerivativeSolver(DS), ActionSolver(AS), Phi(_NumOp.FermionGrid()) {};
virtual std::string action_name(){return "TwoFlavourRatioPseudoFermionAction";}
virtual std::string LogParameters(){
std::stringstream sstream;
sstream << GridLogMessage << "["<<action_name()<<"] has no parameters" << std::endl;
return sstream.str();
}
virtual void refresh(const GaugeField &U, GridParallelRNG& pRNG) {
// P(phi) = e^{- phi^dag V (MdagM)^-1 Vdag phi}

50
lib/qcd/action/scalar/Scalar.h Normal file
View File

@ -0,0 +1,50 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/Scalar.h
Copyright (C) 2017
Author: Guido Cossu <guido.cossu@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_QCD_SCALAR_H
#define GRID_QCD_SCALAR_H
#include <Grid/qcd/action/scalar/ScalarImpl.h>
#include <Grid/qcd/action/scalar/ScalarAction.h>
#include <Grid/qcd/action/scalar/ScalarInteractionAction.h>
namespace Grid {
namespace QCD {
typedef ScalarAction<ScalarImplR> ScalarActionR;
typedef ScalarAction<ScalarImplF> ScalarActionF;
typedef ScalarAction<ScalarImplD> ScalarActionD;
template <int Colours, int Dimensions> using ScalarAdjActionR = ScalarInteractionAction<ScalarNxNAdjImplR<Colours>, Dimensions>;
template <int Colours, int Dimensions> using ScalarAdjActionF = ScalarInteractionAction<ScalarNxNAdjImplF<Colours>, Dimensions>;
template <int Colours, int Dimensions> using ScalarAdjActionD = ScalarInteractionAction<ScalarNxNAdjImplD<Colours>, Dimensions>;
}
}
#endif // GRID_QCD_SCALAR_H

83
lib/qcd/action/scalar/ScalarAction.h Normal file
View File

@ -0,0 +1,83 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/WilsonGaugeAction.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>
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 SCALAR_ACTION_H
#define SCALAR_ACTION_H
namespace Grid {
// FIXME drop the QCD namespace everywhere here
template <class Impl>
class ScalarAction : public QCD::Action<typename Impl::Field> {
public:
INHERIT_FIELD_TYPES(Impl);
private:
RealD mass_square;
RealD lambda;
public:
ScalarAction(RealD ms, RealD l) : mass_square(ms), lambda(l) {}
virtual std::string LogParameters() {
std::stringstream sstream;
sstream << GridLogMessage << "[ScalarAction] lambda : " << lambda << std::endl;
sstream << GridLogMessage << "[ScalarAction] mass_square : " << mass_square << std::endl;
return sstream.str();
}
virtual std::string action_name() {return "ScalarAction";}
virtual void refresh(const Field &U, GridParallelRNG &pRNG) {} // noop as no pseudoferms
virtual RealD S(const Field &p) {
return (mass_square * 0.5 + QCD::Nd) * ScalarObs<Impl>::sumphisquared(p) +
(lambda / 24.) * ScalarObs<Impl>::sumphifourth(p) +
ScalarObs<Impl>::sumphider(p);
};
virtual void deriv(const Field &p,
Field &force) {
Field tmp(p._grid);
Field p2(p._grid);
ScalarObs<Impl>::phisquared(p2, p);
tmp = -(Cshift(p, 0, -1) + Cshift(p, 0, 1));
for (int mu = 1; mu < QCD::Nd; mu++) tmp -= Cshift(p, mu, -1) + Cshift(p, mu, 1);
force =+(mass_square + 2. * QCD::Nd) * p + (lambda / 6.) * p2 * p + tmp;
}
};
} // namespace Grid
#endif // SCALAR_ACTION_H

162
lib/qcd/action/scalar/ScalarImpl.h Normal file
View File

@ -0,0 +1,162 @@
#ifndef SCALAR_IMPL
#define SCALAR_IMPL
namespace Grid {
//namespace QCD {
template <class S>
class ScalarImplTypes {
public:
typedef S Simd;
template <typename vtype>
using iImplField = iScalar<iScalar<iScalar<vtype> > >;
typedef iImplField<Simd> SiteField;
typedef SiteField SitePropagator;
typedef SiteField SiteComplex;
typedef Lattice<SiteField> Field;
typedef Field ComplexField;
typedef Field FermionField;
typedef Field PropagatorField;
static inline void generate_momenta(Field& P, GridParallelRNG& pRNG){
gaussian(pRNG, P);
}
static inline Field projectForce(Field& P){return P;}
static inline void update_field(Field& P, Field& U, double ep) {
U += P*ep;
}
static inline RealD FieldSquareNorm(Field& U) {
return (- sum(trace(U*U))/2.0);
}
static inline void HotConfiguration(GridParallelRNG &pRNG, Field &U) {
gaussian(pRNG, U);
}
static inline void TepidConfiguration(GridParallelRNG &pRNG, Field &U) {
gaussian(pRNG, U);
}
static inline void ColdConfiguration(GridParallelRNG &pRNG, Field &U) {
U = 1.0;
}
static void MomentumSpacePropagator(Field &out, RealD m)
{
GridBase *grid = out._grid;
Field kmu(grid), one(grid);
const unsigned int nd = grid->_ndimension;
std::vector<int> &l = grid->_fdimensions;
one = Complex(1.0,0.0);
out = m*m;
for(int mu = 0; mu < nd; mu++)
{
Real twoPiL = M_PI*2./l[mu];
LatticeCoordinate(kmu,mu);
kmu = 2.*sin(.5*twoPiL*kmu);
out = out + kmu*kmu;
}
out = one/out;
}
static void FreePropagator(const Field &in, Field &out,
const Field &momKernel)
{
FFT fft((GridCartesian *)in._grid);
Field inFT(in._grid);
fft.FFT_all_dim(inFT, in, FFT::forward);
inFT = inFT*momKernel;
fft.FFT_all_dim(out, inFT, FFT::backward);
}
static void FreePropagator(const Field &in, Field &out, RealD m)
{
Field momKernel(in._grid);
MomentumSpacePropagator(momKernel, m);
FreePropagator(in, out, momKernel);
}
};
template <class S, unsigned int N>
class ScalarAdjMatrixImplTypes {
public:
typedef S Simd;
typedef QCD::SU<N> Group;
template <typename vtype>
using iImplField = iScalar<iScalar<iMatrix<vtype, N>>>;
template <typename vtype>
using iImplComplex = iScalar<iScalar<iScalar<vtype>>>;
typedef iImplField<Simd> SiteField;
typedef SiteField SitePropagator;
typedef iImplComplex<Simd> SiteComplex;
typedef Lattice<SiteField> Field;
typedef Lattice<SiteComplex> ComplexField;
typedef Field FermionField;
typedef Field PropagatorField;
static inline void generate_momenta(Field& P, GridParallelRNG& pRNG) {
Group::GaussianFundamentalLieAlgebraMatrix(pRNG, P);
}
static inline Field projectForce(Field& P) {return P;}
static inline void update_field(Field& P, Field& U, double ep) {
U += P*ep;
}
static inline RealD FieldSquareNorm(Field& U) {
return (TensorRemove(sum(trace(U*U))).real());
}
static inline void HotConfiguration(GridParallelRNG &pRNG, Field &U) {
Group::GaussianFundamentalLieAlgebraMatrix(pRNG, U);
}
static inline void TepidConfiguration(GridParallelRNG &pRNG, Field &U) {
Group::GaussianFundamentalLieAlgebraMatrix(pRNG, U, 0.01);
}
static inline void ColdConfiguration(GridParallelRNG &pRNG, Field &U) {
U = zero;
}
};
typedef ScalarImplTypes<vReal> ScalarImplR;
typedef ScalarImplTypes<vRealF> ScalarImplF;
typedef ScalarImplTypes<vRealD> ScalarImplD;
typedef ScalarImplTypes<vComplex> ScalarImplCR;
typedef ScalarImplTypes<vComplexF> ScalarImplCF;
typedef ScalarImplTypes<vComplexD> ScalarImplCD;
// Hardcoding here the size of the matrices
typedef ScalarAdjMatrixImplTypes<vComplex, QCD::Nc> ScalarAdjImplR;
typedef ScalarAdjMatrixImplTypes<vComplexF, QCD::Nc> ScalarAdjImplF;
typedef ScalarAdjMatrixImplTypes<vComplexD, QCD::Nc> ScalarAdjImplD;
template <int Colours > using ScalarNxNAdjImplR = ScalarAdjMatrixImplTypes<vComplex, Colours >;
template <int Colours > using ScalarNxNAdjImplF = ScalarAdjMatrixImplTypes<vComplexF, Colours >;
template <int Colours > using ScalarNxNAdjImplD = ScalarAdjMatrixImplTypes<vComplexD, Colours >;
//}
}
#endif

