diff --git a/TODO b/TODO index c37cbf8b..83bfda5e 100644 --- a/TODO +++ b/TODO @@ -3,19 +3,19 @@ TODO: Large item work list: -1)- BG/Q port and check +1)- BG/Q port and check ; Andrew says ok. 2)- Christoph's local basis expansion Lanczos -3)- Precision conversion and sort out localConvert <-- partial - - - Consistent linear solver flop count/rate -- PARTIAL, time but no flop/s yet +-- +3a)- RNG I/O in ILDG/SciDAC (minor) +3b)- Precision conversion and sort out localConvert <-- partial/easy +3c)- Consistent linear solver flop count/rate -- PARTIAL, time but no flop/s yet 4)- Physical propagator interface 5)- Conserved currents 6)- Multigrid Wilson and DWF, compare to other Multigrid implementations 7)- HDCR resume Recent DONE - --- MultiRHS with spread out extra dim -- Go through filesystem with SciDAC I/O. <--- DONE +-- MultiRHS with spread out extra dim -- Go through filesystem with SciDAC I/O ; <-- DONE ; bmark cori -- Lanczos Remove DenseVector, DenseMatrix; Use Eigen instead. <-- DONE -- GaugeFix into central location <-- DONE -- Scidac and Ildg metadata handling <-- DONE diff --git a/benchmarks/Benchmark_ITT.cc b/benchmarks/Benchmark_ITT.cc index c0ce451f..666e4830 100644 --- a/benchmarks/Benchmark_ITT.cc +++ b/benchmarks/Benchmark_ITT.cc @@ -232,9 +232,13 @@ public: std::cout<Barrier(); for(int i=0;iBarrier(); double t1=usecond(); - // uint64_t ncall = (uint64_t) 2.5*1000.0*1000.0*nwarm/(t1-t0); - // if (ncall < 500) ncall = 500; - uint64_t ncall = 500; - sFGrid->Broadcast(0,&ncall,sizeof(ncall)); - - // std::cout << GridLogMessage << " Estimate " << ncall << " calls per second"< t_time(ncall); for(uint64_t i=0;i L_list({8,12,16,24}); +#else + int sel=1; + std::vector L_list({8,12}); +#endif + int selm1=sel-1; + std::vector robust_list; + std::vector wilson; std::vector dwf4; std::vector dwf5; @@ -722,7 +721,8 @@ int main (int argc, char ** argv) std::cout<1) ) { + std::cout< latt4 = GridDefaultLatt(); - const int Ls=16; + int Ls=16; + for(int i=0;i> Ls; + } + + GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi()); GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid); diff --git a/benchmarks/Benchmark_gparity.cc b/benchmarks/Benchmark_gparity.cc new file mode 100644 index 00000000..f6036aa8 --- /dev/null +++ b/benchmarks/Benchmark_gparity.cc @@ -0,0 +1,190 @@ +#include +#include +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +template +struct scal { + d internal; +}; + + Gamma::Algebra Gmu [] = { + Gamma::Algebra::GammaX, + Gamma::Algebra::GammaY, + Gamma::Algebra::GammaZ, + Gamma::Algebra::GammaT + }; + +typedef typename GparityDomainWallFermionF::FermionField GparityLatticeFermionF; +typedef typename GparityDomainWallFermionD::FermionField GparityLatticeFermionD; + + + +int main (int argc, char ** argv) +{ + Grid_init(&argc,&argv); + + int Ls=16; + for(int i=0;i> Ls; + } + + + int threads = GridThread::GetThreads(); + std::cout<_Nprocessors; + RealD NN = UGrid->NodeCount(); + + std::cout << GridLogMessage<< "*****************************************************************" <Barrier(); + Dw.ZeroCounters(); + Dw.Dhop(src,result,0); + std::cout<Barrier(); + + double volume=Ls; for(int mu=0;muBarrier(); + DwH.ZeroCounters(); + DwH.Dhop(src,result,0); + double t0=usecond(); + for(int i=0;iBarrier(); + + double volume=Ls; for(int mu=0;muBarrier(); + DwD.ZeroCounters(); + DwD.Dhop(src_d,result_d,0); + std::cout<Barrier(); + + double volume=Ls; for(int mu=0;mu simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< #include #include #include +#include #include #include @@ -44,30 +45,16 @@ Author: Peter Boyle #include #include #include - -// Lanczos support -//#include +#include +#include #include #include #include -// Eigen/lanczos // EigCg -// MCR // Pcg -// Multishift CG // Hdcg // GCR // etc.. -// integrator/Leapfrog -// integrator/Omelyan -// integrator/ForceGradient - -// montecarlo/hmc -// montecarlo/rhmc -// montecarlo/metropolis -// etc... - - #endif diff --git a/lib/algorithms/FFT.h b/lib/algorithms/FFT.h index 240f338b..ec558ad9 100644 --- a/lib/algorithms/FFT.h +++ b/lib/algorithms/FFT.h @@ -230,6 +230,7 @@ namespace Grid { // Barrel shift and collect global pencil std::vector lcoor(Nd), gcoor(Nd); result = source; + int pc = processor_coor[dim]; for(int p=0;plSites();idx++) { sgrid->LocalIndexToLocalCoor(idx,cbuf); peekLocalSite(s,result,cbuf); - cbuf[dim]+=p*L; + cbuf[dim]+=((pc+p) % processors[dim])*L; + // cbuf[dim]+=p*L; pokeLocalSite(s,pgbuf,cbuf); } } @@ -278,7 +280,6 @@ namespace Grid { flops+= flops_call*NN; // writing out result - int pc = processor_coor[dim]; PARALLEL_REGION { std::vector clbuf(Nd), cgbuf(Nd); diff --git a/lib/algorithms/LinearOperator.h b/lib/algorithms/LinearOperator.h index 6cb77296..f1b8820e 100644 --- a/lib/algorithms/LinearOperator.h +++ b/lib/algorithms/LinearOperator.h @@ -162,15 +162,10 @@ namespace Grid { _Mat.M(in,out); } void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ - ComplexD dot; - _Mat.M(in,out); - dot= innerProduct(in,out); - n1=real(dot); - - dot = innerProduct(out,out); - n2=real(dot); + ComplexD dot= innerProduct(in,out); n1=real(dot); + n2=norm2(out); } void HermOp(const Field &in, Field &out){ _Mat.M(in,out); @@ -192,10 +187,10 @@ namespace Grid { ni=Mpc(in,tmp); no=MpcDag(tmp,out); } - void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ + virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ MpcDagMpc(in,out,n1,n2); } - void HermOp(const Field &in, Field &out){ + virtual void HermOp(const Field &in, Field &out){ RealD n1,n2; HermOpAndNorm(in,out,n1,n2); } @@ -212,7 +207,6 @@ namespace Grid { void OpDir (const Field &in, Field &out,int dir,int disp) { assert(0); } - }; template class SchurDiagMooeeOperator : public SchurOperatorBase { @@ -270,7 +264,6 @@ namespace Grid { return axpy_norm(out,-1.0,tmp,in); } }; - template class SchurDiagTwoOperator : public SchurOperatorBase { protected: @@ -299,6 +292,45 @@ namespace Grid { return axpy_norm(out,-1.0,tmp,in); } }; + /////////////////////////////////////////////////////////////////////////////////////////////////// + // Left handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) psi = eta --> ( 1 - Moo^-1 Moe Mee^-1 Meo ) psi = Moo^-1 eta + // Right handed Moo^-1 ; (Moo - Moe Mee^-1 Meo) Moo^-1 Moo psi = eta --> ( 1 - Moe Mee^-1 Meo ) Moo^-1 phi=eta ; psi = Moo^-1 phi + /////////////////////////////////////////////////////////////////////////////////////////////////// + template using SchurDiagOneRH = SchurDiagTwoOperator ; + template using SchurDiagOneLH = SchurDiagOneOperator ; + /////////////////////////////////////////////////////////////////////////////////////////////////// + // Staggered use + /////////////////////////////////////////////////////////////////////////////////////////////////// + template + class SchurStaggeredOperator : public SchurOperatorBase { + protected: + Matrix &_Mat; + public: + SchurStaggeredOperator (Matrix &Mat): _Mat(Mat){}; + virtual void HermOpAndNorm(const Field &in, Field &out,RealD &n1,RealD &n2){ + n2 = Mpc(in,out); + ComplexD dot= innerProduct(in,out); + n1 = real(dot); + } + virtual void HermOp(const Field &in, Field &out){ + Mpc(in,out); + } + virtual RealD Mpc (const Field &in, Field &out) { + Field tmp(in._grid); + _Mat.Meooe(in,tmp); + _Mat.MooeeInv(tmp,out); + _Mat.MeooeDag(out,tmp); + _Mat.Mooee(in,out); + return axpy_norm(out,-1.0,tmp,out); + } + virtual RealD MpcDag (const Field &in, Field &out){ + return Mpc(in,out); + } + virtual void MpcDagMpc(const Field &in, Field &out,RealD &ni,RealD &no) { + assert(0);// Never need with staggered + } + }; + template using SchurStagOperator = SchurStaggeredOperator; ///////////////////////////////////////////////////////////// diff --git a/lib/algorithms/approx/Chebyshev.h b/lib/algorithms/approx/Chebyshev.h index 2793f138..f8c21a05 100644 --- a/lib/algorithms/approx/Chebyshev.h +++ b/lib/algorithms/approx/Chebyshev.h @@ -8,6 +8,7 @@ Author: Peter Boyle Author: paboyle +Author: Christoph Lehner 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 @@ -193,6 +194,47 @@ namespace Grid { return sum; }; + RealD approxD(RealD x) + { + RealD Un; + RealD Unm; + RealD Unp; + + RealD y=( x-0.5*(hi+lo))/(0.5*(hi-lo)); + + RealD U0=1; + RealD U1=2*y; + + RealD sum; + sum = Coeffs[1]*U0; + sum+= Coeffs[2]*U1*2.0; + + Un =U1; + Unm=U0; + for(int i=2;i::quiet_NaN(); + } + // Implement the required interface void operator() (LinearOperatorBase &Linop, const Field &in, Field &out) { diff --git a/lib/algorithms/approx/Forecast.h b/lib/algorithms/approx/Forecast.h new file mode 100644 index 00000000..87eb84a6 --- /dev/null +++ b/lib/algorithms/approx/Forecast.h @@ -0,0 +1,152 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/algorithms/approx/Forecast.h + +Copyright (C) 2015 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 INCLUDED_FORECAST_H +#define INCLUDED_FORECAST_H + +namespace Grid { + + // Abstract base class. + // Takes a matrix (Mat), a source (phi), and a vector of Fields (chi) + // and returns a forecasted solution to the system D*psi = phi (psi). + template + class Forecast + { + public: + virtual Field operator()(Matrix &Mat, const Field& phi, const std::vector& chi) = 0; + }; + + // Implementation of Brower et al.'s chronological inverter (arXiv:hep-lat/9509012), + // used to forecast solutions across poles of the EOFA heatbath. + // + // Modified from CPS (cps_pp/src/util/dirac_op/d_op_base/comsrc/minresext.C) + template + class ChronoForecast : public Forecast + { + public: + Field operator()(Matrix &Mat, const Field& phi, const std::vector& prev_solns) + { + int degree = prev_solns.size(); + Field chi(phi); // forecasted solution + + // Trivial cases + if(degree == 0){ chi = zero; return chi; } + else if(degree == 1){ return prev_solns[0]; } + + RealD dot; + ComplexD xp; + Field r(phi); // residual + Field Mv(phi); + std::vector v(prev_solns); // orthonormalized previous solutions + std::vector MdagMv(degree,phi); + + // Array to hold the matrix elements + std::vector> G(degree, std::vector(degree)); + + // Solution and source vectors + std::vector a(degree); + std::vector b(degree); + + // Orthonormalize the vector basis + for(int i=0; i std::abs(G[k][k])){ k = j; } } + if(k != i){ + xp = b[k]; + b[k] = b[i]; + b[i] = xp; + for(int j=0; j=0; i--){ + a[i] = 0.0; + for(int j=i+1; j +Author: paboyle +Author: Chulwoo Jung +Author: Christoph Lehner + + 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_BIRL_H +#define GRID_BIRL_H + +#include //memset + +#include +#include + +#include +#include +#include + +namespace Grid { + +///////////////////////////////////////////////////////////// +// Implicitly restarted lanczos +///////////////////////////////////////////////////////////// + + template + class BlockImplicitlyRestartedLanczos { + + const RealD small = 1.0e-16; +public: + int lock; + int get; + int Niter; + int converged; + + int Nminres; // Minimum number of restarts; only check for convergence after + int Nstop; // Number of evecs checked for convergence + int Nk; // Number of converged sought + int Np; // Np -- Number of spare vecs in kryloc space + int Nm; // Nm -- total number of vectors + + int orth_period; + + RealD OrthoTime; + + RealD eresid, betastp; + SortEigen _sort; + LinearFunction &_HermOp; + LinearFunction &_HermOpTest; + ///////////////////////// + // Constructor + ///////////////////////// + + BlockImplicitlyRestartedLanczos( + LinearFunction & HermOp, + LinearFunction & HermOpTest, + int _Nstop, // sought vecs + int _Nk, // sought vecs + int _Nm, // spare vecs + RealD _eresid, // resid in lmdue deficit + RealD _betastp, // if beta(k) < betastp: converged + int _Niter, // Max iterations + int _Nminres, int _orth_period = 1) : + _HermOp(HermOp), + _HermOpTest(HermOpTest), + Nstop(_Nstop), + Nk(_Nk), + Nm(_Nm), + eresid(_eresid), + betastp(_betastp), + Niter(_Niter), + Nminres(_Nminres), + orth_period(_orth_period) + { + Np = Nm-Nk; assert(Np>0); + }; + + BlockImplicitlyRestartedLanczos( + LinearFunction & HermOp, + LinearFunction & HermOpTest, + int _Nk, // sought vecs + int _Nm, // spare vecs + RealD _eresid, // resid in lmdue deficit + RealD _betastp, // if beta(k) < betastp: converged + int _Niter, // Max iterations + int _Nminres, + int _orth_period = 1) : + _HermOp(HermOp), + _HermOpTest(HermOpTest), + Nstop(_Nk), + Nk(_Nk), + Nm(_Nm), + eresid(_eresid), + betastp(_betastp), + Niter(_Niter), + Nminres(_Nminres), + orth_period(_orth_period) + { + Np = Nm-Nk; assert(Np>0); + }; + + +/* Saad PP. 195 +1. Choose an initial vector v1 of 2-norm unity. Set β1 ≡ 0, v0 ≡ 0 +2. For k = 1,2,...,m Do: +3. wk:=Avk−βkv_{k−1} +4. αk:=(wk,vk) // +5. wk:=wk−αkvk // wk orthog vk +6. βk+1 := ∥wk∥2. If βk+1 = 0 then Stop +7. vk+1 := wk/βk+1 +8. EndDo + */ + void step(std::vector& lmd, + std::vector& lme, + BasisFieldVector& evec, + Field& w,int Nm,int k) + { + assert( k< Nm ); + + GridStopWatch gsw_op,gsw_o; + + Field& evec_k = evec[k]; + + gsw_op.Start(); + _HermOp(evec_k,w); + gsw_op.Stop(); + + if(k>0){ + w -= lme[k-1] * evec[k-1]; + } + + ComplexD zalph = innerProduct(evec_k,w); // 4. αk:=(wk,vk) + RealD alph = real(zalph); + + w = w - alph * evec_k;// 5. wk:=wk−αkvk + + RealD beta = normalise(w); // 6. βk+1 := ∥wk∥2. If βk+1 = 0 then Stop + // 7. vk+1 := wk/βk+1 + + std::cout<0 && k % orth_period == 0) { + orthogonalize(w,evec,k); // orthonormalise + } + gsw_o.Stop(); + + if(k < Nm-1) { + evec[k+1] = w; + } + + std::cout << GridLogMessage << "Timing: operator=" << gsw_op.Elapsed() << + " orth=" << gsw_o.Elapsed() << std::endl; + + } + + void qr_decomp(std::vector& lmd, + std::vector& lme, + int Nk, + int Nm, + std::vector& Qt, + RealD Dsh, + int kmin, + int kmax) + { + int k = kmin-1; + RealD x; + + RealD Fden = 1.0/hypot(lmd[k]-Dsh,lme[k]); + RealD c = ( lmd[k] -Dsh) *Fden; + RealD s = -lme[k] *Fden; + + RealD tmpa1 = lmd[k]; + RealD tmpa2 = lmd[k+1]; + RealD tmpb = lme[k]; + + lmd[k] = c*c*tmpa1 +s*s*tmpa2 -2.0*c*s*tmpb; + lmd[k+1] = s*s*tmpa1 +c*c*tmpa2 +2.0*c*s*tmpb; + lme[k] = c*s*(tmpa1-tmpa2) +(c*c-s*s)*tmpb; + x =-s*lme[k+1]; + lme[k+1] = c*lme[k+1]; + + for(int i=0; i& lmd, + std::vector& lme, + int N1, + int N2, + std::vector& Qt, + GridBase *grid){ + + std::cout << GridLogMessage << "diagonalize_lapack start\n"; + GridStopWatch gsw; + + const int size = Nm; + // tevals.resize(size); + // tevecs.resize(size); + LAPACK_INT NN = N1; + std::vector evals_tmp(NN); + std::vector evec_tmp(NN*NN); + memset(&evec_tmp[0],0,sizeof(double)*NN*NN); + // double AA[NN][NN]; + std::vector DD(NN); + std::vector EE(NN); + for (int i = 0; i< NN; i++) + for (int j = i - 1; j <= i + 1; j++) + if ( j < NN && j >= 0 ) { + if (i==j) DD[i] = lmd[i]; + if (i==j) evals_tmp[i] = lmd[i]; + if (j==(i-1)) EE[j] = lme[j]; + } + LAPACK_INT evals_found; + LAPACK_INT lwork = ( (18*NN) > (1+4*NN+NN*NN)? (18*NN):(1+4*NN+NN*NN)) ; + LAPACK_INT liwork = 3+NN*10 ; + std::vector iwork(liwork); + std::vector work(lwork); + std::vector isuppz(2*NN); + char jobz = 'V'; // calculate evals & evecs + char range = 'I'; // calculate all evals + // char range = 'A'; // calculate all evals + char uplo = 'U'; // refer to upper half of original matrix + char compz = 'I'; // Compute eigenvectors of tridiagonal matrix + std::vector ifail(NN); + LAPACK_INT info; + // int total = QMP_get_number_of_nodes(); + // int node = QMP_get_node_number(); + // GridBase *grid = evec[0]._grid; + int total = grid->_Nprocessors; + int node = grid->_processor; + int interval = (NN/total)+1; + double vl = 0.0, vu = 0.0; + LAPACK_INT il = interval*node+1 , iu = interval*(node+1); + if (iu > NN) iu=NN; + double tol = 0.0; + if (1) { + memset(&evals_tmp[0],0,sizeof(double)*NN); + if ( il <= NN){ + std::cout << GridLogMessage << "dstegr started" << std::endl; + gsw.Start(); + dstegr(&jobz, &range, &NN, + (double*)&DD[0], (double*)&EE[0], + &vl, &vu, &il, &iu, // these four are ignored if second parameteris 'A' + &tol, // tolerance + &evals_found, &evals_tmp[0], (double*)&evec_tmp[0], &NN, + &isuppz[0], + &work[0], &lwork, &iwork[0], &liwork, + &info); + gsw.Stop(); + std::cout << GridLogMessage << "dstegr completed in " << gsw.Elapsed() << std::endl; + for (int i = iu-1; i>= il-1; i--){ + evals_tmp[i] = evals_tmp[i - (il-1)]; + if (il>1) evals_tmp[i-(il-1)]=0.; + for (int j = 0; j< NN; j++){ + evec_tmp[i*NN + j] = evec_tmp[(i - (il-1)) * NN + j]; + if (il>1) evec_tmp[(i-(il-1)) * NN + j]=0.; + } + } + } + { + // QMP_sum_double_array(evals_tmp,NN); + // QMP_sum_double_array((double *)evec_tmp,NN*NN); + grid->GlobalSumVector(&evals_tmp[0],NN); + grid->GlobalSumVector(&evec_tmp[0],NN*NN); + } + } + // cheating a bit. It is better to sort instead of just reversing it, but the document of the routine says evals are sorted in increasing order. qr gives evals in decreasing order. + for(int i=0;i& lmd, + std::vector& lme, + int N2, + int N1, + std::vector& Qt, + GridBase *grid) + { + +#ifdef USE_LAPACK_IRL + const int check_lapack=0; // just use lapack if 0, check against lapack if 1 + + if(!check_lapack) + return diagonalize_lapack(lmd,lme,N2,N1,Qt,grid); + + std::vector lmd2(N1); + std::vector lme2(N1); + std::vector Qt2(N1*N1); + for(int k=0; k= kmin; --j){ + RealD dds = fabs(lmd[j-1])+fabs(lmd[j]); + if(fabs(lme[j-1])+dds > dds){ + kmax = j+1; + goto continued; + } + } + Niter = iter; +#ifdef USE_LAPACK_IRL + if(check_lapack){ + const double SMALL=1e-8; + diagonalize_lapack(lmd2,lme2,N2,N1,Qt2,grid); + std::vector lmd3(N2); + for(int k=0; kSMALL) std::cout<SMALL) std::cout<SMALL) std::cout< dds){ + kmin = j+1; + break; + } + } + } + std::cout< + static RealD normalise(T& v) + { + RealD nn = norm2(v); + nn = sqrt(nn); + v = v * (1.0/nn); + return nn; + } + + void orthogonalize(Field& w, + BasisFieldVector& evec, + int k) + { + double