portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-04 03:05:55 +01:00
Code Releases Activity

Merge branch 'develop' of https://github.com/paboyle/Grid into develop

Browse Source
This commit is contained in:
paboyle 2016-07-15 19:26:06 +01:00
commit f4dd5062d7
62 changed files with 12722 additions and 2805 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

BIN
benchmarks/Benchmark_dwf_ntpf Executable file
View File

Binary file not shown.

BIN
benchmarks/Benchmark_zmm Executable file
View File

Binary file not shown.

8277
configure vendored Executable file
View File

File diff suppressed because it is too large Load Diff

3
lib/Algorithms.h
View File

@ -43,7 +43,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
#include <Grid/algorithms/iterative/NormalEquations.h>
#include <Grid/algorithms/iterative/SchurRedBlack.h>
#include <Grid/algorithms/iterative/ConjugateGradientMultiShift.h>
#include <algorithms/iterative/ConjugateGradientMultiShift.h>
#include <algorithms/iterative/ConjugateGradientMixedPrec.h>
// Lanczos support
#include <Grid/algorithms/iterative/MatrixUtils.h>

42
lib/Init.cc
View File

@ -193,7 +193,7 @@ void Grid_init(int *argc,char ***argv)
std::cout<<GridLogMessage<<"--mpi n.n.n.n : default MPI decomposition"<<std::endl;
std::cout<<GridLogMessage<<"--threads n : default number of OMP threads"<<std::endl;
std::cout<<GridLogMessage<<"--grid n.n.n.n : default Grid size"<<std::endl;
std::cout<<GridLogMessage<<"--log list : comma separted list of streams from Error,Warning,Message,Performance,Iterative,Integrator,Debug"<<std::endl;
std::cout<<GridLogMessage<<"--log list : comma separted list of streams from Error,Warning,Message,Performance,Iterative,Integrator,Debug,Colours"<<std::endl;
exit(EXIT_SUCCESS);
}
@ -234,24 +234,33 @@ void Grid_init(int *argc,char ***argv)
std::cout<<GridLogMessage<<"\tvComplexD : "<<sizeof(vComplexD)*8 <<"bits ; " <<GridCmdVectorIntToString(GridDefaultSimd(4,vComplexD::Nsimd()))<<std::endl;
}
std::string COL_RED = GridLogColours.colour["RED"];
std::string COL_PURPLE = GridLogColours.colour["PURPLE"];
std::string COL_BLACK = GridLogColours.colour["BLACK"];
std::string COL_GREEN = GridLogColours.colour["GREEN"];
std::string COL_BLUE = GridLogColours.colour["BLUE"];
std::string COL_YELLOW = GridLogColours.colour["YELLOW"];
std::string COL_BACKGROUND = GridLogColours.colour["NORMAL"];
std::cout <<std::endl;
std::cout << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl;
std::cout << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl;
std::cout << "__|_ | | | | | | | | | | | | _|__"<<std::endl;
std::cout << "__|_ _|__"<<std::endl;
std::cout << "__|_ GGGG RRRR III DDDD _|__"<<std::endl;
std::cout << "__|_ G R R I D D _|__"<<std::endl;
std::cout << "__|_ G R R I D D _|__"<<std::endl;
std::cout << "__|_ G GG RRRR I D D _|__"<<std::endl;
std::cout << "__|_ G G R R I D D _|__"<<std::endl;
std::cout << "__|_ GGGG R R III DDDD _|__"<<std::endl;
std::cout << "__|_ _|__"<<std::endl;
std::cout << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl;
std::cout << "__|__|__|__|__|__|__|__|__|__|__|__|__|__|__"<<std::endl;
std::cout << " | | | | | | | | | | | | | | "<<std::endl;
std::cout << std::endl;
std::cout <<COL_RED << "__|__|__|__|__"<< "|__|__|_"<<COL_PURPLE<<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<COL_RED << "__|__|__|__|__"<< "|__|__|_"<<COL_PURPLE<<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<COL_RED << "__|__| | | "<< "| | | "<<COL_PURPLE<<" | | |"<< " | | | _|__"<<std::endl;
std::cout <<COL_RED << "__|__ "<< " "<<COL_PURPLE<<" "<< " _|__"<<std::endl;
std::cout <<COL_RED << "__|_ "<<COL_GREEN<<" GGGG "<<COL_RED<<" RRRR "<<COL_BLUE <<" III "<<COL_PURPLE<<"DDDD "<<COL_PURPLE<<" _|__"<<std::endl;
std::cout <<COL_RED << "__|_ "<<COL_GREEN<<"G "<<COL_RED<<" R R "<<COL_BLUE <<" I "<<COL_PURPLE<<"D D "<<COL_PURPLE<<" _|__"<<std::endl;
std::cout <<COL_RED << "__|_ "<<COL_GREEN<<"G "<<COL_RED<<" R R "<<COL_BLUE <<" I "<<COL_PURPLE<<"D D"<<COL_PURPLE<<" _|__"<<std::endl;
std::cout <<COL_BLUE << "__|_ "<<COL_GREEN<<"G GG "<<COL_RED<<" RRRR "<<COL_BLUE <<" I "<<COL_PURPLE<<"D D"<<COL_GREEN <<" _|__"<<std::endl;
std::cout <<COL_BLUE << "__|_ "<<COL_GREEN<<"G G "<<COL_RED<<" R R "<<COL_BLUE <<" I "<<COL_PURPLE<<"D D "<<COL_GREEN <<" _|__"<<std::endl;
std::cout <<COL_BLUE << "__|_ "<<COL_GREEN<<" GGGG "<<COL_RED<<" R R "<<COL_BLUE <<" III "<<COL_PURPLE<<"DDDD "<<COL_GREEN <<" _|__"<<std::endl;
std::cout <<COL_BLUE << "__|__ "<< " "<<COL_GREEN <<" "<< " _|__"<<std::endl;
std::cout <<COL_BLUE << "__|__|__|__|__"<< "|__|__|_"<<COL_GREEN <<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<COL_BLUE << "__|__|__|__|__"<< "|__|__|_"<<COL_GREEN <<"_|__|__|"<< "__|__|__|__|__"<<std::endl;
std::cout <<COL_BLUE << " | | | | "<< "| | | "<<COL_GREEN <<" | | |"<< " | | | | "<<std::endl;
std::cout << std::endl;
std::cout << std::endl;
std::cout <<COL_YELLOW<< std::endl;
std::cout << "Copyright (C) 2015 Peter Boyle, Azusa Yamaguchi, Guido Cossu, Antonin Portelli and other authors"<<std::endl;
std::cout << std::endl;
std::cout << "This program is free software; you can redistribute it and/or modify"<<std::endl;
@ -263,6 +272,7 @@ void Grid_init(int *argc,char ***argv)
std::cout << "but WITHOUT ANY WARRANTY; without even the implied warranty of"<<std::endl;
std::cout << "MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the"<<std::endl;
std::cout << "GNU General Public License for more details."<<std::endl;
std::cout << COL_BACKGROUND <<std::endl;
std::cout << std::endl;
}

72
lib/Log.cc
View File

@ -25,7 +25,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#include <Grid.h>
@ -34,74 +35,43 @@ namespace Grid {
GridStopWatch Logger::StopWatch;
std::ostream Logger::devnull(0);
std::string Logger::BLACK("\033[30m");
std::string Logger::RED("\033[31m");
std::string Logger::GREEN("\033[32m");
std::string Logger::YELLOW("\033[33m");
std::string Logger::BLUE("\033[34m");
std::string Logger::PURPLE("\033[35m");
std::string Logger::CYAN("\033[36m");
std::string Logger::WHITE("\033[37m");
std::string Logger::NORMAL("\033[0;39m");
std::string EMPTY("");
#if 0
GridLogger GridLogError (1,"Error",Logger::RED);
GridLogger GridLogWarning (1,"Warning",Logger::YELLOW);
GridLogger GridLogMessage (1,"Message",Logger::BLACK);
GridLogger GridLogDebug (1,"Debug",Logger::PURPLE);
GridLogger GridLogPerformance(1,"Performance",Logger::GREEN);
GridLogger GridLogIterative (1,"Iterative",Logger::BLUE);
GridLogger GridLogIntegrator (1,"Integrator",Logger::BLUE);
#else
GridLogger GridLogError (1,"Error",EMPTY);
GridLogger GridLogWarning (1,"Warning",EMPTY);
GridLogger GridLogMessage (1,"Message",EMPTY);
GridLogger GridLogDebug (1,"Debug",EMPTY);
GridLogger GridLogPerformance(1,"Performance",EMPTY);
GridLogger GridLogIterative (1,"Iterative",EMPTY);
GridLogger GridLogIntegrator (1,"Integrator",EMPTY);
#endif
Colours GridLogColours(0);
GridLogger GridLogError(1, "Error", GridLogColours, "RED");
GridLogger GridLogWarning(1, "Warning", GridLogColours, "YELLOW");
GridLogger GridLogMessage(1, "Message", GridLogColours, "NORMAL");
GridLogger GridLogDebug(1, "Debug", GridLogColours, "PURPLE");
GridLogger GridLogPerformance(1, "Performance", GridLogColours, "GREEN");
GridLogger GridLogIterative(1, "Iterative", GridLogColours, "BLUE");
GridLogger GridLogIntegrator(1, "Integrator", GridLogColours, "BLUE");
void GridLogConfigure(std::vector<std::string> &logstreams)
{
void GridLogConfigure(std::vector<std::string> &logstreams) {
GridLogError.Active(0);
GridLogWarning.Active(0);
GridLogMessage.Active(0);
GridLogMessage.Active(1); // at least the messages should be always on
GridLogIterative.Active(0);
GridLogDebug.Active(0);
GridLogPerformance.Active(0);
GridLogIntegrator.Active(0);
int blackAndWhite = 1;
if(blackAndWhite){
Logger::BLACK = std::string("");
Logger::RED =Logger::BLACK;
Logger::GREEN =Logger::BLACK;
Logger::YELLOW =Logger::BLACK;
Logger::BLUE =Logger::BLACK;
Logger::PURPLE =Logger::BLACK;
Logger::CYAN =Logger::BLACK;
Logger::WHITE =Logger::BLACK;
Logger::NORMAL =Logger::BLACK;
}
GridLogColours.Active(0);
for (int i = 0; i < logstreams.size(); i++) {
if (logstreams[i] == std::string("Error")) GridLogError.Active(1);
if (logstreams[i] == std::string("Warning")) GridLogWarning.Active(1);
if ( logstreams[i]== std::string("Message") ) GridLogMessage.Active(1);
if (logstreams[i] == std::string("NoMessage")) GridLogMessage.Active(0);
if (logstreams[i] == std::string("Iterative")) GridLogIterative.Active(1);
if (logstreams[i] == std::string("Debug")) GridLogDebug.Active(1);
if ( logstreams[i]== std::string("Performance") ) GridLogPerformance.Active(1);
if (logstreams[i] == std::string("Performance"))
GridLogPerformance.Active(1);
if (logstreams[i] == std::string("Integrator")) GridLogIntegrator.Active(1);
if (logstreams[i] == std::string("Colours")) GridLogColours.Active(1);
}
}
////////////////////////////////////////////////////////////
// Verbose limiter on MPI tasks
////////////////////////////////////////////////////////////
void Grid_quiesce_nodes(void)
{
void Grid_quiesce_nodes(void) {
int me = 0;
#ifdef GRID_COMMS_MPI
MPI_Comm_rank(MPI_COMM_WORLD, &me);
@ -114,13 +84,9 @@ void Grid_quiesce_nodes(void)
}
}
void Grid_unquiesce_nodes(void)
{
void Grid_unquiesce_nodes(void) {
#ifdef GRID_COMMS_MPI
std::cout.clear();
#endif
}
}

84
lib/Log.h
View File

@ -27,6 +27,9 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#include <map>
#ifndef GRID_LOG_H
#define GRID_LOG_H
@ -40,39 +43,81 @@ namespace Grid {
int Rank(void); // used for early stage debug before library init
class Colours{
protected:
bool is_active;
public:
std::map<std::string, std::string> colour;
Colours(bool activate=false){
Active(activate);
};
void Active(bool activate){
is_active=activate;
if (is_active){
colour["BLACK"] ="\033[30m";
colour["RED"] ="\033[31m";
colour["GREEN"] ="\033[32m";
colour["YELLOW"] ="\033[33m";
colour["BLUE"] ="\033[34m";
colour["PURPLE"] ="\033[35m";
colour["CYAN"] ="\033[36m";
colour["WHITE"] ="\033[37m";
colour["NORMAL"] ="\033[0;39m";
} else {
colour["BLACK"] ="";
colour["RED"] ="";
colour["GREEN"] ="";
colour["YELLOW"]="";
colour["BLUE"] ="";
colour["PURPLE"]="";
colour["CYAN"] ="";
colour["WHITE"] ="";
colour["NORMAL"]="";
}
};
};
class Logger {
protected:
Colours &Painter;
int active;
std::string name, topName, COLOUR;
std::string name, topName;
std::string COLOUR;
public:
static GridStopWatch StopWatch;
static std::ostream devnull;
static std::string BLACK;
static std::string RED ;
static std::string GREEN;
static std::string YELLOW;
static std::string BLUE ;
static std::string PURPLE;
static std::string CYAN ;
static std::string WHITE ;
static std::string NORMAL;
std::string background() {return Painter.colour["NORMAL"];}
std::string evidence() {return Painter.colour["YELLOW"];}
std::string colour() {return Painter.colour[COLOUR];}
Logger(std::string topNm, int on, std::string nm,std::string col)
: active(on), name(nm), topName(topNm), COLOUR(col) {};
Logger(std::string topNm, int on, std::string nm, Colours& col_class, std::string col)
: active(on),
name(nm),
topName(topNm),
Painter(col_class),
COLOUR(col){} ;
void Active(int on) {active = on;};
int isActive(void) {return active;};
friend std::ostream& operator<< (std::ostream& stream, const Logger& log){
friend std::ostream& operator<< (std::ostream& stream, Logger& log){
if ( log.active ) {
StopWatch.Stop();
GridTime now = StopWatch.Elapsed();
StopWatch.Start();
stream << BLACK <<std::setw(8) << std::left << log.topName << BLACK<< " : ";
stream << log.COLOUR <<std::setw(11) << log.name << BLACK << " : ";
stream << YELLOW <<std::setw(6) << now <<BLACK << " : " ;
stream << log.COLOUR;
stream << log.background()<< log.topName << log.background()<< " : ";
stream << log.colour() <<std::setw(14) << std::left << log.name << log.background() << " : ";
stream << log.evidence()<< now << log.background() << " : " << log.colour();
return stream;
} else {
return devnull;
@ -83,7 +128,8 @@ public:
class GridLogger: public Logger {
public:
GridLogger(int on, std::string nm, std::string col = Logger::BLACK): Logger("Grid", on, nm, col){};
GridLogger(int on, std::string nm, Colours&col_class, std::string col_key = "NORMAL"):
Logger("Grid", on, nm, col_class, col_key){};
};
void GridLogConfigure(std::vector<std::string> &logstreams);
@ -95,6 +141,7 @@ extern GridLogger GridLogDebug ;
extern GridLogger GridLogPerformance;
extern GridLogger GridLogIterative ;
extern GridLogger GridLogIntegrator ;
extern Colours GridLogColours;
#define _NBACKTRACE (256)
@ -128,5 +175,6 @@ extern void * Grid_backtrace_buffer[_NBACKTRACE];
#define BACKTRACE() BACKTRACEFP(stdout)
}
#endif

2
lib/Make.inc
View File

File diff suppressed because one or more lines are too long

11
lib/Simd.h
View File

@ -24,7 +24,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_SIMD_H
@ -118,6 +119,14 @@ namespace Grid {
inline ComplexD timesI(const ComplexD &r) { return(r*ComplexD(0.0,1.0));}
inline ComplexF timesMinusI(const ComplexF &r){ return(r*ComplexF(0.0,-1.0));}
inline ComplexD timesMinusI(const ComplexD &r){ return(r*ComplexD(0.0,-1.0));}
// define projections to real and imaginay parts
inline ComplexF projReal(const ComplexF &r){return( ComplexF(std::real(r), 0.0));}
inline ComplexD projReal(const ComplexD &r){return( ComplexD(std::real(r), 0.0));}
inline ComplexF projImag(const ComplexF &r){return (ComplexF(std::imag(r), 0.0 ));}
inline ComplexD projImag(const ComplexD &r){return (ComplexD(std::imag(r), 0.0));}
// define auxiliary functions for complex computations
inline void timesI(ComplexF &ret,const ComplexF &r) { ret = timesI(r);}
inline void timesI(ComplexD &ret,const ComplexD &r) { ret = timesI(r);}
inline void timesMinusI(ComplexF &ret,const ComplexF &r){ ret = timesMinusI(r);}

5
lib/algorithms/iterative/ConjugateGradient.h
View File

@ -40,9 +40,10 @@ namespace Grid {
template<class Field>
class ConjugateGradient : public OperatorFunction<Field> {
public:
bool ErrorOnNoConverge; //throw an assert when the CG fails to converge. Defaults true.
RealD Tolerance;
Integer MaxIterations;
ConjugateGradient(RealD tol,Integer maxit) : Tolerance(tol), MaxIterations(maxit) {
ConjugateGradient(RealD tol,Integer maxit, bool err_on_no_conv = true) : Tolerance(tol), MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv){
};
@ -137,12 +138,14 @@ public:
std::cout<<GridLogMessage<<"Time elapsed: Total "<< SolverTimer.Elapsed() << " Matrix "<<MatrixTimer.Elapsed() << " Linalg "<<LinalgTimer.Elapsed();
std::cout<<std::endl;
if(ErrorOnNoConverge)
assert(true_residual/Tolerance < 1000.0);
return;
}
}
std::cout<<GridLogMessage<<"ConjugateGradient did NOT converge"<<std::endl;
if(ErrorOnNoConverge)
assert(0);
}
};

142
lib/algorithms/iterative/ConjugateGradientMixedPrec.h Normal file
View File

@ -0,0 +1,142 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./lib/algorithms/iterative/ConjugateGradientMixedPrec.h
Copyright (C) 2015
Author: Christopher Kelly <ckelly@phys.columbia.edu>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
See the full license in the file "LICENSE" in the top level distribution directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_CONJUGATE_GRADIENT_MIXED_PREC_H
#define GRID_CONJUGATE_GRADIENT_MIXED_PREC_H
namespace Grid {
//Mixed precision restarted defect correction CG
template<class FieldD,class FieldF, typename std::enable_if< getPrecision<FieldD>::value == 2, int>::type = 0,typename std::enable_if< getPrecision<FieldF>::value == 1, int>::type = 0>
class MixedPrecisionConjugateGradient : public LinearFunction<FieldD> {
public:
RealD Tolerance;
Integer MaxInnerIterations;
Integer MaxOuterIterations;
GridBase* SinglePrecGrid; //Grid for single-precision fields
RealD OuterLoopNormMult; //Stop the outer loop and move to a final double prec solve when the residual is OuterLoopNormMult * Tolerance
LinearOperatorBase<FieldF> &Linop_f;
LinearOperatorBase<FieldD> &Linop_d;
//Option to speed up *inner single precision* solves using a LinearFunction that produces a guess
LinearFunction<FieldF> *guesser;
MixedPrecisionConjugateGradient(RealD tol, Integer maxinnerit, Integer maxouterit, GridBase* _sp_grid, LinearOperatorBase<FieldF> &_Linop_f, LinearOperatorBase<FieldD> &_Linop_d) :
Linop_f(_Linop_f), Linop_d(_Linop_d),
Tolerance(tol), MaxInnerIterations(maxinnerit), MaxOuterIterations(maxouterit), SinglePrecGrid(_sp_grid),
OuterLoopNormMult(100.), guesser(NULL){ };
void useGuesser(LinearFunction<FieldF> &g){
guesser = &g;
}
void operator() (const FieldD &src_d_in, FieldD &sol_d){
GridStopWatch TotalTimer;
TotalTimer.Start();
int cb = src_d_in.checkerboard;
sol_d.checkerboard = cb;
RealD src_norm = norm2(src_d_in);
RealD stop = src_norm * Tolerance*Tolerance;
GridBase* DoublePrecGrid = src_d_in._grid;
FieldD tmp_d(DoublePrecGrid);
tmp_d.checkerboard = cb;
FieldD tmp2_d(DoublePrecGrid);
tmp2_d.checkerboard = cb;
FieldD src_d(DoublePrecGrid);
src_d = src_d_in; //source for next inner iteration, computed from residual during operation
RealD inner_tol = Tolerance;
FieldF src_f(SinglePrecGrid);
src_f.checkerboard = cb;
FieldF sol_f(SinglePrecGrid);
sol_f.checkerboard = cb;
ConjugateGradient<FieldF> CG_f(inner_tol, MaxInnerIterations);
CG_f.ErrorOnNoConverge = false;
GridStopWatch InnerCGtimer;
GridStopWatch PrecChangeTimer;
for(Integer outer_iter = 0; outer_iter < MaxOuterIterations; outer_iter++){
//Compute double precision rsd and also new RHS vector.
Linop_d.HermOp(sol_d, tmp_d);
RealD norm = axpy_norm(src_d, -1., tmp_d, src_d_in); //src_d is residual vector
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration " <<outer_iter<<" residual "<< norm<< " target "<< stop<<std::endl;
if(norm < OuterLoopNormMult * stop){
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Outer iteration converged on iteration " <<outer_iter <<std::endl;
break;
}
while(norm * inner_tol * inner_tol < stop) inner_tol *= 2; // inner_tol = sqrt(stop/norm) ??
PrecChangeTimer.Start();
precisionChange(src_f, src_d);
PrecChangeTimer.Stop();
zeroit(sol_f);
//Optionally improve inner solver guess (eg using known eigenvectors)
if(guesser != NULL)
(*guesser)(src_f, sol_f);
//Inner CG
CG_f.Tolerance = inner_tol;
InnerCGtimer.Start();
CG_f(Linop_f, src_f, sol_f);
InnerCGtimer.Stop();
//Convert sol back to double and add to double prec solution
PrecChangeTimer.Start();
precisionChange(tmp_d, sol_f);
PrecChangeTimer.Stop();
axpy(sol_d, 1.0, tmp_d, sol_d);
}
//Final trial CG
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Starting final patch-up double-precision solve"<<std::endl;
ConjugateGradient<FieldD> CG_d(Tolerance, MaxInnerIterations);
CG_d(Linop_d, src_d_in, sol_d);
TotalTimer.Stop();
std::cout<<GridLogMessage<<"MixedPrecisionConjugateGradient: Total " << TotalTimer.Elapsed() << " Precision change " << PrecChangeTimer.Elapsed() << " Inner CG total " << InnerCGtimer.Elapsed() << std::endl;
}
};
}
#endif

6
lib/cartesian/Cartesian_base.h
View File

@ -128,6 +128,12 @@ public:
//////////////////////////////////////////////////////////
// SIMD lane addressing
//////////////////////////////////////////////////////////
virtual int iIndex(std::vector<int> &lcoor)
{
int idx=0;
for(int d=0;d<_ndimension;d++) idx+=_istride[d]*(lcoor[d]/_rdimensions[d]);
return idx;
}
inline void iCoorFromIindex(std::vector<int> &coor,int lane)
{
Lexicographic::CoorFromIndex(coor,lane,_simd_layout);