148
lib/qcd/action/scalar/ScalarInteractionAction.h Normal file
View File

@ -0,0 +1,148 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/WilsonGaugeAction.h
Copyright (C) 2015
Author: Guido Cossu <guido,cossu@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 SCALAR_INT_ACTION_H
#define SCALAR_INT_ACTION_H
// Note: this action can completely absorb the ScalarAction for real float fields
// use the scalarObjs to generalise the structure
namespace Grid {
// FIXME drop the QCD namespace everywhere here
template <class Impl, int Ndim >
class ScalarInteractionAction : public QCD::Action<typename Impl::Field> {
public:
INHERIT_FIELD_TYPES(Impl);
private:
RealD mass_square;
RealD lambda;
typedef typename Field::vector_object vobj;
typedef CartesianStencil<vobj,vobj> Stencil;
SimpleCompressor<vobj> compressor;
int npoint = 2*Ndim;
std::vector<int> directions;// = {0,1,2,3,0,1,2,3}; // forcing 4 dimensions
std::vector<int> displacements;// = {1,1,1,1, -1,-1,-1,-1};
public:
ScalarInteractionAction(RealD ms, RealD l) : mass_square(ms), lambda(l), displacements(2*Ndim,0), directions(2*Ndim,0){
for (int mu = 0 ; mu < Ndim; mu++){
directions[mu] = mu; directions[mu+Ndim] = mu;
displacements[mu] = 1; displacements[mu+Ndim] = -1;
}
}
virtual std::string LogParameters() {
std::stringstream sstream;
sstream << GridLogMessage << "[ScalarAction] lambda : " << lambda << std::endl;
sstream << GridLogMessage << "[ScalarAction] mass_square : " << mass_square << std::endl;
return sstream.str();
}
virtual std::string action_name() {return "ScalarAction";}
virtual void refresh(const Field &U, GridParallelRNG &pRNG) {}
virtual RealD S(const Field &p) {
assert(p._grid->Nd() == Ndim);
static Stencil phiStencil(p._grid, npoint, 0, directions, displacements);
phiStencil.HaloExchange(p, compressor);
Field action(p._grid), pshift(p._grid), phisquared(p._grid);
phisquared = p*p;
action = (2.0*Ndim + mass_square)*phisquared - lambda/24.*phisquared*phisquared;
for (int mu = 0; mu < Ndim; mu++) {
// pshift = Cshift(p, mu, +1); // not efficient, implement with stencils
parallel_for (int i = 0; i < p._grid->oSites(); i++) {
int permute_type;
StencilEntry *SE;
vobj temp2;
const vobj *temp, *t_p;
SE = phiStencil.GetEntry(permute_type, mu, i);
t_p = &p._odata[i];
if ( SE->_is_local ) {
temp = &p._odata[SE->_offset];
if ( SE->_permute ) {
permute(temp2, *temp, permute_type);
action._odata[i] -= temp2*(*t_p) + (*t_p)*temp2;
} else {
action._odata[i] -= (*temp)*(*t_p) + (*t_p)*(*temp);
}
} else {
action._odata[i] -= phiStencil.CommBuf()[SE->_offset]*(*t_p) + (*t_p)*phiStencil.CommBuf()[SE->_offset];
}
}
// action -= pshift*p + p*pshift;
}
// NB the trace in the algebra is normalised to 1/2
// minus sign coming from the antihermitian fields
return -(TensorRemove(sum(trace(action)))).real();
};
virtual void deriv(const Field &p, Field &force) {
assert(p._grid->Nd() == Ndim);
force = (2.0*Ndim + mass_square)*p - lambda/12.*p*p*p;
// move this outside
static Stencil phiStencil(p._grid, npoint, 0, directions, displacements);
phiStencil.HaloExchange(p, compressor);
//for (int mu = 0; mu < QCD::Nd; mu++) force -= Cshift(p, mu, -1) + Cshift(p, mu, 1);
for (int point = 0; point < npoint; point++) {
parallel_for (int i = 0; i < p._grid->oSites(); i++) {
const vobj *temp;
vobj temp2;
int permute_type;
StencilEntry *SE;
SE = phiStencil.GetEntry(permute_type, point, i);
if ( SE->_is_local ) {
temp = &p._odata[SE->_offset];
if ( SE->_permute ) {
permute(temp2, *temp, permute_type);
force._odata[i] -= temp2;
} else {
force._odata[i] -= *temp;
}
} else {
force._odata[i] -= phiStencil.CommBuf()[SE->_offset];
}
}
}
}
};
} // namespace Grid
#endif // SCALAR_INT_ACTION_H

219
lib/qcd/hmc/GenericHMCrunner.h Normal file
View File

@ -0,0 +1,219 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/GenericHmcRunner.h
Copyright (C) 2015
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Guido Cossu <guido.cossu@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_GENERIC_HMC_RUNNER
#define GRID_GENERIC_HMC_RUNNER
#include <unordered_map>
namespace Grid {
namespace QCD {
// very ugly here but possibly resolved if we had a base Reader class
template < class ReaderClass >
class HMCRunnerBase {
public:
virtual void Run() = 0;
virtual void initialize(ReaderClass& ) = 0;
};
template <class Implementation,
template <typename, typename, typename> class Integrator,
class RepresentationsPolicy = NoHirep, class ReaderClass = XmlReader>
class HMCWrapperTemplate: public HMCRunnerBase<ReaderClass> {
public:
INHERIT_FIELD_TYPES(Implementation);
typedef Implementation ImplPolicy; // visible from outside
template <typename S = NoSmearing<Implementation> >
using IntegratorType = Integrator<Implementation, S, RepresentationsPolicy>;
HMCparameters Parameters;
std::string ParameterFile;
HMCResourceManager<Implementation> Resources;
// The set of actions (keep here for lower level users, for now)
ActionSet<Field, RepresentationsPolicy> TheAction;
HMCWrapperTemplate() = default;
HMCWrapperTemplate(HMCparameters Par){
Parameters = Par;
}
void initialize(ReaderClass & TheReader){
std::cout << "Initialization of the HMC" << std::endl;
Resources.initialize(TheReader);
// eventually add smearing
Resources.GetActionSet(TheAction);
}
void ReadCommandLine(int argc, char **argv) {
std::string arg;
if (GridCmdOptionExists(argv, argv + argc, "--StartingType")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--StartingType");
if (arg != "HotStart" && arg != "ColdStart" && arg != "TepidStart" &&
arg != "CheckpointStart") {
std::cout << GridLogError << "Unrecognized option in --StartingType\n";
std::cout
<< GridLogError
<< "Valid [HotStart, ColdStart, TepidStart, CheckpointStart]\n";
exit(1);
}
Parameters.StartingType = arg;
}
if (GridCmdOptionExists(argv, argv + argc, "--StartingTrajectory")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--StartingTrajectory");
std::vector<int> ivec(0);
GridCmdOptionIntVector(arg, ivec);
Parameters.StartTrajectory = ivec[0];
}
if (GridCmdOptionExists(argv, argv + argc, "--Trajectories")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--Trajectories");
std::vector<int> ivec(0);
GridCmdOptionIntVector(arg, ivec);
Parameters.Trajectories = ivec[0];
}
if (GridCmdOptionExists(argv, argv + argc, "--Thermalizations")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--Thermalizations");
std::vector<int> ivec(0);
GridCmdOptionIntVector(arg, ivec);
Parameters.NoMetropolisUntil = ivec[0];
}
if (GridCmdOptionExists(argv, argv + argc, "--ParameterFile")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--ParameterFile");
ParameterFile = arg;
}
}
template <class SmearingPolicy>
void Run(SmearingPolicy &S) {
Runner(S);
}
void Run(){
NoSmearing<Implementation> S;
Runner(S);
}
//////////////////////////////////////////////////////////////////
private:
template <class SmearingPolicy>
void Runner(SmearingPolicy &Smearing) {
auto UGrid = Resources.GetCartesian();
Resources.AddRNGs();
Field U(UGrid);
// Can move this outside?
typedef IntegratorType<SmearingPolicy> TheIntegrator;
TheIntegrator MDynamics(UGrid, Parameters.MD, TheAction, Smearing);
if (Parameters.StartingType == "HotStart") {
// Hot start
Resources.SeedFixedIntegers();
Implementation::HotConfiguration(Resources.GetParallelRNG(), U);
} else if (Parameters.StartingType == "ColdStart") {
// Cold start
Resources.SeedFixedIntegers();
Implementation::ColdConfiguration(Resources.GetParallelRNG(), U);
} else if (Parameters.StartingType == "TepidStart") {
// Tepid start
Resources.SeedFixedIntegers();
Implementation::TepidConfiguration(Resources.GetParallelRNG(), U);
} else if (Parameters.StartingType == "CheckpointStart") {
// CheckpointRestart
Resources.GetCheckPointer()->CheckpointRestore(Parameters.StartTrajectory, U,
Resources.GetSerialRNG(),
Resources.GetParallelRNG());
}
Smearing.set_Field(U);
HybridMonteCarlo<TheIntegrator> HMC(Parameters, MDynamics,
Resources.GetSerialRNG(),
Resources.GetParallelRNG(),
Resources.GetObservables(), U);
// Run it
HMC.evolve();
}
};
// These are for gauge fields, default integrator MinimumNorm2
template <template <typename, typename, typename> class Integrator>
using GenericHMCRunner = HMCWrapperTemplate<PeriodicGimplR, Integrator>;
template <template <typename, typename, typename> class Integrator>
using GenericHMCRunnerF = HMCWrapperTemplate<PeriodicGimplF, Integrator>;
template <template <typename, typename, typename> class Integrator>
using GenericHMCRunnerD = HMCWrapperTemplate<PeriodicGimplD, Integrator>;
// These are for gauge fields, default integrator MinimumNorm2
template <template <typename, typename, typename> class Integrator>
using ConjugateHMCRunner = HMCWrapperTemplate<ConjugateGimplR, Integrator>;
template <template <typename, typename, typename> class Integrator>
using ConjugateHMCRunnerF = HMCWrapperTemplate<ConjugateGimplF, Integrator>;
template <template <typename, typename, typename> class Integrator>
using ConjugateHMCRunnerD = HMCWrapperTemplate<ConjugateGimplD, Integrator>;
template <class RepresentationsPolicy,
template <typename, typename, typename> class Integrator>
using GenericHMCRunnerHirep =
HMCWrapperTemplate<PeriodicGimplR, Integrator, RepresentationsPolicy>;
template <class Implementation, class RepresentationsPolicy,
template <typename, typename, typename> class Integrator>
using GenericHMCRunnerTemplate = HMCWrapperTemplate<Implementation, Integrator, RepresentationsPolicy>;
typedef HMCWrapperTemplate<ScalarImplR, MinimumNorm2, ScalarFields>
ScalarGenericHMCRunner;
typedef HMCWrapperTemplate<ScalarAdjImplR, MinimumNorm2, ScalarMatrixFields>
ScalarAdjGenericHMCRunner;
template <int Colours>
using ScalarNxNAdjGenericHMCRunner = HMCWrapperTemplate < ScalarNxNAdjImplR<Colours>, MinimumNorm2, ScalarNxNMatrixFields<Colours> >;
} // namespace QCD
} // namespace Grid
#endif // GRID_GENERIC_HMC_RUNNER