t0=-usecond()/1e6; + + evec.orthogonalize(w,k); + + normalise(w); + t0+=usecond()/1e6; + OrthoTime +=t0; + } + + void setUnit_Qt(int Nm, std::vector &Qt) { + for(int i=0; i K P = M − K † +Compute the factorization AVM = VM HM + fM eM +repeat + Q=I + for i = 1,...,P do + QiRi =HM −θiI Q = QQi + H M = Q †i H M Q i + end for + βK =HM(K+1,K) σK =Q(M,K) + r=vK+1βK +rσK + VK =VM(1:M)Q(1:M,1:K) + HK =HM(1:K,1:K) + →AVK =VKHK +fKe†K † Extend to an M = K + P step factorization AVM = VMHM + fMeM +until convergence +*/ + + void calc(std::vector& eval, + BasisFieldVector& evec, + const Field& src, + int& Nconv, + bool reverse, + int SkipTest) + { + + GridBase *grid = evec._v[0]._grid;//evec.get(0 + evec_offset)._grid; + assert(grid == src._grid); + + std::cout< lme(Nm); + std::vector lme2(Nm); + std::vector eval2(Nm); + std::vector eval2_copy(Nm); + std::vector Qt(Nm*Nm); + + + Field f(grid); + Field v(grid); + + int k1 = 1; + int k2 = Nk; + + Nconv = 0; + + RealD beta_k; + + // Set initial vector + evec[0] = src; + normalise(evec[0]); + std:: cout<0); + evec.rotate(Qt,k1-1,k2+1,0,Nm,Nm); + + t1=usecond()/1e6; + std::cout<= Nminres) { + std::cout << GridLogMessage << "Rotation to test convergence " << std::endl; + + Field ev0_orig(grid); + ev0_orig = evec[0]; + + evec.rotate(Qt,0,Nk,0,Nk,Nm); + + { + std::cout << GridLogMessage << "Test convergence" << std::endl; + Field B(grid); + + for(int j = 0; j=Nstop || beta_k < betastp){ + goto converged; + } + + std::cout << GridLogMessage << "Rotate back" << std::endl; + //B[j] +=Qt[k+_Nm*j] * _v[k]._odata[ss]; + { + Eigen::MatrixXd qm = Eigen::MatrixXd::Zero(Nk,Nk); + for (int k=0;k QtI(Nm*Nm); + for (int k=0;k +class BlockProjector { +public: + + BasisFieldVector& _evec; + BlockedGrid& _bgrid; + + BlockProjector(BasisFieldVector& evec, BlockedGrid& bgrid) : _evec(evec), _bgrid(bgrid) { + } + + void createOrthonormalBasis(RealD thres = 0.0) { + + GridStopWatch sw; + sw.Start(); + + int cnt = 0; + +#pragma omp parallel shared(cnt) + { + int lcnt = 0; + +#pragma omp for + for (int b=0;b<_bgrid._o_blocks;b++) { + + for (int i=0;i<_evec._Nm;i++) { + + auto nrm0 = _bgrid.block_sp(b,_evec._v[i],_evec._v[i]); + + // |i> -= |j> + for (int j=0;j + void coarseToFine(const CoarseField& in, Field& out) { + + out = zero; + out.checkerboard = _evec._v[0].checkerboard; + + int Nbasis = sizeof(in._odata[0]._internal._internal) / sizeof(in._odata[0]._internal._internal[0]); + assert(Nbasis == _evec._Nm); + +#pragma omp parallel for + for (int b=0;b<_bgrid._o_blocks;b++) { + for (int j=0;j<_evec._Nm;j++) { + _bgrid.block_caxpy(b,out,in._odata[b]._internal._internal[j],_evec._v[j],out); + } + } + + } + + template + void fineToCoarse(const Field& in, CoarseField& out) { + + out = zero; + + int Nbasis = sizeof(out._odata[0]._internal._internal) / sizeof(out._odata[0]._internal._internal[0]); + assert(Nbasis == _evec._Nm); + + + Field tmp(_bgrid._grid); + tmp = in; + +#pragma omp parallel for + for (int b=0;b<_bgrid._o_blocks;b++) { + for (int j=0;j<_evec._Nm;j++) { + // |rhs> -= |j> + auto c = _bgrid.block_sp(b,_evec._v[j],tmp); + _bgrid.block_caxpy(b,tmp,-c,_evec._v[j],tmp); // may make this more numerically stable + out._odata[b]._internal._internal[j] = c; + } + } + + } + + template + void deflateFine(BasisFieldVector& _coef,const std::vector& eval,int N,const Field& src_orig,Field& result) { + result = zero; + for (int i=0;i + void deflateCoarse(BasisFieldVector& _coef,const std::vector& eval,int N,const Field& src_orig,Field& result) { + CoarseField src_coarse(_coef._v[0]._grid); + CoarseField result_coarse = src_coarse; + result_coarse = zero; + fineToCoarse(src_orig,src_coarse); + for (int i=0;i + void deflate(BasisFieldVector& _coef,const std::vector& eval,int N,const Field& src_orig,Field& result) { + // Deflation on coarse Grid is much faster, so use it by default. Deflation on fine Grid is kept for legacy reasons for now. + deflateCoarse(_coef,eval,N,src_orig,result); + } + +}; +} diff --git a/lib/algorithms/iterative/BlockImplicitlyRestartedLanczos/BlockedGrid.h b/lib/algorithms/iterative/BlockImplicitlyRestartedLanczos/BlockedGrid.h new file mode 100644 index 00000000..821272de --- /dev/null +++ b/lib/algorithms/iterative/BlockImplicitlyRestartedLanczos/BlockedGrid.h @@ -0,0 +1,401 @@ +namespace Grid { + +template +class BlockedGrid { +public: + GridBase* _grid; + typedef typename Field::scalar_type Coeff_t; + typedef typename Field::vector_type vCoeff_t; + + std::vector _bs; // block size + std::vector _nb; // number of blocks + std::vector _l; // local dimensions irrespective of cb + std::vector _l_cb; // local dimensions of checkerboarded vector + std::vector _l_cb_o; // local dimensions of inner checkerboarded vector + std::vector _bs_cb; // block size in checkerboarded vector + std::vector _nb_o; // number of blocks of simd o-sites + + int _nd, _blocks, _cf_size, _cf_block_size, _cf_o_block_size, _o_blocks, _block_sites; + + BlockedGrid(GridBase* grid, const std::vector& block_size) : + _grid(grid), _bs(block_size), _nd((int)_bs.size()), + _nb(block_size), _l(block_size), _l_cb(block_size), _nb_o(block_size), + _l_cb_o(block_size), _bs_cb(block_size) { + + _blocks = 1; + _o_blocks = 1; + _l = grid->FullDimensions(); + _l_cb = grid->LocalDimensions(); + _l_cb_o = grid->_rdimensions; + + _cf_size = 1; + _block_sites = 1; + for (int i=0;i<_nd;i++) { + _l[i] /= grid->_processors[i]; + + assert(!(_l[i] % _bs[i])); // lattice must accommodate choice of blocksize + + int r = _l[i] / _l_cb[i]; + assert(!(_bs[i] % r)); // checkerboarding must accommodate choice of blocksize + _bs_cb[i] = _bs[i] / r; + _block_sites *= _bs_cb[i]; + _nb[i] = _l[i] / _bs[i]; + _nb_o[i] = _nb[i] / _grid->_simd_layout[i]; + if (_nb[i] % _grid->_simd_layout[i]) { // simd must accommodate choice of blocksize + std::cout << GridLogMessage << "Problem: _nb[" << i << "] = " << _nb[i] << " _grid->_simd_layout[" << i << "] = " << _grid->_simd_layout[i] << std::endl; + assert(0); + } + _blocks *= _nb[i]; + _o_blocks *= _nb_o[i]; + _cf_size *= _l[i]; + } + + _cf_size *= 12 / 2; + _cf_block_size = _cf_size / _blocks; + _cf_o_block_size = _cf_size / _o_blocks; + + std::cout << GridLogMessage << "BlockedGrid:" << std::endl; + std::cout << GridLogMessage << " _l = " << _l << std::endl; + std::cout << GridLogMessage << " _l_cb = " << _l_cb << std::endl; + std::cout << GridLogMessage << " _l_cb_o = " << _l_cb_o << std::endl; + std::cout << GridLogMessage << " _bs = " << _bs << std::endl; + std::cout << GridLogMessage << " _bs_cb = " << _bs_cb << std::endl; + + std::cout << GridLogMessage << " _nb = " << _nb << std::endl; + std::cout << GridLogMessage << " _nb_o = " << _nb_o << std::endl; + std::cout << GridLogMessage << " _blocks = " << _blocks << std::endl; + std::cout << GridLogMessage << " _o_blocks = " << _o_blocks << std::endl; + std::cout << GridLogMessage << " sizeof(vCoeff_t) = " << sizeof(vCoeff_t) << std::endl; + std::cout << GridLogMessage << " _cf_size = " << _cf_size << std::endl; + std::cout << GridLogMessage << " _cf_block_size = " << _cf_block_size << std::endl; + std::cout << GridLogMessage << " _block_sites = " << _block_sites << std::endl; + std::cout << GridLogMessage << " _grid->oSites() = " << _grid->oSites() << std::endl; + + // _grid->Barrier(); + //abort(); + } + + void block_to_coor(int b, std::vector& x0) { + + std::vector bcoor; + bcoor.resize(_nd); + x0.resize(_nd); + assert(b < _o_blocks); + Lexicographic::CoorFromIndex(bcoor,b,_nb_o); + int i; + + for (i=0;i<_nd;i++) { + x0[i] = bcoor[i]*_bs_cb[i]; + } + + //std::cout << GridLogMessage << "Map block b -> " << x0 << std::endl; + + } + + void block_site_to_o_coor(const std::vector& x0, std::vector& coor, int i) { + Lexicographic::CoorFromIndex(coor,i,_bs_cb); + for (int j=0;j<_nd;j++) + coor[j] += x0[j]; + } + + int block_site_to_o_site(const std::vector& x0, int i) { + std::vector coor; coor.resize(_nd); + block_site_to_o_coor(x0,coor,i); + Lexicographic::IndexFromCoor(coor,i,_l_cb_o); + return i; + } + + vCoeff_t block_sp(int b, const Field& x, const Field& y) { + + std::vector x0; + block_to_coor(b,x0); + + vCoeff_t ret = 0.0; + for (int i=0;i<_block_sites;i++) { // only odd sites + int ss = block_site_to_o_site(x0,i); + ret += TensorRemove(innerProduct(x._odata[ss],y._odata[ss])); + } + + return ret; + + } + + vCoeff_t block_sp(int b, const Field& x, const std::vector< ComplexD >& y) { + + std::vector x0; + block_to_coor(b,x0); + + constexpr int nsimd = sizeof(vCoeff_t) / sizeof(Coeff_t); + int lsize = _cf_o_block_size / _block_sites; + + std::vector< ComplexD > ret(nsimd); + for (int i=0;i + void vcaxpy(iScalar& r,const vCoeff_t& a,const iScalar& x,const iScalar& y) { + vcaxpy(r._internal,a,x._internal,y._internal); + } + + template + void vcaxpy(iVector& r,const vCoeff_t& a,const iVector& x,const iVector& y) { + for (int i=0;i x0; + block_to_coor(b,x0); + + for (int i=0;i<_block_sites;i++) { // only