226
lib/lattice/Lattice_ET.h
View File

@ -24,16 +24,17 @@ Author: neo <cossu@post.kek.jp>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_LATTICE_ET_H
#define GRID_LATTICE_ET_H
#include <iostream>
#include <vector>
#include <tuple>
#include <typeinfo>
#include <vector>
namespace Grid {
@ -41,8 +42,8 @@ namespace Grid {
// Predicated where support
////////////////////////////////////////////////////
template <class iobj, class vobj, class robj>
inline vobj predicatedWhere(const iobj &predicate,const vobj &iftrue,const robj &iffalse) {
inline vobj predicatedWhere(const iobj &predicate, const vobj &iftrue,
const robj &iffalse) {
typename std::remove_const<vobj>::type ret;
typedef typename vobj::scalar_object scalar_object;
@ -75,50 +76,69 @@ namespace Grid {
// from tuple is hideous; C++14 introduces std::make_index_sequence for this
////////////////////////////////////////////
// leaf eval of lattice ; should enable if protect using traits
template <typename T> using is_lattice = std::is_base_of<LatticeBase,T >;
template <typename T>
using is_lattice = std::is_base_of<LatticeBase, T>;
template <typename T>
using is_lattice_expr = std::is_base_of<LatticeExpressionBase, T>;
template <typename T> using is_lattice_expr = std::is_base_of<LatticeExpressionBase,T >;
//Specialization of getVectorType for lattices
template<typename T>
struct getVectorType<Lattice<T> >{
typedef typename Lattice<T>::vector_object type;
};
template<class sobj>
inline sobj eval(const unsigned int ss, const sobj &arg)
{
return arg;
}
template <class lobj>
inline const lobj &eval(const unsigned int ss, const Lattice<lobj> &arg)
{
inline const lobj &eval(const unsigned int ss, const Lattice<lobj> &arg) {
return arg._odata[ss];
}
// handle nodes in syntax tree
template <typename Op, typename T1>
auto inline eval(const unsigned int ss, const LatticeUnaryExpression<Op,T1 > &expr) // eval one operand
-> decltype(expr.first.func(eval(ss,std::get<0>(expr.second))))
{
auto inline eval(
const unsigned int ss,
const LatticeUnaryExpression<Op, T1> &expr) // eval one operand
-> decltype(expr.first.func(eval(ss, std::get<0>(expr.second)))) {
return expr.first.func(eval(ss, std::get<0>(expr.second)));
}
template <typename Op, typename T1, typename T2>
auto inline eval(const unsigned int ss, const LatticeBinaryExpression<Op,T1,T2> &expr) // eval two operands
-> decltype(expr.first.func(eval(ss,std::get<0>(expr.second)),eval(ss,std::get<1>(expr.second))))
{
return expr.first.func(eval(ss,std::get<0>(expr.second)),eval(ss,std::get<1>(expr.second)));
auto inline eval(
const unsigned int ss,
const LatticeBinaryExpression<Op, T1, T2> &expr) // eval two operands
-> decltype(expr.first.func(eval(ss, std::get<0>(expr.second)),
eval(ss, std::get<1>(expr.second)))) {
return expr.first.func(eval(ss, std::get<0>(expr.second)),
eval(ss, std::get<1>(expr.second)));
}
template <typename Op, typename T1, typename T2, typename T3>
auto inline eval(const unsigned int ss, const LatticeTrinaryExpression<Op,T1,T2,T3 > &expr) // eval three operands
-> decltype(expr.first.func(eval(ss,std::get<0>(expr.second)),eval(ss,std::get<1>(expr.second)),eval(ss,std::get<2>(expr.second))))
{
return expr.first.func(eval(ss,std::get<0>(expr.second)),eval(ss,std::get<1>(expr.second)),eval(ss,std::get<2>(expr.second)) );
auto inline eval(const unsigned int ss,
const LatticeTrinaryExpression<Op, T1, T2, T3>
&expr) // eval three operands
-> decltype(expr.first.func(eval(ss, std::get<0>(expr.second)),
eval(ss, std::get<1>(expr.second)),
eval(ss, std::get<2>(expr.second)))) {
return expr.first.func(eval(ss, std::get<0>(expr.second)),
eval(ss, std::get<1>(expr.second)),
eval(ss, std::get<2>(expr.second)));
}
//////////////////////////////////////////////////////////////////////////
// Obtain the grid from an expression, ensuring conformable. This must follow a tree recursion
// Obtain the grid from an expression, ensuring conformable. This must follow a
// tree recursion
//////////////////////////////////////////////////////////////////////////
template<class T1, typename std::enable_if<is_lattice<T1>::value, T1>::type * =nullptr >
template <class T1,
typename std::enable_if<is_lattice<T1>::value, T1>::type * = nullptr>
inline void GridFromExpression(GridBase *&grid, const T1 &lat) // Lattice leaf
{
if (grid) {
@ -126,35 +146,37 @@ inline void GridFromExpression(GridBase * &grid,const T1& lat) // Lattice leaf
}
grid = lat._grid;
}
template<class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr >
inline void GridFromExpression(GridBase * &grid,const T1& notlat) // non-lattice leaf
{
}
template <class T1,
typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
inline void GridFromExpression(GridBase *&grid,
const T1 &notlat) // non-lattice leaf
{}
template <typename Op, typename T1>
inline void GridFromExpression(GridBase * &grid,const LatticeUnaryExpression<Op,T1 > &expr)
{
inline void GridFromExpression(GridBase *&grid,
const LatticeUnaryExpression<Op, T1> &expr) {
GridFromExpression(grid, std::get<0>(expr.second)); // recurse
}
template <typename Op, typename T1, typename T2>
inline void GridFromExpression(GridBase * &grid,const LatticeBinaryExpression<Op,T1,T2> &expr)
{
inline void GridFromExpression(
GridBase *&grid, const LatticeBinaryExpression<Op, T1, T2> &expr) {
GridFromExpression(grid, std::get<0>(expr.second)); // recurse
GridFromExpression(grid, std::get<1>(expr.second));
}
template <typename Op, typename T1, typename T2, typename T3>
inline void GridFromExpression( GridBase * &grid,const LatticeTrinaryExpression<Op,T1,T2,T3 > &expr)
{
inline void GridFromExpression(
GridBase *&grid, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr) {
GridFromExpression(grid, std::get<0>(expr.second)); // recurse
GridFromExpression(grid, std::get<1>(expr.second));
GridFromExpression(grid, std::get<2>(expr.second));
}
//////////////////////////////////////////////////////////////////////////
// Obtain the CB from an expression, ensuring conformable. This must follow a tree recursion
// Obtain the CB from an expression, ensuring conformable. This must follow a
// tree recursion
//////////////////////////////////////////////////////////////////////////
template<class T1, typename std::enable_if<is_lattice<T1>::value, T1>::type * =nullptr >
template <class T1,
typename std::enable_if<is_lattice<T1>::value, T1>::type * = nullptr>
inline void CBFromExpression(int &cb, const T1 &lat) // Lattice leaf
{
if ((cb == Odd) || (cb == Even)) {
@ -163,28 +185,29 @@ inline void CBFromExpression(int &cb,const T1& lat) // Lattice leaf
cb = lat.checkerboard;
// std::cout<<GridLogMessage<<"Lattice leaf cb "<<cb<<std::endl;
}
template<class T1,typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr >
template <class T1,
typename std::enable_if<!is_lattice<T1>::value, T1>::type * = nullptr>
inline void CBFromExpression(int &cb, const T1 &notlat) // non-lattice leaf
{
// std::cout<<GridLogMessage<<"Non lattice leaf cb"<<cb<<std::endl;
}
template <typename Op, typename T1>
inline void CBFromExpression(int &cb,const LatticeUnaryExpression<Op,T1 > &expr)
{
inline void CBFromExpression(int &cb,
const LatticeUnaryExpression<Op, T1> &expr) {
CBFromExpression(cb, std::get<0>(expr.second)); // recurse
// std::cout<<GridLogMessage<<"Unary node cb "<<cb<<std::endl;
}
template <typename Op, typename T1, typename T2>
inline void CBFromExpression(int &cb,const LatticeBinaryExpression<Op,T1,T2> &expr)
{
inline void CBFromExpression(int &cb,
const LatticeBinaryExpression<Op, T1, T2> &expr) {
CBFromExpression(cb, std::get<0>(expr.second)); // recurse
CBFromExpression(cb, std::get<1>(expr.second));
// std::cout<<GridLogMessage<<"Binary node cb "<<cb<<std::endl;
}
template <typename Op, typename T1, typename T2, typename T3>
inline void CBFromExpression( int &cb,const LatticeTrinaryExpression<Op,T1,T2,T3 > &expr)
{
inline void CBFromExpression(
int &cb, const LatticeTrinaryExpression<Op, T1, T2, T3> &expr) {
CBFromExpression(cb, std::get<0>(expr.second)); // recurse
CBFromExpression(cb, std::get<1>(expr.second));
CBFromExpression(cb, std::get<2>(expr.second));
@ -195,8 +218,8 @@ inline void CBFromExpression( int &cb,const LatticeTrinaryExpression<Op,T1,T2,T3
// Unary operators and funcs
////////////////////////////////////////////
#define GridUnopClass(name, ret) \
template <class arg> struct name\
{\
template <class arg> \
struct name { \
static auto inline func(const arg a) -> decltype(ret) { return ret; } \
};
@ -212,11 +235,15 @@ GridUnopClass(UnaryReal,real(a));
GridUnopClass(UnaryImag, imag(a));
GridUnopClass(UnaryToReal, toReal(a));
GridUnopClass(UnaryToComplex, toComplex(a));
GridUnopClass(UnaryTimesI, timesI(a));
GridUnopClass(UnaryTimesMinusI, timesMinusI(a));
GridUnopClass(UnaryAbs, abs(a));
GridUnopClass(UnarySqrt, sqrt(a));
GridUnopClass(UnaryRsqrt, rsqrt(a));
GridUnopClass(UnarySin, sin(a));
GridUnopClass(UnaryCos, cos(a));
GridUnopClass(UnaryAsin, asin(a));
GridUnopClass(UnaryAcos, acos(a));
GridUnopClass(UnaryLog, log(a));
GridUnopClass(UnaryExp, exp(a));
@ -225,10 +252,9 @@ GridUnopClass(UnaryExp,exp(a));
////////////////////////////////////////////
#define GridBinOpClass(name, combination) \
template <class left, class right> \
struct name\
{\
static auto inline func(const left &lhs,const right &rhs)-> decltype(combination) const \
{\
struct name { \
static auto inline func(const left &lhs, const right &rhs) \
-> decltype(combination) const { \
return combination; \
} \
}
@ -246,17 +272,18 @@ GridBinOpClass(BinaryOrOr ,lhs||rhs);
////////////////////////////////////////////////////
#define GridTrinOpClass(name, combination) \
template <class predicate, class left, class right> \
struct name\
{\
static auto inline func(const predicate &pred,const left &lhs,const right &rhs)-> decltype(combination) const \
{\
struct name { \
static auto inline func(const predicate &pred, const left &lhs, \
const right &rhs) -> decltype(combination) const { \
return combination; \
} \
}
GridTrinOpClass(TrinaryWhere,(predicatedWhere<predicate, \
typename std::remove_reference<left>::type, \
typename std::remove_reference<right>::type> (pred,lhs,rhs)));
GridTrinOpClass(
TrinaryWhere,
(predicatedWhere<predicate, typename std::remove_reference<left>::type,
typename std::remove_reference<right>::type>(pred, lhs,
rhs)));
////////////////////////////////////////////
// Operator syntactical glue
@ -264,49 +291,66 @@ GridTrinOpClass(TrinaryWhere,(predicatedWhere<predicate, \
#define GRID_UNOP(name) name<decltype(eval(0, arg))>
#define GRID_BINOP(name) name<decltype(eval(0, lhs)), decltype(eval(0, rhs))>
#define GRID_TRINOP(name) name<decltype(eval(0, pred)), decltype(eval(0, lhs)), decltype(eval(0, rhs))>
#define GRID_TRINOP(name) \
name<decltype(eval(0, pred)), decltype(eval(0, lhs)), decltype(eval(0, rhs))>
#define GRID_DEF_UNOP(op, name) \
template <typename T1, \
typename std::enable_if<is_lattice<T1>::value||is_lattice_expr<T1>::value, T1>::type* = nullptr> inline auto op(const T1 &arg) \
-> decltype(LatticeUnaryExpression<GRID_UNOP(name),const T1&>(std::make_pair(GRID_UNOP(name)(),std::forward_as_tuple(arg)))) \
{ return LatticeUnaryExpression<GRID_UNOP(name), const T1 &>(std::make_pair(GRID_UNOP(name)(),std::forward_as_tuple(arg))); }
typename std::enable_if<is_lattice<T1>::value || \
is_lattice_expr<T1>::value, \
T1>::type * = nullptr> \
inline auto op(const T1 &arg) \
->decltype(LatticeUnaryExpression<GRID_UNOP(name), const T1 &>( \
std::make_pair(GRID_UNOP(name)(), std::forward_as_tuple(arg)))) { \
return LatticeUnaryExpression<GRID_UNOP(name), const T1 &>( \
std::make_pair(GRID_UNOP(name)(), std::forward_as_tuple(arg))); \
}
#define GRID_BINOP_LEFT(op, name) \
template <typename T1, typename T2, \
typename std::enable_if<is_lattice<T1>::value||is_lattice_expr<T1>::value, T1>::type* = nullptr>\
typename std::enable_if<is_lattice<T1>::value || \
is_lattice_expr<T1>::value, \
T1>::type * = nullptr> \
inline auto op(const T1 &lhs, const T2 &rhs) \
-> decltype(LatticeBinaryExpression<GRID_BINOP(name),const T1&,const T2 &>(std::make_pair(GRID_BINOP(name)(),\
std::forward_as_tuple(lhs, rhs)))) \
{\
return LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>(std::make_pair(GRID_BINOP(name)(),\
std::forward_as_tuple(lhs, rhs))); \
->decltype( \
LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>( \
std::make_pair(GRID_BINOP(name)(), \
std::forward_as_tuple(lhs, rhs)))) { \
return LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>( \
std::make_pair(GRID_BINOP(name)(), std::forward_as_tuple(lhs, rhs))); \
}
#define GRID_BINOP_RIGHT(op, name) \
template <typename T1, typename T2, \
typename std::enable_if<!is_lattice<T1>::value && !is_lattice_expr<T1>::value, T1>::type* = nullptr,\
typename std::enable_if< is_lattice<T2>::value || is_lattice_expr<T2>::value, T2>::type* = nullptr> \
typename std::enable_if<!is_lattice<T1>::value && \
!is_lattice_expr<T1>::value, \
T1>::type * = nullptr, \
typename std::enable_if<is_lattice<T2>::value || \
is_lattice_expr<T2>::value, \
T2>::type * = nullptr> \
inline auto op(const T1 &lhs, const T2 &rhs) \
-> decltype(LatticeBinaryExpression<GRID_BINOP(name),const T1&,const T2 &>(std::make_pair(GRID_BINOP(name)(),\
std::forward_as_tuple(lhs, rhs)))) \
{\
return LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>(std::make_pair(GRID_BINOP(name)(),\
std::forward_as_tuple(lhs, rhs))); \
->decltype( \
LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>( \
std::make_pair(GRID_BINOP(name)(), \
std::forward_as_tuple(lhs, rhs)))) { \
return LatticeBinaryExpression<GRID_BINOP(name), const T1 &, const T2 &>( \
std::make_pair(GRID_BINOP(name)(), std::forward_as_tuple(lhs, rhs))); \
}
#define GRID_DEF_BINOP(op, name) \
GRID_BINOP_LEFT(op, name); \
GRID_BINOP_RIGHT(op, name);
#define GRID_DEF_TRINOP(op, name) \
template <typename T1,typename T2,typename T3> inline auto op(const T1 &pred,const T2&lhs,const T3 &rhs) \
-> decltype(LatticeTrinaryExpression<GRID_TRINOP(name),const T1&,const T2 &,const T3&>(std::make_pair(GRID_TRINOP(name)(),\
std::forward_as_tuple(pred,lhs,rhs)))) \
{\
return LatticeTrinaryExpression<GRID_TRINOP(name), const T1 &, const T2 &,const T3&>(std::make_pair(GRID_TRINOP(name)(), \
std::forward_as_tuple(pred,lhs, rhs))); \
template <typename T1, typename T2, typename T3> \
inline auto op(const T1 &pred, const T2 &lhs, const T3 &rhs) \
->decltype( \
LatticeTrinaryExpression<GRID_TRINOP(name), const T1 &, const T2 &, \
const T3 &>(std::make_pair( \
GRID_TRINOP(name)(), std::forward_as_tuple(pred, lhs, rhs)))) { \
return LatticeTrinaryExpression<GRID_TRINOP(name), const T1 &, const T2 &, \
const T3 &>(std::make_pair( \
GRID_TRINOP(name)(), std::forward_as_tuple(pred, lhs, rhs))); \
}
////////////////////////
// Operator definitions
@ -325,11 +369,16 @@ GRID_DEF_UNOP(real,UnaryReal);
GRID_DEF_UNOP(imag, UnaryImag);
GRID_DEF_UNOP(toReal, UnaryToReal);
GRID_DEF_UNOP(toComplex, UnaryToComplex);
GRID_DEF_UNOP(abs ,UnaryAbs); //abs overloaded in cmath C++98; DON'T do the abs-fabs-dabs-labs thing
GRID_DEF_UNOP(timesI, UnaryTimesI);
GRID_DEF_UNOP(timesMinusI, UnaryTimesMinusI);
GRID_DEF_UNOP(abs, UnaryAbs); // abs overloaded in cmath C++98; DON'T do the
// abs-fabs-dabs-labs thing
GRID_DEF_UNOP(sqrt, UnarySqrt);
GRID_DEF_UNOP(rsqrt, UnaryRsqrt);
GRID_DEF_UNOP(sin, UnarySin);
GRID_DEF_UNOP(cos, UnaryCos);
GRID_DEF_UNOP(asin, UnaryAsin);
GRID_DEF_UNOP(acos, UnaryAcos);
GRID_DEF_UNOP(log, UnaryLog);
GRID_DEF_UNOP(exp, UnaryExp);
@ -349,29 +398,29 @@ GRID_DEF_TRINOP(where,TrinaryWhere);
/////////////////////////////////////////////////////////////
template <class Op, class T1>
auto closure(const LatticeUnaryExpression<Op, T1> &expr)
-> Lattice<decltype(expr.first.func(eval(0,std::get<0>(expr.second))))>
{
Lattice<decltype(expr.first.func(eval(0,std::get<0>(expr.second))))> ret(expr);
-> Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second))))> {
Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second))))> ret(
expr);
return ret;
}
template <class Op, class T1, class T2>
auto closure(const LatticeBinaryExpression<Op, T1, T2> &expr)
-> Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
eval(0,std::get<1>(expr.second))))>
{
eval(0, std::get<1>(expr.second))))> {
Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
eval(0,std::get<1>(expr.second))))> ret(expr);
eval(0, std::get<1>(expr.second))))>
ret(expr);
return ret;
}
template <class Op, class T1, class T2, class T3>
auto closure(const LatticeTrinaryExpression<Op, T1, T2, T3> &expr)
-> Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
eval(0, std::get<1>(expr.second)),
eval(0,std::get<2>(expr.second))))>
{
eval(0, std::get<2>(expr.second))))> {
Lattice<decltype(expr.first.func(eval(0, std::get<0>(expr.second)),
eval(0, std::get<1>(expr.second)),
eval(0,std::get<2>(expr.second))))> ret(expr);
eval(0, std::get<2>(expr.second))))>
ret(expr);
return ret;
}
@ -382,7 +431,6 @@ template<class Op,class T1, class T2, class T3>
#undef GRID_DEF_UNOP
#undef GRID_DEF_BINOP
#undef GRID_DEF_TRINOP
}
#if 0

17
lib/lattice/Lattice_base.h
View File

@ -24,7 +24,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_LATTICE_BASE_H
@ -256,6 +257,18 @@ PARALLEL_FOR_LOOP
checkerboard=0;
}
Lattice(const Lattice& r){ // copy constructor
_grid = r._grid;
checkerboard = r.checkerboard;
_odata.resize(_grid->oSites());// essential
PARALLEL_FOR_LOOP
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){
@ -268,7 +281,7 @@ PARALLEL_FOR_LOOP
template<class robj> strong_inline Lattice<vobj> & operator = (const Lattice<robj> & r){
this->checkerboard = r.checkerboard;
conformable(*this,r);
std::cout<<GridLogMessage<<"Lattice operator ="<<std::endl;
PARALLEL_FOR_LOOP
for(int ss=0;ss<_grid->oSites();ss++){
this->_odata[ss]=r._odata[ss];

2
lib/lattice/Lattice_reduction.h
View File

@ -40,7 +40,7 @@ namespace Grid {
////////////////////////////////////////////////////////////////////////////////////////////////////
template<class vobj> inline RealD norm2(const Lattice<vobj> &arg){
ComplexD nrm = innerProduct(arg,arg);
return real(nrm);
return std::real(nrm);
}
template<class vobj>

94
lib/lattice/Lattice_transfer.h
View File

@ -386,7 +386,7 @@ void InsertSlice(Lattice<vobj> &lowDim,Lattice<vobj> & higherDim,int slice, int
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
//PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -428,7 +428,7 @@ void ExtractSlice(Lattice<vobj> &lowDim, Lattice<vobj> & higherDim,int slice, in
}
}
// the above should guarantee that the operations are local
PARALLEL_FOR_LOOP
//PARALLEL_FOR_LOOP
for(int idx=0;idx<lg->lSites();idx++){
std::vector<int> lcoor(nl);
std::vector<int> hcoor(nh);
@ -482,6 +482,96 @@ 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){
typedef typename vobj::vector_type vtype;
GridBase* in_grid = in._grid;
out.resize(in_grid->lSites());
int ndim = in_grid->Nd();
int in_nsimd = vtype::Nsimd();
std::vector<std::vector<int> > in_icoor(in_nsimd);
for(int lane=0; lane < in_nsimd; lane++){
in_icoor[lane].resize(ndim);
in_grid->iCoorFromIindex(in_icoor[lane], lane);
}
PARALLEL_FOR_LOOP
for(int in_oidx = 0; in_oidx < in_grid->oSites(); in_oidx++){ //loop over outer index
//Assemble vector of pointers to output elements
std::vector<sobj*> out_ptrs(in_nsimd);
std::vector<int> in_ocoor(ndim);
in_grid->oCoorFromOindex(in_ocoor, in_oidx);
std::vector<int> lcoor(in_grid->Nd());
for(int lane=0; lane < in_nsimd; lane++){
for(int mu=0;mu<ndim;mu++)
lcoor[mu] = in_ocoor[mu] + in_grid->_rdimensions[mu]*in_icoor[lane][mu];
int lex;
Lexicographic::IndexFromCoor(lcoor, lex, in_grid->_ldimensions);
out_ptrs[lane] = &out[lex];
}
//Unpack into those ptrs
const vobj & in_vobj = in._odata[in_oidx];
extract1(in_vobj, out_ptrs, 0);
}
}
//Convert a Lattice from one precision to another
template<class VobjOut, class VobjIn>
void precisionChange(Lattice<VobjOut> &out, const Lattice<VobjIn> &in){
assert(out._grid->Nd() == in._grid->Nd());
out.checkerboard = in.checkerboard;
GridBase *in_grid=in._grid;
GridBase *out_grid = out._grid;
typedef typename VobjOut::scalar_object SobjOut;
typedef typename VobjIn::scalar_object SobjIn;
int ndim = out._grid->Nd();
int out_nsimd = out_grid->Nsimd();
std::vector<std::vector<int> > out_icoor(out_nsimd);
for(int lane=0; lane < out_nsimd; lane++){
out_icoor[lane].resize(ndim);
out_grid->iCoorFromIindex(out_icoor[lane], lane);
}
std::vector<SobjOut> in_slex_conv(in_grid->lSites());
unvectorizeToLexOrdArray(in_slex_conv, in);
PARALLEL_FOR_LOOP
for(int out_oidx=0;out_oidx<out_grid->oSites();out_oidx++){
std::vector<int> out_ocoor(ndim);
out_grid->oCoorFromOindex(out_ocoor, out_oidx);
std::vector<SobjOut*> ptrs(out_nsimd);
std::vector<int> lcoor(out_grid->Nd());
for(int lane=0; lane < out_nsimd; lane++){
for(int mu=0;mu<ndim;mu++)
lcoor[mu] = out_ocoor[mu] + out_grid->_rdimensions[mu]*out_icoor[lane][mu];
int llex; Lexicographic::IndexFromCoor(lcoor, llex, out_grid->_ldimensions);
ptrs[lane] = &in_slex_conv[llex];
}
merge(out._odata[out_oidx], ptrs, 0);
}
}
}
#endif

0
lib/pugixml/.dirstamp
View File

1
lib/qcd/QCD.h
View File

@ -501,5 +501,6 @@ namespace QCD {
#include <Grid/qcd/hmc/integrators/Integrator_algorithm.h>
#include <Grid/qcd/hmc/HMC.h>
#include <qcd/smearing/Smearing.h>
#endif

1
lib/qcd/action/ActionBase.h
View File

@ -35,6 +35,7 @@ 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

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

@ -75,7 +75,7 @@ namespace Grid {
//
//
// template<class Impl>
// class MyOp : pubic<Impl> {
// class MyOp : public<Impl> {
// public:
//
// INHERIT_ALL_IMPL_TYPES(Impl);