199
lib/qcd/hmc/HMC.h
View File

@ -34,13 +34,15 @@ directory
* @brief Classes for Hybrid Monte Carlo update
*
* @author Guido Cossu
* Time-stamp: <2015-07-30 16:58:26 neo>
*/
//--------------------------------------------------------------------
#ifndef HMC_INCLUDED
#define HMC_INCLUDED
#include <string>
#include <list>
#include <Grid/qcd/hmc/integrators/Integrator.h>
#include <Grid/qcd/hmc/integrators/Integrator_algorithm.h>
@ -48,91 +50,64 @@ directory
namespace Grid {
namespace QCD {
struct HMCparameters {
Integer StartTrajectory;
Integer Trajectories; /* @brief Number of sweeps in this run */
bool MetropolisTest;
Integer NoMetropolisUntil;
struct HMCparameters: Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(HMCparameters,
Integer, StartTrajectory,
Integer, Trajectories, /* @brief Number of sweeps in this run */
bool, MetropolisTest,
Integer, NoMetropolisUntil,
std::string, StartingType,
IntegratorParameters, MD)
HMCparameters() {
////////////////////////////// Default values
MetropolisTest = true;
MetropolisTest = true;
NoMetropolisUntil = 10;
StartTrajectory = 0;
Trajectories = 200;
StartTrajectory = 0;
Trajectories = 10;
StartingType = "HotStart";
/////////////////////////////////
}
void print() const {
std::cout << GridLogMessage << "[HMC parameter] Trajectories : " << Trajectories << "\n";
std::cout << GridLogMessage << "[HMC parameter] Start trajectory : " << StartTrajectory << "\n";
std::cout << GridLogMessage << "[HMC parameter] Metropolis test (on/off): " << MetropolisTest << "\n";
std::cout << GridLogMessage << "[HMC parameter] Thermalization trajs : " << NoMetropolisUntil << "\n";
template <class ReaderClass >
HMCparameters(Reader<ReaderClass> & TheReader){
initialize(TheReader);
}
template < class ReaderClass >
void initialize(Reader<ReaderClass> &TheReader){
std::cout << GridLogMessage << "Reading HMC\n";
read(TheReader, "HMC", *this);
}
void print_parameters() const {
std::cout << GridLogMessage << "[HMC parameters] Trajectories : " << Trajectories << "\n";
std::cout << GridLogMessage << "[HMC parameters] Start trajectory : " << StartTrajectory << "\n";
std::cout << GridLogMessage << "[HMC parameters] Metropolis test (on/off): " << std::boolalpha << MetropolisTest << "\n";
std::cout << GridLogMessage << "[HMC parameters] Thermalization trajs : " << NoMetropolisUntil << "\n";
std::cout << GridLogMessage << "[HMC parameters] Starting type : " << StartingType << "\n";
MD.print_parameters();
}
};
template <class GaugeField>
class HmcObservable {
public:
virtual void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) = 0;
};
template <class Gimpl>
class PlaquetteLogger : public HmcObservable<typename Gimpl::GaugeField> {
private:
std::string Stem;
public:
INHERIT_GIMPL_TYPES(Gimpl);
PlaquetteLogger(std::string cf) { Stem = cf; };
void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
std::string file;
{
std::ostringstream os;
os << Stem << "." << traj;
file = os.str();
}
std::ofstream of(file);
RealD peri_plaq = WilsonLoops<PeriodicGimplR>::avgPlaquette(U);
RealD peri_rect = WilsonLoops<PeriodicGimplR>::avgRectangle(U);
RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(U);
RealD impl_rect = WilsonLoops<Gimpl>::avgRectangle(U);
of << traj << " " << impl_plaq << " " << impl_rect << " " << peri_plaq
<< " " << peri_rect << std::endl;
std::cout << GridLogMessage << "traj"
<< " "
<< "plaq "
<< " "
<< " rect "
<< " "
<< "peri_plaq"
<< " "
<< "peri_rect" << std::endl;
std::cout << GridLogMessage << traj << " " << impl_plaq << " " << impl_rect
<< " " << peri_plaq << " " << peri_rect << std::endl;
}
};
// template <class GaugeField, class Integrator, class Smearer, class
// Boundary>
template <class GaugeField, class IntegratorType>
template <class IntegratorType>
class HybridMonteCarlo {
private:
const HMCparameters Params;
GridSerialRNG &sRNG; // Fixme: need a RNG management strategy.
GridParallelRNG &pRNG; // Fixme: need a RNG management strategy.
GaugeField &Ucur;
typedef typename IntegratorType::Field Field;
typedef std::vector< HmcObservable<Field> * > ObsListType;
//pass these from the resource manager
GridSerialRNG &sRNG;
GridParallelRNG &pRNG;
Field &Ucur;
IntegratorType &TheIntegrator;
std::vector<HmcObservable<GaugeField> *> Observables;
ObsListType Observables;
/////////////////////////////////////////////////////////
// Metropolis step
@ -167,13 +142,13 @@ class HybridMonteCarlo {
/////////////////////////////////////////////////////////
// Evolution
/////////////////////////////////////////////////////////
RealD evolve_step(GaugeField &U) {
RealD evolve_hmc_step(Field &U) {
TheIntegrator.refresh(U, pRNG); // set U and initialize P and phi's
RealD H0 = TheIntegrator.S(U); // initial state action
std::streamsize current_precision = std::cout.precision();
std::cout.precision(17);
std::cout.precision(15);
std::cout << GridLogMessage << "Total H before trajectory = " << H0 << "\n";
std::cout.precision(current_precision);
@ -181,64 +156,96 @@ class HybridMonteCarlo {
RealD H1 = TheIntegrator.S(U); // updated state action
std::cout.precision(17);
std::cout << GridLogMessage << "Total H after trajectory = " << H1
<< " dH = " << H1 - H0 << "\n";
std::cout.precision(current_precision);
///////////////////////////////////////////////////////////
if(0){
std::cout << "------------------------- Reversibility test" << std::endl;
TheIntegrator.reverse_momenta();
TheIntegrator.integrate(U);
H1 = TheIntegrator.S(U); // updated state action
std::cout << "--------------------------------------------" << std::endl;
}
///////////////////////////////////////////////////////////
std::cout.precision(15);
std::cout << GridLogMessage << "Total H after trajectory = " << H1
<< " dH = " << H1 - H0 << "\n";
std::cout.precision(current_precision);
return (H1 - H0);
}
public:
/////////////////////////////////////////
// Constructor
/////////////////////////////////////////
HybridMonteCarlo(HMCparameters Pams, IntegratorType &_Int,
GridSerialRNG &_sRNG, GridParallelRNG &_pRNG, GaugeField &_U)
: Params(Pams), TheIntegrator(_Int), sRNG(_sRNG), pRNG(_pRNG), Ucur(_U) {}
HybridMonteCarlo(HMCparameters _Pams, IntegratorType &_Int,
GridSerialRNG &_sRNG, GridParallelRNG &_pRNG,
ObsListType _Obs, Field &_U)
: Params(_Pams), TheIntegrator(_Int), sRNG(_sRNG), pRNG(_pRNG), Observables(_Obs), Ucur(_U) {}
~HybridMonteCarlo(){};
void AddObservable(HmcObservable<GaugeField> *obs) {
Observables.push_back(obs);
}
void evolve(void) {
Real DeltaH;
GaugeField Ucopy(Ucur._grid);
Field Ucopy(Ucur._grid);
Params.print();
Params.print_parameters();
TheIntegrator.print_actions();
// Actual updates (evolve a copy Ucopy then copy back eventually)
for (int traj = Params.StartTrajectory;
traj < Params.Trajectories + Params.StartTrajectory; ++traj) {
unsigned int FinalTrajectory = Params.Trajectories + Params.NoMetropolisUntil + Params.StartTrajectory;
for (int traj = Params.StartTrajectory; traj < FinalTrajectory; ++traj) {
std::cout << GridLogMessage << "-- # Trajectory = " << traj << "\n";
if (traj < Params.StartTrajectory + Params.NoMetropolisUntil) {
std::cout << GridLogMessage << "-- Thermalization" << std::endl;
}
double t0=usecond();
Ucopy = Ucur;
DeltaH = evolve_step(Ucopy);
DeltaH = evolve_hmc_step(Ucopy);
// Metropolis-Hastings test
bool accept = true;
if (traj >= Params.NoMetropolisUntil) {
if (traj >= Params.StartTrajectory + Params.NoMetropolisUntil) {
accept = metropolis_test(DeltaH);
} else {
std::cout << GridLogMessage << "Skipping Metropolis test" << std::endl;
}
if (accept) {
Ucur = Ucopy;
}
if (accept)
Ucur = Ucopy;
double t1=usecond();
std::cout << GridLogMessage << "Total time for trajectory (s): " << (t1-t0)/1e6 << std::endl;
for (int obs = 0; obs < Observables.size(); obs++) {
std::cout << GridLogDebug << "Observables # " << obs << std::endl;
std::cout << GridLogDebug << "Observables total " << Observables.size() << std::endl;
std::cout << GridLogDebug << "Observables pointer " << Observables[obs] << std::endl;
Observables[obs]->TrajectoryComplete(traj + 1, Ucur, sRNG, pRNG);
}
std::cout << GridLogMessage << ":::::::::::::::::::::::::::::::::::::::::::" << std::endl;
}
}
};
} // QCD
} // Grid
#include <Grid/parallelIO/NerscIO.h>
#include <Grid/qcd/hmc/NerscCheckpointer.h>
#include <Grid/qcd/hmc/HmcRunner.h>
// april 11 2017 merge, Guido, commenting out
//#include <Grid/parallelIO/NerscIO.h>
//#include <Grid/qcd/hmc/NerscCheckpointer.h>
//#include <Grid/qcd/hmc/HmcRunner.h>
#endif

111
lib/qcd/hmc/HMCModules.h Normal file
View File

@ -0,0 +1,111 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/GenericHmcRunner.h
Copyright (C) 2015
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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_HMC_MODULES
#define GRID_HMC_MODULES
#include "HMC_GridModules.h"
namespace Grid {
namespace QCD {
////////////////////////////////////////////////////////////////////
struct RNGModuleParameters: Serializable {
GRID_SERIALIZABLE_CLASS_MEMBERS(RNGModuleParameters,
std::string, serial_seeds,
std::string, parallel_seeds,);
std::vector<int> getSerialSeeds(){return strToVec<int>(serial_seeds);}
std::vector<int> getParallelSeeds(){return strToVec<int>(parallel_seeds);}
RNGModuleParameters(): serial_seeds("1"), parallel_seeds("1"){}
template <class ReaderClass >
RNGModuleParameters(Reader<ReaderClass>& Reader){
read(Reader, "RandomNumberGenerator", *this);
}
};
// Random number generators module
class RNGModule{
GridSerialRNG sRNG_;
std::unique_ptr<GridParallelRNG> pRNG_;
RNGModuleParameters Params_;
public:
RNGModule(){};
void set_pRNG(GridParallelRNG* pRNG){
pRNG_.reset(pRNG);
}
void set_RNGSeeds(RNGModuleParameters& Params) {
Params_ = Params;
}
GridSerialRNG& get_sRNG() { return sRNG_; }
GridParallelRNG& get_pRNG() { return *pRNG_.get(); }
void seed() {
auto SerialSeeds = Params_.getSerialSeeds();
auto ParallelSeeds = Params_.getParallelSeeds();
if (SerialSeeds.size() == 0 && ParallelSeeds.size() == 0) {
std::cout << GridLogError << "Seeds not initialized" << std::endl;
exit(1);
}
sRNG_.SeedFixedIntegers(SerialSeeds);
pRNG_->SeedFixedIntegers(ParallelSeeds);
}
};
/*
///////////////////////////////////////////////////////////////////
/// Smearing module
template <class ImplementationPolicy>
class SmearingModule{
virtual void get_smearing();
};
template <class ImplementationPolicy>
class StoutSmearingModule: public SmearingModule<ImplementationPolicy>{
SmearedConfiguration<ImplementationPolicy> SmearingPolicy;
};
*/
} // namespace QCD
} // namespace Grid
#endif // GRID_HMC_MODULES