odd sites + int ss = block_site_to_o_site(x0,i); + vcaxpy(ret._odata[ss],a,x._odata[ss],y._odata[ss]); + } + + } + + void block_caxpy(int b, std::vector< ComplexD >& ret, const vCoeff_t& a, const Field& x, const std::vector< ComplexD >& y) { + std::vector x0; + block_to_coor(b,x0); + + constexpr int nsimd = sizeof(vCoeff_t) / sizeof(Coeff_t); + int lsize = _cf_o_block_size / _block_sites; + + for (int i=0;i<_block_sites;i++) { // only odd sites + int ss = block_site_to_o_site(x0,i); + + int n = lsize / nsimd; + for (int l=0;l& x) { + std::vector x0; + block_to_coor(b,x0); + + int lsize = _cf_o_block_size / _block_sites; + + for (int i=0;i<_block_sites;i++) { // only odd sites + int ss = block_site_to_o_site(x0,i); + + for (int l=0;l& x) { + std::vector x0; + block_to_coor(b,x0); + + int lsize = _cf_o_block_size / _block_sites; + + for (int i=0;i<_block_sites;i++) { // only odd sites + int ss = block_site_to_o_site(x0,i); + + for (int l=0;l + void vcscale(iScalar& r,const vCoeff_t& a,const iScalar& x) { + vcscale(r._internal,a,x._internal); + } + + template + void vcscale(iVector& r,const vCoeff_t& a,const iVector& x) { + for (int i=0;i x0; + block_to_coor(b,x0); + + for (int i=0;i<_block_sites;i++) { // only odd sites + int ss = block_site_to_o_site(x0,i); + vcscale(ret._odata[ss],a,ret._odata[ss]); + } + } + + void getCanonicalBlockOffset(int cb, std::vector& x0) { + const int ndim = 5; + assert(_nb.size() == ndim); + std::vector _nbc = { _nb[1], _nb[2], _nb[3], _nb[4], _nb[0] }; + std::vector _bsc = { _bs[1], _bs[2], _bs[3], _bs[4], _bs[0] }; + x0.resize(ndim); + + assert(cb >= 0); + assert(cb < _nbc[0]*_nbc[1]*_nbc[2]*_nbc[3]*_nbc[4]); + + Lexicographic::CoorFromIndex(x0,cb,_nbc); + int i; + + for (i=0;i& buf) { + std::vector _bsc = { _bs[1], _bs[2], _bs[3], _bs[4], _bs[0] }; + std::vector ldim = v._grid->LocalDimensions(); + std::vector cldim = { ldim[1], ldim[2], ldim[3], ldim[4], ldim[0] }; + const int _nbsc = _bs_cb[0]*_bs_cb[1]*_bs_cb[2]*_bs_cb[3]*_bs_cb[4]; + // take canonical block cb of v and put it in canonical ordering in buf + std::vector cx0; + getCanonicalBlockOffset(cb,cx0); + +#pragma omp parallel + { + std::vector co0,cl0; + co0=cx0; cl0=cx0; + +#pragma omp for + for (int i=0;i<_nbsc;i++) { + Lexicographic::CoorFromIndex(co0,2*i,_bsc); // 2* for eo + for (int j=0;j<(int)_bsc.size();j++) + cl0[j] = cx0[j] + co0[j]; + + std::vector l0 = { cl0[4], cl0[0], cl0[1], cl0[2], cl0[3] }; + int oi = v._grid->oIndex(l0); + int ii = v._grid->iIndex(l0); + int lti = i; + + //if (cb < 2 && i<2) + // std::cout << GridLogMessage << "Map: " << cb << ", " << i << " To: " << cl0 << ", " << cx0 << ", " << oi << ", " << ii << std::endl; + + for (int s=0;s<4;s++) + for (int c=0;c<3;c++) { + Coeff_t& ld = ((Coeff_t*)&v._odata[oi]._internal._internal[s]._internal[c])[ii]; + int ti = 12*lti + 3*s + c; + ld = Coeff_t(buf[2*ti+0], buf[2*ti+1]); + } + } + } + } + + void peekBlockOfVectorCanonical(int cb,const Field& v,std::vector& buf) { + std::vector _bsc = { _bs[1], _bs[2], _bs[3], _bs[4], _bs[0] }; + std::vector ldim = v._grid->LocalDimensions(); + std::vector cldim = { ldim[1], ldim[2], ldim[3], ldim[4], ldim[0] }; + const int _nbsc = _bs_cb[0]*_bs_cb[1]*_bs_cb[2]*_bs_cb[3]*_bs_cb[4]; + // take canonical block cb of v and put it in canonical ordering in buf + std::vector cx0; + getCanonicalBlockOffset(cb,cx0); + + buf.resize(_cf_block_size * 2); + +#pragma omp parallel + { + std::vector co0,cl0; + co0=cx0; cl0=cx0; + +#pragma omp for + for (int i=0;i<_nbsc;i++) { + Lexicographic::CoorFromIndex(co0,2*i,_bsc); // 2* for eo + for (int j=0;j<(int)_bsc.size();j++) + cl0[j] = cx0[j] + co0[j]; + + std::vector l0 = { cl0[4], cl0[0], cl0[1], cl0[2], cl0[3] }; + int oi = v._grid->oIndex(l0); + int ii = v._grid->iIndex(l0); + int lti = i; + + //if (cb < 2 && i<2) + // std::cout << GridLogMessage << "Map: " << cb << ", " << i << " To: " << cl0 << ", " << cx0 << ", " << oi << ", " << ii << std::endl; + + for (int s=0;s<4;s++) + for (int c=0;c<3;c++) { + Coeff_t& ld = ((Coeff_t*)&v._odata[oi]._internal._internal[s]._internal[c])[ii]; + int ti = 12*lti + 3*s + c; + buf[2*ti+0] = ld.real(); + buf[2*ti+1] = ld.imag(); + } + } + } + } + + int globalToLocalCanonicalBlock(int slot,const std::vector& src_nodes,int nb) { + // processor coordinate + int _nd = (int)src_nodes.size(); + std::vector _src_nodes = src_nodes; + std::vector pco(_nd); + Lexicographic::CoorFromIndex(pco,slot,_src_nodes); + std::vector cpco = { pco[1], pco[2], pco[3], pco[4], pco[0] }; + + // get local block + std::vector _nbc = { _nb[1], _nb[2], _nb[3], _nb[4], _nb[0] }; + assert(_nd == 5); + std::vector c_src_local_blocks(_nd); + for (int i=0;i<_nd;i++) { + assert(_grid->_fdimensions[i] % (src_nodes[i] * _bs[i]) == 0); + c_src_local_blocks[(i+4) % 5] = _grid->_fdimensions[i] / src_nodes[i] / _bs[i]; + } + std::vector cbcoor(_nd); // coordinate of block in slot in canonical form + Lexicographic::CoorFromIndex(cbcoor,nb,c_src_local_blocks); + + // cpco, cbcoor + std::vector clbcoor(_nd); + for (int i=0;i<_nd;i++) { + int cgcoor = cpco[i] * c_src_local_blocks[i] + cbcoor[i]; // global block coordinate + int pcoor = cgcoor / _nbc[i]; // processor coordinate in my Grid + int tpcoor = _grid->_processor_coor[(i+1)%5]; + if (pcoor != tpcoor) + return -1; + clbcoor[i] = cgcoor - tpcoor * _nbc[i]; // canonical local block coordinate for canonical dimension i + } + + int lnb; + Lexicographic::IndexFromCoor(clbcoor,lnb,_nbc); + //std::cout << "Mapped slot = " << slot << " nb = " << nb << " to " << lnb << std::endl; + return lnb; + } + + + }; + +} diff --git a/lib/algorithms/iterative/BlockImplicitlyRestartedLanczos/FieldBasisVector.h b/lib/algorithms/iterative/BlockImplicitlyRestartedLanczos/FieldBasisVector.h new file mode 100644 index 00000000..e715fc25 --- /dev/null +++ b/lib/algorithms/iterative/BlockImplicitlyRestartedLanczos/FieldBasisVector.h @@ -0,0 +1,163 @@ +namespace Grid { + +template +class BasisFieldVector { + public: + int _Nm; + + typedef typename Field::scalar_type Coeff_t; + typedef typename Field::vector_type vCoeff_t; + typedef typename Field::vector_object vobj; + typedef typename vobj::scalar_object sobj; + + std::vector _v; // _Nfull vectors + + void report(int n,GridBase* value) { + + std::cout << GridLogMessage << "BasisFieldVector allocated:\n"; + std::cout << GridLogMessage << " Delta N = " << n << "\n"; + std::cout << GridLogMessage << " Size of full vectors (size) = " << + ((double)n*sizeof(vobj)*value->oSites() / 1024./1024./1024.) << " GB\n"; + std::cout << GridLogMessage << " Size = " << _v.size() << " Capacity = " << _v.capacity() << std::endl; + + value->Barrier(); + + if (value->IsBoss()) { + system("cat /proc/meminfo"); + } + + value->Barrier(); + + } + + BasisFieldVector(int Nm,GridBase* value) : _Nm(Nm), _v(Nm,value) { + report(Nm,value); + } + + ~BasisFieldVector() { + } + + Field& operator[](int i) { + return _v[i]; + } + + void orthogonalize(Field& w, int k) { + for(int j=0; j& Qt,int j0, int j1, int k0,int k1,int Nm) { + + GridBase* grid = _v[0]._grid; + +#pragma omp parallel + { + std::vector < vobj > B(Nm); + +#pragma omp for + for(int ss=0;ss < grid->oSites();ss++){ + for(int j=j0; j _Nm) + _v.reserve(n); + + _v.resize(n,_v[0]._grid); + + if (n < _Nm) + _v.shrink_to_fit(); + + report(n - _Nm,_v[0]._grid); + + _Nm = n; + } + + std::vector getIndex(std::vector& sort_vals) { + + std::vector idx(sort_vals.size()); + iota(idx.begin(), idx.end(), 0); + + // sort indexes based on comparing values in v + sort(idx.begin(), idx.end(), + [&sort_vals](int i1, int i2) {return ::fabs(sort_vals[i1]) < ::fabs(sort_vals[i2]);}); + + return idx; + } + + void reorderInPlace(std::vector& sort_vals, std::vector& idx) { + GridStopWatch gsw; + gsw.Start(); + + int nswaps = 0; + for (size_t i=0;i& sort_vals, bool reverse) { + + std::vector idx = getIndex(sort_vals); + if (reverse) + std::reverse(idx.begin(), idx.end()); + + reorderInPlace(sort_vals,idx); + + } + + void deflate(const std::vector& eval,const Field& src_orig,Field& result) { + result = zero; + int N = (int)_v.size(); + for (int i=0;i { MaxIterations(maxit), ErrorOnNoConverge(err_on_no_conv){}; - void operator()(LinearOperatorBase &Linop, const Field &src, - Field &psi) { + void operator()(LinearOperatorBase &Linop, const Field &src, Field &psi) { + psi.checkerboard = src.checkerboard; conformable(psi, src); diff --git a/lib/algorithms/iterative/ConjugateGradientReliableUpdate.h b/lib/algorithms/iterative/ConjugateGradientReliableUpdate.h new file mode 100644 index 00000000..13666f97 --- /dev/null +++ b/lib/algorithms/iterative/ConjugateGradientReliableUpdate.h @@ -0,0 +1,256 @@ + /************************************************************************************* + + Grid physics library, www.github.com/paboyle/Grid + + Source file: ./lib/algorithms/iterative/ConjugateGradientReliableUpdate.h + + Copyright (C) 2015 + +Author: Christopher Kelly + + 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_RELIABLE_UPDATE_H +#define GRID_CONJUGATE_GRADIENT_RELIABLE_UPDATE_H + +namespace Grid { + + template::value == 2, int>::type = 0,typename std::enable_if< getPrecision::value == 1, int>::type = 0> + class ConjugateGradientReliableUpdate : public LinearFunction { + public: + bool ErrorOnNoConverge; // throw an assert when the CG fails to converge. + // Defaults true. + RealD Tolerance; + Integer MaxIterations; + Integer IterationsToComplete; //Number of iterations the CG took to finish. Filled in upon completion + Integer ReliableUpdatesPerformed; + + bool DoFinalCleanup; //Final DP cleanup, defaults to true + Integer IterationsToCleanup; //Final DP cleanup step iterations + + LinearOperatorBase &Linop_f; + LinearOperatorBase &Linop_d; + GridBase* SinglePrecGrid; + RealD Delta; //reliable update parameter + + //Optional ability to switch to a different linear operator once the tolerance reaches a certain point. Useful for single/half -> single/single + LinearOperatorBase *Linop_fallback; + RealD fallback_transition_tol; + + + ConjugateGradientReliableUpdate(RealD tol, Integer maxit, RealD _delta, GridBase* _sp_grid, LinearOperatorBase &_Linop_f, LinearOperatorBase &_Linop_d, bool err_on_no_conv = true) + : Tolerance(tol), + MaxIterations(maxit), + Delta(_delta), + Linop_f(_Linop_f), + Linop_d(_Linop_d), + SinglePrecGrid(_sp_grid), + ErrorOnNoConverge(err_on_no_conv), + DoFinalCleanup(true), + Linop_fallback(NULL) + {}; + + void setFallbackLinop(LinearOperatorBase &_Linop_fallback, const RealD _fallback_transition_tol){ + Linop_fallback = &_Linop_fallback; + fallback_transition_tol = _fallback_transition_tol; + } + + void operator()(const FieldD &src, FieldD &psi) { + LinearOperatorBase *Linop_f_use = &Linop_f; + bool using_fallback = false; + + psi.checkerboard = src.checkerboard; + conformable(psi, src); + + RealD cp, c, a, d, b, ssq, qq, b_pred; + + FieldD p(src); + FieldD mmp(src); + FieldD r(src); + + // Initial residual computation & set up + RealD guess = norm2(psi); + assert(std::isnan(guess) == 0); + + Linop_d.HermOpAndNorm(psi, mmp, d, b); + + r = src - mmp; + p = r; + + a = norm2(p); + cp = a; + ssq = norm2(src); + + std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: guess " << guess << std::endl; + std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: src " << ssq << std::endl; + std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: mp " << d << std::endl; + std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: mmp " << b << std::endl; + std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: cp,r " << cp << std::endl; + std::cout << GridLogIterative << std::setprecision(4) << "ConjugateGradientReliableUpdate: p " << a << std::endl; + + RealD rsq = Tolerance * Tolerance * ssq; + + // Check if guess is really REALLY good :) + if (cp <= rsq) { + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate guess was REALLY good\n"; + std::cout << GridLogMessage << "\tComputed residual " << sqrt(cp / ssq)<HermOpAndNorm(p_f, mmp_f, d, qq); + MatrixTimer.Stop(); + + LinalgTimer.Start(); + + a = c / d; + b_pred = a * (a * qq - d) / c; + + cp = axpy_norm(r_f, -a, mmp_f, r_f); + b = cp / c; + + // Fuse these loops ; should be really easy + psi_f = a * p_f + psi_f; + //p_f = p_f * b + r_f; + + LinalgTimer.Stop(); + + std::cout << GridLogIterative << "ConjugateGradientReliableUpdate: Iteration " << k + << " residual " << cp << " target " << rsq << std::endl; + std::cout << GridLogDebug << "a = "<< a << " b_pred = "<< b_pred << " b = "<< b << std::endl; + std::cout << GridLogDebug << "qq = "<< qq << " d = "<< d << " c = "<< c << std::endl; + + if(cp > MaxResidSinceLastRelUp){ + std::cout << GridLogIterative << "ConjugateGradientReliableUpdate: updating MaxResidSinceLastRelUp : " << MaxResidSinceLastRelUp << " -> " << cp << std::endl; + MaxResidSinceLastRelUp = cp; + } + + // Stopping condition + if (cp <= rsq) { + //Although not written in the paper, I assume that I have to add on the final solution + precisionChange(mmp, psi_f); + psi = psi + mmp; + + + SolverTimer.Stop(); + Linop_d.HermOpAndNorm(psi, mmp, d, qq); + p = mmp - src; + + RealD srcnorm = sqrt(norm2(src)); + RealD resnorm = sqrt(norm2(p)); + RealD true_residual = resnorm / srcnorm; + + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate Converged on iteration " << k << " after " << l << " reliable updates" << std::endl; + std::cout << GridLogMessage << "\tComputed residual " << sqrt(cp / ssq)< CG(Tolerance,MaxIterations); + CG.ErrorOnNoConverge = ErrorOnNoConverge; + CG(Linop_d,src,psi); + IterationsToCleanup = CG.IterationsToComplete; + } + else if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0); + + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate complete.\n"; + return; + } + else if(cp < Delta * MaxResidSinceLastRelUp) { //reliable update + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate " + << cp << "(residual) < " << Delta << "(Delta) * " << MaxResidSinceLastRelUp << "(MaxResidSinceLastRelUp) on iteration " << k << " : performing reliable update\n"; + precisionChange(mmp, psi_f); + psi = psi + mmp; + + Linop_d.HermOpAndNorm(psi, mmp, d, qq); + r = src - mmp; + + psi_f = zero; + precisionChange(r_f, r); + cp = norm2(r); + MaxResidSinceLastRelUp = cp; + + b = cp/c; + + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate new residual " << cp << std::endl; + + l = l+1; + } + + p_f = p_f * b + r_f; //update search vector after reliable update appears to help convergence + + if(!using_fallback && Linop_fallback != NULL && cp < fallback_transition_tol){ + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate switching to fallback linear operator on iteration " << k << " at residual " << cp << std::endl; + Linop_f_use = Linop_fallback; + using_fallback = true; + } + + + } + std::cout << GridLogMessage << "ConjugateGradientReliableUpdate did NOT converge" + << std::endl; + + if (ErrorOnNoConverge) assert(0); + IterationsToComplete = k; + ReliableUpdatesPerformed = l; + } + }; + + +}; + + + +#endif diff --git a/lib/algorithms/iterative/SchurRedBlack.h b/lib/algorithms/iterative/SchurRedBlack.h index 5caabb4b..a309386b 100644 --- a/lib/algorithms/iterative/SchurRedBlack.h +++ b/lib/algorithms/iterative/SchurRedBlack.h @@ -53,16 +53,110 @@ Author: Peter Boyle * M psi = eta *********************** *Odd - * i) (D_oo)^{\dag} D_oo psi_o = (D_oo)^dag L^{-1} eta_o + * i) D_oo psi_o = L^{-1} eta_o * eta_o' = (D_oo)^dag (eta_o - Moe Mee^{-1} eta_e) + * (D_oo)^{\dag} D_oo psi_o = (D_oo)^dag L^{-1} eta_o *Even * ii) Mee psi_e + Meo psi_o = src_e * * => sol_e = M_ee^-1 * ( src_e - Meo sol_o )... * + * + * TODO: Other options: + * + * a) change checkerboards for Schur e<->o + * + * Left precon by Moo^-1 + * b) Doo^{dag} M_oo^-dag Moo^-1 Doo psi_0 = (D_oo)^dag M_oo^-dag Moo^-1 L^{-1} eta_o + * eta_o' = (D_oo)^dag M_oo^-dag Moo^-1 (eta_o - Moe Mee^{-1} eta_e) + * + * Right precon by Moo^-1 + * c) M_oo^-dag Doo^{dag} Doo Moo^-1 phi_0 = M_oo^-dag (D_oo)^dag L^{-1} eta_o + * eta_o' = M_oo^-dag (D_oo)^dag (eta_o - Moe Mee^{-1} eta_e) + * psi_o = M_oo^-1 phi_o + * TODO: Deflation */ namespace Grid { + /////////////////////////////////////////////////////////////////////////////////////////////////////// + // Take a matrix and form a Red Black solver calling a Herm solver + // Use of RB info prevents making SchurRedBlackSolve conform to standard interface + /////////////////////////////////////////////////////////////////////////////////////////////////////// + + template class SchurRedBlackStaggeredSolve { + private: + OperatorFunction & _HermitianRBSolver; + int CBfactorise; + public: + + ///////////////////////////////////////////////////// + // Wrap the usual normal equations Schur trick + ///////////////////////////////////////////////////// + SchurRedBlackStaggeredSolve(OperatorFunction &HermitianRBSolver) : + _HermitianRBSolver(HermitianRBSolver) + { + CBfactorise=0; + }; + + template + void operator() (Matrix & _Matrix,const Field &in, Field &out){ + + // FIXME CGdiagonalMee not implemented virtual function + // FIXME use CBfactorise to control schur decomp + GridBase *grid = _Matrix.RedBlackGrid(); + GridBase *fgrid= _Matrix.Grid(); + + SchurStaggeredOperator _HermOpEO(_Matrix); + + Field src_e(grid); + Field src_o(grid); + Field sol_e(grid); + Field sol_o(grid); + Field tmp(grid); + Field Mtmp(grid); + Field resid(fgrid); + + pickCheckerboard(Even,src_e,in); + pickCheckerboard(Odd ,src_o,in); + pickCheckerboard(Even,sol_e,out); + pickCheckerboard(Odd ,sol_o,out); + + ///////////////////////////////////////////////////// + // src_o = Mdag * (source_o - Moe MeeInv source_e) + ///////////////////////////////////////////////////// + _Matrix.MooeeInv(src_e,tmp); assert( tmp.checkerboard ==Even); + _Matrix.Meooe (tmp,Mtmp); assert( Mtmp.checkerboard ==Odd); + tmp=src_o-Mtmp; assert( tmp.checkerboard ==Odd); + + _Matrix.Mooee(tmp,src_o); assert(src_o.checkerboard ==Odd); + + ////////////////////////////////////////////////////////////// + // Call the red-black solver + ////////////////////////////////////////////////////////////// + std::cout< using SchurRedBlackStagSolve = SchurRedBlackStaggeredSolve; + /////////////////////////////////////////////////////////////////////////////////////////////////////// // Take a matrix and form a Red Black solver calling a Herm solver // Use of RB info prevents making SchurRedBlackSolve conform to standard interface @@ -76,12 +170,10 @@ namespace Grid { ///////////////////////////////////////////////////// // Wrap the usual normal equations Schur trick ///////////////////////////////////////////////////// - SchurRedBlackDiagMooeeSolve(OperatorFunction &HermitianRBSolver) : - _HermitianRBSolver(HermitianRBSolver) - { - CBfactorise=0; - }; - + SchurRedBlackDiagMooeeSolve(OperatorFunction &HermitianRBSolver,int cb=0) : _HermitianRBSolver(HermitianRBSolver) + { + CBfactorise=cb; + }; template void operator() (Matrix & _Matrix,const Field &in, Field &out){ @@ -141,5 +233,166 @@ namespace Grid { } }; + + /////////////////////////////////////////////////////////////////////////////////////////////////////// + // Take a matrix and form a Red Black solver calling a Herm solver + // Use of RB info prevents making SchurRedBlackSolve conform to standard interface + /////////////////////////////////////////////////////////////////////////////////////////////////////// + template class SchurRedBlackDiagTwoSolve { + private: + OperatorFunction & _HermitianRBSolver; + int CBfactorise; + public: + + ///////////////////////////////////////////////////// + // Wrap the usual normal equations Schur trick + ///////////////////////////////////////////////////// + SchurRedBlackDiagTwoSolve(OperatorFunction &HermitianRBSolver) : + _HermitianRBSolver(HermitianRBSolver) + { + CBfactorise=0; + }; + + template + void operator() (Matrix & _Matrix,const Field &in, Field &out){ + + // FIXME CGdiagonalMee not implemented virtual function + // FIXME use CBfactorise to control schur decomp + GridBase *grid = _Matrix.RedBlackGrid(); + GridBase *fgrid= _Matrix.Grid(); + + SchurDiagTwoOperator _HermOpEO(_Matrix); + + Field src_e(grid); + Field src_o(grid); + Field sol_e(grid); + Field sol_o(grid); + Field tmp(grid); + Field Mtmp(grid); + Field resid(fgrid); + + pickCheckerboard(Even,src_e,in); + pickCheckerboard(Odd ,src_o,in); + pickCheckerboard(Even,sol_e,out); + pickCheckerboard(Odd ,sol_o,out); + + ///////////////////////////////////////////////////// + // src_o = Mdag * (source_o - Moe MeeInv source_e) + ///////////////////////////////////////////////////// + _Matrix.MooeeInv(src_e,tmp); assert( tmp.checkerboard ==Even); + _Matrix.Meooe (tmp,Mtmp); assert( Mtmp.checkerboard ==Odd); + tmp=src_o-Mtmp; assert( tmp.checkerboard ==Odd); + + // get the right MpcDag + _HermOpEO.MpcDag(tmp,src_o); assert(src_o.checkerboard ==Odd); + + ////////////////////////////////////////////////////////////// + // Call the red-black solver + ////////////////////////////////////////////////////////////// + std::cout< class SchurRedBlackDiagTwoMixed { + private: + LinearFunction & _HermitianRBSolver; + int CBfactorise; + public: + + ///////////////////////////////////////////////////// + // Wrap the usual normal equations Schur trick + ///////////////////////////////////////////////////// + SchurRedBlackDiagTwoMixed(LinearFunction &HermitianRBSolver) : + _HermitianRBSolver(HermitianRBSolver) + { + CBfactorise=0; + }; + + template + void operator() (Matrix & _Matrix,const Field &in, Field &out){ + + // FIXME CGdiagonalMee not implemented virtual function + // FIXME use CBfactorise to control schur decomp + GridBase *grid = _Matrix.RedBlackGrid(); + GridBase *fgrid= _Matrix.Grid(); + + SchurDiagTwoOperator _HermOpEO(_Matrix); + + Field src_e(grid); + Field src_o(grid); + Field sol_e(grid); + Field sol_o(grid); + Field tmp(grid); + Field Mtmp(grid); + Field resid(fgrid); + + pickCheckerboard(Even,src_e,in); + pickCheckerboard(Odd ,src_o,in); + pickCheckerboard(Even,sol_e,out); + pickCheckerboard(Odd ,sol_o,out); + + ///////////////////////////////////////////////////// + // src_o = Mdag * (source_o - Moe MeeInv source_e) + ///////////////////////////////////////////////////// + _Matrix.MooeeInv(src_e,tmp); assert( tmp.checkerboard ==Even); + _Matrix.Meooe (tmp,Mtmp); assert( Mtmp.checkerboard ==Odd); + tmp=src_o-Mtmp; assert( tmp.checkerboard ==Odd); + + // get the right MpcDag + _HermOpEO.MpcDag(tmp,src_o); assert(src_o.checkerboard ==Odd); + + ////////////////////////////////////////////////////////////// + // Call the red-black solver + ////////////////////////////////////////////////////////////// + std::cout< friend class Lattice; GridBase(const std::vector & processor_grid) : CartesianCommunicator(processor_grid) {}; + GridBase(const std::vector & processor_grid, + const CartesianCommunicator &parent) : CartesianCommunicator(processor_grid,parent) {}; // Physics Grid information. std::vector _simd_layout;// Which dimensions get relayed out over simd lanes. @@ -210,9 +212,6 @@ public: assert(lidx & gcoor,int & gidx){ gidx=0; int mult=1; diff --git a/lib/cartesian/Cartesian_full.h b/lib/cartesian/Cartesian_full.h index 815e3b22..a6a85ab7 100644 --- a/lib/cartesian/Cartesian_full.h +++ b/lib/cartesian/Cartesian_full.h @@ -61,9 +61,29 @@ public: virtual int CheckerBoardShift(int source_cb,int dim,int shift, int osite){ return shift; } + ///////////////////////////////////////////////////////////////////////// + // Constructor takes a parent grid and possibly subdivides communicator. + ///////////////////////////////////////////////////////////////////////// GridCartesian(const std::vector &dimensions, - const std::vector &simd_layout, - const std::vector &processor_grid) : GridBase(processor_grid) + const std::vector &simd_layout, + const std::vector &processor_grid, + const GridCartesian &parent) : GridBase(processor_grid,parent) + { + Init(dimensions,simd_layout,processor_grid); + } + ///////////////////////////////////////////////////////////////////////// + // Construct from comm world + ///////////////////////////////////////////////////////////////////////// + GridCartesian(const std::vector &dimensions, + const std::vector &simd_layout, + const std::vector &processor_grid) : GridBase(processor_grid) + { + Init(dimensions,simd_layout,processor_grid); + } + + void Init(const std::vector &dimensions, + const std::vector &simd_layout, + const std::vector &processor_grid) { /////////////////////// // Grid information diff --git a/lib/cartesian/Cartesian_red_black.h b/lib/cartesian/Cartesian_red_black.h index b1a5b9ef..f89cacc5 100644 --- a/lib/cartesian/Cartesian_red_black.h +++ b/lib/cartesian/Cartesian_red_black.h @@ -112,24 +112,57 @@ public: } }; - GridRedBlackCartesian(const GridBase *base) : GridRedBlackCartesian(base->_fdimensions,base->_simd_layout,base->_processors) {}; + //////////////////////////////////////////////////////////// + // Create Redblack from original grid; require full grid pointer ? + //////////////////////////////////////////////////////////// + GridRedBlackCartesian(const GridBase *base) : GridBase(base->_processors,*base) + { + int dims = base->_ndimension; + std::vector checker_dim_mask(dims,1); + int checker_dim = 0; + Init(base->_fdimensions,base->_simd_layout,base->_processors,checker_dim_mask,checker_dim); + }; - GridRedBlackCartesian(const std::vector &dimensions, + //////////////////////////////////////////////////////////// + // Create redblack from original grid, with non-trivial checker dim mask + //////////////////////////////////////////////////////////// + GridRedBlackCartesian(const GridBase *base, + const std::vector &checker_dim_mask, + int checker_dim + ) : GridBase(base->_processors,*base) + { + Init(base->_fdimensions,base->_simd_layout,base->_processors,checker_dim_mask,checker_dim) ; + } +#if 0 + //////////////////////////////////////////////////////////// + // Create redblack grid ;; deprecate these. Should not + // need direct creation of redblack without a full grid to base on + //////////////////////////////////////////////////////////// + GridRedBlackCartesian(const GridBase *base, + const std::vector &dimensions, const std::vector &simd_layout, const std::vector &processor_grid, const std::vector &checker_dim_mask, int checker_dim - ) : GridBase(processor_grid) + ) : GridBase(processor_grid,*base) { Init(dimensions,simd_layout,processor_grid,checker_dim_mask,checker_dim); } - GridRedBlackCartesian(const std::vector &dimensions, + + //////////////////////////////////////////////////////////// + // Create redblack grid + //////////////////////////////////////////////////////////// + GridRedBlackCartesian(const GridBase *base, + const