107
lib/qcd/action/gauge/GaugeImpl.h
View File

@ -22,7 +22,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_QCD_GAUGE_IMPL_H
@ -32,7 +33,6 @@ namespace Grid {
namespace QCD {
////////////////////////////////////////////////////////////////////////
// Implementation dependent gauge types
////////////////////////////////////////////////////////////////////////
@ -46,110 +46,120 @@ template<class Gimpl> class WilsonLoops;
typedef typename GImpl::SiteGaugeField SiteGaugeField; \
typedef typename GImpl::SiteGaugeLink SiteGaugeLink;
//
template<class S,int Nrepresentation=Nc>
class GaugeImplTypes {
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 >;
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<SiteGaugeLink> GaugeLinkField; // bit ugly naming; polarised
// gauge field, lorentz... all
// ugly
typedef Lattice<SiteGaugeField> GaugeField;
// Move this elsewhere?
static inline void AddGaugeLink(GaugeField &U, GaugeLinkField &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;
}
}
};
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
template<class GimplTypes>
class PeriodicGaugeImpl : public GimplTypes {
template <class GimplTypes> class PeriodicGaugeImpl : public GimplTypes {
public:
INHERIT_GIMPL_TYPES(GimplTypes);
////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Support needed for the assembly of loops including all boundary condition effects such as conjugate bcs
// Support needed for the assembly of loops including all boundary condition
// effects such as conjugate bcs
////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class covariant> static inline
Lattice<covariant> CovShiftForward (const GaugeLinkField &Link, int mu, const Lattice<covariant> &field) {
template <class covariant>
static inline Lattice<covariant>
CovShiftForward(const GaugeLinkField &Link, int mu,
const Lattice<covariant> &field) {
return PeriodicBC::CovShiftForward(Link, mu, field);
}
template<class covariant> static inline
Lattice<covariant> CovShiftBackward(const GaugeLinkField &Link, int mu,const Lattice<covariant> &field) {
template <class covariant>
static inline Lattice<covariant>
CovShiftBackward(const GaugeLinkField &Link, int mu,
const Lattice<covariant> &field) {
return PeriodicBC::CovShiftBackward(Link, mu, field);
}
static inline
GaugeLinkField CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
static inline GaugeLinkField
CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
return Cshift(adj(Link), mu, -1);
}
static inline
GaugeLinkField CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
static inline GaugeLinkField
CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
return Link;
}
static inline
GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu) {
static inline GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu) {
return Cshift(Link, mu, 1);
}
static inline bool isPeriodicGaugeField(void) {
return true;
}
static inline bool isPeriodicGaugeField(void) { return true; }
};
// Composition with smeared link, bc's etc.. probably need multiple inheritance
// Variable precision "S" and variable Nc
template<class GimplTypes>
class ConjugateGaugeImpl : public GimplTypes {
template <class GimplTypes> class ConjugateGaugeImpl : public GimplTypes {
public:
INHERIT_GIMPL_TYPES(GimplTypes);
////////////////////////////////////////////////////////////////////////////////////////////////////////////
// Support needed for the assembly of loops including all boundary condition effects such as Gparity.
// Support needed for the assembly of loops including all boundary condition
// effects such as Gparity.
////////////////////////////////////////////////////////////////////////////////////////////////////////////
template<class covariant> static
Lattice<covariant> CovShiftForward (const GaugeLinkField &Link, int mu, const Lattice<covariant> &field) {
template <class covariant>
static Lattice<covariant> CovShiftForward(const GaugeLinkField &Link, int mu,
const Lattice<covariant> &field) {
return ConjugateBC::CovShiftForward(Link, mu, field);
}
template<class covariant> static
Lattice<covariant> CovShiftBackward(const GaugeLinkField &Link, int mu,const Lattice<covariant> &field) {
template <class covariant>
static Lattice<covariant> CovShiftBackward(const GaugeLinkField &Link, int mu,
const Lattice<covariant> &field) {
return ConjugateBC::CovShiftBackward(Link, mu, field);
}
static inline
GaugeLinkField CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
static inline GaugeLinkField
CovShiftIdentityBackward(const GaugeLinkField &Link, int mu) {
GridBase *grid = Link._grid;
int Lmu = grid->GlobalDimensions()[mu] - 1;
Lattice<iScalar<vInteger> > coor(grid); LatticeCoordinate(coor,mu);
Lattice<iScalar<vInteger>> coor(grid);
LatticeCoordinate(coor, mu);
GaugeLinkField tmp(grid);
tmp = adj(Link);
tmp = where(coor == Lmu, conjugate(tmp), tmp);
return Cshift(tmp, mu, -1); // moves towards positive mu
}
static inline
GaugeLinkField CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
static inline GaugeLinkField
CovShiftIdentityForward(const GaugeLinkField &Link, int mu) {
return Link;
}
static inline
GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu) {
static inline GaugeLinkField ShiftStaple(const GaugeLinkField &Link, int mu) {
GridBase *grid = Link._grid;
int Lmu = grid->GlobalDimensions()[mu] - 1;
Lattice<iScalar<vInteger> > coor(grid); LatticeCoordinate(coor,mu);
Lattice<iScalar<vInteger>> coor(grid);
LatticeCoordinate(coor, mu);
GaugeLinkField tmp(grid);
tmp = Cshift(Link, mu, 1);
@ -157,10 +167,7 @@ template<class Gimpl> class WilsonLoops;
return tmp;
}
static inline bool isPeriodicGaugeField(void) {
return false;
}
static inline bool isPeriodicGaugeField(void) { return false; }
};
typedef GaugeImplTypes<vComplex, Nc> GimplTypesR;
@ -171,10 +178,10 @@ template<class Gimpl> class WilsonLoops;
typedef PeriodicGaugeImpl<GimplTypesF> PeriodicGimplF; // Float
typedef PeriodicGaugeImpl<GimplTypesD> PeriodicGimplD; // Double
typedef ConjugateGaugeImpl<GimplTypesR> ConjugateGimplR; // Real.. whichever prec
typedef ConjugateGaugeImpl<GimplTypesR>
ConjugateGimplR; // Real.. whichever prec
typedef ConjugateGaugeImpl<GimplTypesF> ConjugateGimplF; // Float
typedef ConjugateGaugeImpl<GimplTypesD> ConjugateGimplD; // Double
}
}

56
lib/qcd/action/pseudofermion/OneFlavourEvenOddRational.h
View File

@ -22,7 +22,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef QCD_PSEUDOFERMION_ONE_FLAVOUR_EVEN_ODD_RATIONAL_H
@ -42,7 +43,8 @@ namespace Grid{
//
template <class Impl>
class OneFlavourEvenOddRationalPseudoFermionAction : public Action<typename Impl::GaugeField> {
class OneFlavourEvenOddRationalPseudoFermionAction
: public Action<typename Impl::GaugeField> {
public:
INHERIT_IMPL_TYPES(Impl);
@ -55,41 +57,40 @@ namespace Grid{
MultiShiftFunction PowerNegQuarter;
private:
FermionOperator<Impl> &FermOp; // the basic operator
// NOT using "Nroots"; IroIro is -- perhaps later, but this wasn't good for us historically
// NOT using "Nroots"; IroIro is -- perhaps later, but this wasn't good for us
// historically
// and hasenbusch works better
FermionField PhiEven; // the pseudo fermion field for this trajectory
FermionField PhiOdd; // the pseudo fermion field for this trajectory
public:
OneFlavourEvenOddRationalPseudoFermionAction(FermionOperator<Impl> &Op,
Params & p ) : FermOp(Op),
Params &p)
: FermOp(Op),
PhiEven(Op.FermionRedBlackGrid()),
PhiOdd(Op.FermionRedBlackGrid()),
param(p)
{
param(p) {
AlgRemez remez(param.lo, param.hi, param.precision);
// MdagM^(+- 1/2)
std::cout<<GridLogMessage << "Generating degree "<<param.degree<<" for x^(1/2)"<<std::endl;
std::cout << GridLogMessage << "Generating degree " << param.degree
<< " for x^(1/2)" << std::endl;
remez.generateApprox(param.degree, 1, 2);
PowerHalf.Init(remez, param.tolerance, false);
PowerNegHalf.Init(remez, param.tolerance, true);
// MdagM^(+- 1/4)
std::cout<<GridLogMessage << "Generating degree "<<param.degree<<" for x^(1/4)"<<std::endl;
std::cout << GridLogMessage << "Generating degree " << param.degree
<< " for x^(1/4)" << std::endl;
remez.generateApprox(param.degree, 1, 4);
PowerQuarter.Init(remez, param.tolerance, false);
PowerNegQuarter.Init(remez, param.tolerance, true);
};
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}
// Phi = MpcdagMpc^{1/4} eta
@ -106,7 +107,8 @@ namespace Grid{
FermionField etaOdd(FermOp.FermionRedBlackGrid());
FermionField etaEven(FermOp.FermionRedBlackGrid());
gaussian(pRNG,eta); eta=eta*scale;
gaussian(pRNG, eta);
eta = eta * scale;
pickCheckerboard(Even, etaEven, eta);
pickCheckerboard(Odd, etaOdd, eta);
@ -124,26 +126,27 @@ namespace Grid{
assert(FermOp.ConstEE() == 1);
PhiEven = zero;
};
//////////////////////////////////////////////////////
// S = phi^dag (Mdag M)^-1/2 phi
//////////////////////////////////////////////////////
virtual RealD S(const GaugeField &U) {
FermOp.ImportGauge(U);
FermionField Y(FermOp.FermionRedBlackGrid());
SchurDifferentiableOperator<Impl> Mpc(FermOp);
ConjugateGradientMultiShift<FermionField> msCG(param.MaxIter,PowerNegQuarter);
ConjugateGradientMultiShift<FermionField> msCG(param.MaxIter,
PowerNegQuarter);
msCG(Mpc, PhiOdd, Y);
RealD action = norm2(Y);
std::cout << GridLogMessage << "Pseudofermion action FIXME -- is -1/4 solve or -1/2 solve faster??? "<<action<<std::endl;
std::cout << GridLogMessage << "Pseudofermion action FIXME -- is -1/4 "
"solve or -1/2 solve faster??? "
<< action << std::endl;
return action;
};
@ -158,7 +161,8 @@ namespace Grid{
//
// d[N/D] is then
//
// \sum_k -ak [M^dagM+bk]^{-1} [ dM^dag M + M^dag dM ] [M^dag M + bk]^{-1}
// \sum_k -ak [M^dagM+bk]^{-1} [ dM^dag M + M^dag dM ] [M^dag M +
// bk]^{-1}
//
// Need
// Mf Phi_k = [MdagM+bk]^{-1} Phi
@ -166,11 +170,11 @@ namespace Grid{
//
// With these building blocks
//
// dS/dU = \sum_k -ak Mf Phi_k^dag [ dM^dag M + M^dag dM ] Mf Phi_k
// dS/dU = \sum_k -ak Mf Phi_k^dag [ dM^dag M + M^dag dM ] Mf
// Phi_k
// S = innerprodReal(Phi,Mf Phi);
//////////////////////////////////////////////////////
virtual void deriv(const GaugeField &U, GaugeField &dSdU) {
const int Npole = PowerNegHalf.poles.size();
std::vector<FermionField> MPhi_k(Npole, FermOp.FermionRedBlackGrid());
@ -190,23 +194,21 @@ namespace Grid{
dSdU = zero;
for (int k = 0; k < Npole; k++) {
RealD ak = PowerNegHalf.residues[k];
X = MPhi_k[k];
Mpc.Mpc(X, Y);
Mpc.MpcDeriv (tmp , Y, X ); dSdU=dSdU+ak*tmp;
Mpc.MpcDagDeriv(tmp , X, Y ); dSdU=dSdU+ak*tmp;
Mpc.MpcDeriv(tmp, Y, X);
dSdU = dSdU + ak * tmp;
Mpc.MpcDagDeriv(tmp, X, Y);
dSdU = dSdU + ak * tmp;
}
dSdU = Ta(dSdU);
// dSdU = Ta(dSdU);
};
};
}
}
#endif

2
lib/qcd/action/pseudofermion/OneFlavourEvenOddRationalRatio.h
View File

@ -256,7 +256,7 @@ namespace Grid{
}
dSdU = Ta(dSdU);
//dSdU = Ta(dSdU);
};
};

2
lib/qcd/action/pseudofermion/OneFlavourRational.h
View File

@ -186,7 +186,7 @@ namespace Grid{
}
dSdU = Ta(dSdU);
//dSdU = Ta(dSdU);
};
};

2
lib/qcd/action/pseudofermion/OneFlavourRationalRatio.h
View File

@ -242,7 +242,7 @@ namespace Grid{
}
dSdU = Ta(dSdU);
//dSdU = Ta(dSdU);
};
};

2
lib/qcd/action/pseudofermion/TwoFlavour.h
View File

@ -137,7 +137,7 @@ namespace Grid{
FermOp.MDeriv(tmp , Y, X,DaggerNo ); dSdU=tmp;
FermOp.MDeriv(tmp , X, Y,DaggerYes); dSdU=dSdU+tmp;
dSdU = Ta(dSdU);
//dSdU = Ta(dSdU);
};

2
lib/qcd/action/pseudofermion/TwoFlavourEvenOdd.h
View File

@ -173,7 +173,7 @@ namespace Grid{
FermOp.MeeDeriv(tmp , X, Y,DaggerYes); dSdU=dSdU+tmp;
*/
dSdU = Ta(dSdU);
//dSdU = Ta(dSdU);
};

3
lib/qcd/action/pseudofermion/TwoFlavourEvenOddRatio.h
View File

@ -188,7 +188,8 @@ namespace Grid{
assert(NumOp.ConstEE() == 1);
assert(DenOp.ConstEE() == 1);
dSdU = -Ta(dSdU);
//dSdU = -Ta(dSdU);
dSdU = -dSdU;
};
};

3
lib/qcd/action/pseudofermion/TwoFlavourRatio.h
View File

@ -155,7 +155,8 @@ namespace Grid{
DenOp.MDeriv(force,Y,X,DaggerNo); dSdU=dSdU-force;
DenOp.MDeriv(force,X,Y,DaggerYes); dSdU=dSdU-force;
dSdU = - Ta(dSdU);
dSdU *= -1.0;
//dSdU = - Ta(dSdU);
};
};

104
lib/qcd/hmc/HMC.h
View File

@ -25,7 +25,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
//--------------------------------------------------------------------
@ -41,13 +42,10 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
#include <string>
namespace Grid {
namespace QCD {
struct HMCparameters {
Integer StartTrajectory;
Integer Trajectories; /* @brief Number of sweeps in this run */
bool MetropolisTest;
@ -61,27 +59,40 @@ namespace Grid{
Trajectories = 200;
/////////////////////////////////
}
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 GaugeField>
class HmcObservable {
public:
virtual void TrajectoryComplete (int traj, GaugeField &U, GridSerialRNG &sRNG, GridParallelRNG & pRNG )=0;
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;
};
PlaquetteLogger(std::string cf) { Stem = cf; };
void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG, GridParallelRNG & pRNG )
void TrajectoryComplete(int traj, GaugeField &U, GridSerialRNG &sRNG,
GridParallelRNG &pRNG) {
std::string file;
{
std::string file; { std::ostringstream os; os << Stem <<"."<< traj; file = os.str(); }
std::ostringstream os;
os << Stem << "." << traj;
file = os.str();
}
std::ofstream of(file);
RealD peri_plaq = WilsonLoops<PeriodicGimplR>::avgPlaquette(U);
@ -90,17 +101,27 @@ namespace Grid{
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;
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 Integrator, class Smearer, class
// Boundary>
template <class GaugeField, class IntegratorType>
class HybridMonteCarlo {
private:
const HMCparameters Params;
GridSerialRNG &sRNG; // Fixme: need a RNG management strategy.
@ -114,60 +135,64 @@ namespace Grid{
// Metropolis step
/////////////////////////////////////////////////////////
bool metropolis_test(const RealD DeltaH) {
RealD rn_test;
RealD prob = std::exp(-DeltaH);
random(sRNG, rn_test);
std::cout<<GridLogMessage<< "--------------------------------------------\n";
std::cout<<GridLogMessage<< "dH = "<<DeltaH << " Random = "<< rn_test <<"\n";
std::cout<<GridLogMessage<< "Acc. Probability = " << ((prob<1.0)? prob: 1.0)<< " ";
std::cout << GridLogMessage
<< "--------------------------------------------------\n";
std::cout << GridLogMessage << "exp(-dH) = " << prob
<< " Random = " << rn_test << "\n";
std::cout << GridLogMessage
<< "Acc. Probability = " << ((prob < 1.0) ? prob : 1.0) << "\n";
if ((prob > 1.0) || (rn_test <= prob)) { // accepted
std::cout<<GridLogMessage <<"-- ACCEPTED\n";
std::cout << GridLogMessage << "Metropolis_test -- ACCEPTED\n";
std::cout << GridLogMessage
<< "--------------------------------------------------\n";
return true;
} else { // rejected
std::cout<<GridLogMessage <<"-- REJECTED\n";
std::cout << GridLogMessage << "Metropolis_test -- REJECTED\n";
std::cout << GridLogMessage
<< "--------------------------------------------------\n";
return false;
}
}
/////////////////////////////////////////////////////////
// Evolution
/////////////////////////////////////////////////////////
RealD evolve_step(GaugeField &U) {
TheIntegrator.refresh(U, pRNG); // set U and initialize P and phi's
RealD H0 = TheIntegrator.S(U); // initial state action
std::cout<<GridLogMessage<<"Total H before = "<< H0 << "\n";
std::streamsize current_precision = std::cout.precision();
std::cout.precision(17);
std::cout << GridLogMessage << "Total H before trajectory = " << H0 << "\n";
std::cout.precision(current_precision);
TheIntegrator.integrate(U);
RealD H1 = TheIntegrator.S(U); // updated state action
std::cout<<GridLogMessage<<"Total H after = "<< H1 << "\n";
std::cout.precision(17);
std::cout << GridLogMessage << "Total H after trajectory = " << H1
<< " dH = " << H1 - H0 << "\n";
std::cout.precision(current_precision);
return (H1 - H0);
}
public:
/////////////////////////////////////////
// Constructor
/////////////////////////////////////////
HybridMonteCarlo(HMCparameters Pms, IntegratorType &_Int, GridSerialRNG &_sRNG, GridParallelRNG &_pRNG, GaugeField &_U ) :
Params(Pms),
TheIntegrator(_Int),
sRNG(_sRNG),
pRNG(_pRNG),
Ucur(_U)
{
}
HybridMonteCarlo(HMCparameters Pams, IntegratorType &_Int,
GridSerialRNG &_sRNG, GridParallelRNG &_pRNG, GaugeField &_U)
: Params(Pams), TheIntegrator(_Int), sRNG(_sRNG), pRNG(_pRNG), Ucur(_U) {}
~HybridMonteCarlo(){};
void AddObservable(HmcObservable<GaugeField> *obs) {
@ -175,21 +200,22 @@ namespace Grid{
}
void evolve(void) {
Real DeltaH;
GaugeField Ucopy(Ucur._grid);
// Actual updates (evolve a copy Ucopy then copy back eventually)
for(int traj=Params.StartTrajectory; traj < Params.Trajectories+Params.StartTrajectory; ++traj){
Params.print();
// Actual updates (evolve a copy Ucopy then copy back eventually)
for (int traj = Params.StartTrajectory;
traj < Params.Trajectories + Params.StartTrajectory; ++traj) {
std::cout << GridLogMessage << "-- # Trajectory = " << traj << "\n";
Ucopy = Ucur;
DeltaH = evolve_step(Ucopy);
bool accept = true;
if ( traj > Params.NoMetropolisUntil) {
if (traj >= Params.NoMetropolisUntil) {
accept = metropolis_test(DeltaH);
}
@ -200,7 +226,6 @@ namespace Grid{
for (int obs = 0; obs < Observables.size(); obs++) {
Observables[obs]->TrajectoryComplete(traj + 1, Ucur, sRNG, pRNG);
}
}
}
};
@ -208,5 +233,4 @@ namespace Grid{
} // QCD
} // Grid
#endif

51
lib/qcd/hmc/HmcRunner.h
View File

@ -47,7 +47,7 @@ public:
GridRedBlackCartesian * UrbGrid ;
GridRedBlackCartesian * FrbGrid ;
virtual void BuildTheAction (int argc, char **argv) = 0;
virtual void BuildTheAction (int argc, char **argv) = 0; // necessary?
void Run (int argc, char **argv){
@ -81,10 +81,37 @@ public:
NumTraj = ivec[0];
}
// Create integrator
typedef MinimumNorm2<GaugeField> IntegratorType;// change here to change the algorithm
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
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";
//////////////
typedef MinimumNorm2<GaugeField, SmearedConfiguration<Gimpl> > IntegratorType;// change here to change the algorithm
IntegratorParameters MDpar(20);
IntegratorType MDynamics(UGrid,MDpar, TheAction);
IntegratorType MDynamics(UGrid, MDpar, TheAction, SmearingPolicy);
// Checkpoint strategy
NerscHmcCheckpointer<Gimpl> Checkpoint(std::string("ckpoint_lat"),std::string("ckpoint_rng"),1);
@ -93,42 +120,38 @@ public:
HMCparameters HMCpar;
HMCpar.StartTrajectory = StartTraj;
HMCpar.Trajectories = NumTraj;
HMCpar.NoMetropolisUntil = NumThermalizations;
GridSerialRNG sRNG;
GridParallelRNG pRNG(UGrid);
LatticeGaugeField U(UGrid);
std::vector<int> SerSeed({1,2,3,4,5});
std::vector<int> ParSeed({6,7,8,9,10});
if ( StartType == HotStart ) {
// Hot start
HMCpar.NoMetropolisUntil =10;
HMCpar.MetropolisTest = true;
sRNG.SeedFixedIntegers(SerSeed);
pRNG.SeedFixedIntegers(ParSeed);
SU3::HotConfiguration(pRNG, U);
} else if ( StartType == ColdStart ) {
// Cold start
HMCpar.NoMetropolisUntil =10;
HMCpar.MetropolisTest = true;
sRNG.SeedFixedIntegers(SerSeed);
pRNG.SeedFixedIntegers(ParSeed);
SU3::ColdConfiguration(pRNG, U);
} else if ( StartType == TepidStart ) {
// Tepid start
HMCpar.NoMetropolisUntil =10;
HMCpar.MetropolisTest = true;
sRNG.SeedFixedIntegers(SerSeed);
pRNG.SeedFixedIntegers(ParSeed);
SU3::TepidConfiguration(pRNG, U);
} else if ( StartType == CheckpointStart ) {
HMCpar.NoMetropolisUntil =10;
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);

38
lib/qcd/hmc/integrators/Integrator.h
View File

@ -68,7 +68,7 @@ namespace Grid{
};
/*! @brief Class for Molecular Dynamics management */
template<class GaugeField>
template<class GaugeField, class SmearingPolicy>
class Integrator {
protected:
@ -85,6 +85,8 @@ namespace Grid{
GaugeField P;
SmearingPolicy &Smearer;
// 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){
@ -111,10 +113,17 @@ namespace Grid{
}
void update_P(GaugeField &Mom,GaugeField&U, int level,double ep){
// input U actually not used...
for(int a=0; a<as[level].actions.size(); ++a){
GaugeField force(U._grid);
as[level].actions.at(a)->deriv(U,force);
Mom = Mom - force*ep;
GaugeField& 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;
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;
}
}
@ -137,6 +146,8 @@ namespace Grid{
ProjectOnGroup(Umu);
PokeIndex<LorentzIndex>(U, Umu, mu);
}
// Update the smeared fields, can be implemented as observer
Smearer.set_GaugeField(U);
}
virtual void step (GaugeField& U,int level, int first,int last)=0;
@ -145,14 +156,17 @@ namespace Grid{
Integrator(GridBase* grid,
IntegratorParameters Par,
ActionSet<GaugeField> & Aset):
ActionSet<GaugeField> & Aset,
SmearingPolicy &Sm):
Params(Par),
as(Aset),
P(grid),
levels(Aset.size())
levels(Aset.size()),
Smearer(Sm)
{
t_P.resize(levels,0.0);
t_U=0.0;
// initialization of smearer delegated outside of Integrator
};
virtual ~Integrator(){}
@ -163,13 +177,16 @@ namespace Grid{
generate_momenta(P,pRNG);
for(int level=0; level< as.size(); ++level){
for(int actionID=0; actionID<as[level].actions.size(); ++actionID){
as[level].actions.at(actionID)->refresh(U, pRNG);
// get gauge field from the SmearingPolicy and
// based on the boolean is_smeared in actionID
GaugeField& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
as[level].actions.at(actionID)->refresh(Us, pRNG);
}
}
}
// Calculate action
RealD S(GaugeField& U){
RealD S(GaugeField& U){// here also U not used
LatticeComplex Hloc(U._grid); Hloc = zero;
// Momenta
@ -186,8 +203,11 @@ namespace Grid{
// Actions
for(int level=0; level<as.size(); ++level){
for(int actionID=0; actionID<as[level].actions.size(); ++actionID){
Hterm = as[level].actions.at(actionID)->S(U);
std::cout<<GridLogMessage << "Level "<<level<<" term "<<actionID<<" H = "<<Hterm<<std::endl;
// get gauge field from the SmearingPolicy and
// based on the boolean is_smeared in actionID
GaugeField& 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 "<<actionID<<" H = "<<Hterm<<std::endl;
H += Hterm;
}
}

23
lib/qcd/hmc/integrators/Integrator_algorithm.h
View File

@ -91,14 +91,17 @@ namespace Grid{
* P 1/2 P 1/2
*/
template<class GaugeField> class LeapFrog : public Integrator<GaugeField> {
template<class GaugeField, class SmearingPolicy> class LeapFrog :
public Integrator<GaugeField, SmearingPolicy> {
public:
typedef LeapFrog<GaugeField> Algorithm;
typedef LeapFrog<GaugeField, SmearingPolicy> Algorithm;
LeapFrog(GridBase* grid,
IntegratorParameters Par,
ActionSet<GaugeField> & Aset): Integrator<GaugeField>(grid,Par,Aset) {};
ActionSet<GaugeField> & Aset,
SmearingPolicy & Sm):
Integrator<GaugeField, SmearingPolicy>(grid,Par,Aset,Sm) {};
void step (GaugeField& U, int level,int _first, int _last){
@ -135,7 +138,8 @@ namespace Grid{
}
};
template<class GaugeField> class MinimumNorm2 : public Integrator<GaugeField> {
template<class GaugeField, class SmearingPolicy> class MinimumNorm2 :
public Integrator<GaugeField, SmearingPolicy> {
private:
const RealD lambda = 0.1931833275037836;
@ -143,7 +147,9 @@ namespace Grid{
MinimumNorm2(GridBase* grid,
IntegratorParameters Par,
ActionSet<GaugeField> & Aset): Integrator<GaugeField>(grid,Par,Aset) {};
ActionSet<GaugeField> & Aset,
SmearingPolicy& Sm):
Integrator<GaugeField, SmearingPolicy>(grid,Par,Aset,Sm) {};
void step (GaugeField& U, int level, int _first,int _last){
@ -191,7 +197,8 @@ namespace Grid{
};
template<class GaugeField> class ForceGradient : public Integrator<GaugeField> {
template<class GaugeField, class SmearingPolicy> class ForceGradient :
public Integrator<GaugeField, SmearingPolicy> {
private:
const RealD lambda = 1.0/6.0;;
const RealD chi = 1.0/72.0;
@ -202,7 +209,9 @@ namespace Grid{
// Looks like dH scales as dt^4. tested wilson/wilson 2 level.
ForceGradient(GridBase* grid,
IntegratorParameters Par,
ActionSet<GaugeField> & Aset): Integrator<GaugeField>(grid,Par,Aset) {};
ActionSet<GaugeField> & Aset,
SmearingPolicy &Sm):
Integrator<GaugeField, SmearingPolicy>(grid,Par,Aset, Sm) {};
void FG_update_P(GaugeField&U, int level,double fg_dt,double ep){