301
lib/qcd/hmc/HMCResourceManager.h Normal file
View File

@ -0,0 +1,301 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/GenericHmcRunner.h
Copyright (C) 2015
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 HMC_RESOURCE_MANAGER_H
#define HMC_RESOURCE_MANAGER_H
#include <unordered_map>
// One function per Checkpointer, use a macro to simplify
#define RegisterLoadCheckPointerFunction(NAME) \
void Load##NAME##Checkpointer(const CheckpointerParameters& Params_) { \
if (!have_CheckPointer) { \
std::cout << GridLogDebug << "Loading Checkpointer " << #NAME \
<< std::endl; \
CP = std::unique_ptr<CheckpointerBaseModule>( \
new NAME##CPModule<ImplementationPolicy>(Params_)); \
have_CheckPointer = true; \
} else { \
std::cout << GridLogError << "Checkpointer already loaded " \
<< std::endl; \
exit(1); \
} \
}
namespace Grid {
namespace QCD {
// HMC Resource manager
template <class ImplementationPolicy>
class HMCResourceManager {
typedef HMCModuleBase< QCD::BaseHmcCheckpointer<ImplementationPolicy> > CheckpointerBaseModule;
typedef HMCModuleBase< QCD::HmcObservable<typename ImplementationPolicy::Field> > ObservableBaseModule;
typedef ActionModuleBase< QCD::Action<typename ImplementationPolicy::Field>, GridModule > ActionBaseModule;
// Named storage for grid pairs (std + red-black)
std::unordered_map<std::string, GridModule> Grids;
RNGModule RNGs;
// SmearingModule<ImplementationPolicy> Smearing;
std::unique_ptr<CheckpointerBaseModule> CP;
// A vector of HmcObservable modules
std::vector<std::unique_ptr<ObservableBaseModule> > ObservablesList;
// A vector of HmcObservable modules
std::multimap<int, std::unique_ptr<ActionBaseModule> > ActionsList;
std::vector<int> multipliers;
bool have_RNG;
bool have_CheckPointer;
// NOTE: operator << is not overloaded for std::vector<string>
// so thsi function is necessary
void output_vector_string(const std::vector<std::string> &vs){
for (auto &i: vs)
std::cout << i << " ";
std::cout << std::endl;
}
public:
HMCResourceManager() : have_RNG(false), have_CheckPointer(false) {}
template <class ReaderClass, class vector_type = vComplex >
void initialize(ReaderClass &Read){
// assumes we are starting from the main node
// Geometry
GridModuleParameters GridPar(Read);
GridFourDimModule<vector_type> GridMod( GridPar) ;
AddGrid("gauge", GridMod);
// Checkpointer
auto &CPfactory = HMC_CPModuleFactory<cp_string, ImplementationPolicy, ReaderClass >::getInstance();
Read.push("Checkpointer");
std::string cp_type;
read(Read,"name", cp_type);
std::cout << "Registered types " << std::endl;
output_vector_string(CPfactory.getBuilderList());
CP = CPfactory.create(cp_type, Read);
CP->print_parameters();
Read.pop();
have_CheckPointer = true;
RNGModuleParameters RNGpar(Read);
SetRNGSeeds(RNGpar);
// Observables
auto &ObsFactory = HMC_ObservablesModuleFactory<observable_string, typename ImplementationPolicy::Field, ReaderClass>::getInstance();
Read.push(observable_string);// here must check if existing...
do {
std::string obs_type;
read(Read,"name", obs_type);
std::cout << "Registered types " << std::endl;
output_vector_string(ObsFactory.getBuilderList() );
ObservablesList.emplace_back(ObsFactory.create(obs_type, Read));
ObservablesList[ObservablesList.size() - 1]->print_parameters();
} while (Read.nextElement(observable_string));
Read.pop();
// Loop on levels
if(!Read.push("Actions")){
std::cout << "Actions not found" << std::endl;
exit(1);
}
if(!Read.push("Level")){// push must check if the node exist
std::cout << "Level not found" << std::endl;
exit(1);
}
do
{
fill_ActionsLevel(Read);
}
while(Read.push("Level"));
Read.pop();
}
template <class RepresentationPolicy>
void GetActionSet(ActionSet<typename ImplementationPolicy::Field, RepresentationPolicy>& Aset){
Aset.resize(multipliers.size());
for(auto it = ActionsList.begin(); it != ActionsList.end(); it++){
(*it).second->acquireResource(Grids["gauge"]);
Aset[(*it).first-1].push_back((*it).second->getPtr());
}
}
//////////////////////////////////////////////////////////////
// Grids
//////////////////////////////////////////////////////////////
void AddGrid(std::string s, GridModule& M) {
// Check for name clashes
auto search = Grids.find(s);
if (search != Grids.end()) {
std::cout << GridLogError << "Grid with name \"" << search->first
<< "\" already present. Terminating\n";
exit(1);
}
Grids[s] = std::move(M);
}
// Add a named grid set, 4d shortcut
void AddFourDimGrid(std::string s) {
GridFourDimModule<vComplex> Mod;
AddGrid(s, Mod);
}
GridCartesian* GetCartesian(std::string s = "") {
if (s.empty()) s = Grids.begin()->first;
std::cout << GridLogDebug << "Getting cartesian grid from: " << s
<< std::endl;
return Grids[s].get_full();
}
GridRedBlackCartesian* GetRBCartesian(std::string s = "") {
if (s.empty()) s = Grids.begin()->first;
std::cout << GridLogDebug << "Getting rb-cartesian grid from: " << s
<< std::endl;
return Grids[s].get_rb();
}
//////////////////////////////////////////////////////
// Random number generators
//////////////////////////////////////////////////////
void AddRNGs(std::string s = "") {
// Couple the RNGs to the GridModule tagged by s
// the default is the first grid registered
assert(Grids.size() > 0 && !have_RNG);
if (s.empty()) s = Grids.begin()->first;
std::cout << GridLogDebug << "Adding RNG to grid: " << s << std::endl;
RNGs.set_pRNG(new GridParallelRNG(GetCartesian(s)));
have_RNG = true;
}
void SetRNGSeeds(RNGModuleParameters& Params) { RNGs.set_RNGSeeds(Params); }
GridSerialRNG& GetSerialRNG() { return RNGs.get_sRNG(); }
GridParallelRNG& GetParallelRNG() {
assert(have_RNG);
return RNGs.get_pRNG();
}
void SeedFixedIntegers() {
assert(have_RNG);
RNGs.seed();
}
//////////////////////////////////////////////////////
// Checkpointers
//////////////////////////////////////////////////////
BaseHmcCheckpointer<ImplementationPolicy>* GetCheckPointer() {
if (have_CheckPointer)
return CP->getPtr();
else {
std::cout << GridLogError << "Error: no checkpointer defined"
<< std::endl;
exit(1);
}
}
RegisterLoadCheckPointerFunction(Binary);
RegisterLoadCheckPointerFunction(Nersc);
#ifdef HAVE_LIME
RegisterLoadCheckPointerFunction(ILDG);
#endif
////////////////////////////////////////////////////////
// Observables
////////////////////////////////////////////////////////
template<class T, class... Types>
void AddObservable(Types&&... Args){
ObservablesList.push_back(std::unique_ptr<T>(new T(std::forward<Types>(Args)...)));
ObservablesList.back()->print_parameters();
}
std::vector<HmcObservable<typename ImplementationPolicy::Field>* > GetObservables(){
std::vector<HmcObservable<typename ImplementationPolicy::Field>* > out;
for (auto &i : ObservablesList){
out.push_back(i->getPtr());
}
// Add the checkpointer to the observables
out.push_back(GetCheckPointer());
return out;
}
private:
// this private
template <class ReaderClass >
void fill_ActionsLevel(ReaderClass &Read){
// Actions set
int m;
Read.readDefault("multiplier",m);
multipliers.push_back(m);
std::cout << "Level : " << multipliers.size() << " with multiplier : " << m << std::endl;
// here gauge
Read.push("Action");
do{
auto &ActionFactory = HMC_ActionModuleFactory<gauge_string, typename ImplementationPolicy::Field, ReaderClass>::getInstance();
std::string action_type;
Read.readDefault("name", action_type);
output_vector_string(ActionFactory.getBuilderList() );
ActionsList.emplace(m, ActionFactory.create(action_type, Read));
} while (Read.nextElement("Action"));
ActionsList.find(m)->second->print_parameters();
Read.pop();
}
};
}
}
#endif // HMC_RESOURCE_MANAGER_H

137
lib/qcd/hmc/HMCRunnerModule.h Normal file
View File

@ -0,0 +1,137 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/GenericHmcRunner.h
Copyright (C) 2015
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 HMC_RUNNER_MODULE
#define HMC_RUNNER_MODULE
namespace Grid {
// the reader class is necessary here for the automatic initialization of the resources
// if we had a virtual reader would have been unecessary
template <class HMCType, class ReaderClass >
class HMCModule
: public Parametrized< QCD::HMCparameters >,
public HMCModuleBase< QCD::HMCRunnerBase<ReaderClass> > {
public:
typedef HMCModuleBase< QCD::HMCRunnerBase<ReaderClass> > Base;
typedef typename Base::Product Product;
std::unique_ptr<HMCType> HMCPtr;
HMCModule(QCD::HMCparameters Par) : Parametrized<QCD::HMCparameters>(Par) {}
template <class ReaderCl>
HMCModule(Reader<ReaderCl>& R) : Parametrized<QCD::HMCparameters>(R, "HMC"){};
Product* getPtr() {
if (!HMCPtr) initialize();
return HMCPtr.get();
}
private:
virtual void initialize() = 0;
};
// Factory
template <char const *str, class ReaderClass >
class HMCRunnerModuleFactory
: public Factory < HMCModuleBase< QCD::HMCRunnerBase<ReaderClass> > , Reader<ReaderClass> > {
public:
typedef Reader<ReaderClass> TheReader;
// use SINGLETON FUNCTOR MACRO HERE
HMCRunnerModuleFactory(const HMCRunnerModuleFactory& e) = delete;
void operator=(const HMCRunnerModuleFactory& e) = delete;
static HMCRunnerModuleFactory& getInstance(void) {
static HMCRunnerModuleFactory e;
return e;
}
private:
HMCRunnerModuleFactory(void) = default;
std::string obj_type() const {
return std::string(str);
}
};
///////////////
// macro for these
template < class ImplementationPolicy, class RepresentationPolicy, class ReaderClass >
class HMCLeapFrog: public HMCModule< QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::LeapFrog>, ReaderClass >{
typedef HMCModule< QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::LeapFrog>, ReaderClass > HMCBaseMod;
using HMCBaseMod::HMCBaseMod;
// aquire resource
virtual void initialize(){
this->HMCPtr.reset(new QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::LeapFrog>(this->Par_) );
}
};
template < class ImplementationPolicy, class RepresentationPolicy, class ReaderClass >
class HMCMinimumNorm2: public HMCModule< QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::MinimumNorm2>, ReaderClass >{
typedef HMCModule< QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::MinimumNorm2>, ReaderClass > HMCBaseMod;
using HMCBaseMod::HMCBaseMod;
// aquire resource
virtual void initialize(){
this->HMCPtr.reset(new QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::MinimumNorm2>(this->Par_));
}
};
template < class ImplementationPolicy, class RepresentationPolicy, class ReaderClass >
class HMCForceGradient: public HMCModule< QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::ForceGradient>, ReaderClass >{
typedef HMCModule< QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::ForceGradient>, ReaderClass > HMCBaseMod;
using HMCBaseMod::HMCBaseMod;
// aquire resource
virtual void initialize(){
this->HMCPtr.reset(new QCD::GenericHMCRunnerTemplate<ImplementationPolicy, RepresentationPolicy, QCD::ForceGradient>(this->Par_) );
}
};
extern char hmc_string[];
//////////////////////////////////////////////////////////////
}
#endif