std::vector &dimensions, const std::vector &simd_layout, - const std::vector &processor_grid) : GridBase(processor_grid) + const std::vector &processor_grid) : GridBase(processor_grid,*base) { std::vector checker_dim_mask(dimensions.size(),1); - Init(dimensions,simd_layout,processor_grid,checker_dim_mask,0); + int checker_dim = 0; + Init(dimensions,simd_layout,processor_grid,checker_dim_mask,checker_dim); } +#endif + void Init(const std::vector &dimensions, const std::vector &simd_layout, const std::vector &processor_grid, diff --git a/lib/communicator/Communicator_base.cc b/lib/communicator/Communicator_base.cc index 956de0d2..ce9a3cf0 100644 --- a/lib/communicator/Communicator_base.cc +++ b/lib/communicator/Communicator_base.cc @@ -37,7 +37,7 @@ namespace Grid { // Info that is setup once and indept of cartesian layout /////////////////////////////////////////////////////////////// void * CartesianCommunicator::ShmCommBuf; -uint64_t CartesianCommunicator::MAX_MPI_SHM_BYTES = 128*1024*1024; +uint64_t CartesianCommunicator::MAX_MPI_SHM_BYTES = 1024LL*1024LL*1024LL; CartesianCommunicator::CommunicatorPolicy_t CartesianCommunicator::CommunicatorPolicy= CartesianCommunicator::CommunicatorPolicyConcurrent; int CartesianCommunicator::nCommThreads = -1; @@ -67,7 +67,7 @@ void CartesianCommunicator::ShmBufferFreeAll(void) { ///////////////////////////////// // Grid information queries ///////////////////////////////// -int CartesianCommunicator::Dimensions(void) { return _ndimension; }; +int CartesianCommunicator::Dimensions(void) { return _ndimension; }; int CartesianCommunicator::IsBoss(void) { return _processor==0; }; int CartesianCommunicator::BossRank(void) { return 0; }; int CartesianCommunicator::ThisRank(void) { return _processor; }; @@ -96,6 +96,113 @@ void CartesianCommunicator::GlobalSumVector(ComplexD *c,int N) GlobalSumVector((double *)c,2*N); } + +#if defined( GRID_COMMS_MPI) || defined (GRID_COMMS_MPIT) + +CartesianCommunicator::CartesianCommunicator(const std::vector &processors,const CartesianCommunicator &parent) +{ + _ndimension = processors.size(); + assert(_ndimension = parent._ndimension); + + ////////////////////////////////////////////////////////////////////////////////////////////////////// + // split the communicator + ////////////////////////////////////////////////////////////////////////////////////////////////////// + int Nparent; + MPI_Comm_size(parent.communicator,&Nparent); + + int childsize=1; + for(int d=0;d ccoor(_ndimension); // coor within subcommunicator + std::vector scoor(_ndimension); // coor of split within parent + std::vector ssize(_ndimension); // coor of split within parent + + for(int d=0;d<_ndimension;d++){ + ccoor[d] = parent._processor_coor[d] % processors[d]; + scoor[d] = parent._processor_coor[d] / processors[d]; + ssize[d] = parent._processors[d]/ processors[d]; + } + int crank,srank; // rank within subcomm ; rank of subcomm within blocks of subcomms + Lexicographic::IndexFromCoor(ccoor,crank,processors); + Lexicographic::IndexFromCoor(scoor,srank,ssize); + + MPI_Comm comm_split; + if ( Nchild > 1 ) { + + // std::cout << GridLogMessage<<"Child communicator of "<< std::hex << parent.communicator << std::dec< &processors, MPI_Comm communicator_base) +{ + // if ( communicator_base != communicator_world ) { + // std::cout << "Cartesian communicator created with a non-world communicator"< periodic(_ndimension,1); + MPI_Cart_create(communicator_base, _ndimension,&_processors[0],&periodic[0],1,&communicator); + MPI_Comm_rank(communicator,&_processor); + MPI_Cart_coords(communicator,_processor,_ndimension,&_processor_coor[0]); + + int Size; + MPI_Comm_size(communicator,&Size); + +#ifdef GRID_COMMS_MPIT + communicator_halo.resize (2*_ndimension); + for(int i=0;i<_ndimension*2;i++){ + MPI_Comm_dup(communicator,&communicator_halo[i]); + } +#endif + + assert(Size==_Nprocessors); +} + +CartesianCommunicator::CartesianCommunicator(const std::vector &processors) +{ + InitFromMPICommunicator(processors,communicator_world); +} + +#endif + #if !defined( GRID_COMMS_MPI3) int CartesianCommunicator::NodeCount(void) { return ProcessorCount();}; @@ -147,8 +254,13 @@ void *CartesianCommunicator::ShmBufferTranslate(int rank,void * local_p) { } void CartesianCommunicator::ShmInitGeneric(void){ #if 1 - - int mmap_flag = MAP_SHARED | MAP_ANONYMOUS; + int mmap_flag =0; +#ifdef MAP_ANONYMOUS + mmap_flag = mmap_flag| MAP_SHARED | MAP_ANONYMOUS; +#endif +#ifdef MAP_ANON + mmap_flag = mmap_flag| MAP_SHARED | MAP_ANON; +#endif #ifdef MAP_HUGETLB if ( Hugepages ) mmap_flag |= MAP_HUGETLB; #endif @@ -157,6 +269,9 @@ void CartesianCommunicator::ShmInitGeneric(void){ perror("mmap failed "); exit(EXIT_FAILURE); } +#ifdef MADV_HUGEPAGE + if (!Hugepages ) madvise(ShmCommBuf,MAX_MPI_SHM_BYTES,MADV_HUGEPAGE); +#endif #else ShmBufStorageVector.resize(MAX_MPI_SHM_BYTES); ShmCommBuf=(void *)&ShmBufStorageVector[0]; diff --git a/lib/communicator/Communicator_base.h b/lib/communicator/Communicator_base.h index ac866ced..8ff22dbd 100644 --- a/lib/communicator/Communicator_base.h +++ b/lib/communicator/Communicator_base.h @@ -83,6 +83,7 @@ class CartesianCommunicator { std::vector communicator_halo; typedef MPI_Request CommsRequest_t; + #else typedef int CommsRequest_t; #endif @@ -147,11 +148,23 @@ class CartesianCommunicator { // Must call in Grid startup //////////////////////////////////////////////// static void Init(int *argc, char ***argv); - + //////////////////////////////////////////////// - // Constructor of any given grid + // Constructors to sub-divide a parent communicator + // and default to comm world //////////////////////////////////////////////// + CartesianCommunicator(const std::vector &processors,const CartesianCommunicator &parent); CartesianCommunicator(const std::vector &pdimensions_in); + + private: +#if defined (GRID_COMMS_MPI) || defined (GRID_COMMS_MPIT) + //////////////////////////////////////////////// + // Private initialise from an MPI communicator + // Can use after an MPI_Comm_split, but hidden from user so private + //////////////////////////////////////////////// + void InitFromMPICommunicator(const std::vector &processors, MPI_Comm communicator_base); +#endif + public: //////////////////////////////////////////////////////////////////////////////////////// // Wraps MPI_Cart routines, or implements equivalent on other impls @@ -249,6 +262,23 @@ class CartesianCommunicator { // Broadcast a buffer and composite larger //////////////////////////////////////////////////////////// void Broadcast(int root,void* data, int bytes); + + //////////////////////////////////////////////////////////// + // All2All down one dimension + //////////////////////////////////////////////////////////// + template void AllToAll(int dim,std::vector &in, std::vector &out){ + assert(dim>=0); + assert(dim<_ndimension); + int numnode = _processors[dim]; + // std::cerr << " AllToAll in.size() "< void Broadcast(int root,obj &data) { diff --git a/lib/communicator/Communicator_mpi.cc b/lib/communicator/Communicator_mpi.cc index bd2a62fb..678e4517 100644 --- a/lib/communicator/Communicator_mpi.cc +++ b/lib/communicator/Communicator_mpi.cc @@ -52,29 +52,6 @@ void CartesianCommunicator::Init(int *argc, char ***argv) { MPI_Comm_dup (MPI_COMM_WORLD,&communicator_world); ShmInitGeneric(); } - -CartesianCommunicator::CartesianCommunicator(const std::vector &processors) -{ - _ndimension = processors.size(); - std::vector periodic(_ndimension,1); - - _Nprocessors=1; - _processors = processors; - _processor_coor.resize(_ndimension); - - MPI_Cart_create(communicator_world, _ndimension,&_processors[0],&periodic[0],1,&communicator); - MPI_Comm_rank(communicator,&_processor); - MPI_Cart_coords(communicator,_processor,_ndimension,&_processor_coor[0]); - - for(int i=0;i<_ndimension;i++){ - _Nprocessors*=_processors[i]; - } - - int Size; - MPI_Comm_size(communicator,&Size); - - assert(Size==_Nprocessors); -} void CartesianCommunicator::GlobalSum(uint32_t &u){ int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator); assert(ierr==0); @@ -210,6 +187,21 @@ void CartesianCommunicator::Broadcast(int root,void* data, int bytes) root, communicator); assert(ierr==0); +} +void CartesianCommunicator::AllToAll(int dim,void *in,void *out,int bytes) +{ + std::vector row(_ndimension,1); + assert(dim>=0 && dim<_ndimension); + + // Split the communicator + row[dim] = _processors[dim]; + + CartesianCommunicator Comm(row,*this); + Comm.AllToAll(in,out,bytes); +} +void CartesianCommunicator::AllToAll(void *in,void *out,int bytes) +{ + MPI_Alltoall(in ,bytes,MPI_BYTE,out,bytes,MPI_BYTE,communicator); } /////////////////////////////////////////////////////// // Should only be used prior to Grid Init finished. @@ -230,5 +222,7 @@ void CartesianCommunicator::BroadcastWorld(int root,void* data, int bytes) assert(ierr==0); } + + } diff --git a/lib/communicator/Communicator_mpi3.cc b/lib/communicator/Communicator_mpi3.cc index cb7fa390..dce9588a 100644 --- a/lib/communicator/Communicator_mpi3.cc +++ b/lib/communicator/Communicator_mpi3.cc @@ -198,7 +198,46 @@ void CartesianCommunicator::Init(int *argc, char ***argv) { ShmCommBuf = 0; ShmCommBufs.resize(ShmSize); -#if 1 + //////////////////////////////////////////////////////////////////////////////////////////// + // Hugetlbf and others map filesystems as mappable huge pages + //////////////////////////////////////////////////////////////////////////////////////////// +#ifdef GRID_MPI3_SHMMMAP + char shm_name [NAME_MAX]; + for(int r=0;r shmids(ShmSize); if ( ShmRank == 0 ) { for(int r=0;r &c assert(lr!