130
lib/qcd/smearing/APEsmearing.h Normal file
View File

@ -0,0 +1,130 @@
/*!
@brief Declaration of Smear_APE class for APE smearing
*/
#ifndef APE_SMEAR_
#define APE_SMEAR_
namespace Grid {
namespace QCD {
/*! @brief APE type smearing of link variables. */
template <class Gimpl>
class Smear_APE: public Smear<Gimpl>{
private:
const std::vector<double> rho;/*!< Array of weights */
//This member must be private - we do not want to control from outside
std::vector<double> set_rho(const double common_rho) const {
std::vector<double> res;
for(int mn=0; mn<Nd*Nd; ++mn) res.push_back(common_rho);
for(int mu=0; mu<Nd; ++mu) res[mu + mu*Nd] = 0.0;
return res;
}
public:
// Defines the gauge field types
INHERIT_GIMPL_TYPES(Gimpl)
// Constructors and destructors
Smear_APE(const std::vector<double>& rho_):rho(rho_){} // check vector size
Smear_APE(double rho_val):rho(set_rho(rho_val)){}
Smear_APE():rho(set_rho(1.0)){}
~Smear_APE(){}
///////////////////////////////////////////////////////////////////////////////
void smear(GaugeField& u_smr, const GaugeField& U)const{
GridBase *grid = U._grid;
GaugeLinkField Cup(grid), tmp_stpl(grid);
WilsonLoops<Gimpl> WL;
u_smr = zero;
for(int mu=0; mu<Nd; ++mu){
Cup = zero;
for(int nu=0; nu<Nd; ++nu){
if (nu != mu) {
// get the staple in direction mu, nu
WL.Staple(tmp_stpl, U, mu, nu); //nb staple conventions of IroIro and Grid differ by a dagger
Cup += tmp_stpl*rho[mu + Nd * nu];
}
}
// save the Cup link-field on the u_smr gauge-field
pokeLorentz(u_smr, adj(Cup), mu); // u_smr[mu] = Cup^dag see conventions for Staple
}
}
////////////////////////////////////////////////////////////////////////////////
void derivative(GaugeField& SigmaTerm,
const GaugeField& iLambda,
const GaugeField& U)const{
// Reference
// Morningstar, Peardon, Phys.Rev.D69,054501(2004)
// Equation 75
// Computing Sigma_mu, derivative of S[fat links] with respect to the thin links
// Output SigmaTerm
GridBase *grid = U._grid;
WilsonLoops<Gimpl> WL;
GaugeLinkField staple(grid), u_tmp(grid);
GaugeLinkField iLambda_mu(grid), iLambda_nu(grid);
GaugeLinkField U_mu(grid), U_nu(grid);
GaugeLinkField sh_field(grid), temp_Sigma(grid);
Real rho_munu, rho_numu;
for(int mu = 0; mu < Nd; ++mu){
U_mu = peekLorentz( U, mu);
iLambda_mu = peekLorentz(iLambda, mu);
for(int nu = 0; nu < Nd; ++nu){
if(nu==mu) continue;
U_nu = peekLorentz( U, nu);
iLambda_nu = peekLorentz(iLambda, nu);
rho_munu = rho[mu + Nd * nu];
rho_numu = rho[nu + Nd * mu];
WL.StapleUpper(staple, U, mu, nu);
temp_Sigma = -rho_numu*staple*iLambda_nu; //ok
//-r_numu*U_nu(x+mu)*Udag_mu(x+nu)*Udag_nu(x)*Lambda_nu(x)
Gimpl::AddGaugeLink(SigmaTerm, temp_Sigma, mu);
sh_field = Cshift(iLambda_nu, mu, 1);// general also for Gparity?
temp_Sigma = rho_numu*sh_field*staple; //ok
//r_numu*Lambda_nu(mu)*U_nu(x+mu)*Udag_mu(x+nu)*Udag_nu(x)
Gimpl::AddGaugeLink(SigmaTerm, temp_Sigma, mu);
sh_field = Cshift(iLambda_mu, nu, 1);
temp_Sigma = -rho_munu*staple*U_nu*sh_field*adj(U_nu); //ok
//-r_munu*U_nu(x+mu)*Udag_mu(x+nu)*Lambda_mu(x+nu)*Udag_nu(x)
Gimpl::AddGaugeLink(SigmaTerm, temp_Sigma, mu);
staple = zero;
sh_field = Cshift(U_nu, mu, 1);
temp_Sigma = -rho_munu*adj(sh_field)*adj(U_mu)*iLambda_mu*U_nu;
temp_Sigma += rho_numu*adj(sh_field)*adj(U_mu)*iLambda_nu*U_nu;
u_tmp = adj(U_nu)*iLambda_nu;
sh_field = Cshift(u_tmp, mu, 1);
temp_Sigma += -rho_numu*sh_field*adj(U_mu)*U_nu;
sh_field = Cshift(temp_Sigma, nu, -1);
Gimpl::AddGaugeLink(SigmaTerm, sh_field, mu);
}
}
}
};
}// namespace QCD
}//namespace Grid
#endif

17
lib/qcd/smearing/BaseSmearing.h Normal file
View File

@ -0,0 +1,17 @@
/*
@brief Declares base smearing class Smear
*/
#ifndef BASE_SMEAR_
#define BASE_SMEAR_
template <class Gimpl>
class Smear{
public:
INHERIT_GIMPL_TYPES(Gimpl) // inherits the types for the gauge fields
virtual ~Smear(){}
virtual void smear (GaugeField&,const GaugeField&)const = 0;
virtual void derivative(GaugeField&,
const GaugeField&,const GaugeField&) const = 0;
};
#endif

262
lib/qcd/smearing/GaugeConfiguration.h Normal file
View File

@ -0,0 +1,262 @@
/*!
@file GaugeConfiguration.h
@brief Declares the GaugeConfiguration class
*/
#ifndef GAUGE_CONFIG_
#define GAUGE_CONFIG_
namespace Grid {
namespace QCD {
/*!
@brief Smeared configuration container
It will behave like a configuration from the point of view of
the HMC update and integrators.
An "advanced configuration" object that can provide not only the
data to store the gauge configuration but also operations to manipulate
it, like smearing.
It stores a list of smeared configurations.
*/
template <class Gimpl>
class SmearedConfiguration {
public:
INHERIT_GIMPL_TYPES(Gimpl);
private:
const unsigned int smearingLevels;
Smear_Stout<Gimpl> StoutSmearing;
std::vector<GaugeField> SmearedSet;
// Member functions
//====================================================================
void fill_smearedSet(GaugeField& U) {
ThinLinks = &U; // attach the smearing routine to the field U
// check the pointer is not null
if (ThinLinks == NULL)
std::cout << GridLogError
<< "[SmearedConfiguration] Error in ThinLinks pointer\n";
if (smearingLevels > 0) {
std::cout << GridLogDebug
<< "[SmearedConfiguration] Filling SmearedSet\n";
GaugeField previous_u(ThinLinks->_grid);
previous_u = *ThinLinks;
for (int smearLvl = 0; smearLvl < smearingLevels; ++smearLvl) {
StoutSmearing.smear(SmearedSet[smearLvl], previous_u);
previous_u = SmearedSet[smearLvl];
// For debug purposes
RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(previous_u);
std::cout << GridLogDebug
<< "[SmearedConfiguration] Plaq: " << impl_plaq << std::endl;
}
}
}
//====================================================================
GaugeField AnalyticSmearedForce(const GaugeField& SigmaKPrime,
const GaugeField& GaugeK) const {
GridBase* grid = GaugeK._grid;
GaugeField C(grid), SigmaK(grid), iLambda(grid);
GaugeLinkField iLambda_mu(grid);
GaugeLinkField iQ(grid), e_iQ(grid);
GaugeLinkField SigmaKPrime_mu(grid);
GaugeLinkField GaugeKmu(grid), Cmu(grid);
StoutSmearing.BaseSmear(C, GaugeK);
SigmaK = zero;
iLambda = zero;
for (int mu = 0; mu < Nd; mu++) {
Cmu = peekLorentz(C, mu);
GaugeKmu = peekLorentz(GaugeK, mu);
SigmaKPrime_mu = peekLorentz(SigmaKPrime, mu);
iQ = Ta(Cmu * adj(GaugeKmu));
set_iLambda(iLambda_mu, e_iQ, iQ, SigmaKPrime_mu, GaugeKmu);
pokeLorentz(SigmaK, SigmaKPrime_mu * e_iQ + adj(Cmu) * iLambda_mu, mu);
pokeLorentz(iLambda, iLambda_mu, mu);
}
StoutSmearing.derivative(SigmaK, iLambda,
GaugeK); // derivative of SmearBase
return SigmaK;
}
/*! @brief Returns smeared configuration at level 'Level' */
const GaugeField& get_smeared_conf(int Level) const {
return SmearedSet[Level];
}
//====================================================================
void set_iLambda(GaugeLinkField& iLambda, GaugeLinkField& e_iQ,
const GaugeLinkField& iQ, const GaugeLinkField& Sigmap,
const GaugeLinkField& GaugeK) const {
GridBase* grid = iQ._grid;
GaugeLinkField iQ2(grid), iQ3(grid), B1(grid), B2(grid), USigmap(grid);
GaugeLinkField unity(grid);
unity = 1.0;
LatticeComplex u(grid), w(grid);
LatticeComplex f0(grid), f1(grid), f2(grid);
LatticeComplex xi0(grid), xi1(grid), tmp(grid);
LatticeComplex u2(grid), w2(grid), cosw(grid);
LatticeComplex emiu(grid), e2iu(grid), qt(grid), fden(grid);
LatticeComplex r01(grid), r11(grid), r21(grid), r02(grid), r12(grid);
LatticeComplex r22(grid), tr1(grid), tr2(grid);
LatticeComplex b10(grid), b11(grid), b12(grid), b20(grid), b21(grid),
b22(grid);
LatticeComplex LatticeUnitComplex(grid);
LatticeUnitComplex = 1.0;
// Exponential
iQ2 = iQ * iQ;
iQ3 = iQ * iQ2;
StoutSmearing.set_uw(u, w, iQ2, iQ3);
StoutSmearing.set_fj(f0, f1, f2, u, w);
e_iQ = f0 * unity + timesMinusI(f1) * iQ - f2 * iQ2;
// Getting B1, B2, Gamma and Lambda
// simplify this part, reduntant calculations in set_fj
xi0 = StoutSmearing.func_xi0(w);
xi1 = StoutSmearing.func_xi1(w);
u2 = u * u;
w2 = w * w;
cosw = cos(w);
emiu = cos(u) - timesI(sin(u));
e2iu = cos(2.0 * u) + timesI(sin(2.0 * u));
r01 = (2.0 * u + timesI(2.0 * (u2 - w2))) * e2iu +
emiu * ((16.0 * u * cosw + 2.0 * u * (3.0 * u2 + w2) * xi0) +
timesI(-8.0 * u2 * cosw + 2.0 * (9.0 * u2 + w2) * xi0));
r11 = (2.0 * LatticeUnitComplex + timesI(4.0 * u)) * e2iu +
emiu * ((-2.0 * cosw + (3.0 * u2 - w2) * xi0) +
timesI((2.0 * u * cosw + 6.0 * u * xi0)));
r21 =
2.0 * timesI(e2iu) + emiu * (-3.0 * u * xi0 + timesI(cosw - 3.0 * xi0));
r02 = -2.0 * e2iu +
emiu * (-8.0 * u2 * xi0 +
timesI(2.0 * u * (cosw + xi0 + 3.0 * u2 * xi1)));
r12 = emiu * (2.0 * u * xi0 + timesI(-cosw - xi0 + 3.0 * u2 * xi1));
r22 = emiu * (xi0 - timesI(3.0 * u * xi1));
fden = LatticeUnitComplex / (2.0 * (9.0 * u2 - w2) * (9.0 * u2 - w2));
b10 = 2.0 * u * r01 + (3.0 * u2 - w2) * r02 - (30.0 * u2 + 2.0 * w2) * f0;
b11 = 2.0 * u * r11 + (3.0 * u2 - w2) * r12 - (30.0 * u2 + 2.0 * w2) * f1;
b12 = 2.0 * u * r21 + (3.0 * u2 - w2) * r22 - (30.0 * u2 + 2.0 * w2) * f2;
b20 = r01 - (3.0 * u) * r02 - (24.0 * u) * f0;
b21 = r11 - (3.0 * u) * r12 - (24.0 * u) * f1;
b22 = r21 - (3.0 * u) * r22 - (24.0 * u) * f2;
b10 *= fden;
b11 *= fden;
b12 *= fden;
b20 *= fden;
b21 *= fden;
b22 *= fden;
B1 = b10 * unity + timesMinusI(b11) * iQ - b12 * iQ2;
B2 = b20 * unity + timesMinusI(b21) * iQ - b22 * iQ2;
USigmap = GaugeK * Sigmap;
tr1 = trace(USigmap * B1);
tr2 = trace(USigmap * B2);
GaugeLinkField QUS = iQ * USigmap;
GaugeLinkField USQ = USigmap * iQ;
GaugeLinkField iGamma = tr1 * iQ - timesI(tr2) * iQ2 +
timesI(f1) * USigmap + f2 * QUS + f2 * USQ;
iLambda = Ta(iGamma);
}
//====================================================================
public:
GaugeField*
ThinLinks; /*!< @brief Pointer to the thin
links configuration */
/*! @brief Standard constructor */
SmearedConfiguration(GridCartesian* UGrid, unsigned int Nsmear,
Smear_Stout<Gimpl>& Stout)
: smearingLevels(Nsmear), StoutSmearing(Stout), ThinLinks(NULL) {
for (unsigned int i = 0; i < smearingLevels; ++i)
SmearedSet.push_back(*(new GaugeField(UGrid)));
}
/*! For just thin links */
SmearedConfiguration()
: smearingLevels(0), StoutSmearing(), SmearedSet(), ThinLinks(NULL) {}
// attach the smeared routines to the thin links U and fill the smeared set
void set_GaugeField(GaugeField& U) { fill_smearedSet(U); }
//====================================================================
void smeared_force(GaugeField& SigmaTilde) const {
if (smearingLevels > 0) {
GaugeField force = SigmaTilde; // actually = U*SigmaTilde
GaugeLinkField tmp_mu(SigmaTilde._grid);
for (int mu = 0; mu < Nd; mu++) {
// to get just SigmaTilde
tmp_mu = adj(peekLorentz(SmearedSet[smearingLevels - 1], mu)) *
peekLorentz(force, mu);
pokeLorentz(force, tmp_mu, mu);
}
for (int ismr = smearingLevels - 1; ismr > 0; --ismr)
force = AnalyticSmearedForce(force, get_smeared_conf(ismr - 1));
force = AnalyticSmearedForce(force, *ThinLinks);
for (int mu = 0; mu < Nd; mu++) {
tmp_mu = peekLorentz(*ThinLinks, mu) * peekLorentz(force, mu);
pokeLorentz(SigmaTilde, tmp_mu, mu);
}
} // if smearingLevels = 0 do nothing
}
//====================================================================
GaugeField& get_SmearedU() { return SmearedSet[smearingLevels - 1]; }
GaugeField& get_U(bool smeared = false) {
// get the config, thin links by default
if (smeared) {
if (smearingLevels) {
RealD impl_plaq =
WilsonLoops<Gimpl>::avgPlaquette(SmearedSet[smearingLevels - 1]);
std::cout << GridLogDebug << "getting Usmr Plaq: " << impl_plaq
<< std::endl;
return get_SmearedU();
} else {
RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(*ThinLinks);
std::cout << GridLogDebug << "getting Thin Plaq: " << impl_plaq
<< std::endl;
return *ThinLinks;
}
} else {
RealD impl_plaq = WilsonLoops<Gimpl>::avgPlaquette(*ThinLinks);
std::cout << GridLogDebug << "getting Thin Plaq: " << impl_plaq
<< std::endl;
return *ThinLinks;
}
}
};
}
}
#endif

9
lib/qcd/smearing/Smearing.h Normal file
View File

@ -0,0 +1,9 @@
#ifndef GRID_QCD_SMEARING_H
#define GRID_QCD_SMEARING_H
#include <qcd/smearing/BaseSmearing.h>
#include <qcd/smearing/APEsmearing.h>
#include <qcd/smearing/StoutSmearing.h>
#include <qcd/smearing/GaugeConfiguration.h>
#endif

160
lib/qcd/smearing/StoutSmearing.h Normal file
View File

@ -0,0 +1,160 @@
/*
@file stoutSmear.hpp
@brief Declares Stout smearing class
*/
#ifndef STOUT_SMEAR_
#define STOUT_SMEAR_
namespace Grid {
namespace QCD {
/*! @brief Stout smearing of link variable. */
template <class Gimpl>
class Smear_Stout : public Smear<Gimpl> {
private:
const Smear<Gimpl>* SmearBase;
public:
INHERIT_GIMPL_TYPES(Gimpl)
Smear_Stout(Smear<Gimpl>* base) : SmearBase(base) {
static_assert(Nc == 3,
"Stout smearing currently implemented only for Nc==3");
}
/*! Default constructor */
Smear_Stout(double rho = 1.0) : SmearBase(new Smear_APE<Gimpl>(rho)) {
static_assert(Nc == 3,
"Stout smearing currently implemented only for Nc==3");
}
~Smear_Stout() {} // delete SmearBase...
void smear(GaugeField& u_smr, const GaugeField& U) const {
GaugeField C(U._grid);
GaugeLinkField tmp(U._grid), iq_mu(U._grid), Umu(U._grid);
std::cout << GridLogDebug << "Stout smearing started\n";
// Smear the configurations
SmearBase->smear(C, U);
for (int mu = 0; mu < Nd; mu++) {
tmp = peekLorentz(C, mu);
Umu = peekLorentz(U, mu);
iq_mu = Ta(
tmp *
adj(Umu)); // iq_mu = Ta(Omega_mu) to match the signs with the paper
exponentiate_iQ(tmp, iq_mu);
pokeLorentz(u_smr, tmp * Umu, mu); // u_smr = exp(iQ_mu)*U_mu
}
std::cout << GridLogDebug << "Stout smearing completed\n";
};
void derivative(GaugeField& SigmaTerm, const GaugeField& iLambda,
const GaugeField& Gauge) const {
SmearBase->derivative(SigmaTerm, iLambda, Gauge);
};
void BaseSmear(GaugeField& C, const GaugeField& U) const {
SmearBase->smear(C, U);
};
void exponentiate_iQ(GaugeLinkField& e_iQ, const GaugeLinkField& iQ) const {
// Put this outside
// only valid for SU(3) matrices
// only one Lorentz direction at a time
// notice that it actually computes
// exp ( input matrix )
// the i sign is coming from outside
// input matrix is anti-hermitian NOT hermitian
GridBase* grid = iQ._grid;
GaugeLinkField unity(grid);
unity = 1.0;
GaugeLinkField iQ2(grid), iQ3(grid);
LatticeComplex u(grid), w(grid);
LatticeComplex f0(grid), f1(grid), f2(grid);
iQ2 = iQ * iQ;
iQ3 = iQ * iQ2;
set_uw(u, w, iQ2, iQ3);
set_fj(f0, f1, f2, u, w);
e_iQ = f0 * unity + timesMinusI(f1) * iQ - f2 * iQ2;
};
void set_uw(LatticeComplex& u, LatticeComplex& w, GaugeLinkField& iQ2,
GaugeLinkField& iQ3) const {
Complex one_over_three = 1.0 / 3.0;
Complex one_over_two = 1.0 / 2.0;
GridBase* grid = u._grid;
LatticeComplex c0(grid), c1(grid), tmp(grid), c0max(grid), theta(grid);
// sign in c0 from the conventions on the Ta
c0 = -imag(trace(iQ3)) * one_over_three;
c1 = -real(trace(iQ2)) * one_over_two;
// Cayley Hamilton checks to machine precision, tested
tmp = c1 * one_over_three;
c0max = 2.0 * pow(tmp, 1.5);
theta = acos(c0 / c0max) *
one_over_three; // divide by three here, now leave as it is
u = sqrt(tmp) * cos(theta);
w = sqrt(c1) * sin(theta);
}
void set_fj(LatticeComplex& f0, LatticeComplex& f1, LatticeComplex& f2,
const LatticeComplex& u, const LatticeComplex& w) const {
GridBase* grid = u._grid;
LatticeComplex xi0(grid), u2(grid), w2(grid), cosw(grid);
LatticeComplex fden(grid);
LatticeComplex h0(grid), h1(grid), h2(grid);
LatticeComplex e2iu(grid), emiu(grid), ixi0(grid), qt(grid);
LatticeComplex unity(grid);
unity = 1.0;
xi0 = func_xi0(w);
u2 = u * u;
w2 = w * w;
cosw = cos(w);
ixi0 = timesI(xi0);
emiu = cos(u) - timesI(sin(u));
e2iu = cos(2.0 * u) + timesI(sin(2.0 * u));
h0 = e2iu * (u2 - w2) +
emiu * ((8.0 * u2 * cosw) + (2.0 * u * (3.0 * u2 + w2) * ixi0));
h1 = e2iu * (2.0 * u) - emiu * ((2.0 * u * cosw) - (3.0 * u2 - w2) * ixi0);
h2 = e2iu - emiu * (cosw + (3.0 * u) * ixi0);
fden = unity / (9.0 * u2 - w2); // reals
f0 = h0 * fden;
f1 = h1 * fden;
f2 = h2 * fden;
}
LatticeComplex func_xi0(const LatticeComplex& w) const {
// Define a function to do the check
// if( w < 1e-4 ) std::cout << GridLogWarning<< "[Smear_stout] w too small:
// "<< w <<"\n";
return sin(w) / w;
}
LatticeComplex func_xi1(const LatticeComplex& w) const {
// Define a function to do the check
// if( w < 1e-4 ) std::cout << GridLogWarning << "[Smear_stout] w too small:
// "<< w <<"\n";
return cos(w) / (w * w) - sin(w) / (w * w * w);
}
};
}
}
#endif

39
lib/qcd/utils/SUn.h
View File

@ -552,15 +552,24 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
}
// reunitarise??
static void LieRandomize(GridParallelRNG &pRNG,LatticeMatrix &out,double scale=1.0){
template<typename LatticeMatrixType>
static void LieRandomize(GridParallelRNG &pRNG,LatticeMatrixType &out,double scale=1.0){
GridBase *grid = out._grid;
LatticeComplex ca (grid);
LatticeMatrix lie(grid);
LatticeMatrix la (grid);
Complex ci(0.0,scale);
Complex cone(1.0,0.0);
Matrix ta;
typedef typename LatticeMatrixType::vector_type vector_type;
typedef typename LatticeMatrixType::scalar_type scalar_type;
typedef iSinglet<vector_type> vTComplexType;
typedef Lattice<vTComplexType> LatticeComplexType;
typedef typename GridTypeMapper<typename LatticeMatrixType::vector_object>::scalar_object MatrixType;
LatticeComplexType ca (grid);
LatticeMatrixType lie(grid);
LatticeMatrixType la (grid);
ComplexD ci(0.0,scale);
ComplexD cone(1.0,0.0);
MatrixType ta;
lie=zero;
for(int a=0;a<generators();a++){
@ -596,9 +605,13 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
}
template<typename GaugeField>
static void HotConfiguration(GridParallelRNG &pRNG,GaugeField &out){
typedef typename GaugeField::vector_type vector_type;
typedef iSUnMatrix<vector_type> vMatrixType;
typedef Lattice<vMatrixType> LatticeMatrixType;
static void HotConfiguration(GridParallelRNG &pRNG,LatticeGaugeField &out){
LatticeMatrix Umu(out._grid);
LatticeMatrixType Umu(out._grid);
for(int mu=0;mu<Nd;mu++){
LieRandomize(pRNG,Umu,1.0);
PokeIndex<LorentzIndex>(out,Umu,mu);
@ -622,13 +635,15 @@ Note that in step D setting B ~ X - A and using B in place of A in step E will g
static void taProj( const LatticeMatrix &in, LatticeMatrix &out){
out = Ta(in);
}
static void taExp( const LatticeMatrix &x, LatticeMatrix &ex){
template<typename LatticeMatrixType>
static void taExp( const LatticeMatrixType &x, LatticeMatrixType &ex){
typedef typename LatticeMatrixType::scalar_type ComplexType;
LatticeMatrix xn(x._grid);
LatticeMatrixType xn(x._grid);
RealD nfac = 1.0;
xn = x;
ex =xn+Complex(1.0); // 1+x
ex =xn+ComplexType(1.0); // 1+x
// Do a 12th order exponentiation
for(int i=2; i <= 12; ++i)