133
lib/qcd/hmc/HMC_GridModules.h Normal file
View File

@ -0,0 +1,133 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/HMC_GridModules.h
Copyright (C) 2015
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 HMC_GRID_MODULES
#define HMC_GRID_MODULES
namespace Grid {
// Resources
// Modules for grids
// Introduce another namespace HMCModules?
class GridModuleParameters: Serializable{
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(GridModuleParameters,
std::string, lattice,
std::string, mpi);
std::vector<int> getLattice(){return strToVec<int>(lattice);}
std::vector<int> getMpi() {return strToVec<int>(mpi);}
void check(){
if (getLattice().size() != getMpi().size()) {
std::cout << GridLogError
<< "Error in GridModuleParameters: lattice and mpi dimensions "
"do not match"
<< std::endl;
exit(1);
}
}
template <class ReaderClass>
GridModuleParameters(Reader<ReaderClass>& Reader, std::string n = "LatticeGrid"):name(n) {
read(Reader, name, *this);
check();
}
// Save on file
template< class WriterClass>
void save(Writer<WriterClass>& Writer){
check();
write(Writer, name, *this);
}
private:
std::string name;
};
// Lower level class
class GridModule {
public:
GridCartesian* get_full() {
std::cout << GridLogDebug << "Getting cartesian in module"<< std::endl;
return grid_.get(); }
GridRedBlackCartesian* get_rb() {
std::cout << GridLogDebug << "Getting rb-cartesian in module"<< std::endl;
return rbgrid_.get(); }
void set_full(GridCartesian* grid) { grid_.reset(grid); }
void set_rb(GridRedBlackCartesian* rbgrid) { rbgrid_.reset(rbgrid); }
protected:
std::unique_ptr<GridCartesian> grid_;
std::unique_ptr<GridRedBlackCartesian> rbgrid_;
};
////////////////////////////////////
// Classes for the user
////////////////////////////////////
// Note: the space time grid should be out of the QCD namespace
template< class vector_type>
class GridFourDimModule : public GridModule {
public:
GridFourDimModule() {
using namespace QCD;
set_full(SpaceTimeGrid::makeFourDimGrid(
GridDefaultLatt(), GridDefaultSimd(4, vector_type::Nsimd()),
GridDefaultMpi()));
set_rb(SpaceTimeGrid::makeFourDimRedBlackGrid(grid_.get()));
}
GridFourDimModule(GridModuleParameters Params) {
using namespace QCD;
Params.check();
std::vector<int> lattice_v = Params.getLattice();
std::vector<int> mpi_v = Params.getMpi();
if (lattice_v.size() == 4) {
set_full(SpaceTimeGrid::makeFourDimGrid(
lattice_v, GridDefaultSimd(4, vector_type::Nsimd()),
mpi_v));
set_rb(SpaceTimeGrid::makeFourDimRedBlackGrid(grid_.get()));
} else {
std::cout << GridLogError
<< "Error in GridFourDimModule: lattice dimension different from 4"
<< std::endl;
exit(1);
}
}
};
typedef GridFourDimModule<vComplex> GridDefaultFourDimModule;
} // namespace Grid
#endif // HMC_GRID_MODULES

14
lib/qcd/hmc/HMC_aggregate.h
View File

@ -33,10 +33,20 @@ directory
#include <string>
#include <Grid/qcd/observables/hmc_observable.h>
#include <Grid/qcd/hmc/HMC.h>
// annoying location; should move this ?
#include <Grid/parallelIO/IldgIOtypes.h>
#include <Grid/parallelIO/IldgIO.h>
#include <Grid/parallelIO/NerscIO.h>
#include <Grid/qcd/hmc/NerscCheckpointer.h>
#include <Grid/qcd/hmc/HmcRunner.h>
#include <Grid/qcd/hmc/checkpointers/CheckPointers.h>
#include <Grid/qcd/hmc/HMCModules.h>
#include <Grid/qcd/modules/mods.h>
#include <Grid/qcd/hmc/HMCResourceManager.h>
#include <Grid/qcd/hmc/GenericHMCrunner.h>
#include <Grid/qcd/hmc/HMCRunnerModule.h>
#endif

191
lib/qcd/hmc/HmcRunner.h
View File

@ -1,191 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/HmcRunner.h
Copyright (C) 2015
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 HMC_RUNNER
#define HMC_RUNNER
namespace Grid {
namespace QCD {
template <class Gimpl, class RepresentationsPolicy = NoHirep >
class NerscHmcRunnerTemplate {
public:
INHERIT_GIMPL_TYPES(Gimpl);
enum StartType_t { ColdStart, HotStart, TepidStart, CheckpointStart };
ActionSet<GaugeField, RepresentationsPolicy> TheAction;
GridCartesian *UGrid;
GridCartesian *FGrid;
GridRedBlackCartesian *UrbGrid;
GridRedBlackCartesian *FrbGrid;
virtual void BuildTheAction(int argc, char **argv) = 0; // necessary?
void Run(int argc, char **argv) {
StartType_t StartType = HotStart;
std::string arg;
if (GridCmdOptionExists(argv, argv + argc, "--StartType")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--StartType");
if (arg == "HotStart") {
StartType = HotStart;
} else if (arg == "ColdStart") {
StartType = ColdStart;
} else if (arg == "TepidStart") {
StartType = TepidStart;
} else if (arg == "CheckpointStart") {
StartType = CheckpointStart;
} else {
std::cout << GridLogError << "Unrecognized option in --StartType\n";
std::cout << GridLogError << "Valid [HotStart, ColdStart, TepidStart, CheckpointStart]\n";
assert(0);
}
}
int StartTraj = 0;
if (GridCmdOptionExists(argv, argv + argc, "--StartTrajectory")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--StartTrajectory");
std::vector<int> ivec(0);
GridCmdOptionIntVector(arg, ivec);
StartTraj = ivec[0];
}
int NumTraj = 1;
if (GridCmdOptionExists(argv, argv + argc, "--Trajectories")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--Trajectories");
std::vector<int> ivec(0);
GridCmdOptionIntVector(arg, ivec);
NumTraj = ivec[0];
}
int NumThermalizations = 10;
if (GridCmdOptionExists(argv, argv + argc, "--Thermalizations")) {
arg = GridCmdOptionPayload(argv, argv + argc, "--Thermalizations");
std::vector<int> ivec(0);
GridCmdOptionIntVector(arg, ivec);
NumThermalizations = ivec[0];
}
GridSerialRNG sRNG;
GridParallelRNG pRNG(UGrid);
LatticeGaugeField U(UGrid); // change this to an extended field (smearing class)?
std::vector<int> SerSeed({1, 2, 3, 4, 5});
std::vector<int> ParSeed({6, 7, 8, 9, 10});
// Create integrator, including the smearing policy
// Smearing policy, only defined for Nc=3
/*
std::cout << GridLogDebug << " Creating the Stout class\n";
double rho = 0.1; // smearing parameter, now hardcoded
int Nsmear = 1; // number of smearing levels
Smear_Stout<Gimpl> Stout(rho);
std::cout << GridLogDebug << " Creating the SmearedConfiguration class\n";
//SmearedConfiguration<Gimpl> SmearingPolicy(UGrid, Nsmear, Stout);
std::cout << GridLogDebug << " done\n";
*/
//////////////
NoSmearing<Gimpl> SmearingPolicy;
// change here to change the algorithm
typedef MinimumNorm2<GaugeField, NoSmearing<Gimpl>, RepresentationsPolicy > IntegratorType;
IntegratorParameters MDpar(40, 1.0);
IntegratorType MDynamics(UGrid, MDpar, TheAction, SmearingPolicy);
// Checkpoint strategy
NerscHmcCheckpointer<Gimpl> Checkpoint(std::string("ckpoint_lat"),
std::string("ckpoint_rng"), 1);
PlaquetteLogger<Gimpl> PlaqLog(std::string("plaq"));
HMCparameters HMCpar;
HMCpar.StartTrajectory = StartTraj;
HMCpar.Trajectories = NumTraj;
HMCpar.NoMetropolisUntil = NumThermalizations;
if (StartType == HotStart) {
// Hot start
HMCpar.MetropolisTest = true;
sRNG.SeedFixedIntegers(SerSeed);
pRNG.SeedFixedIntegers(ParSeed);
SU<Nc>::HotConfiguration(pRNG, U);
} else if (StartType == ColdStart) {
// Cold start
HMCpar.MetropolisTest = true;
sRNG.SeedFixedIntegers(SerSeed);
pRNG.SeedFixedIntegers(ParSeed);
SU<Nc>::ColdConfiguration(pRNG, U);
} else if (StartType == TepidStart) {
// Tepid start
HMCpar.MetropolisTest = true;
sRNG.SeedFixedIntegers(SerSeed);
pRNG.SeedFixedIntegers(ParSeed);
SU<Nc>::TepidConfiguration(pRNG, U);
} else if (StartType == CheckpointStart) {
HMCpar.MetropolisTest = true;
// CheckpointRestart
Checkpoint.CheckpointRestore(StartTraj, U, sRNG, pRNG);
}
// Attach the gauge field to the smearing Policy and create the fill the
// smeared set
// notice that the unit configuration is singular in this procedure
std::cout << GridLogMessage << "Filling the smeared set\n";
SmearingPolicy.set_GaugeField(U);
HybridMonteCarlo<GaugeField, IntegratorType> HMC(HMCpar, MDynamics, sRNG,
pRNG, U);
HMC.AddObservable(&Checkpoint);
HMC.AddObservable(&PlaqLog);
// Run it
HMC.evolve();
}
};
typedef NerscHmcRunnerTemplate<PeriodicGimplR> NerscHmcRunner;
typedef NerscHmcRunnerTemplate<PeriodicGimplF> NerscHmcRunnerF;
typedef NerscHmcRunnerTemplate<PeriodicGimplD> NerscHmcRunnerD;
typedef NerscHmcRunnerTemplate<PeriodicGimplR> PeriodicNerscHmcRunner;
typedef NerscHmcRunnerTemplate<PeriodicGimplF> PeriodicNerscHmcRunnerF;
typedef NerscHmcRunnerTemplate<PeriodicGimplD> PeriodicNerscHmcRunnerD;
typedef NerscHmcRunnerTemplate<ConjugateGimplR> ConjugateNerscHmcRunner;
typedef NerscHmcRunnerTemplate<ConjugateGimplF> ConjugateNerscHmcRunnerF;
typedef NerscHmcRunnerTemplate<ConjugateGimplD> ConjugateNerscHmcRunnerD;
template <class RepresentationsPolicy>
using NerscHmcRunnerHirep = NerscHmcRunnerTemplate<PeriodicGimplR, RepresentationsPolicy>;
}
}
#endif