286
lib/qcd/utils/WilsonLoops.h
View File

@ -25,7 +25,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef QCD_UTILS_WILSON_LOOPS_H
@ -34,10 +35,8 @@ namespace Grid {
namespace QCD {
// Common wilson loop observables
template<class Gimpl>
class WilsonLoops : public Gimpl {
template <class Gimpl> class WilsonLoops : public Gimpl {
public:
INHERIT_GIMPL_TYPES(Gimpl);
typedef typename Gimpl::GaugeLinkField GaugeMat;
@ -46,21 +45,25 @@ public:
//////////////////////////////////////////////////
// directed plaquette oriented in mu,nu plane
//////////////////////////////////////////////////
static void dirPlaquette(GaugeMat &plaq,const std::vector<GaugeMat> &U, const int mu, const int nu)
{
// Annoyingly, must use either scope resolution to find dependent base class,
// or this-> ; there is no "this" in a static method. This forces explicit Gimpl scope
// resolution throughout the usage in this file, and rather defeats the purpose of deriving
static void dirPlaquette(GaugeMat &plaq, const std::vector<GaugeMat> &U,
const int mu, const int nu) {
// Annoyingly, must use either scope resolution to find dependent base
// class,
// or this-> ; there is no "this" in a static method. This forces explicit
// Gimpl scope
// resolution throughout the usage in this file, and rather defeats the
// purpose of deriving
// from Gimpl.
plaq= Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[nu],nu,
Gimpl::CovShiftForward (U[mu],mu,U[nu])));
plaq = Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftBackward(
U[nu], nu, Gimpl::CovShiftForward(U[mu], mu, U[nu])));
}
//////////////////////////////////////////////////
// trace of directed plaquette oriented in mu,nu plane
//////////////////////////////////////////////////
static void traceDirPlaquette(LatticeComplex &plaq, const std::vector<GaugeMat> &U, const int mu, const int nu)
{
static void traceDirPlaquette(LatticeComplex &plaq,
const std::vector<GaugeMat> &U, const int mu,
const int nu) {
GaugeMat sp(U[0]._grid);
dirPlaquette(sp, U, mu, nu);
plaq = trace(sp);
@ -68,8 +71,8 @@ public:
//////////////////////////////////////////////////
// sum over all planes of plaquette
//////////////////////////////////////////////////
static void sitePlaquette(LatticeComplex &Plaq,const std::vector<GaugeMat> &U)
{
static void sitePlaquette(LatticeComplex &Plaq,
const std::vector<GaugeMat> &U) {
LatticeComplex sitePlaq(U[0]._grid);
Plaq = zero;
for (int mu = 1; mu < Nd; mu++) {
@ -83,7 +86,7 @@ public:
// sum over all x,y,z,t and over all planes of plaquette
//////////////////////////////////////////////////
static RealD sumPlaquette(const GaugeLorentz &Umu) {
std::vector<GaugeMat> U(Nd,Umu._grid);
std::vector<GaugeMat> U(4, Umu._grid);
for (int mu = 0; mu < Nd; mu++) {
U[mu] = PeekIndex<LorentzIndex>(Umu, mu);
@ -101,19 +104,20 @@ public:
// average over all x,y,z,t and over all planes of plaquette
//////////////////////////////////////////////////
static RealD avgPlaquette(const GaugeLorentz &Umu) {
RealD sumplaq = sumPlaquette(Umu);
double vol = Umu._grid->gSites();
double faces = (1.0 * Nd * (Nd - 1)) / 2.0;
return sumplaq / vol / faces / Nc; // Nd , Nc dependent... FIXME
}
static RealD linkTrace(const GaugeLorentz &Umu){
std::vector<GaugeMat> U(Nd,Umu._grid);
LatticeComplex Tr(Umu._grid); Tr=zero;
//////////////////////////////////////////////////
// average over traced single links
//////////////////////////////////////////////////
static RealD linkTrace(const GaugeLorentz &Umu) {
std::vector<GaugeMat> U(4, Umu._grid);
LatticeComplex Tr(Umu._grid);
Tr = zero;
for (int mu = 0; mu < Nd; mu++) {
U[mu] = PeekIndex<LorentzIndex>(Umu, mu);
Tr = Tr + trace(U[mu]);
@ -124,8 +128,53 @@ public:
double vol = Umu._grid->gSites();
return p.real()/vol/((double)(Nd*(Nd-1)));
return p.real() / vol / 4.0 / 3.0;
};
//////////////////////////////////////////////////
// the sum over all staples on each site in direction mu,nu
//////////////////////////////////////////////////
static void Staple(GaugeMat &staple, const GaugeLorentz &Umu, int mu,
int nu) {
GridBase *grid = Umu._grid;
std::vector<GaugeMat> U(4, grid);
for (int d = 0; d < Nd; d++) {
U[d] = PeekIndex<LorentzIndex>(Umu, d);
}
staple = zero;
if (nu != mu) {
// mu
// ^
// |__> nu
// __
// |
// __|
//
staple += Gimpl::ShiftStaple(
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftIdentityBackward(U[nu], nu))),
mu);
// __
// |
// |__
//
//
staple += Gimpl::ShiftStaple(
Gimpl::CovShiftBackward(U[nu], nu,
Gimpl::CovShiftBackward(U[mu], mu, U[nu])),
mu);
}
}
//////////////////////////////////////////////////
// the sum over all staples on each site
//////////////////////////////////////////////////
@ -140,7 +189,6 @@ public:
staple = zero;
GaugeMat tmp(grid);
for (int nu = 0; nu < Nd; nu++) {
if (nu != mu) {
@ -155,9 +203,11 @@ public:
//
staple += Gimpl::ShiftStaple(
Gimpl::CovShiftForward (U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftIdentityBackward(U[nu],nu))),mu);
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftIdentityBackward(U[nu], nu))),
mu);
// __
// |
@ -166,30 +216,70 @@ public:
//
staple += Gimpl::ShiftStaple(
Gimpl::CovShiftBackward(U[nu], nu,
Gimpl::CovShiftBackward(U[mu],mu,U[nu])),mu);
Gimpl::CovShiftBackward(U[mu], mu, U[nu])),
mu);
}
}
}
//////////////////////////////////////////////////
// the sum over all staples on each site in direction mu,nu, upper part
//////////////////////////////////////////////////
static void StapleUpper(GaugeMat &staple, const GaugeLorentz &Umu, int mu,
int nu) {
staple = zero;
if (nu != mu) {
GridBase *grid = Umu._grid;
std::vector<GaugeMat> U(4, grid);
for (int d = 0; d < Nd; d++) {
U[d] = PeekIndex<LorentzIndex>(Umu, d);
}
// mu
// ^
// |__> nu
// __
// |
// __|
//
staple += Gimpl::ShiftStaple(
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftIdentityBackward(U[nu], nu))),
mu);
}
}
//////////////////////////////////////////////////////
// Similar to above for rectangle is required
//////////////////////////////////////////////////////
static void dirRectangle(GaugeMat &rect,const std::vector<GaugeMat> &U, const int mu, const int nu)
{
rect = Gimpl::CovShiftForward(U[mu],mu,Gimpl::CovShiftForward(U[mu],mu,U[nu]))* // ->->|
adj(Gimpl::CovShiftForward(U[nu],nu,Gimpl::CovShiftForward(U[mu],mu,U[mu]))) ;
static void dirRectangle(GaugeMat &rect, const std::vector<GaugeMat> &U,
const int mu, const int nu) {
rect = Gimpl::CovShiftForward(
U[mu], mu, Gimpl::CovShiftForward(U[mu], mu, U[nu])) * // ->->|
adj(Gimpl::CovShiftForward(
U[nu], nu, Gimpl::CovShiftForward(U[mu], mu, U[mu])));
rect = rect +
Gimpl::CovShiftForward(U[mu],mu,Gimpl::CovShiftForward(U[nu],nu,U[nu]))* // ->||
adj(Gimpl::CovShiftForward(U[nu],nu,Gimpl::CovShiftForward(U[nu],nu,U[mu]))) ;
Gimpl::CovShiftForward(
U[mu], mu, Gimpl::CovShiftForward(U[nu], nu, U[nu])) * // ->||
adj(Gimpl::CovShiftForward(
U[nu], nu, Gimpl::CovShiftForward(U[nu], nu, U[mu])));
}
static void traceDirRectangle(LatticeComplex &rect, const std::vector<GaugeMat> &U, const int mu, const int nu)
{
static void traceDirRectangle(LatticeComplex &rect,
const std::vector<GaugeMat> &U, const int mu,
const int nu) {
GaugeMat sp(U[0]._grid);
dirRectangle(sp, U, mu, nu);
rect = trace(sp);
}
static void siteRectangle(LatticeComplex &Rect,const std::vector<GaugeMat> &U)
{
static void siteRectangle(LatticeComplex &Rect,
const std::vector<GaugeMat> &U) {
LatticeComplex siteRect(U[0]._grid);
Rect = zero;
for (int mu = 1; mu < Nd; mu++) {
@ -199,6 +289,7 @@ public:
}
}
}
//////////////////////////////////////////////////
// sum over all x,y,z,t and over all planes of plaquette
//////////////////////////////////////////////////
@ -239,11 +330,14 @@ public:
}
////////////////////////////////////////////////////////////////////////////
// Hop by two optimisation strategy does not work nicely with Gparity. (could do,
// Hop by two optimisation strategy does not work nicely with Gparity. (could
// do,
// but need to track two deep where cross boundary and apply a conjugation).
// Must differentiate this in Gimpl, and use Gimpl::isPeriodicGaugeField to do so .
// Must differentiate this in Gimpl, and use Gimpl::isPeriodicGaugeField to do
// so .
////////////////////////////////////////////////////////////////////////////
static void RectStapleOptimised(GaugeMat &Stap,std::vector<GaugeMat> &U2,std::vector<GaugeMat> &U,int mu){
static void RectStapleOptimised(GaugeMat &Stap, std::vector<GaugeMat> &U2,
std::vector<GaugeMat> &U, int mu) {
Stap = zero;
@ -263,14 +357,12 @@ public:
Staple2x1 = Gimpl::CovShiftForward(U[nu], nu, tmp);
// Down staple
// |___ ___|
//
tmp = adj(U2[mu]) * U[nu];
Staple2x1 += Gimpl::CovShiftBackward(U[nu], nu, Cshift(tmp, mu, -2));
// ___ ___
// | ___|
// |___ ___|
@ -285,7 +377,8 @@ public:
// tmp= Staple2x1* Cshift(U[mu],mu,-2);
// Stap+= Cshift(tmp,mu,1) ;
Stap+= Cshift(Staple2x1,mu,1)*Cshift(U[mu],mu,-1); ;
Stap += Cshift(Staple2x1, mu, 1) * Cshift(U[mu], mu, -1);
;
// --
// | |
@ -306,19 +399,16 @@ public:
tmp = adj(U2[nu]) * tmp;
tmp = Cshift(tmp, nu, -2);
Stap += Cshift(tmp, mu, 1);
}}
}
}
}
static void RectStaple(GaugeMat &Stap,const GaugeLorentz & Umu,int mu)
{
static void RectStaple(GaugeMat &Stap, const GaugeLorentz &Umu, int mu) {
RectStapleUnoptimised(Stap, Umu, mu);
}
static void RectStaple(const GaugeLorentz &Umu, GaugeMat &Stap,
std::vector<GaugeMat> &U2,
std::vector<GaugeMat> &U, int mu)
{
std::vector<GaugeMat> &U2, std::vector<GaugeMat> &U,
int mu) {
if (Gimpl::isPeriodicGaugeField()) {
RectStapleOptimised(Stap, U2, U, mu);
} else {
@ -326,7 +416,8 @@ public:
}
}
static void RectStapleUnoptimised(GaugeMat &Stap,const GaugeLorentz &Umu,int mu){
static void RectStapleUnoptimised(GaugeMat &Stap, const GaugeLorentz &Umu,
int mu) {
GridBase *grid = Umu._grid;
std::vector<GaugeMat> U(Nd, grid);
@ -342,38 +433,52 @@ public:
// | __ |
//
Stap += Gimpl::ShiftStaple(
Gimpl::CovShiftForward (U[mu],mu,
Gimpl::CovShiftForward (U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftIdentityBackward(U[nu],nu))))) , mu);
Gimpl::CovShiftForward(
U[mu], mu,
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu,
Gimpl::CovShiftBackward(
U[mu], mu,
Gimpl::CovShiftIdentityBackward(U[nu], nu))))),
mu);
// __
// |__ __ |
Stap += Gimpl::ShiftStaple(
Gimpl::CovShiftForward (U[mu],mu,
Gimpl::CovShiftBackward(U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[mu],mu, U[nu])))) , mu);
Gimpl::CovShiftForward(
U[mu], mu,
Gimpl::CovShiftBackward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftBackward(U[mu], mu, U[nu])))),
mu);
// __
// |__ __ |
Stap += Gimpl::ShiftStaple(
Gimpl::CovShiftBackward(U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftForward(U[nu],nu,U[mu])))) , mu);
Gimpl::CovShiftBackward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftForward(U[nu], nu, U[mu])))),
mu);
// __ ___
// |__ |
Stap += Gimpl::ShiftStaple(
Gimpl::CovShiftForward (U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[nu],nu,U[mu])))) , mu);
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftBackward(U[nu], nu, U[mu])))),
mu);
// --
// | |
@ -381,12 +486,16 @@ public:
// | |
Stap += Gimpl::ShiftStaple(
Gimpl::CovShiftForward(U[nu],nu,
Gimpl::CovShiftForward(U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftBackward(U[nu],nu,
Gimpl::CovShiftIdentityBackward(U[nu],nu))))) , mu);
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftForward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu,
Gimpl::CovShiftBackward(
U[nu], nu,
Gimpl::CovShiftIdentityBackward(U[nu], nu))))),
mu);
// | |
//
@ -394,22 +503,23 @@ public:
// --
Stap += Gimpl::ShiftStaple(
Gimpl::CovShiftBackward(U[nu],nu,
Gimpl::CovShiftBackward(U[nu],nu,
Gimpl::CovShiftBackward(U[mu],mu,
Gimpl::CovShiftForward (U[nu],nu,U[nu])))) , mu);
}}
Gimpl::CovShiftBackward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[nu], nu,
Gimpl::CovShiftBackward(
U[mu], mu, Gimpl::CovShiftForward(U[nu], nu, U[nu])))),
mu);
}
}
}
};
typedef WilsonLoops<PeriodicGimplR> ColourWilsonLoops;
typedef WilsonLoops<PeriodicGimplR> U1WilsonLoops;
typedef WilsonLoops<PeriodicGimplR> SU2WilsonLoops;
typedef WilsonLoops<PeriodicGimplR> SU3WilsonLoops;
}}
}
}
#endif

0
lib/serialisation/.dirstamp
View File

419
lib/simd/Grid_vector_types.h
View File

@ -25,7 +25,8 @@ Author: neo <cossu@post.kek.jp>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
//---------------------------------------------------------------------------
@ -64,10 +65,12 @@ namespace Grid {
//////////////////////////////////////
// To take the floating point type of real/complex type
//////////////////////////////////////
template <typename T> struct RealPart {
template <typename T>
struct RealPart {
typedef T type;
};
template <typename T> struct RealPart< std::complex<T> >{
template <typename T>
struct RealPart<std::complex<T> > {
typedef T type;
};
@ -76,24 +79,36 @@ namespace Grid {
//////////////////////////////////////
// type alias used to simplify the syntax of std::enable_if
template <typename T> using Invoke = typename T::type;
template <typename Condition, typename ReturnType> using EnableIf = Invoke<std::enable_if<Condition::value, ReturnType> >;
template <typename Condition, typename ReturnType> using NotEnableIf= Invoke<std::enable_if<!Condition::value, ReturnType> >;
template <typename T>
using Invoke = typename T::type;
template <typename Condition, typename ReturnType>
using EnableIf = Invoke<std::enable_if<Condition::value, ReturnType> >;
template <typename Condition, typename ReturnType>
using NotEnableIf = Invoke<std::enable_if<!Condition::value, ReturnType> >;
////////////////////////////////////////////////////////
// Check for complexity with type traits
template <typename T> struct is_complex : public std::false_type {};
template <> struct is_complex<std::complex<double> >: public std::true_type {};
template <> struct is_complex<std::complex<float> > : public std::true_type {};
template <typename T>
struct is_complex : public std::false_type {};
template <>
struct is_complex<std::complex<double> > : public std::true_type {};
template <>
struct is_complex<std::complex<float> > : public std::true_type {};
template <typename T> using IfReal = Invoke<std::enable_if<std::is_floating_point<T>::value,int> > ;
template <typename T> using IfComplex = Invoke<std::enable_if<is_complex<T>::value,int> > ;
template <typename T> using IfInteger = Invoke<std::enable_if<std::is_integral<T>::value,int> > ;
template <typename T>
using IfReal = Invoke<std::enable_if<std::is_floating_point<T>::value, int> >;
template <typename T>
using IfComplex = Invoke<std::enable_if<is_complex<T>::value, int> >;
template <typename T>
using IfInteger = Invoke<std::enable_if<std::is_integral<T>::value, int> >;
template <typename T> using IfNotReal = Invoke<std::enable_if<!std::is_floating_point<T>::value,int> > ;
template <typename T> using IfNotComplex = Invoke<std::enable_if<!is_complex<T>::value,int> > ;
template <typename T> using IfNotInteger = Invoke<std::enable_if<!std::is_integral<T>::value,int> > ;
template <typename T>
using IfNotReal =
Invoke<std::enable_if<!std::is_floating_point<T>::value, int> >;
template <typename T>
using IfNotComplex = Invoke<std::enable_if<!is_complex<T>::value, int> >;
template <typename T>
using IfNotInteger = Invoke<std::enable_if<!std::is_integral<T>::value, int> >;
////////////////////////////////////////////////////////
// Define the operation templates functors
@ -111,33 +126,112 @@ namespace Grid {
}
///////////////////////////////////////////////
/*
@brief Grid_simd class for the SIMD vector type operations
*/
template <class Scalar_type, class Vector_type>
class Grid_simd {
public:
typedef typename RealPart<Scalar_type>::type Real;
typedef Vector_type vector_type;
typedef Scalar_type scalar_type;
typedef union conv_t_union {
Vector_type v;
Scalar_type s[sizeof(Vector_type) / sizeof(Scalar_type)];
conv_t_union(){};
} conv_t;
////////////////////////////////////////////////////////////////////
// General permute; assumes vector length is same across
// all subtypes; may not be a good assumption, but could
// add the vector width as a template param for BG/Q for example
////////////////////////////////////////////////////////////////////
friend inline void permute0(Grid_simd &y,Grid_simd b){
y.v = Optimization::Permute::Permute0(b.v);
}
friend inline void permute1(Grid_simd &y,Grid_simd b){
y.v = Optimization::Permute::Permute1(b.v);
}
friend inline void permute2(Grid_simd &y,Grid_simd b){
y.v = Optimization::Permute::Permute2(b.v);
}
friend inline void permute3(Grid_simd &y,Grid_simd b){
y.v = Optimization::Permute::Permute3(b.v);
}
friend inline void permute(Grid_simd &y,Grid_simd b,int perm)
{
if ( perm & RotateBit ) {
int dist = perm&0xF;
y=rotate(b,dist);
return;
} else {
switch(perm){
case 3: permute3(y,b); break;
case 2: permute2(y,b); break;
case 1: permute1(y,b); break;
case 0: permute0(y,b); break;
default: assert(0);
}
}
}
Vector_type v;
};// end of Grid_simd class definition
static inline int Nsimd(void) { return sizeof(Vector_type)/sizeof(Scalar_type);}
static inline int Nsimd(void) {
return sizeof(Vector_type) / sizeof(Scalar_type);
}
template <class S, class V, IfNotComplex<S> =0>
inline void rotate( Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
{
nrot = nrot % Grid_simd<S,V>::Nsimd();
// std::cout << "Rotate Real by "<<nrot<<std::endl;
ret.v = Optimization::Rotate::rotate(b.v,nrot);
}
template <class S, class V, IfComplex<S> =0>
inline void rotate(Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
{
nrot = nrot % Grid_simd<S,V>::Nsimd();
// std::cout << "Rotate Complex by "<<nrot<<std::endl;
ret.v = Optimization::Rotate::rotate(b.v,2*nrot);
}
Grid_simd& operator=(const Grid_simd&& rhs){v=rhs.v;return *this;};
Grid_simd& operator=(const Grid_simd& rhs){v=rhs.v;return *this;}; //faster than not declaring it and leaving to the compiler
///////////////////////
// Splat
///////////////////////
// this is only for the complex version
template <class S, class V, IfComplex<S> =0, class ABtype>
inline void vsplat(Grid_simd<S,V> &ret,ABtype a, ABtype b){
ret.v = binary<V>(a, b, VsplatSIMD());
}
template <class S, class V>
inline void vbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
S* typepun =(S*) &src;
vsplat(ret,typepun[lane]);
}
// overload if complex
template <class S,class V> inline void vsplat(Grid_simd<S,V> &ret, EnableIf<is_complex < S >, S> c) {
vsplat(ret,real(c),imag(c));
}
//if real fill with a, if complex fill with a in the real part (first function above)
template <class S,class V>
inline void vsplat(Grid_simd<S,V> &ret,NotEnableIf<is_complex< S>,S> a){
ret.v = unary<V>(a, VsplatSIMD());
}
//////////////////////////
=======
Grid_simd &operator=(const Grid_simd &&rhs) {
v = rhs.v;
return *this;
};
Grid_simd &operator=(const Grid_simd &rhs) {
v = rhs.v;
return *this;
}; // faster than not declaring it and leaving to the compiler
Grid_simd() = default;
Grid_simd(const Grid_simd &rhs) : v(rhs.v){}; // compiles in movaps
Grid_simd(const Grid_simd &&rhs) : v(rhs.v){};
@ -156,31 +250,77 @@ namespace Grid {
vsplat(*this, a);
};
Grid_simd(const Real a){
vsplat(*this,Scalar_type(a));
};
Grid_simd(const Real a) { vsplat(*this, Scalar_type(a)); };
>>>>>>> 8b9301a74cde2f28a321baf4bca854d950585b56
///////////////////////////////////////////////
// mac, mult, sub, add, adj
///////////////////////////////////////////////
// FIXME -- alias this to an inline MAC struct.
friend inline void mac (Grid_simd * __restrict__ y,const Grid_simd * __restrict__ a,const Grid_simd *__restrict__ x){ *y = (*a)*(*x)+(*y); };
friend inline void mac(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ a,
const Grid_simd *__restrict__ x) {
*y = (*a) * (*x) + (*y);
};
friend inline void mult(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Grid_simd *__restrict__ r) {
*y = (*l) * (*r);
}
friend inline void sub(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Grid_simd *__restrict__ r) {
*y = (*l) - (*r);
}
friend inline void add(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Grid_simd *__restrict__ r) {
*y = (*l) + (*r);
}
friend inline void mult(Grid_simd * __restrict__ y,const Grid_simd * __restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) * (*r); }
friend inline void sub (Grid_simd * __restrict__ y,const Grid_simd * __restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) - (*r); }
friend inline void add (Grid_simd * __restrict__ y,const Grid_simd * __restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) + (*r); }
friend inline void mac(Grid_simd *__restrict__ y,
const Scalar_type *__restrict__ a,
const Grid_simd *__restrict__ x) {
*y = (*a) * (*x) + (*y);
};
friend inline void mult(Grid_simd *__restrict__ y,
const Scalar_type *__restrict__ l,
const Grid_simd *__restrict__ r) {
*y = (*l) * (*r);
}
friend inline void sub(Grid_simd *__restrict__ y,
const Scalar_type *__restrict__ l,
const Grid_simd *__restrict__ r) {
*y = (*l) - (*r);
}
friend inline void add(Grid_simd *__restrict__ y,
const Scalar_type *__restrict__ l,
const Grid_simd *__restrict__ r) {
*y = (*l) + (*r);
}
friend inline void mac (Grid_simd *__restrict__ y,const Scalar_type *__restrict__ a,const Grid_simd *__restrict__ x){ *y = (*a)*(*x)+(*y); };
friend inline void mult(Grid_simd *__restrict__ y,const Scalar_type *__restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) * (*r); }
friend inline void sub (Grid_simd *__restrict__ y,const Scalar_type *__restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) - (*r); }
friend inline void add (Grid_simd *__restrict__ y,const Scalar_type *__restrict__ l,const Grid_simd *__restrict__ r){ *y = (*l) + (*r); }
friend inline void mac (Grid_simd *__restrict__ y,const Grid_simd *__restrict__ a,const Scalar_type *__restrict__ x){ *y = (*a)*(*x)+(*y); };
friend inline void mult(Grid_simd *__restrict__ y,const Grid_simd *__restrict__ l,const Scalar_type *__restrict__ r){ *y = (*l) * (*r); }
friend inline void sub (Grid_simd *__restrict__ y,const Grid_simd *__restrict__ l,const Scalar_type *__restrict__ r){ *y = (*l) - (*r); }
friend inline void add (Grid_simd *__restrict__ y,const Grid_simd *__restrict__ l,const Scalar_type *__restrict__ r){ *y = (*l) + (*r); }
friend inline void mac(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ a,
const Scalar_type *__restrict__ x) {
*y = (*a) * (*x) + (*y);
};
friend inline void mult(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Scalar_type *__restrict__ r) {
*y = (*l) * (*r);
}
friend inline void sub(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Scalar_type *__restrict__ r) {
*y = (*l) - (*r);
}
friend inline void add(Grid_simd *__restrict__ y,
const Grid_simd *__restrict__ l,
const Scalar_type *__restrict__ r) {
*y = (*l) + (*r);
}
////////////////////////////////////////////////////////////////////////
// FIXME: gonna remove these load/store, get, set, prefetch
@ -199,16 +339,14 @@ namespace Grid {
///////////////////////
// Vprefetch
///////////////////////
friend inline void vprefetch(const Grid_simd &v)
{
friend inline void vprefetch(const Grid_simd &v) {
prefetch_HINT_T0((const char *)&v.v);
}
///////////////////////
// Reduce
///////////////////////
friend inline Scalar_type Reduce(const Grid_simd & in)
{
friend inline Scalar_type Reduce(const Grid_simd &in) {
return unary<Scalar_type>(in.v, ReduceSIMD<Scalar_type, Vector_type>());
}
@ -255,7 +393,8 @@ namespace Grid {
// provides support
///////////////////////////////////////
template<class functor> friend inline Grid_simd SimdApply (const functor &func,const Grid_simd &v) {
template <class functor>
friend inline Grid_simd SimdApply(const functor &func, const Grid_simd &v) {
Grid_simd ret;
Grid_simd::conv_t conv;
@ -266,7 +405,10 @@ namespace Grid {
ret.v = conv.v;
return ret;
}
template<class functor> friend inline Grid_simd SimdApplyBinop (const functor &func,const Grid_simd &x,const Grid_simd &y) {
template <class functor>
friend inline Grid_simd SimdApplyBinop(const functor &func,
const Grid_simd &x,
const Grid_simd &y) {
Grid_simd ret;
Grid_simd::conv_t cx;
Grid_simd::conv_t cy;
@ -297,20 +439,27 @@ namespace Grid {
friend inline void permute3(Grid_simd &y, Grid_simd b) {
y.v = Optimization::Permute::Permute3(b.v);
}
friend inline void permute(Grid_simd &y,Grid_simd b,int perm)
{
friend inline void permute(Grid_simd &y, Grid_simd b, int perm) {
if (perm & RotateBit) {
int dist = perm & 0xF;
y = rotate(b, dist);
return;
} else {
switch(perm){
case 3: permute3(y,b); break;
case 2: permute2(y,b); break;
case 1: permute1(y,b); break;
case 0: permute0(y,b); break;
default: assert(0);
}
switch (perm) {
case 3:
permute3(y, b);
break;
case 2:
permute2(y, b);
break;
case 1:
permute1(y, b);
break;
case 0:
permute0(y, b);
break;
default:
assert(0);
}
}
@ -320,8 +469,7 @@ namespace Grid {
// General rotate
////////////////////////////////////////////////////////////////////
template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S,V> rotate(Grid_simd<S,V> b,int nrot)
{
inline Grid_simd<S, V> rotate(Grid_simd<S, V> b, int nrot) {
nrot = nrot % Grid_simd<S, V>::Nsimd();
Grid_simd<S, V> ret;
// std::cout << "Rotate Real by "<<nrot<<std::endl;
@ -329,28 +477,13 @@ namespace Grid {
return ret;
}
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S,V> rotate(Grid_simd<S,V> b,int nrot)
{
inline Grid_simd<S, V> rotate(Grid_simd<S, V> b, int nrot) {
nrot = nrot % Grid_simd<S, V>::Nsimd();
Grid_simd<S, V> ret;
// std::cout << "Rotate Complex by "<<nrot<<std::endl;
ret.v = Optimization::Rotate::rotate(b.v, 2 * nrot);
return ret;
}
template <class S, class V, IfNotComplex<S> =0>
inline void rotate( Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
{
nrot = nrot % Grid_simd<S,V>::Nsimd();
// std::cout << "Rotate Real by "<<nrot<<std::endl;
ret.v = Optimization::Rotate::rotate(b.v,nrot);
}
template <class S, class V, IfComplex<S> =0>
inline void rotate(Grid_simd<S,V> &ret,Grid_simd<S,V> b,int nrot)
{
nrot = nrot % Grid_simd<S,V>::Nsimd();
// std::cout << "Rotate Complex by "<<nrot<<std::endl;
ret.v = Optimization::Rotate::rotate(b.v,2*nrot);
}
///////////////////////
// Splat
@ -362,18 +495,14 @@ namespace Grid {
ret.v = binary<V>(a, b, VsplatSIMD());
}
template <class S, class V>
inline void vbroadcast(Grid_simd<S,V> &ret,const Grid_simd<S,V> &src,int lane){
S* typepun =(S*) &src;
vsplat(ret,typepun[lane]);
}
// overload if complex
template <class S,class V> inline void vsplat(Grid_simd<S,V> &ret, EnableIf<is_complex < S >, S> c) {
template <class S, class V>
inline void vsplat(Grid_simd<S, V> &ret, EnableIf<is_complex<S>, S> c) {
vsplat(ret, real(c), imag(c));
}
//if real fill with a, if complex fill with a in the real part (first function above)
// if real fill with a, if complex fill with a in the real part (first function
// above)
template <class S, class V>
inline void vsplat(Grid_simd<S, V> &ret, NotEnableIf<is_complex<S>, S> a) {
ret.v = unary<V>(a, VsplatSIMD());
@ -384,24 +513,60 @@ namespace Grid {
// Initialise to 1,0,i for the correct types
///////////////////////////////////////////////
// For complex types
template <class S,class V, IfComplex<S> = 0 > inline void vone(Grid_simd<S,V> &ret) { vsplat(ret,S(1.0,0.0)); }
template <class S,class V, IfComplex<S> = 0 > inline void vzero(Grid_simd<S,V> &ret) { vsplat(ret,S(0.0,0.0)); }// use xor?
template <class S,class V, IfComplex<S> = 0 > inline void vcomplex_i(Grid_simd<S,V> &ret){ vsplat(ret,S(0.0,1.0));}
template <class S, class V, IfComplex<S> = 0>
inline void vone(Grid_simd<S, V> &ret) {
vsplat(ret, S(1.0, 0.0));
}
template <class S, class V, IfComplex<S> = 0>
inline void vzero(Grid_simd<S, V> &ret) {
vsplat(ret, S(0.0, 0.0));
} // use xor?
template <class S, class V, IfComplex<S> = 0>
inline void vcomplex_i(Grid_simd<S, V> &ret) {
vsplat(ret, S(0.0, 1.0));
}
template <class S,class V, IfComplex<S> = 0 > inline void visign(Grid_simd<S,V> &ret){ vsplat(ret,S(1.0,-1.0));}
template <class S,class V, IfComplex<S> = 0 > inline void vrsign(Grid_simd<S,V> &ret){ vsplat(ret,S(-1.0,1.0));}
template <class S, class V, IfComplex<S> = 0>
inline void visign(Grid_simd<S, V> &ret) {
vsplat(ret, S(1.0, -1.0));
}
template <class S, class V, IfComplex<S> = 0>
inline void vrsign(Grid_simd<S, V> &ret) {
vsplat(ret, S(-1.0, 1.0));
}
// if not complex overload here
template <class S,class V, IfReal<S> = 0 > inline void vone (Grid_simd<S,V> &ret){ vsplat(ret,S(1.0)); }
template <class S,class V, IfReal<S> = 0 > inline void vzero(Grid_simd<S,V> &ret){ vsplat(ret,S(0.0)); }
template <class S, class V, IfReal<S> = 0>
inline void vone(Grid_simd<S, V> &ret) {
vsplat(ret, S(1.0));
}
template <class S, class V, IfReal<S> = 0>
inline void vzero(Grid_simd<S, V> &ret) {
vsplat(ret, S(0.0));
}
// For integral types
template <class S,class V,IfInteger<S> = 0 > inline void vone(Grid_simd<S,V> &ret) {vsplat(ret,1); }
template <class S,class V,IfInteger<S> = 0 > inline void vzero(Grid_simd<S,V> &ret) {vsplat(ret,0); }
template <class S,class V,IfInteger<S> = 0 > inline void vtrue (Grid_simd<S,V> &ret){vsplat(ret,0xFFFFFFFF);}
template <class S,class V,IfInteger<S> = 0 > inline void vfalse(Grid_simd<S,V> &ret){vsplat(ret,0);}
template <class S, class V, IfInteger<S> = 0>
inline void vone(Grid_simd<S, V> &ret) {
vsplat(ret, 1);
}
template <class S, class V, IfInteger<S> = 0>
inline void vzero(Grid_simd<S, V> &ret) {
vsplat(ret, 0);
}
template <class S, class V, IfInteger<S> = 0>
inline void vtrue(Grid_simd<S, V> &ret) {
vsplat(ret, 0xFFFFFFFF);
}
template <class S, class V, IfInteger<S> = 0>
inline void vfalse(Grid_simd<S, V> &ret) {
vsplat(ret, 0);
}
template<class S,class V> inline void zeroit(Grid_simd<S,V> &z){ vzero(z);}
template <class S, class V>
inline void zeroit(Grid_simd<S, V> &z) {
vzero(z);
}
///////////////////////
// Vstream
@ -424,33 +589,36 @@ namespace Grid {
////////////////////////////////////
// Arithmetic operator overloads +,-,*
////////////////////////////////////
template<class S,class V> inline Grid_simd<S,V> operator + (Grid_simd<S,V> a, Grid_simd<S,V> b) {
template <class S, class V>
inline Grid_simd<S, V> operator+(Grid_simd<S, V> a, Grid_simd<S, V> b) {
Grid_simd<S, V> ret;
ret.v = binary<V>(a.v, b.v, SumSIMD());
return ret;
};
template<class S,class V> inline Grid_simd<S,V> operator - (Grid_simd<S,V> a, Grid_simd<S,V> b) {
template <class S, class V>
inline Grid_simd<S, V> operator-(Grid_simd<S, V> a, Grid_simd<S, V> b) {
Grid_simd<S, V> ret;
ret.v = binary<V>(a.v, b.v, SubSIMD());
return ret;
};
// Distinguish between complex types and others
template<class S,class V, IfComplex<S> = 0 > inline Grid_simd<S,V> operator * (Grid_simd<S,V> a, Grid_simd<S,V> b) {
template <class S, class V, IfComplex<S> = 0>
inline Grid_simd<S, V> operator*(Grid_simd<S, V> a, Grid_simd<S, V> b) {
Grid_simd<S, V> ret;
ret.v = binary<V>(a.v, b.v, MultComplexSIMD());
return ret;
};
// Real/Integer types
template<class S,class V, IfNotComplex<S> = 0 > inline Grid_simd<S,V> operator * (Grid_simd<S,V> a, Grid_simd<S,V> b) {
template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S, V> operator*(Grid_simd<S, V> a, Grid_simd<S, V> b) {
Grid_simd<S, V> ret;
ret.v = binary<V>(a.v, b.v, MultSIMD());
return ret;
};
///////////////////////
// Conjugate
///////////////////////
@ -460,13 +628,16 @@ namespace Grid {
ret.v = unary<V>(in.v, ConjSIMD());
return ret;
}
template <class S,class V, IfNotComplex<S> = 0 > inline Grid_simd<S,V> conjugate(const Grid_simd<S,V> &in){
template <class S, class V, IfNotComplex<S> = 0>
inline Grid_simd<S, V> conjugate(const Grid_simd<S, V> &in) {
return in; // for real objects
}
// Suppress adj for integer types... // odd; why conjugate above but not adj??
template <class S, class V, IfNotInteger<S> = 0>
inline Grid_simd<S,V> adj(const Grid_simd<S,V> &in){ return conjugate(in); }
inline Grid_simd<S, V> adj(const Grid_simd<S, V> &in) {
return conjugate(in);
}
///////////////////////
// timesMinusI
@ -513,14 +684,14 @@ namespace Grid {
/////////////////////
template <class S, class V>
inline Grid_simd< S, V> innerProduct(const Grid_simd< S, V> & l, const Grid_simd< S, V> & r)
{
inline Grid_simd<S, V> innerProduct(const Grid_simd<S, V> &l,
const Grid_simd<S, V> &r) {
return conjugate(l) * r;
}
template <class S, class V>
inline Grid_simd< S, V> outerProduct(const Grid_simd< S, V> &l, const Grid_simd< S, V> & r)
{
inline Grid_simd<S, V> outerProduct(const Grid_simd<S, V> &l,
const Grid_simd<S, V> &r) {
return l * conjugate(r);
}
@ -529,7 +700,6 @@ namespace Grid {
return arg;
}
////////////////////////////////////////////////////////////
// copy/splat complex real parts into real;
// insert real into complex and zero imag;
@ -537,12 +707,11 @@ namespace Grid {
// real = toReal( complex )
template <class S, class V, IfReal<S> = 0>
inline Grid_simd<S,V> toReal(const Grid_simd<std::complex<S>,V> &in)
{
inline Grid_simd<S, V> toReal(const Grid_simd<std::complex<S>, V> &in) {
typedef Grid_simd<S, V> simd;
simd ret;
typename simd::conv_t conv;
conv.v = in.v;
conv.v = in.v; // copy the vector content (bytewise)
for (int i = 0; i < simd::Nsimd(); i += 2) {
conv.s[i + 1] = conv.s[i]; // duplicate (r,r);(r,r);(r,r); etc...
}
@ -552,16 +721,17 @@ namespace Grid {
// complex = toComplex( real )
template <class R, class V, IfReal<R> = 0> // must be a real arg
inline Grid_simd<std::complex<R>,V> toComplex (const Grid_simd<R,V> &in)
{
inline Grid_simd<std::complex<R>, V> toComplex(const Grid_simd<R, V> &in) {
typedef Grid_simd<R, V> Rsimd;
typedef Grid_simd<std::complex<R>, V> Csimd;
typename Rsimd::conv_t conv; // address as real
conv.v = in.v;
for (int i = 0; i < Rsimd::Nsimd(); i += 2) {
assert(conv.s[i+1]==conv.s[i]); // trap any cases where real was not duplicated
// indicating the SIMD grids of real and imag assignment did not correctly match
assert(conv.s[i + 1] ==
conv.s[i]); // trap any cases where real was not duplicated
// indicating the SIMD grids of real and imag assignment did not correctly
// match
conv.s[i + 1] = 0.0; // zero imaginary parts
}
Csimd ret;
@ -581,16 +751,23 @@ namespace Grid {
/////////////////////////////////////////
// Some traits to recognise the types
/////////////////////////////////////////
template <typename T> struct is_simd : public std::false_type{};
template <> struct is_simd<vRealF> : public std::true_type {};
template <> struct is_simd<vRealD> : public std::true_type {};
template <> struct is_simd<vComplexF>: public std::true_type {};
template <> struct is_simd<vComplexD>: public std::true_type {};
template <> struct is_simd<vInteger> : public std::true_type {};
template <typename T> using IfSimd = Invoke<std::enable_if< is_simd<T>::value,int> > ;
template <typename T> using IfNotSimd = Invoke<std::enable_if<!is_simd<T>::value,unsigned> > ;
template <typename T>
struct is_simd : public std::false_type {};
template <>
struct is_simd<vRealF> : public std::true_type {};
template <>
struct is_simd<vRealD> : public std::true_type {};
template <>
struct is_simd<vComplexF> : public std::true_type {};
template <>
struct is_simd<vComplexD> : public std::true_type {};
template <>
struct is_simd<vInteger> : public std::true_type {};
template <typename T>
using IfSimd = Invoke<std::enable_if<is_simd<T>::value, int> >;
template <typename T>
using IfNotSimd = Invoke<std::enable_if<!is_simd<T>::value, unsigned> >;
}
#endif

157
lib/simd/Grid_vector_unops.h
View File

@ -25,7 +25,8 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_VECTOR_UNOPS
@ -35,85 +36,84 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
namespace Grid {
template<class scalar> struct SqrtRealFunctor {
scalar operator()(const scalar &a) const {
return sqrt(real(a));
}
template <class scalar>
struct SqrtRealFunctor {
scalar operator()(const scalar &a) const { return sqrt(real(a)); }
};
template<class scalar> struct RSqrtRealFunctor {
template <class scalar>
struct RSqrtRealFunctor {
scalar operator()(const scalar &a) const {
return scalar(1.0 / sqrt(real(a)));
}
};
template<class scalar> struct CosRealFunctor {
scalar operator()(const scalar &a) const {
return cos(real(a));
}
template <class scalar>
struct CosRealFunctor {
scalar operator()(const scalar &a) const { return cos(real(a)); }
};
template<class scalar> struct SinRealFunctor {
scalar operator()(const scalar &a) const {
return sin(real(a));
}
template <class scalar>
struct SinRealFunctor {
scalar operator()(const scalar &a) const { return sin(real(a)); }
};
template<class scalar> struct LogRealFunctor {
scalar operator()(const scalar &a) const {
return log(real(a));
}
template <class scalar>
struct AcosRealFunctor {
scalar operator()(const scalar &a) const { return acos(real(a)); }
};
template<class scalar> struct ExpRealFunctor {
scalar operator()(const scalar &a) const {
return exp(real(a));
}
};
template<class scalar> struct NotFunctor {
scalar operator()(const scalar &a) const {
return (!a);
}
};
template<class scalar> struct AbsRealFunctor {
scalar operator()(const scalar &a) const {
return std::abs(real(a));
}
template <class scalar>
struct AsinRealFunctor {
scalar operator()(const scalar &a) const { return asin(real(a)); }
};
template<class scalar> struct PowRealFunctor {
template <class scalar>
struct LogRealFunctor {
scalar operator()(const scalar &a) const { return log(real(a)); }
};
template <class scalar>
struct ExpRealFunctor {
scalar operator()(const scalar &a) const { return exp(real(a)); }
};
template <class scalar>
struct NotFunctor {
scalar operator()(const scalar &a) const { return (!a); }
};
template <class scalar>
struct AbsRealFunctor {
scalar operator()(const scalar &a) const { return std::abs(real(a)); }
};
template <class scalar>
struct PowRealFunctor {
double y;
PowRealFunctor(double _y) : y(_y){};
scalar operator()(const scalar &a) const {
return pow(real(a),y);
}
scalar operator()(const scalar &a) const { return pow(real(a), y); }
};
template<class scalar> struct ModIntFunctor {
template <class scalar>
struct ModIntFunctor {
Integer y;
ModIntFunctor(Integer _y) : y(_y){};
scalar operator()(const scalar &a) const {
return Integer(a)%y;
}
scalar operator()(const scalar &a) const { return Integer(a) % y; }
};
template<class scalar> struct DivIntFunctor {
template <class scalar>
struct DivIntFunctor {
Integer y;
DivIntFunctor(Integer _y) : y(_y){};
scalar operator()(const scalar &a) const {
return Integer(a)/y;
}
scalar operator()(const scalar &a) const { return Integer(a) / y; }
};
template<class scalar> struct RealFunctor {
scalar operator()(const scalar &a) const {
return real(a);
}
template <class scalar>
struct RealFunctor {
scalar operator()(const scalar &a) const { return std::real(a); }
};
template<class scalar> struct ImagFunctor {
scalar operator()(const scalar &a) const {
return imag(a);
}
template <class scalar>
struct ImagFunctor {
scalar operator()(const scalar &a) const { return std::imag(a); }
};
template <class S, class V>
inline Grid_simd<S, V> real(const Grid_simd<S, V> &r) {
@ -123,7 +123,6 @@ namespace Grid {
inline Grid_simd<S, V> imag(const Grid_simd<S, V> &r) {
return SimdApply(ImagFunctor<S>(), r);
}
template <class S, class V>
inline Grid_simd<S, V> sqrt(const Grid_simd<S, V> &r) {
return SimdApply(SqrtRealFunctor<S>(), r);
@ -146,6 +145,14 @@ namespace Grid {
return SimdApply(SinRealFunctor<S>(), r);
}
template <class S, class V>
inline Grid_simd<S, V> acos(const Grid_simd<S, V> &r) {
return SimdApply(AcosRealFunctor<S>(), r);
}
template <class S, class V>
inline Grid_simd<S, V> asin(const Grid_simd<S, V> &r) {
return SimdApply(AsinRealFunctor<S>(), r);
}
template <class S, class V>
inline Grid_simd<S, V> log(const Grid_simd<S, V> &r) {
return SimdApply(LogRealFunctor<S>(), r);
}
@ -177,51 +184,51 @@ namespace Grid {
// Allows us to assign into **conformable** real vectors from complex
////////////////////////////////////////////////////////////////////////////
// template < class S, class V >
// inline auto ComplexRemove(const Grid_simd<S,V> &c) -> Grid_simd<Grid_simd<S,V>::Real,V> {
// inline auto ComplexRemove(const Grid_simd<S,V> &c) ->
// Grid_simd<Grid_simd<S,V>::Real,V> {
// Grid_simd<Grid_simd<S,V>::Real,V> ret;
// ret.v = c.v;
// return ret;
// }
template<class scalar> struct AndFunctor {
scalar operator()(const scalar &x, const scalar &y) const {
return x & y;
}
template <class scalar>
struct AndFunctor {
scalar operator()(const scalar &x, const scalar &y) const { return x & y; }
};
template<class scalar> struct OrFunctor {
scalar operator()(const scalar &x, const scalar &y) const {
return x | y;
}
template <class scalar>
struct OrFunctor {
scalar operator()(const scalar &x, const scalar &y) const { return x | y; }
};
template<class scalar> struct AndAndFunctor {
scalar operator()(const scalar &x, const scalar &y) const {
return x && y;
}
template <class scalar>
struct AndAndFunctor {
scalar operator()(const scalar &x, const scalar &y) const { return x && y; }
};
template<class scalar> struct OrOrFunctor {
scalar operator()(const scalar &x, const scalar &y) const {
return x || y;
}
template <class scalar>
struct OrOrFunctor {
scalar operator()(const scalar &x, const scalar &y) const { return x || y; }
};
////////////////////////////////
// Calls to simd binop functors
////////////////////////////////
template <class S, class V>
inline Grid_simd<S,V> operator &(const Grid_simd<S,V> &x,const Grid_simd<S,V> &y) {
inline Grid_simd<S, V> operator&(const Grid_simd<S, V> &x,
const Grid_simd<S, V> &y) {
return SimdApplyBinop(AndFunctor<S>(), x, y);
}
template <class S, class V>
inline Grid_simd<S,V> operator &&(const Grid_simd<S,V> &x,const Grid_simd<S,V> &y) {
inline Grid_simd<S, V> operator&&(const Grid_simd<S, V> &x,
const Grid_simd<S, V> &y) {
return SimdApplyBinop(AndAndFunctor<S>(), x, y);
}
template <class S, class V>
inline Grid_simd<S,V> operator |(const Grid_simd<S,V> &x,const Grid_simd<S,V> &y) {
inline Grid_simd<S, V> operator|(const Grid_simd<S, V> &x,
const Grid_simd<S, V> &y) {
return SimdApplyBinop(OrFunctor<S>(), x, y);
}
template <class S, class V>
inline Grid_simd<S,V> operator ||(const Grid_simd<S,V> &x,const Grid_simd<S,V> &y) {
inline Grid_simd<S, V> operator||(const Grid_simd<S, V> &x,
const Grid_simd<S, V> &y) {
return SimdApplyBinop(OrOrFunctor<S>(), x, y);
}
}
#endif