80
lib/qcd/hmc/NerscCheckpointer.h
View File

@ -1,80 +0,0 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/NerscCheckpointer.h
Copyright (C) 2015
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 NERSC_CHECKPOINTER
#define NERSC_CHECKPOINTER
#include <string>
#include <iostream>
#include <sstream>
namespace Grid{
namespace QCD{
template<class Gimpl>
class NerscHmcCheckpointer : public HmcObservable<typename Gimpl::GaugeField> {
private:
std::string configStem;
std::string rngStem;
int SaveInterval;
public:
INHERIT_GIMPL_TYPES(Gimpl);
NerscHmcCheckpointer(std::string cf, std::string rn,int savemodulo) {
configStem = cf;
rngStem = rn;
SaveInterval= savemodulo;
};
void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG, GridParallelRNG & pRNG )
{
if ( (traj % SaveInterval)== 0 ) {
std::string rng; { std::ostringstream os; os << rngStem <<"."<< traj; rng = os.str(); }
std::string config;{ std::ostringstream os; os << configStem <<"."<< traj; config = os.str();}
int precision32=1;
int tworow =0;
NerscIO::writeRNGState(sRNG,pRNG,rng);
NerscIO::writeConfiguration(U,config,tworow,precision32);
}
};
void CheckpointRestore(int traj, GaugeField &U, GridSerialRNG &sRNG, GridParallelRNG & pRNG ){
std::string rng; { std::ostringstream os; os << rngStem <<"."<< traj; rng = os.str(); }
std::string config;{ std::ostringstream os; os << configStem <<"."<< traj; config = os.str();}
NerscField header;
NerscIO::readRNGState(sRNG,pRNG,header,rng);
NerscIO::readConfiguration(U,header,config);
};
};
}}
#endif

107
lib/qcd/hmc/UsingHMC.md Normal file
View File

@ -0,0 +1,107 @@
Using HMC in Grid version 0.5.1
These are the instructions to use the Generalised HMC on Grid version 0.5.1.
Disclaimer: GRID is still under active development so any information here can be changed in future releases.
Command line options
===================
(relevant file GenericHMCrunner.h)
The initial configuration can be changed at the command line using
--StartType <your choice>
valid choices, one among these
HotStart, ColdStart, TepidStart, CheckpointStart
default: HotStart
example
./My_hmc_exec --StartType HotStart
The CheckpointStart option uses the prefix for the configurations and rng seed files defined in your executable and the initial configuration is specified by
--StartTrajectory <integer>
default: 0
The number of trajectories for a specific run are specified at command line by
--Trajectories <integer>
default: 1
The number of thermalization steps (i.e. steps when the Metropolis acceptance check is turned off) is specified by
--Thermalizations <integer>
default: 10
Any other parameter is defined in the source for the executable.
HMC controls
===========
The lines
std::vector<int> SerSeed({1, 2, 3, 4, 5});
std::vector<int> ParSeed({6, 7, 8, 9, 10});
define the seeds for the serial and the parallel RNG.
The line
TheHMC.MDparameters.set(20, 1.0);// MDsteps, traj length
declares the number of molecular dynamics steps and the total trajectory length.
Actions
======
Action names are defined in the file
lib/qcd/Actions.h
Gauge actions list:
WilsonGaugeActionR;
WilsonGaugeActionF;
WilsonGaugeActionD;
PlaqPlusRectangleActionR;
PlaqPlusRectangleActionF;
PlaqPlusRectangleActionD;
IwasakiGaugeActionR;
IwasakiGaugeActionF;
IwasakiGaugeActionD;
SymanzikGaugeActionR;
SymanzikGaugeActionF;
SymanzikGaugeActionD;
ConjugateWilsonGaugeActionR;
ConjugateWilsonGaugeActionF;
ConjugateWilsonGaugeActionD;
ConjugatePlaqPlusRectangleActionR;
ConjugatePlaqPlusRectangleActionF;
ConjugatePlaqPlusRectangleActionD;
ConjugateIwasakiGaugeActionR;
ConjugateIwasakiGaugeActionF;
ConjugateIwasakiGaugeActionD;
ConjugateSymanzikGaugeActionR;
ConjugateSymanzikGaugeActionF;
ConjugateSymanzikGaugeActionD;
ScalarActionR;
ScalarActionF;
ScalarActionD;
each of these action accept one single parameter at creation time (beta).
Example for creating a Symanzik action with beta=4.0
SymanzikGaugeActionR(4.0)
The suffixes R,F,D in the action names refer to the Real
(the precision is defined at compile time by the --enable-precision flag in the configure),
Float and Double, that force the precision of the action to be 32, 64 bit respectively.

89
lib/qcd/hmc/checkpointers/BaseCheckpointer.h Normal file
View File

@ -0,0 +1,89 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/BaseCheckpointer.h
Copyright (C) 2015
Author: Guido Cossu <guido.cossu@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 BASE_CHECKPOINTER
#define BASE_CHECKPOINTER
namespace Grid {
namespace QCD {
class CheckpointerParameters : Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(CheckpointerParameters,
std::string, config_prefix,
std::string, rng_prefix,
int, saveInterval,
std::string, format, );
CheckpointerParameters(std::string cf = "cfg", std::string rn = "rng",
int savemodulo = 1, const std::string &f = "IEEE64BIG")
: config_prefix(cf),
rng_prefix(rn),
saveInterval(savemodulo),
format(f){};
template <class ReaderClass >
CheckpointerParameters(Reader<ReaderClass> &Reader) {
read(Reader, "Checkpointer", *this);
}
};
//////////////////////////////////////////////////////////////////////////////
// Base class for checkpointers
template <class Impl>
class BaseHmcCheckpointer : public HmcObservable<typename Impl::Field> {
public:
void build_filenames(int traj, CheckpointerParameters &Params,
std::string &conf_file, std::string &rng_file) {
{
std::ostringstream os;
os << Params.rng_prefix << "." << traj;
rng_file = os.str();
}
{
std::ostringstream os;
os << Params.config_prefix << "." << traj;
conf_file = os.str();
}
}
virtual void initialize(const CheckpointerParameters &Params) = 0;
virtual void CheckpointRestore(int traj, typename Impl::Field &U,
GridSerialRNG &sRNG,
GridParallelRNG &pRNG) = 0;
}; // class BaseHmcCheckpointer
///////////////////////////////////////////////////////////////////////////////
}
}
#endif

113
lib/qcd/hmc/checkpointers/BinaryCheckpointer.h Normal file
View File

@ -0,0 +1,113 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/BinaryCheckpointer.h
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 BINARY_CHECKPOINTER
#define BINARY_CHECKPOINTER
#include <iostream>
#include <sstream>
#include <string>
namespace Grid {
namespace QCD {
// Simple checkpointer, only binary file
template <class Impl>
class BinaryHmcCheckpointer : public BaseHmcCheckpointer<Impl> {
private:
CheckpointerParameters Params;
public:
INHERIT_FIELD_TYPES(Impl); // Gets the Field type, a Lattice object
// Extract types from the Field
typedef typename Field::vector_object vobj;
typedef typename vobj::scalar_object sobj;
typedef typename getPrecision<sobj>::real_scalar_type sobj_stype;
typedef typename sobj::DoublePrecision sobj_double;
BinaryHmcCheckpointer(const CheckpointerParameters &Params_) {
initialize(Params_);
}
void initialize(const CheckpointerParameters &Params_) { Params = Params_; }
void truncate(std::string file) {
std::ofstream fout(file, std::ios::out);
fout.close();
}
void TrajectoryComplete(int traj, Field &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG) {
if ((traj % Params.saveInterval) == 0) {
std::string config, rng;
this->build_filenames(traj, Params, config, rng);
uint32_t nersc_csum;
uint32_t scidac_csuma;
uint32_t scidac_csumb;
BinarySimpleUnmunger<sobj_double, sobj> munge;
truncate(rng);
BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
truncate(config);
BinaryIO::writeLatticeObject<vobj, sobj_double>(U, config, munge, 0, Params.format,
nersc_csum,scidac_csuma,scidac_csumb);
std::cout << GridLogMessage << "Written Binary Configuration " << config
<< " checksum " << std::hex
<< nersc_csum <<"/"
<< scidac_csuma <<"/"
<< scidac_csumb
<< std::dec << std::endl;
}
};
void CheckpointRestore(int traj, Field &U, GridSerialRNG &sRNG, GridParallelRNG &pRNG) {
std::string config, rng;
this->build_filenames(traj, Params, config, rng);
BinarySimpleMunger<sobj_double, sobj> munge;
uint32_t nersc_csum;
uint32_t scidac_csuma;
uint32_t scidac_csumb;
BinaryIO::readRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
BinaryIO::readLatticeObject<vobj, sobj_double>(U, config, munge, 0, Params.format,
nersc_csum,scidac_csuma,scidac_csumb);
std::cout << GridLogMessage << "Read Binary Configuration " << config
<< " checksums " << std::hex << nersc_csum<<"/"<<scidac_csuma<<"/"<<scidac_csumb
<< std::dec << std::endl;
};
};
}
}
#endif

158
lib/qcd/hmc/checkpointers/CheckPointerModules.h Normal file
View File

@ -0,0 +1,158 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/WilsonGaugeAction.h
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 CP_MODULES_H
#define CP_MODULES_H
// FIXME Reorganize QCD namespace
namespace Grid {
////////////////////////////////////////////////////////////////////////
// Checkpoint module, owns the Checkpointer
////////////////////////////////////////////////////////////////////////
template <class ImplementationPolicy>
class CheckPointerModule: public Parametrized<QCD::CheckpointerParameters>, public HMCModuleBase< QCD::BaseHmcCheckpointer<ImplementationPolicy> > {
public:
std::unique_ptr<QCD::BaseHmcCheckpointer<ImplementationPolicy> > CheckPointPtr;
typedef QCD::CheckpointerParameters APar;
typedef HMCModuleBase< QCD::BaseHmcCheckpointer<ImplementationPolicy> > Base;
typedef typename Base::Product Product;
CheckPointerModule(APar Par): Parametrized<APar>(Par) {}
template <class ReaderClass>
CheckPointerModule(Reader<ReaderClass>& Reader) : Parametrized<APar>(Reader){};
virtual void print_parameters(){
std::cout << this->Par_ << std::endl;
}
Product* getPtr() {
if (!CheckPointPtr) initialize();
return CheckPointPtr.get();
}
private:
virtual void initialize() = 0;
};
template <char const *str, class ImplementationPolicy, class ReaderClass >
class HMC_CPModuleFactory
: public Factory < HMCModuleBase< QCD::BaseHmcCheckpointer<ImplementationPolicy> > , Reader<ReaderClass> > {
public:
typedef Reader<ReaderClass> TheReader;
// use SINGLETON FUNCTOR MACRO HERE
HMC_CPModuleFactory(const HMC_CPModuleFactory& e) = delete;
void operator=(const HMC_CPModuleFactory& e) = delete;
static HMC_CPModuleFactory& getInstance(void) {
static HMC_CPModuleFactory e;
return e;
}
private:
HMC_CPModuleFactory(void) = default;
std::string obj_type() const {
return std::string(str);
}
};
/////////////////////////////////////////////////////////////////////
// Concrete classes
/////////////////////////////////////////////////////////////////////
namespace QCD{
template<class ImplementationPolicy>
class BinaryCPModule: public CheckPointerModule< ImplementationPolicy> {
typedef CheckPointerModule< ImplementationPolicy> CPBase;
using CPBase::CPBase; // for constructors
// acquire resource
virtual void initialize(){
this->CheckPointPtr.reset(new BinaryHmcCheckpointer<ImplementationPolicy>(this->Par_));
}
};
template<class ImplementationPolicy>
class NerscCPModule: public CheckPointerModule< ImplementationPolicy> {
typedef CheckPointerModule< ImplementationPolicy> CPBase;
using CPBase::CPBase; // for constructors inheritance
// acquire resource
virtual void initialize(){
this->CheckPointPtr.reset(new NerscHmcCheckpointer<ImplementationPolicy>(this->Par_));
}
};
#ifdef HAVE_LIME
template<class ImplementationPolicy>
class ILDGCPModule: public CheckPointerModule< ImplementationPolicy> {
typedef CheckPointerModule< ImplementationPolicy> CPBase;
using CPBase::CPBase; // for constructors
// acquire resource
virtual void initialize(){
this->CheckPointPtr.reset(new ILDGHmcCheckpointer<ImplementationPolicy>(this->Par_));
}
};
#endif
}// QCD temporarily here
extern char cp_string[];
/*
// use macros?
static Registrar<QCD::BinaryCPModule<QCD::PeriodicGimplR>, HMC_CPModuleFactory<cp_string, QCD::PeriodicGimplR, XmlReader> > __CPBinarymodXMLInit("Binary");
static Registrar<QCD::NerscCPModule<QCD::PeriodicGimplR> , HMC_CPModuleFactory<cp_string, QCD::PeriodicGimplR, XmlReader> > __CPNerscmodXMLInit("Nersc");
#ifdef HAVE_LIME
static Registrar<QCD::ILDGCPModule<QCD::PeriodicGimplR> , HMC_CPModuleFactory<cp_string, QCD::PeriodicGimplR, XmlReader> > __CPILDGmodXMLInit("ILDG");
#endif
*/
}// Grid
#endif //CP_MODULES_H

40
lib/qcd/hmc/checkpointers/CheckPointers.h Normal file
View File

@ -0,0 +1,40 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/action/gauge/WilsonGaugeAction.h
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 CHECKPOINTERS_H
#define CHECKPOINTERS_H
#include <Grid/qcd/hmc/checkpointers/BaseCheckpointer.h>
#include <Grid/qcd/hmc/checkpointers/NerscCheckpointer.h>
#include <Grid/qcd/hmc/checkpointers/BinaryCheckpointer.h>
#include <Grid/qcd/hmc/checkpointers/ILDGCheckpointer.h>
//#include <Grid/qcd/hmc/checkpointers/CheckPointerModules.h>
#endif // CHECKPOINTERS_H

120
lib/qcd/hmc/checkpointers/ILDGCheckpointer.h Normal file
View File

@ -0,0 +1,120 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/ILDGCheckpointer.h
Copyright (C) 2016
Author: Guido Cossu <guido.cossu@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 ILDG_CHECKPOINTER
#define ILDG_CHECKPOINTER
#ifdef HAVE_LIME
#include <iostream>
#include <sstream>
#include <string>
namespace Grid {
namespace QCD {
// Only for Gauge fields
template <class Implementation>
class ILDGHmcCheckpointer : public BaseHmcCheckpointer<Implementation> {
private:
CheckpointerParameters Params;
public:
INHERIT_GIMPL_TYPES(Implementation);
ILDGHmcCheckpointer(const CheckpointerParameters &Params_) { initialize(Params_); }
void initialize(const CheckpointerParameters &Params_) {
Params = Params_;
// check here that the format is valid
int ieee32big = (Params.format == std::string("IEEE32BIG"));
int ieee32 = (Params.format == std::string("IEEE32"));
int ieee64big = (Params.format == std::string("IEEE64BIG"));
int ieee64 = (Params.format == std::string("IEEE64"));
if (!(ieee64big || ieee32 || ieee32big || ieee64)) {
std::cout << GridLogError << "Unrecognized file format " << Params.format
<< std::endl;
std::cout << GridLogError
<< "Allowed: IEEE32BIG | IEEE32 | IEEE64BIG | IEEE64"
<< std::endl;
exit(1);
}
}
void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
if ((traj % Params.saveInterval) == 0) {
std::string config, rng;
this->build_filenames(traj, Params, config, rng);
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::writeRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
IldgWriter _IldgWriter;
_IldgWriter.open(config);
_IldgWriter.writeConfiguration(U, traj, config, config);
_IldgWriter.close();
std::cout << GridLogMessage << "Written ILDG Configuration on " << config
<< " checksum " << std::hex
<< nersc_csum<<"/"
<< scidac_csuma<<"/"
<< scidac_csumb
<< std::dec << std::endl;
}
};
void CheckpointRestore(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
std::string config, rng;
this->build_filenames(traj, Params, config, rng);
uint32_t nersc_csum,scidac_csuma,scidac_csumb;
BinaryIO::readRNG(sRNG, pRNG, rng, 0,nersc_csum,scidac_csuma,scidac_csumb);
FieldMetaData header;
IldgReader _IldgReader;
_IldgReader.open(config);
_IldgReader.readConfiguration(U,header); // format from the header
_IldgReader.close();
std::cout << GridLogMessage << "Read ILDG Configuration from " << config
<< " checksum " << std::hex
<< nersc_csum<<"/"
<< scidac_csuma<<"/"
<< scidac_csumb
<< std::dec << std::endl;
};
};
}
}
#endif // HAVE_LIME
#endif // ILDG_CHECKPOINTER

80
lib/qcd/hmc/checkpointers/NerscCheckpointer.h Normal file
View File

@ -0,0 +1,80 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/qcd/hmc/NerscCheckpointer.h
Copyright (C) 2015
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 NERSC_CHECKPOINTER
#define NERSC_CHECKPOINTER
#include <iostream>
#include <sstream>
#include <string>
namespace Grid {
namespace QCD {
// Only for Gauge fields
template <class Gimpl>
class NerscHmcCheckpointer : public BaseHmcCheckpointer<Gimpl> {
private:
CheckpointerParameters Params;
public:
INHERIT_GIMPL_TYPES(Gimpl); // only for gauge configurations
NerscHmcCheckpointer(const CheckpointerParameters &Params_) { initialize(Params_); }
void initialize(const CheckpointerParameters &Params_) {
Params = Params_;
Params.format = "IEEE64BIG"; // fixed, overwrite any other choice
}
void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
if ((traj % Params.saveInterval) == 0) {
std::string config, rng;
this->build_filenames(traj, Params, config, rng);
int precision32 = 1;
int tworow = 0;
NerscIO::writeRNGState(sRNG, pRNG, rng);
NerscIO::writeConfiguration(U, config, tworow, precision32);
}
};
void CheckpointRestore(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
std::string config, rng;
this->build_filenames(traj, Params, config, rng);
FieldMetaData header;
NerscIO::readRNGState(sRNG, pRNG, header, rng);
NerscIO::readConfiguration(U, header, config);
};
};
}
}
#endif