192
lib/tensors/Tensor_class.h
View File

@ -23,7 +23,8 @@ Author: Peter Boyle <paboyle@ph.ed.ac.uk>
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
See the full license in the file "LICENSE" in the top level distribution
directory
*************************************************************************************/
/* END LEGAL */
#ifndef GRID_MATH_TENSORS_H
@ -38,7 +39,8 @@ namespace Grid {
// It is useful to NOT have any constructors
// so that these classes assert "is_pod<class> == true"
// because then the standard C++ valarray container eliminates fill overhead on new allocation and
// because then the standard C++ valarray container eliminates fill overhead on
// new allocation and
// non-move copying.
//
// However note that doing this eliminates some syntactical sugar such as
@ -46,8 +48,8 @@ namespace Grid {
//
class GridTensorBase {};
template<class vtype> class iScalar
{
template <class vtype>
class iScalar {
public:
vtype _internal;
@ -66,7 +68,8 @@ public:
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1 };
// Scalar no action
// template<int Level> using tensor_reduce_level = typename iScalar<GridTypeMapper<vtype>::tensor_reduce_level<Level> >;
// template<int Level> using tensor_reduce_level = typename
// iScalar<GridTypeMapper<vtype>::tensor_reduce_level<Level> >;
iScalar() = default;
/*
iScalar(const iScalar<vtype> &copyme)=default;
@ -74,14 +77,16 @@ public:
iScalar<vtype> & operator= (const iScalar<vtype> &copyme) = default;
iScalar<vtype> & operator= (iScalar<vtype> &&copyme) = default;
*/
iScalar(scalar_type s) : _internal(s) {};// recurse down and hit the constructor for vector_type
iScalar(scalar_type s)
: _internal(s){}; // recurse down and hit the constructor for vector_type
iScalar(const Zero &z) { *this = zero; };
iScalar<vtype> &operator=(const Zero &hero) {
zeroit(*this);
return *this;
}
friend strong_inline void vstream(iScalar<vtype> &out,const iScalar<vtype> &in){
friend strong_inline void vstream(iScalar<vtype> &out,
const iScalar<vtype> &in) {
vstream(out._internal, in._internal);
}
friend strong_inline void vbroadcast(iScalar<vtype> &out,const iScalar<vtype> &in,int lane){
@ -93,7 +98,8 @@ public:
friend strong_inline void prefetch(iScalar<vtype> &that) {
prefetch(that._internal);
}
friend strong_inline void permute(iScalar<vtype> &out,const iScalar<vtype> &in,int permutetype){
friend strong_inline void permute(iScalar<vtype> &out,
const iScalar<vtype> &in, int permutetype) {
permute(out._internal, in._internal, permutetype);
}
friend strong_inline void rotate(iScalar<vtype> &out,const iScalar<vtype> &in,int rot){
@ -119,28 +125,43 @@ public:
*this = (*this) + r;
return *this;
}
strong_inline vtype & operator ()(void) {
return _internal;
}
strong_inline const vtype & operator ()(void) const {
return _internal;
}
strong_inline vtype &operator()(void) { return _internal; }
strong_inline const vtype &operator()(void) const { return _internal; }
// Type casts meta programmed, must be pure scalar to match TensorRemove
template<class U=vtype,class V=scalar_type,IfComplex<V> = 0,IfNotSimd<U> = 0> operator ComplexF () const { return(TensorRemove(_internal)); };
template<class U=vtype,class V=scalar_type,IfComplex<V> = 0,IfNotSimd<U> = 0> operator ComplexD () const { return(TensorRemove(_internal)); };
// template<class U=vtype,class V=scalar_type,IfComplex<V> = 0,IfNotSimd<U> = 0> operator RealD () const { return(real(TensorRemove(_internal))); }
template<class U=vtype,class V=scalar_type,IfReal<V> = 0,IfNotSimd<U> = 0> operator RealD () const { return TensorRemove(_internal); }
template<class U=vtype,class V=scalar_type,IfInteger<V> = 0,IfNotSimd<U> = 0> operator Integer () const { return Integer(TensorRemove(_internal)); }
template <class U = vtype, class V = scalar_type, IfComplex<V> = 0,
IfNotSimd<U> = 0>
operator ComplexF() const {
return (TensorRemove(_internal));
};
template <class U = vtype, class V = scalar_type, IfComplex<V> = 0,
IfNotSimd<U> = 0>
operator ComplexD() const {
return (TensorRemove(_internal));
};
// template<class U=vtype,class V=scalar_type,IfComplex<V> = 0,IfNotSimd<U> =
// 0> operator RealD () const { return(real(TensorRemove(_internal))); }
template <class U = vtype, class V = scalar_type, IfReal<V> = 0,
IfNotSimd<U> = 0>
operator RealD() const {
return TensorRemove(_internal);
}
template <class U = vtype, class V = scalar_type, IfInteger<V> = 0,
IfNotSimd<U> = 0>
operator Integer() const {
return Integer(TensorRemove(_internal));
}
// convert from a something to a scalar via constructor of something arg
template<class T,typename std::enable_if<!isGridTensor<T>::value, T>::type* = nullptr > strong_inline iScalar<vtype> operator = (T arg)
{
template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type
* = nullptr>
strong_inline iScalar<vtype> operator=(T arg) {
_internal = arg;
return *this;
}
friend std::ostream& operator<< (std::ostream& stream, const iScalar<vtype> &o){
friend std::ostream &operator<<(std::ostream &stream,
const iScalar<vtype> &o) {
stream << "S {" << o._internal << "}";
return stream;
};
@ -148,14 +169,19 @@ public:
///////////////////////////////////////////////////////////
// Allows to turn scalar<scalar<scalar<double>>>> back to double.
///////////////////////////////////////////////////////////
template<class T> strong_inline typename std::enable_if<!isGridTensor<T>::value, T>::type TensorRemove(T arg) { return arg;}
template<class vtype> strong_inline auto TensorRemove(iScalar<vtype> arg) -> decltype(TensorRemove(arg._internal))
{
template <class T>
strong_inline typename std::enable_if<!isGridTensor<T>::value, T>::type
TensorRemove(T arg) {
return arg;
}
template <class vtype>
strong_inline auto TensorRemove(iScalar<vtype> arg)
-> decltype(TensorRemove(arg._internal)) {
return TensorRemove(arg._internal);
}
template<class vtype,int N> class iVector
{
template <class vtype, int N>
class iVector {
public:
vtype _internal[N];
@ -171,11 +197,11 @@ public:
typedef iVector<typename GridTypeMapper<vtype>::Complexified, N> Complexified;
typedef iVector<typename GridTypeMapper<vtype>::Realified, N> Realified;
template<class T,typename std::enable_if<!isGridTensor<T>::value, T>::type* = nullptr > strong_inline auto operator = (T arg) -> iVector<vtype,N>
{
template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type
* = nullptr>
strong_inline auto operator=(T arg) -> iVector<vtype, N> {
zeroit(*this);
for(int i=0;i<N;i++)
_internal[i] = arg;
for (int i = 0; i < N; i++) _internal[i] = arg;
return *this;
}
@ -201,11 +227,13 @@ public:
friend strong_inline void prefetch(iVector<vtype, N> &that) {
for (int i = 0; i < N; i++) prefetch(that._internal[i]);
}
friend strong_inline void vstream(iVector<vtype,N> &out,const iVector<vtype,N> &in){
friend strong_inline void vstream(iVector<vtype, N> &out,
const iVector<vtype, N> &in) {
for (int i = 0; i < N; i++) {
vstream(out._internal[i], in._internal[i]);
}
}
<<<<<<< HEAD
friend strong_inline void vbroadcast(iVector<vtype,N> &out,const iVector<vtype,N> &in,int lane){
for(int i=0;i<N;i++){
vbroadcast(out._internal[i],in._internal[i],lane);
@ -214,6 +242,13 @@ public:
friend strong_inline void permute(iVector<vtype,N> &out,const iVector<vtype,N> &in,int permutetype){
for(int i=0;i<N;i++){
permute(out._internal[i],in._internal[i],permutetype);
=======
friend strong_inline void permute(iVector<vtype, N> &out,
const iVector<vtype, N> &in,
int permutetype) {
for (int i = 0; i < N; i++) {
permute(out._internal[i], in._internal[i], permutetype);
>>>>>>> 8b9301a74cde2f28a321baf4bca854d950585b56
}
}
friend strong_inline void rotate(iVector<vtype,N> &out,const iVector<vtype,N> &in,int rot){
@ -241,13 +276,10 @@ public:
*this = (*this) + r;
return *this;
}
strong_inline vtype & operator ()(int i) {
return _internal[i];
}
strong_inline const vtype & operator ()(int i) const {
return _internal[i];
}
friend std::ostream& operator<< (std::ostream& stream, const iVector<vtype,N> &o){
strong_inline vtype &operator()(int i) { return _internal[i]; }
strong_inline const vtype &operator()(int i) const { return _internal[i]; }
friend std::ostream &operator<<(std::ostream &stream,
const iVector<vtype, N> &o) {
stream << "V<" << N << ">{";
for (int i = 0; i < N; i++) {
stream << o._internal[i];
@ -261,8 +293,8 @@ public:
// }
};
template<class vtype,int N> class iMatrix
{
template <class vtype, int N>
class iMatrix {
public:
vtype _internal[N][N];
@ -282,20 +314,18 @@ public:
enum { TensorLevel = GridTypeMapper<vtype>::TensorLevel + 1 };
iMatrix(const Zero &z) { *this = zero; };
iMatrix() = default;
iMatrix &operator=(const iMatrix &rhs) {
for (int i = 0; i < N; i++)
for(int j=0;j<N;j++)
vstream(_internal[i][j],rhs._internal[i][j]);
for (int j = 0; j < N; j++) vstream(_internal[i][j], rhs._internal[i][j]);
return *this;
};
iMatrix(scalar_type s) { (*this) = s ;};// recurse down and hit the constructor for vector_type
iMatrix(scalar_type s) {
(*this) = s;
}; // recurse down and hit the constructor for vector_type
/*
iMatrix(const iMatrix<vtype,N> &copyme)=default;
@ -304,20 +334,19 @@ public:
iMatrix<vtype,N> & operator= (iMatrix<vtype,N> &&copyme) = default;
*/
iMatrix<vtype, N> &operator=(const Zero &hero) {
zeroit(*this);
return *this;
}
template<class T,typename std::enable_if<!isGridTensor<T>::value, T>::type* = nullptr > strong_inline auto operator = (T arg) -> iMatrix<vtype,N>
{
template <class T, typename std::enable_if<!isGridTensor<T>::value, T>::type
* = nullptr>
strong_inline auto operator=(T arg) -> iMatrix<vtype, N> {
zeroit(*this);
for(int i=0;i<N;i++)
_internal[i][i] = arg;
for (int i = 0; i < N; i++) _internal[i][i] = arg;
return *this;
}
<<<<<<< HEAD
friend strong_inline void zeroit(iMatrix<vtype,N> &that){
for(int i=0;i<N;i++){
for(int j=0;j<N;j++){
@ -347,6 +376,26 @@ public:
for(int j=0;j<N;j++){
permute(out._internal[i][j],in._internal[i][j],permutetype);
}}
=======
friend strong_inline void zeroit(iMatrix<vtype, N> &that) {
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
zeroit(that._internal[i][j]);
}
}
}
friend strong_inline void prefetch(iMatrix<vtype, N> &that) {
for (int i = 0; i < N; i++)
for (int j = 0; j < N; j++) prefetch(that._internal[i][j]);
}
friend strong_inline void vstream(iMatrix<vtype, N> &out,
const iMatrix<vtype, N> &in) {
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
vstream(out._internal[i][j], in._internal[i][j]);
}
}
>>>>>>> 8b9301a74cde2f28a321baf4bca854d950585b56
}
friend strong_inline void rotate(iMatrix<vtype,N> &out,const iMatrix<vtype,N> &in,int rot){
for(int i=0;i<N;i++){
@ -355,6 +404,15 @@ public:
}}
}
friend strong_inline void permute(iMatrix<vtype, N> &out,
const iMatrix<vtype, N> &in,
int permutetype) {
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
permute(out._internal[i][j], in._internal[i][j], permutetype);
}
}
}
// Unary negation
friend strong_inline iMatrix<vtype, N> operator-(const iMatrix<vtype, N> &r) {
@ -362,7 +420,8 @@ public:
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
ret._internal[i][j] = -r._internal[i][j];
}}
}
}
return ret;
}
// *=,+=,-= operators inherit from corresponding "*,-,+" behaviour
@ -383,13 +442,12 @@ public:
}
// returns an lvalue reference
strong_inline vtype & operator ()(int i,int j) {
return _internal[i][j];
}
strong_inline vtype &operator()(int i, int j) { return _internal[i][j]; }
strong_inline const vtype &operator()(int i, int j) const {
return _internal[i][j];
}
friend std::ostream& operator<< (std::ostream& stream, const iMatrix<vtype,N> &o){
friend std::ostream &operator<<(std::ostream &stream,
const iMatrix<vtype, N> &o) {
stream << "M<" << N << ">{";
for (int i = 0; i < N; i++) {
stream << "{";
@ -407,27 +465,25 @@ public:
// strong_inline vtype && operator ()(int i,int j) {
// return _internal[i][j];
// }
};
template<class v> void vprefetch(const iScalar<v> &vv)
{
template <class v>
void vprefetch(const iScalar<v> &vv) {
vprefetch(vv._internal);
}
template<class v,int N> void vprefetch(const iVector<v,N> &vv)
{
template <class v, int N>
void vprefetch(const iVector<v, N> &vv) {
for (int i = 0; i < N; i++) {
vprefetch(vv._internal[i]);
}
}
template<class v,int N> void vprefetch(const iMatrix<v,N> &vv)
{
template <class v, int N>
void vprefetch(const iMatrix<v, N> &vv) {
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
vprefetch(vv._internal[i][j]);
}}
}
}
}
}
#endif

28
lib/tensors/Tensor_extract_merge.h
View File

@ -10,6 +10,7 @@ Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: neo <cossu@post.kek.jp>
Author: paboyle <paboyle@ph.ed.ac.uk>
Author: Christopher Kelly <ckelly@phys.columbia.edu>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@ -167,6 +168,33 @@ void extract(const vobj &vec,std::vector<typename vobj::scalar_object *> &extrac
}
}
////////////////////////////////////////////////////////////////////////
// Extract to a bunch of scalar object pointers of different scalar type, with offset. Useful for precision change
////////////////////////////////////////////////////////////////////////
template<class vobj, class sobj> inline
void extract1(const vobj &vec,std::vector<sobj*> &extracted, int offset)
{
typedef typename vobj::scalar_type vobj_scalar_type ;
typedef typename vobj::vector_type vobj_vector_type ;
typedef typename sobj::scalar_type sobj_scalar_type ;
static const int words=sizeof(vobj)/sizeof(vobj_vector_type);
static const int Nsimd=vobj_vector_type::Nsimd();
int Nextr=extracted.size();
int s = Nsimd/Nextr;
vobj_scalar_type * vp = (vobj_scalar_type *)&vec;
for(int w=0;w<words;w++){
for(int i=0;i<Nextr;i++){
sobj_scalar_type * pointer = (sobj_scalar_type *)& extracted[i][offset];
pointer[w] = vp[i*s+w*Nsimd];
}
}
}
////////////////////////////////////////////////////////////////////////
// Merge a contiguous array of scalar objects
////////////////////////////////////////////////////////////////////////

8
lib/tensors/Tensor_index.h
View File

@ -146,14 +146,14 @@ class TensorIndexRecursion {
}
template<class vtype,int N> inline static
void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal,0)),N> &arg, int i)
void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0],0)),N> &arg, int i)
{
for(int ii=0;ii<N;ii++){
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal[ii],arg._internal[ii],i);
}
}
template<class vtype,int N> inline static
void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal,0)),N> &arg, int i,int j)
void pokeIndex(iVector<vtype,N> &ret, const iVector<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0],0)),N> &arg, int i,int j)
{
for(int ii=0;ii<N;ii++){
TensorIndexRecursion<Level-1>::pokeIndex(ret._internal[ii],arg._internal[ii],i,j);
@ -161,7 +161,7 @@ class TensorIndexRecursion {
}
template<class vtype,int N> inline static
void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal,0)),N> &arg, int i)
void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0][0],0)),N> &arg, int i)
{
for(int ii=0;ii<N;ii++){
for(int jj=0;jj<N;jj++){
@ -169,7 +169,7 @@ class TensorIndexRecursion {
}}
}
template<class vtype,int N> inline static
void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal,0)),N> &arg, int i,int j)
void pokeIndex(iMatrix<vtype,N> &ret, const iMatrix<decltype(TensorIndexRecursion<Level-1>::peekIndex(ret._internal[0][0],0)),N> &arg, int i,int j)
{
for(int ii=0;ii<N;ii++){
for(int jj=0;jj<N;jj++){

30
lib/tensors/Tensor_traits.h
View File

@ -8,6 +8,7 @@
Author: Azusa Yamaguchi <ayamaguc@staffmail.ed.ac.uk>
Author: Peter Boyle <paboyle@ph.ed.ac.uk>
Author: Christopher Kelly <ckelly@phys.columbia.edu>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@ -230,6 +231,35 @@ namespace Grid {
static const bool value = true;
};
//Get the SIMD vector type from a Grid tensor or Lattice<Tensor>
template<typename T>
struct getVectorType{
typedef T type;
};
//Query if a tensor or Lattice<Tensor> is SIMD vector or scalar
template<typename T>
class isSIMDvectorized{
template<typename U>
static typename std::enable_if< !std::is_same< typename GridTypeMapper<typename getVectorType<U>::type>::scalar_type, typename GridTypeMapper<typename getVectorType<U>::type>::vector_type>::value, char>::type test(void *);
template<typename U>
static double test(...);
public:
enum {value = sizeof(test<T>(0)) == sizeof(char) };
};
//Get the precision of a Lattice, tensor or scalar type in units of sizeof(float)
template<typename T>
class getPrecision{
typedef typename getVectorType<T>::type vector_obj; //get the vector_obj (i.e. a grid Tensor) if its a Lattice<vobj>, do nothing otherwise (i.e. if fundamental or grid Tensor)
typedef typename GridTypeMapper<vector_obj>::scalar_type scalar_type; //get the associated scalar type. Works on fundamental and tensor types
typedef typename GridTypeMapper<scalar_type>::Realified real_scalar_type; //remove any std::complex wrapper, should get us to the fundamental type
public:
enum { value = sizeof(real_scalar_type)/sizeof(float) };
};
}
#endif

2
lib/tensors/Tensor_unary.h
View File

@ -86,6 +86,8 @@ UNARY(sqrt);
UNARY(rsqrt);
UNARY(sin);
UNARY(cos);
UNARY(asin);
UNARY(acos);
UNARY(log);
UNARY(exp);
UNARY(abs);

4
scripts/Make.inc Normal file
View File

@ -0,0 +1,4 @@
HFILES=
CCFILES=

3
scripts/filelist
View File

@ -3,7 +3,8 @@
home=`pwd`
cd $home/lib
HFILES=`find . -type f -name '*.h'`
HFILES=`find . -type f -name '*.h' -not -path '*/Old/*'`
HFILES="$HFILES Config.h"
CCFILES=`find . -type f -name '*.cc' -not -name '*ommunicator*.cc'`
echo> Make.inc
echo HFILES=$HFILES >> Make.inc

6
tests/Make.inc
View File

@ -1,11 +1,15 @@
bin_PROGRAMS += Test_cshift Test_simd Test_stencil
bin_PROGRAMS += Test_cshift Test_dwf_mixedcg_prec Test_simd Test_stencil
Test_cshift_SOURCES=Test_cshift.cc
Test_cshift_LDADD=-lGrid
Test_dwf_mixedcg_prec_SOURCES=Test_dwf_mixedcg_prec.cc
Test_dwf_mixedcg_prec_LDADD=-lGrid
Test_simd_SOURCES=Test_simd.cc
Test_simd_LDADD=-lGrid

108
tests/Test_dwf_mixedcg_prec.cc Normal file
View File

@ -0,0 +1,108 @@
/*************************************************************************************
Grid physics library, www.github.com/paboyle/Grid
Source file: ./tests/Test_dwf_cg_prec.cc
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 <Grid.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
template<class d>
struct scal {
d internal;
};
Gamma::GammaMatrix Gmu [] = {
Gamma::GammaX,
Gamma::GammaY,
Gamma::GammaZ,
Gamma::GammaT
};
int main (int argc, char ** argv)
{
Grid_init(&argc,&argv);
const int Ls=8;
GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexD::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid);
GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid);
GridCartesian * UGrid_f = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplexF::Nsimd()),GridDefaultMpi());
GridRedBlackCartesian * UrbGrid_f = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid_f);
GridCartesian * FGrid_f = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid_f);
GridRedBlackCartesian * FrbGrid_f = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid_f);
std::vector<int> seeds4({1,2,3,4});
std::vector<int> seeds5({5,6,7,8});
GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5);
GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4);
LatticeFermionD src(FGrid); random(RNG5,src);
LatticeFermionD result(FGrid); result=zero;
LatticeGaugeFieldD Umu(UGrid);
LatticeGaugeFieldF Umu_f(UGrid_f);
SU3::HotConfiguration(RNG4,Umu);
precisionChange(Umu_f,Umu);
RealD mass=0.1;
RealD M5=1.8;
DomainWallFermionD Ddwf(Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass,M5);
DomainWallFermionF Ddwf_f(Umu_f,*FGrid_f,*FrbGrid_f,*UGrid_f,*UrbGrid_f,mass,M5);
LatticeFermionD src_o(FrbGrid);
LatticeFermionD result_o(FrbGrid);
LatticeFermionD result_o_2(FrbGrid);
pickCheckerboard(Odd,src_o,src);
result_o.checkerboard = Odd;
result_o = zero;
result_o_2.checkerboard = Odd;
result_o_2 = zero;
SchurDiagMooeeOperator<DomainWallFermionD,LatticeFermionD> HermOpEO(Ddwf);
SchurDiagMooeeOperator<DomainWallFermionF,LatticeFermionF> HermOpEO_f(Ddwf_f);
std::cout << "Starting mixed CG" << std::endl;
MixedPrecisionConjugateGradient<LatticeFermionD,LatticeFermionF> mCG(1.0e-8, 10000, 50, FrbGrid_f, HermOpEO_f, HermOpEO);
mCG(src_o,result_o);
std::cout << "Starting regular CG" << std::endl;
ConjugateGradient<LatticeFermionD> CG(1.0e-8,10000);
CG(HermOpEO,src_o,result_o_2);
LatticeFermionD diff_o(FrbGrid);
RealD diff = axpy_norm(diff_o, -1.0, result_o, result_o_2);
std::cout << "Diff between mixed and regular CG: " << diff << std::endl;
Grid_finalize();
}

30
tests/Test_simd.cc
View File

@ -62,6 +62,18 @@ public:
template<class vec> void operator()(vec &rr,vec &i1,vec &i2) const { rr = adj(i1);}
std::string name(void) const { return std::string("Adj"); }
};
class funcImag {
public:
funcImag() {};
template<class vec> void operator()(vec &rr,vec &i1,vec &i2) const { rr = imag(i1);}
std::string name(void) const { return std::string("imag"); }
};
class funcReal {
public:
funcReal() {};
template<class vec> void operator()(vec &rr,vec &i1,vec &i2) const { rr = real(i1);}
std::string name(void) const { return std::string("real"); }
};
class funcTimesI {
public:
@ -141,8 +153,14 @@ void Tester(const functor &func)
}
extract<vec,scal>(v_result,result);
std::cout << GridLogMessage << " " << func.name() << std::endl;
std::cout << GridLogDebug << v_input1 << std::endl;
std::cout << GridLogDebug << v_result << std::endl;
int ok=0;
for(int i=0;i<Nsimd;i++){
if ( abs(reference[i]-result[i])>1.0e-7){
@ -389,6 +407,8 @@ int main (int argc, char ** argv)
Tester<ComplexF,vComplexF>(funcTimes());
Tester<ComplexF,vComplexF>(funcConj());
Tester<ComplexF,vComplexF>(funcAdj());
Tester<ComplexF,vComplexF>(funcReal());
Tester<ComplexF,vComplexF>(funcImag());
Tester<ComplexF,vComplexF>(funcInnerProduct());
ReductionTester<ComplexF,ComplexF,vComplexF>(funcReduce());
@ -421,13 +441,17 @@ int main (int argc, char ** argv)
Tester<ComplexD,vComplexD>(funcTimes());
Tester<ComplexD,vComplexD>(funcConj());
Tester<ComplexD,vComplexD>(funcAdj());
Tester<ComplexD, vComplexD>(funcReal());
Tester<ComplexD, vComplexD>(funcImag());
Tester<ComplexD, vComplexD>(funcInnerProduct());
ReductionTester<ComplexD, ComplexD, vComplexD>(funcReduce());
std::cout<<GridLogMessage << "==================================="<< std::endl;
std::cout << GridLogMessage
<< "===================================" << std::endl;
std::cout << GridLogMessage << "Testing vComplexD permutes " << std::endl;
std::cout<<GridLogMessage << "==================================="<< std::endl;
std::cout << GridLogMessage
<< "===================================" << std::endl;
// Log2 iteration
for (int i = 0; (1 << i) < vComplexD::Nsimd(); i++) {

276
tests/core/Test_main.cc
View File

@ -30,30 +30,32 @@ Author: paboyle <paboyle@ph.ed.ac.uk>
/* END LEGAL */
#include <Grid/Grid.h>
using namespace std;
using namespace Grid;
using namespace Grid::QCD;
/*
Grid_main.cc(232): error: no suitable user-defined conversion from "Grid::iScalar<Grid::iMatrix<Grid::iScalar<Grid::Complex>, 4>>" to "const Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Complex, 3>>>" exists
Grid_main.cc(232): error: no suitable user-defined conversion from
"Grid::iScalar<Grid::iMatrix<Grid::iScalar<Grid::Complex>, 4>>" to "const
Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Complex, 3>>>" exists
c_m = peekIdiot<SpinColourMatrix>(scm,1,2);
*/
template<class vobj> auto peekIdiot(const vobj &rhs,int i,int j) -> decltype(peekIndex<2>(rhs,0,0))
{
template <class vobj>
auto peekIdiot(const vobj &rhs, int i, int j)
-> decltype(peekIndex<2>(rhs, 0, 0)) {
return peekIndex<2>(rhs, i, j);
}
template<class vobj> auto peekDumKopf(const vobj &rhs,int i,int j) -> decltype(peekIndex<3>(rhs,0,0))
{
template <class vobj>
auto peekDumKopf(const vobj &rhs, int i, int j)
-> decltype(peekIndex<3>(rhs, 0, 0)) {
return peekIndex<3>(rhs, i, j);
}
template<class vobj> auto peekDumKopf(const vobj &rhs,int i) -> decltype(peekIndex<3>(rhs,0))
{
template <class vobj>
auto peekDumKopf(const vobj &rhs, int i) -> decltype(peekIndex<3>(rhs, 0)) {
return peekIndex<3>(rhs, i);
}
int main (int argc, char ** argv)
{
int main(int argc, char **argv) {
Grid_init(&argc, &argv);
std::vector<int> latt_size = GridDefaultLatt();
@ -73,7 +75,6 @@ int main (int argc, char ** argv)
#endif
for (int lat = 8; lat <= 16; lat += 40) {
std::cout << "Lat " << lat << std::endl;
latt_size[0] = lat;
@ -132,7 +133,6 @@ int main (int argc, char ** argv)
LatticeLorentzColourMatrix lcMat(&Fine);
LatticeComplex scalar(&Fine);
LatticeReal rscalar(&Fine);
LatticeReal iscalar(&Fine);
@ -154,24 +154,40 @@ int main (int argc, char ** argv)
random(FineRNG, sVec);
random(FineRNG, scVec);
fflush(stdout);
LatticeColourMatrix newFoo = Foo;
// confirm correctness of copy constructor
Bar = Foo - newFoo;
std::cout << "Copy constructor diff check: ";
double test_cc = norm2(Bar);
if (test_cc < 1e-5){
std::cout << "OK\n";
}
else{
std::cout << "fail\n";
abort();
}
TComplex tr = trace(cmat);
cVec = cMat * cVec; // LatticeColourVector = LatticeColourMatrix * LatticeColourVector
sVec = sMat * sVec; // LatticeSpinVector = LatticeSpinMatrix * LatticeSpinVector
scVec= scMat * scVec;// LatticeSpinColourVector = LatticeSpinColourMatrix * LatticeSpinColourVector
scVec= cMat * scVec; // LatticeSpinColourVector = LatticeColourMatrix * LatticeSpinColourVector
scVec= sMat * scVec; // LatticeSpinColourVector = LatticeSpinMatrix * LatticeSpinColourVector
cVec = cMat * cVec; // LatticeColourVector = LatticeColourMatrix
// * LatticeColourVector
sVec = sMat * sVec; // LatticeSpinVector = LatticeSpinMatrix
// * LatticeSpinVector
scVec = scMat * scVec; // LatticeSpinColourVector =
// LatticeSpinColourMatrix *
// LatticeSpinColourVector
scVec = cMat * scVec; // LatticeSpinColourVector = LatticeColourMatrix
// * LatticeSpinColourVector
scVec = sMat * scVec; // LatticeSpinColourVector = LatticeSpinMatrix
// * LatticeSpinColourVector
cMat = outerProduct(cVec, cVec);
scalar = localInnerProduct(cVec, cVec);
cMat = Ta(cMat); // traceless antihermitian
scalar += scalar;
scalar -= scalar;
scalar *= scalar;
@ -193,7 +209,6 @@ int main (int argc, char ** argv)
// scalar =cmplx(rscalar,iscalar);
PokeIndex<ColourIndex>(cVec, scalar, 1);
scalar = transpose(scalar);
scalar = TransposeIndex<ColourIndex>(scalar);
scalar = TraceIndex<SpinIndex>(scalar);
@ -214,9 +229,8 @@ int main (int argc, char ** argv)
// localNorm2
// localInnerProduct
scMat = sMat*scMat; // LatticeSpinColourMatrix = LatticeSpinMatrix * LatticeSpinColourMatrix
scMat = sMat * scMat; // LatticeSpinColourMatrix = LatticeSpinMatrix
// * LatticeSpinColourMatrix
///////////////////////
// Non-lattice (const objects) * Lattice
@ -229,10 +243,13 @@ int main (int argc, char ** argv)
// vSpinColourMatrix vscm;
scMat = cMat * scMat;
scm = cm * scm; // SpinColourMatrix = ColourMatrix * SpinColourMatrix
scm =
cm * scm; // SpinColourMatrix = ColourMatrix * SpinColourMatrix
scm = scm * cm; // SpinColourMatrix = SpinColourMartix * ColourMatrix
scm = GammaFive * scm ; // SpinColourMatrix = SpinMatrix * SpinColourMatrix
scm = scm* GammaFive ; // SpinColourMatrix = SpinColourMatrix * SpinMatrix
scm = GammaFive *
scm; // SpinColourMatrix = SpinMatrix * SpinColourMatrix
scm =
scm * GammaFive; // SpinColourMatrix = SpinColourMatrix * SpinMatrix
scm = scm * cplx;
vscm = vscm * cplx;
@ -261,7 +278,6 @@ int main (int argc, char ** argv)
scm = transpose(scm);
scm = transposeIndex<1>(scm);
random(SerialRNG, cm);
std::cout << GridLogMessage << cm << std::endl;
@ -269,7 +285,6 @@ int main (int argc, char ** argv)
TComplex tracecm = trace(cm);
std::cout << GridLogMessage << cm << std::endl;
cm = Exponentiate(cm, 2.0, 12);
std::cout << GridLogMessage << cm << " " << std::endl;
Complex det = Determinant(cm);
@ -283,11 +298,9 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << cm << " " << std::endl;
std::cout << GridLogMessage << "norm: " << norm2(cm) << std::endl;
// det = Determinant(cm);
// std::cout<<GridLogMessage << "determinant: " << det << std::endl;
// Foo = Foo+scalar; // LatticeColourMatrix+Scalar
// Foo = Foo*scalar; // LatticeColourMatrix*Scalar
// Foo = Foo-scalar; // LatticeColourMatrix-Scalar
@ -308,19 +321,20 @@ int main (int argc, char ** argv)
std::cout << GridLogMessage << cm << " " << std::endl;
std::cout << GridLogMessage << "norm: " << norm2(cm) << std::endl;
std::cout<<GridLogMessage << "norm cMmat : " << norm2(cMat) << std::endl;
std::cout << GridLogMessage << "norm cMmat : " << norm2(cMat)
<< std::endl;
cMat = expMat(cMat, ComplexD(1.0, 0.0));
std::cout<<GridLogMessage << "norm expMat: " << norm2(cMat) << std::endl;
std::cout << GridLogMessage << "norm expMat: " << norm2(cMat)
<< std::endl;
peekSite(cm, cMat, mysite);
std::cout << GridLogMessage << cm << " " << std::endl;
std::cout<<GridLogMessage << "determinant: " << Determinant(cm) << std::endl;
std::cout << GridLogMessage << "determinant: " << Determinant(cm)
<< std::endl;
std::cout << GridLogMessage << "norm: " << norm2(cm) << std::endl;
// LatticeComplex trlcMat(&Fine);
// trlcMat = trace(lcMat); // Trace involving iVector - now generates error
// trlcMat = trace(lcMat); // Trace involving iVector - now generates
// error
{ // Peek-ology and Poke-ology, with a little app-ology
Complex c;
@ -328,8 +342,10 @@ int main (int argc, char ** argv)
SpinMatrix s_m;
SpinColourMatrix sc_m;
s_m = TensorIndexRecursion<ColourIndex>::traceIndex(sc_m); // Map to traceColour
c_m = TensorIndexRecursion<SpinIndex>::traceIndex(sc_m); // map to traceSpin
s_m = TensorIndexRecursion<ColourIndex>::traceIndex(
sc_m); // Map to traceColour
c_m = TensorIndexRecursion<SpinIndex>::traceIndex(
sc_m); // map to traceSpin
c = TensorIndexRecursion<SpinIndex>::traceIndex(s_m);
c = TensorIndexRecursion<ColourIndex>::traceIndex(c_m);
@ -343,8 +359,10 @@ int main (int argc, char ** argv)
printf("c. Level %d\n", c_m().TensorLevel);
printf("c. Level %d\n", c_m()().TensorLevel);
c_m()() = scm()(0,0); //ColourComponents of CM <= ColourComponents of SpinColourMatrix
scm()(1,1) = cm()(); //ColourComponents of CM <= ColourComponents of SpinColourMatrix
c_m()() = scm()(0, 0); // ColourComponents of CM <= ColourComponents of
// SpinColourMatrix
scm()(1, 1) = cm()(); // ColourComponents of CM <= ColourComponents of
// SpinColourMatrix
c = scm()(1, 1)(1, 2);
scm()(1, 1)(2, 1) = c;
@ -364,12 +382,12 @@ int main (int argc, char ** argv)
*/
lex_sites(Foo);
Integer mm[4];
mm[0] = 1;
mm[1] = Fine._rdimensions[0];
mm[2] = Fine._ldimensions[0] * Fine._ldimensions[1];
mm[3]=Fine._ldimensions[0]*Fine._ldimensions[1]*Fine._ldimensions[2];
mm[3] =
Fine._ldimensions[0] * Fine._ldimensions[1] * Fine._ldimensions[2];
LatticeInteger lex(&Fine);
lex = zero;
@ -377,35 +395,37 @@ int main (int argc, char ** argv)
LatticeInteger coor(&Fine);
LatticeCoordinate(coor, d);
lex = lex + coor * mm[d];
}
// Bar = zero;
// Bar = where(lex<Integer(10),Foo,Bar);
cout << "peeking sites..\n";
{
std::vector<int> coor(4);
for (coor[3] = 0; coor[3] < latt_size[3] / mpi_layout[3]; coor[3]++) {
for (coor[2] = 0; coor[2] < latt_size[2] / mpi_layout[2]; coor[2]++) {
for(coor[1]=0;coor[1]<latt_size[1]/mpi_layout[1];coor[1]++){
for(coor[0]=0;coor[0]<latt_size[0]/mpi_layout[0];coor[0]++){
for (coor[1] = 0; coor[1] < latt_size[1] / mpi_layout[1];
coor[1]++) {
for (coor[0] = 0; coor[0] < latt_size[0] / mpi_layout[0];
coor[0]++) {
ColourMatrix bar;
peekSite(bar, Bar, coor);
for (int r = 0; r < 3; r++) {
for (int c = 0; c < 3; c++) {
// cout<<"bar "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<" "<<bar()()(r,c)<<std::endl;
}}
}}}}
// cout<<"bar "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<"
// "<<bar()()(r,c)<<std::endl;
}
}
}
}
}
}
}
// setCheckerboard(ShiftedCheck,rFoo);
// setCheckerboard(ShiftedCheck,bFoo);
// Lattice SU(3) x SU(3)
Fine.Barrier();
FooBar = Foo * Bar;
@ -440,11 +460,15 @@ int main (int argc, char ** argv)
Fine.Barrier();
if (Fine.IsBoss()) {
#ifdef OMP
printf("mult NumThread %d , Lattice size %d , %f us per call\n",omp_get_max_threads(),lat,(t1-t0)/ncall);
printf("mult NumThread %d , Lattice size %d , %f us per call\n",
omp_get_max_threads(), lat, (t1 - t0) / ncall);
#endif
printf("mult NumThread %d , Lattice size %d , %f us per call\n",omp,lat,(t1-t0)/ncall);
printf("mult NumThread %d , Lattice size %d , %f Mflop/s\n",omp,lat,flops/(t1-t0));
printf("mult NumThread %d , Lattice size %d , %f MB/s\n",omp,lat,bytes/(t1-t0));
printf("mult NumThread %d , Lattice size %d , %f us per call\n", omp,
lat, (t1 - t0) / ncall);
printf("mult NumThread %d , Lattice size %d , %f Mflop/s\n", omp, lat,
flops / (t1 - t0));
printf("mult NumThread %d , Lattice size %d , %f MB/s\n", omp, lat,
bytes / (t1 - t0));
}
mult(FooBar, Foo, Bar);
FooBar = Foo * Bar;
@ -464,9 +488,12 @@ int main (int argc, char ** argv)
FooBar = Foo * Bar;
if (Fine.IsBoss()) {
printf("Cshift Mult: NumThread %d , Lattice size %d , %f us per call\n",omp,lat,(t1-t0)/ncall);
printf("Cshift Mult: NumThread %d , Lattice size %d , %f Mflop/s\n",omp,lat,flops/(t1-t0));
printf("Cshift Mult: NumThread %d , Lattice size %d , %f MB/s\n",omp,lat,bytes/(t1-t0));
printf("Cshift Mult: NumThread %d , Lattice size %d , %f us per call\n",
omp, lat, (t1 - t0) / ncall);
printf("Cshift Mult: NumThread %d , Lattice size %d , %f Mflop/s\n",
omp, lat, flops / (t1 - t0));
printf("Cshift Mult: NumThread %d , Lattice size %d , %f MB/s\n", omp,
lat, bytes / (t1 - t0));
}
// pickCheckerboard(0,rFoo,FooBar);
// pickCheckerboard(1,bFoo,FooBar);
@ -475,20 +502,19 @@ int main (int argc, char ** argv)
double nrm = 0;
LatticeColourMatrix deriv(&Fine);
double half = 0.5;
deriv = 0.5 * Cshift(Foo, 0, 1) - 0.5 * Cshift(Foo, 0, -1);
for (int dir = 0; dir < 4; dir++) {
for (int shift = 0; shift < latt_size[dir]; shift++) {
pickCheckerboard(0,rFoo,Foo); // Pick out red or black checkerboards
pickCheckerboard(0, rFoo,
Foo); // Pick out red or black checkerboards
pickCheckerboard(1, bFoo, Foo);
if (Fine.IsBoss()) {
std::cout<<GridLogMessage << "Shifting both parities by "<< shift <<" direction "<< dir <<std::endl;
std::cout << GridLogMessage << "Shifting both parities by " << shift
<< " direction " << dir << std::endl;
}
Shifted = Cshift(Foo, dir, shift); // Shift everything
@ -497,45 +523,48 @@ int main (int argc, char ** argv)
ShiftedCheck = zero;
setCheckerboard(ShiftedCheck, bShifted); // Put them all together
setCheckerboard(ShiftedCheck,rShifted); // and check the results (later)
setCheckerboard(ShiftedCheck,
rShifted); // and check the results (later)
// Check results
std::vector<int> coor(4);
for (coor[3] = 0; coor[3] < latt_size[3] / mpi_layout[3]; coor[3]++) {
for(coor[2]=0;coor[2]<latt_size[2]/mpi_layout[2];coor[2]++){
for(coor[1]=0;coor[1]<latt_size[1]/mpi_layout[1];coor[1]++){
for(coor[0]=0;coor[0]<latt_size[0]/mpi_layout[0];coor[0]++){
for (coor[2] = 0; coor[2] < latt_size[2] / mpi_layout[2];
coor[2]++) {
for (coor[1] = 0; coor[1] < latt_size[1] / mpi_layout[1];
coor[1]++) {
for (coor[0] = 0; coor[0] < latt_size[0] / mpi_layout[0];
coor[0]++) {
std::complex<Grid::Real> diff;
std::vector<int> shiftcoor = coor;
shiftcoor[dir]=(shiftcoor[dir]+shift+latt_size[dir])%(latt_size[dir]/mpi_layout[dir]);
shiftcoor[dir] = (shiftcoor[dir] + shift + latt_size[dir]) %
(latt_size[dir] / mpi_layout[dir]);
std::vector<int> rl(4);
for (int dd = 0; dd < 4; dd++) {
rl[dd] = latt_size[dd] / simd_layout[dd] / mpi_layout[dd];
}
int lex = coor[0]%rl[0]
+ (coor[1]%rl[1])*rl[0]
+ (coor[2]%rl[2])*rl[0]*rl[1]
+ (coor[3]%rl[3])*rl[0]*rl[1]*rl[2];
lex +=
+1000*(coor[0]/rl[0])
+1000*(coor[1]/rl[1])*simd_layout[0]
+1000*(coor[2]/rl[2])*simd_layout[0]*simd_layout[1]
+1000*(coor[3]/rl[3])*simd_layout[0]*simd_layout[1]*simd_layout[2];
int lex = coor[0] % rl[0] + (coor[1] % rl[1]) * rl[0] +
(coor[2] % rl[2]) * rl[0] * rl[1] +
(coor[3] % rl[3]) * rl[0] * rl[1] * rl[2];
lex += +1000 * (coor[0] / rl[0]) +
1000 * (coor[1] / rl[1]) * simd_layout[0] +
1000 * (coor[2] / rl[2]) * simd_layout[0] *
simd_layout[1] +
1000 * (coor[3] / rl[3]) * simd_layout[0] *
simd_layout[1] * simd_layout[2];
int lex_coor = shiftcoor[0]%rl[0]
+ (shiftcoor[1]%rl[1])*rl[0]
+ (shiftcoor[2]%rl[2])*rl[0]*rl[1]
+ (shiftcoor[3]%rl[3])*rl[0]*rl[1]*rl[2];
lex_coor +=
+1000*(shiftcoor[0]/rl[0])
+1000*(shiftcoor[1]/rl[1])*simd_layout[0]
+1000*(shiftcoor[2]/rl[2])*simd_layout[0]*simd_layout[1]
+1000*(shiftcoor[3]/rl[3])*simd_layout[0]*simd_layout[1]*simd_layout[2];
int lex_coor = shiftcoor[0] % rl[0] +
(shiftcoor[1] % rl[1]) * rl[0] +
(shiftcoor[2] % rl[2]) * rl[0] * rl[1] +
(shiftcoor[3] % rl[3]) * rl[0] * rl[1] * rl[2];
lex_coor += +1000 * (shiftcoor[0] / rl[0]) +
1000 * (shiftcoor[1] / rl[1]) * simd_layout[0] +
1000 * (shiftcoor[2] / rl[2]) * simd_layout[0] *
simd_layout[1] +
1000 * (shiftcoor[3] / rl[3]) * simd_layout[0] *
simd_layout[1] * simd_layout[2];
ColourMatrix foo;
ColourMatrix bar;
@ -558,36 +587,46 @@ int main (int argc, char ** argv)
Real Ttr = real(trprod);
double nn = Ttr;
if (nn > 0)
cout<<"Shift real trace fail "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<endl;
cout << "Shift real trace fail " << coor[0] << coor[1]
<< coor[2] << coor[3] << endl;
for (int r = 0; r < 3; r++) {
for (int c = 0; c < 3; c++) {
diff = shifted1()()(r, c) - shifted2()()(r, c);
nn = real(conjugate(diff) * diff);
if (nn > 0)
cout<<"Shift fail (shifted1/shifted2-ref) "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<" "
cout << "Shift fail (shifted1/shifted2-ref) " << coor[0]
<< coor[1] << coor[2] << coor[3] << " "
<< shifted1()()(r, c) << " " << shifted2()()(r, c)
<< " "<< foo()()(r,c)<< " lex expect " << lex_coor << " lex "<<lex<<endl;
<< " " << foo()()(r, c) << " lex expect "
<< lex_coor << " lex " << lex << endl;
else if (0)
cout<<"Shift pass 1vs2 "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<" "
<<shifted1()()(r,c)<<" "<<shifted2()()(r,c)
<< " "<< foo()()(r,c)<< " lex expect " << lex_coor << " lex "<<lex<<endl;
}}
cout << "Shift pass 1vs2 " << coor[0] << coor[1]
<< coor[2] << coor[3] << " " << shifted1()()(r, c)
<< " " << shifted2()()(r, c) << " "
<< foo()()(r, c) << " lex expect " << lex_coor
<< " lex " << lex << endl;
}
}
for (int r = 0; r < 3; r++) {
for (int c = 0; c < 3; c++) {
diff = shifted3()()(r, c) - shifted2()()(r, c);
nn = real(conjugate(diff) * diff);
if (nn > 0)
cout<<"Shift rb fail (shifted3/shifted2-ref) "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<" "
cout << "Shift rb fail (shifted3/shifted2-ref) "
<< coor[0] << coor[1] << coor[2] << coor[3] << " "
<< shifted3()()(r, c) << " " << shifted2()()(r, c)
<< " "<< foo()()(r,c)<< " lex expect " << lex_coor << " lex "<<lex<<endl;
<< " " << foo()()(r, c) << " lex expect "
<< lex_coor << " lex " << lex << endl;
else if (0)
cout<<"Shift rb pass 3vs2 "<<coor[0]<<coor[1]<<coor[2]<<coor[3] <<" "
<<shifted3()()(r,c)<<" "<<shifted2()()(r,c)
<< " "<< foo()()(r,c)<< " lex expect " << lex_coor << " lex "<<lex<<endl;
}}
cout << "Shift rb pass 3vs2 " << coor[0] << coor[1]
<< coor[2] << coor[3] << " " << shifted3()()(r, c)
<< " " << shifted2()()(r, c) << " "
<< foo()()(r, c) << " lex expect " << lex_coor
<< " lex " << lex << endl;
}
}
peekSite(bar, Bar, coor);
peekSite(foobar1, FooBar, coor);
@ -596,18 +635,31 @@ int main (int argc, char ** argv)
for (int c = 0; c < Nc; c++) {
diff = foobar2()()(r, c) - foobar1()()(r, c);
nrm = nrm + real(conjugate(diff) * diff);
}}
}}}}
if( Fine.IsBoss() ){
std::cout<<GridLogMessage << "LatticeColorMatrix * LatticeColorMatrix nrm diff = "<<nrm<<std::endl;
}
}}
}
}
}
}
}
if (Fine.IsBoss()) {
std::cout << GridLogMessage
<< "LatticeColorMatrix * LatticeColorMatrix nrm diff = "
<< nrm << std::endl;
}
}
}
} // loop for lat
} // loop for omp
/*
// Testing Smearing routine compilation, separate in a different file
GridCartesian Fine(latt_size,simd_layout,mpi_layout);
Smear_APE< PeriodicGimplR > APEsmearing; // periodic gauge implemetation
Smear_Stout< PeriodicGimplR > StoutSmearing(&APEsmearing);
SmearedConfiguration< PeriodicGimplR > SmartConf(&Fine, 3, StoutSmearing);
std::cout<<GridLogMessage << sizeof(vComplexF) << std::endl;
*/
Grid_finalize();
}

3
tests/hmc/Test_hmc_EOWilsonFermionGauge.cc
View File

@ -66,6 +66,9 @@ public:
TwoFlavourEvenOddPseudoFermionAction<ImplPolicy> Nf2(FermOp,CG,CG);
//Set smearing (true/false), default: false
Nf2.is_smeared=false;
//Collect actions
ActionLevel<LatticeGaugeField> Level1(1);
Level1.push_back(&Nf2);

3
tests/hmc/Test_hmc_EOWilsonRatio.cc
View File

@ -66,6 +66,9 @@ public:
ConjugateGradient<FermionField> CG(1.0e-8,10000);
TwoFlavourEvenOddRatioPseudoFermionAction<ImplPolicy> Nf2(NumOp, DenOp,CG,CG);
//Set smearing (true/false), default: false
Nf2.is_smeared=true;
//Collect actions
ActionLevel<LatticeGaugeField> Level1;
Level1.push_back(&Nf2);

4
tests/hmc/Test_hmc_WilsonFermionGauge.cc
View File

@ -67,6 +67,10 @@ public:
TwoFlavourPseudoFermionAction<ImplPolicy> Nf2(FermOp,CG,CG);
//Set smearing (true/false), default: false
Nf2.is_smeared = true;
//Collect actions
ActionLevel<LatticeGaugeField> Level1(1);
Level1.push_back(&Nf2);

3
tests/hmc/Test_hmc_WilsonRatio.cc
View File

@ -66,6 +66,9 @@ public:
ConjugateGradient<FermionField> CG(1.0e-8,10000);
TwoFlavourRatioPseudoFermionAction<ImplPolicy> Nf2(NumOp, DenOp,CG,CG);
//Set smearing (true/false), default: false
Nf2.is_smeared=true;
//Collect actions
ActionLevel<LatticeGaugeField> Level1;
Level1.push_back(&Nf2);

32
tests/hmc/Test_rhmc_EOWilson1p1.cc
View File

@ -35,10 +35,8 @@ using namespace Grid::QCD;
namespace Grid {
namespace QCD {
class HmcRunner : public NerscHmcRunner {
public:
void BuildTheAction(int argc, char **argv)
{
@ -46,7 +44,9 @@ public:
typedef WilsonFermionR FermionAction;
typedef typename FermionAction::FermionField FermionField;
UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi());
UGrid = SpaceTimeGrid::makeFourDimGrid(
GridDefaultLatt(), GridDefaultSimd(Nd, vComplex::Nsimd()),
GridDefaultMpi());
UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid);
FGrid = UGrid;
@ -62,9 +62,15 @@ public:
FermionAction FermOp(U, *FGrid, *FrbGrid, mass);
// 1+1 flavour
OneFlavourRationalParams Params(1.0e-4,64.0,1000,1.0e-6);
OneFlavourEvenOddRationalPseudoFermionAction<WilsonImplR> WilsonNf1a(FermOp,Params);
OneFlavourEvenOddRationalPseudoFermionAction<WilsonImplR> WilsonNf1b(FermOp,Params);
OneFlavourRationalParams Params(1.0e-4, 64.0, 2000, 1.0e-6);
OneFlavourEvenOddRationalPseudoFermionAction<WilsonImplR> WilsonNf1a(
FermOp, Params);
OneFlavourEvenOddRationalPseudoFermionAction<WilsonImplR> WilsonNf1b(
FermOp, Params);
//Smearing on/off
WilsonNf1a.is_smeared = true;
WilsonNf1b.is_smeared = true;
// Collect actions
ActionLevel<LatticeGaugeField> Level1;
@ -76,22 +82,18 @@ public:
Run(argc, argv);
};
};
}
}
}}
int main (int argc, char ** argv)
{
int main(int argc, char **argv) {
Grid_init(&argc, &argv);
int threads = GridThread::GetThreads();
std::cout<<GridLogMessage << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout << GridLogMessage << "Grid is setup to use " << threads
<< " threads" << std::endl;
HmcRunner TheHMC;
TheHMC.BuildTheAction(argc, argv);
}

4
tests/hmc/Test_rhmc_Wilson1p1.cc
View File

@ -66,6 +66,10 @@ public:
OneFlavourRationalPseudoFermionAction<WilsonImplR> WilsonNf1a(FermOp,Params);
OneFlavourRationalPseudoFermionAction<WilsonImplR> WilsonNf1b(FermOp,Params);
//Set smearing (true/false), default: false
WilsonNf1a.is_smeared=false;
WilsonNf1b.is_smeared=false;
//Collect actions
ActionLevel<LatticeGaugeField> Level1;
Level1.push_back(&WilsonNf1a);