211
lib/qcd/hmc/integrators/Integrator.h
View File

@ -8,8 +8,7 @@ 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: Guido Cossu <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
@ -30,79 +29,62 @@ directory
*************************************************************************************/
/* END LEGAL */
//--------------------------------------------------------------------
/*! @file Integrator.h
* @brief Classes for the Molecular Dynamics integrator
*
* @author Guido Cossu
* Time-stamp: <2015-07-30 16:21:29 neo>
*/
//--------------------------------------------------------------------
#ifndef INTEGRATOR_INCLUDED
#define INTEGRATOR_INCLUDED
// class Observer;
#include <memory>
namespace Grid {
namespace QCD {
struct IntegratorParameters {
int Nexp;
int MDsteps; // number of outer steps
RealD trajL; // trajectory length
RealD stepsize;
class IntegratorParameters: Serializable {
public:
GRID_SERIALIZABLE_CLASS_MEMBERS(IntegratorParameters,
std::string, name, // name of the integrator
unsigned int, MDsteps, // number of outer steps
RealD, trajL, // trajectory length
)
IntegratorParameters(int MDsteps_, RealD trajL_ = 1.0, int Nexp_ = 12)
: Nexp(Nexp_),
MDsteps(MDsteps_),
trajL(trajL_),
stepsize(trajL / MDsteps){
// empty body constructor
};
IntegratorParameters(int MDsteps_ = 10, RealD trajL_ = 1.0)
: MDsteps(MDsteps_),
trajL(trajL_){
// empty body constructor
};
template <class ReaderClass, typename std::enable_if<isReader<ReaderClass>::value, int >::type = 0 >
IntegratorParameters(ReaderClass & Reader){
std::cout << "Reading integrator\n";
read(Reader, "Integrator", *this);
}
void print_parameters() const {
std::cout << GridLogMessage << "[Integrator] Type : " << name << std::endl;
std::cout << GridLogMessage << "[Integrator] Trajectory length : " << trajL << std::endl;
std::cout << GridLogMessage << "[Integrator] Number of MD steps : " << MDsteps << std::endl;
std::cout << GridLogMessage << "[Integrator] Step size : " << trajL/MDsteps << std::endl;
}
};
/*! @brief Class for Molecular Dynamics management */
template <class GaugeField, class SmearingPolicy, class RepresentationPolicy>
template <class FieldImplementation, class SmearingPolicy, class RepresentationPolicy>
class Integrator {
protected:
typedef IntegratorParameters ParameterType;
typedef typename FieldImplementation::Field MomentaField; //for readability
typedef typename FieldImplementation::Field Field;
int levels; // number of integration levels
double t_U; // Track time passing on each level and for U and for P
std::vector<double> t_P;
MomentaField P;
SmearingPolicy& Smearer;
RepresentationPolicy Representations;
IntegratorParameters Params;
const ActionSet<GaugeField, RepresentationPolicy> as;
const ActionSet<Field, RepresentationPolicy> as;
int levels; //
double t_U; // Track time passing on each level and for U and for P
std::vector<double> t_P; //
GaugeField P;
SmearingPolicy& Smearer;
RepresentationPolicy Representations;
// Should match any legal (SU(n)) gauge field
// Need to use this template to match Ncol to pass to SU<N> class
template <int Ncol, class vec>
void generate_momenta(Lattice<iVector<iScalar<iMatrix<vec, Ncol> >, Nd> >& P,
GridParallelRNG& pRNG) {
typedef Lattice<iScalar<iScalar<iMatrix<vec, Ncol> > > > GaugeLinkField;
GaugeLinkField Pmu(P._grid);
Pmu = zero;
for (int mu = 0; mu < Nd; mu++) {
SU<Ncol>::GaussianFundamentalLieAlgebraMatrix(pRNG, Pmu);
PokeIndex<LorentzIndex>(P, Pmu, mu);
}
}
// ObserverList observers; // not yet
// typedef std::vector<Observer*> ObserverList;
// void register_observers();
// void notify_observers();
void update_P(GaugeField& U, int level, double ep) {
void update_P(Field& U, int level, double ep) {
t_P[level] += ep;
update_P(P, U, level, ep);
@ -120,67 +102,60 @@ class Integrator {
FieldType forceR(U._grid);
// Implement smearing only for the fundamental representation now
repr_set.at(a)->deriv(Rep.U, forceR);
GF force =
Rep.RtoFundamentalProject(forceR); // Ta for the fundamental rep
std::cout << GridLogIntegrator << "Hirep Force average: "
<< norm2(force) / (U._grid->gSites()) << std::endl;
GF force = Rep.RtoFundamentalProject(forceR); // Ta for the fundamental rep
Real force_abs = std::sqrt(norm2(force)/(U._grid->gSites()));
std::cout << GridLogIntegrator << "Hirep Force average: " << force_abs << std::endl;
Mom -= force * ep ;
}
}
} update_P_hireps{};
void update_P(GaugeField& Mom, GaugeField& U, int level, double ep) {
void update_P(MomentaField& Mom, Field& U, int level, double ep) {
// input U actually not used in the fundamental case
// Fundamental updates, include smearing
for (int a = 0; a < as[level].actions.size(); ++a) {
GaugeField force(U._grid);
GaugeField& Us = Smearer.get_U(as[level].actions.at(a)->is_smeared);
for (int a = 0; a < as[level].actions.size(); ++a) {
Field force(U._grid);
conformable(U._grid, Mom._grid);
Field& Us = Smearer.get_U(as[level].actions.at(a)->is_smeared);
as[level].actions.at(a)->deriv(Us, force); // deriv should NOT include Ta
std::cout << GridLogIntegrator
<< "Smearing (on/off): " << as[level].actions.at(a)->is_smeared
<< std::endl;
std::cout << GridLogIntegrator << "Smearing (on/off): " << as[level].actions.at(a)->is_smeared << std::endl;
if (as[level].actions.at(a)->is_smeared) Smearer.smeared_force(force);
force = Ta(force);
std::cout << GridLogIntegrator
<< "Force average: " << norm2(force) / (U._grid->gSites())
<< std::endl;
Mom -= force * ep;
force = FieldImplementation::projectForce(force); // Ta for gauge fields
Real force_abs = std::sqrt(norm2(force)/U._grid->gSites());
std::cout << GridLogIntegrator << "Force average: " << force_abs << std::endl;
Mom -= force * ep;
}
// Force from the other representations
as[level].apply(update_P_hireps, Representations, Mom, U, ep);
}
void update_U(GaugeField& U, double ep) {
void update_U(Field& U, double ep) {
update_U(P, U, ep);
t_U += ep;
int fl = levels - 1;
std::cout << GridLogIntegrator << " "
<< "[" << fl << "] U "
<< " dt " << ep << " : t_U " << t_U << std::endl;
std::cout << GridLogIntegrator << " " << "[" << fl << "] U " << " dt " << ep << " : t_U " << t_U << std::endl;
}
void update_U(GaugeField& Mom, GaugeField& U, double ep) {
// rewrite exponential to deal internally with the lorentz index?
for (int mu = 0; mu < Nd; mu++) {
auto Umu = PeekIndex<LorentzIndex>(U, mu);
auto Pmu = PeekIndex<LorentzIndex>(Mom, mu);
Umu = expMat(Pmu, ep, Params.Nexp) * Umu;
PokeIndex<LorentzIndex>(U, ProjectOnGroup(Umu), mu);
}
void update_U(MomentaField& Mom, Field& U, double ep) {
// exponential of Mom*U in the gauge fields case
FieldImplementation::update_field(Mom, U, ep);
// Update the smeared fields, can be implemented as observer
Smearer.set_GaugeField(U);
Smearer.set_Field(U);
// Update the higher representations fields
Representations.update(U); // void functions if fundamental representation
}
virtual void step(GaugeField& U, int level, int first, int last) = 0;
virtual void step(Field& U, int level, int first, int last) = 0;
public:
Integrator(GridBase* grid, IntegratorParameters Par,
ActionSet<GaugeField, RepresentationPolicy>& Aset,
ActionSet<Field, RepresentationPolicy>& Aset,
SmearingPolicy& Sm)
: Params(Par),
as(Aset),
@ -195,6 +170,31 @@ class Integrator {
virtual ~Integrator() {}
virtual std::string integrator_name() = 0;
void print_parameters(){
std::cout << GridLogMessage << "[Integrator] Name : "<< integrator_name() << std::endl;
Params.print_parameters();
}
void print_actions(){
std::cout << GridLogMessage << ":::::::::::::::::::::::::::::::::::::::::" << std::endl;
std::cout << GridLogMessage << "[Integrator] Action summary: "<<std::endl;
for (int level = 0; level < as.size(); ++level) {
std::cout << GridLogMessage << "[Integrator] ---- Level: "<< level << std::endl;
for (int actionID = 0; actionID < as[level].actions.size(); ++actionID) {
std::cout << GridLogMessage << "["<< as[level].actions.at(actionID)->action_name() << "] ID: " << actionID << std::endl;
std::cout << as[level].actions.at(actionID)->LogParameters();
}
}
std::cout << GridLogMessage << ":::::::::::::::::::::::::::::::::::::::::"<< std::endl;
}
void reverse_momenta(){
P *= -1.0;
}
// to be used by the actionlevel class to iterate
// over the representations
struct _refresh {
@ -210,14 +210,16 @@ class Integrator {
} refresh_hireps{};
// Initialization of momenta and actions
void refresh(GaugeField& U, GridParallelRNG& pRNG) {
void refresh(Field& U, GridParallelRNG& pRNG) {
assert(P._grid == U._grid);
std::cout << GridLogIntegrator << "Integrator refresh\n";
generate_momenta(P, pRNG);
FieldImplementation::generate_momenta(P, pRNG);
// Update the smeared fields, can be implemented as observer
// necessary to keep the fields updated even after a reject
// of the Metropolis
Smearer.set_GaugeField(U);
Smearer.set_Field(U);
// Set the (eventual) representations gauge fields
Representations.update(U);
@ -228,7 +230,7 @@ class Integrator {
for (int actionID = 0; actionID < as[level].actions.size(); ++actionID) {
// get gauge field from the SmearingPolicy and
// based on the boolean is_smeared in actionID
GaugeField& Us =
Field& Us =
Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
as[level].actions.at(actionID)->refresh(Us, pRNG);
}
@ -256,18 +258,9 @@ class Integrator {
} S_hireps{};
// Calculate action
RealD S(GaugeField& U) { // here also U not used
RealD S(Field& U) { // here also U not used
LatticeComplex Hloc(U._grid);
Hloc = zero;
// Momenta
for (int mu = 0; mu < Nd; mu++) {
auto Pmu = PeekIndex<LorentzIndex>(P, mu);
Hloc -= trace(Pmu * Pmu);
}
Complex Hsum = sum(Hloc);
RealD H = Hsum.real();
RealD H = - FieldImplementation::FieldSquareNorm(P); // - trace (P*P)
RealD Hterm;
std::cout << GridLogMessage << "Momentum action H_p = " << H << "\n";
@ -276,7 +269,7 @@ class Integrator {
for (int actionID = 0; actionID < as[level].actions.size(); ++actionID) {
// get gauge field from the SmearingPolicy and
// based on the boolean is_smeared in actionID
GaugeField& Us =
Field& Us =
Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
Hterm = as[level].actions.at(actionID)->S(Us);
std::cout << GridLogMessage << "S Level " << level << " term "
@ -289,7 +282,7 @@ class Integrator {
return H;
}
void integrate(GaugeField& U) {
void integrate(Field& U) {
// reset the clocks
t_U = 0;
for (int level = 0; level < as.size(); ++level) {
@ -311,8 +304,14 @@ class Integrator {
// and that we indeed got to the end of the trajectory
assert(fabs(t_U - Params.trajL) < 1.0e-6);
}
};
}
}
#endif // INTEGRATOR_INCLUDED

Some files were not shown because too many files have changed in this diff Show More