=-1); Lexicographic::CoorFromIndex(coor,lr,_processors); } + +////////////////////////////////// +// Try to subdivide communicator +////////////////////////////////// +CartesianCommunicator::CartesianCommunicator(const std::vector &processors,const CartesianCommunicator &parent) + : CartesianCommunicator(processors) +{ + std::cout << "Attempts to split MPI3 communicators will fail until implemented" < &processors) { int ierr; diff --git a/lib/communicator/Communicator_mpit.cc b/lib/communicator/Communicator_mpit.cc index eb6ef87d..5137c27b 100644 --- a/lib/communicator/Communicator_mpit.cc +++ b/lib/communicator/Communicator_mpit.cc @@ -53,33 +53,6 @@ void CartesianCommunicator::Init(int *argc, char ***argv) { ShmInitGeneric(); } -CartesianCommunicator::CartesianCommunicator(const std::vector &processors) -{ - _ndimension = processors.size(); - std::vector periodic(_ndimension,1); - - _Nprocessors=1; - _processors = processors; - _processor_coor.resize(_ndimension); - - MPI_Cart_create(communicator_world, _ndimension,&_processors[0],&periodic[0],1,&communicator); - MPI_Comm_rank(communicator,&_processor); - MPI_Cart_coords(communicator,_processor,_ndimension,&_processor_coor[0]); - - for(int i=0;i<_ndimension;i++){ - _Nprocessors*=_processors[i]; - } - - communicator_halo.resize (2*_ndimension); - for(int i=0;i<_ndimension*2;i++){ - MPI_Comm_dup(communicator,&communicator_halo[i]); - } - - int Size; - MPI_Comm_size(communicator,&Size); - - assert(Size==_Nprocessors); -} void CartesianCommunicator::GlobalSum(uint32_t &u){ int ierr=MPI_Allreduce(MPI_IN_PLACE,&u,1,MPI_UINT32_T,MPI_SUM,communicator); assert(ierr==0); diff --git a/lib/communicator/Communicator_none.cc b/lib/communicator/Communicator_none.cc index 5319ab93..e9d71a15 100644 --- a/lib/communicator/Communicator_none.cc +++ b/lib/communicator/Communicator_none.cc @@ -38,6 +38,9 @@ void CartesianCommunicator::Init(int *argc, char *** arv) ShmInitGeneric(); } +CartesianCommunicator::CartesianCommunicator(const std::vector &processors,const CartesianCommunicator &parent) + : CartesianCommunicator(processors) {} + CartesianCommunicator::CartesianCommunicator(const std::vector &processors) { _processors = processors; @@ -95,6 +98,10 @@ void CartesianCommunicator::SendToRecvFromComplete(std::vector & { assert(0); } +void CartesianCommunicator::AllToAll(int dim,void *in,void *out,int bytes) +{ + bcopy(in,out,bytes); +} int CartesianCommunicator::RankWorld(void){return 0;} void CartesianCommunicator::Barrier(void){} diff --git a/lib/communicator/Communicator_shmem.cc b/lib/communicator/Communicator_shmem.cc index 3c76c808..ed49285d 100644 --- a/lib/communicator/Communicator_shmem.cc +++ b/lib/communicator/Communicator_shmem.cc @@ -75,6 +75,11 @@ void CartesianCommunicator::Init(int *argc, char ***argv) { ShmInitGeneric(); } +CartesianCommunicator::CartesianCommunicator(const std::vector &processors,const CartesianCommunicator &parent) + : CartesianCommunicator(processors) +{ + std::cout << "Attempts to split SHMEM communicators will fail " < &processors) { _ndimension = processors.size(); diff --git a/lib/json/json.hpp b/lib/json/json.hpp index d5dc111e..9d589120 100644 --- a/lib/json/json.hpp +++ b/lib/json/json.hpp @@ -1,7 +1,7 @@ /* __ _____ _____ _____ __| | __| | | | JSON for Modern C++ -| | |__ | | | | | | version 2.0.10 +| | |__ | | | | | | version 2.1.1 |_____|_____|_____|_|___| https://github.com/nlohmann/json Licensed under the MIT License . @@ -28,43 +28,42 @@ SOFTWARE. #ifndef NLOHMANN_JSON_HPP #define NLOHMANN_JSON_HPP -#include -#include // all_of, for_each, transform + +#include // all_of, copy, fill, find, for_each, generate_n, none_of, remove, reverse, transform #include // array #include // assert -#include // isdigit #include // and, not, or -#include // isfinite, ldexp, signbit +#include // lconv, localeconv +#include // isfinite, labs, ldexp, signbit #include // nullptr_t, ptrdiff_t, size_t #include // int64_t, uint64_t -#include // strtod, strtof, strtold, strtoul -#include // strlen +#include // abort, strtod, strtof, strtold, strtoul, strtoll, strtoull +#include // memcpy, strlen +#include // forward_list #include // function, hash, less #include // initializer_list -#include // setw -#include // istream, ostream -#include // advance, begin, bidirectional_iterator_tag, distance, end, inserter, iterator, iterator_traits, next, random_access_iterator_tag, reverse_iterator +#include // hex +#include // istream, ostream +#include // advance, begin, back_inserter, bidirectional_iterator_tag, distance, end, inserter, iterator, iterator_traits, next, random_access_iterator_tag, reverse_iterator #include // numeric_limits #include // locale #include // map #include // addressof, allocator, allocator_traits, unique_ptr #include // accumulate #include // stringstream -#include // domain_error, invalid_argument, out_of_range #include // getline, stoi, string, to_string -#include // add_pointer, enable_if, is_arithmetic, is_base_of, is_const, is_constructible, is_convertible, is_floating_point, is_integral, is_nothrow_move_assignable, std::is_nothrow_move_constructible, std::is_pointer, std::is_reference, std::is_same, remove_const, remove_pointer, remove_reference +#include // add_pointer, conditional, decay, enable_if, false_type, integral_constant, is_arithmetic, is_base_of, is_const, is_constructible, is_convertible, is_default_constructible, is_enum, is_floating_point, is_integral, is_nothrow_move_assignable, is_nothrow_move_constructible, is_pointer, is_reference, is_same, is_scalar, is_signed, remove_const, remove_cv, remove_pointer, remove_reference, true_type, underlying_type #include // declval, forward, make_pair, move, pair, swap +#include // valarray #include // vector // exclude unsupported compilers #if defined(__clang__) - #define CLANG_VERSION (__clang_major__ * 10000 + __clang_minor__ * 100 + __clang_patchlevel__) - #if CLANG_VERSION < 30400 + #if (__clang_major__ * 10000 + __clang_minor__ * 100 + __clang_patchlevel__) < 30400 #error "unsupported Clang version - see https://github.com/nlohmann/json#supported-compilers" #endif #elif defined(__GNUC__) - #define GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__) - #if GCC_VERSION < 40800 + #if (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__) < 40805 #error "unsupported GCC version - see https://github.com/nlohmann/json#supported-compilers" #endif #endif @@ -90,6 +89,26 @@ SOFTWARE. #define JSON_DEPRECATED #endif +// allow to disable exceptions +#if (defined(__cpp_exceptions) || defined(__EXCEPTIONS) || defined(_CPPUNWIND)) && not defined(JSON_NOEXCEPTION) + #define JSON_THROW(exception) throw exception + #define JSON_TRY try + #define JSON_CATCH(exception) catch(exception) +#else + #define JSON_THROW(exception) std::abort() + #define JSON_TRY if(true) + #define JSON_CATCH(exception) if(false) +#endif + +// manual branch prediction +#if defined(__clang__) || defined(__GNUC__) || defined(__GNUG__) + #define JSON_LIKELY(x) __builtin_expect(!!(x), 1) + #define JSON_UNLIKELY(x) __builtin_expect(!!(x), 0) +#else + #define JSON_LIKELY(x) x + #define JSON_UNLIKELY(x) x +#endif + /*! @brief namespace for Niels Lohmann @see https://github.com/nlohmann @@ -97,39 +116,7128 @@ SOFTWARE. */ namespace nlohmann { +template +struct adl_serializer; + +// forward declaration of basic_json (required to split the class) +template class ObjectType = + std::map, + template class ArrayType = std::vector, + class StringType = std::string, class BooleanType = bool, + class NumberIntegerType = std::int64_t, + class NumberUnsignedType = std::uint64_t, + class NumberFloatType = double, + template class AllocatorType = std::allocator, + template class JSONSerializer = + adl_serializer> +class basic_json; + +// Ugly macros to avoid uglier copy-paste when specializing basic_json +// This is only temporary and will be removed in 3.0 + +#define NLOHMANN_BASIC_JSON_TPL_DECLARATION \ + template class ObjectType, \ + template class ArrayType, \ + class StringType, class BooleanType, class NumberIntegerType, \ + class NumberUnsignedType, class NumberFloatType, \ + template class AllocatorType, \ + template class JSONSerializer> + +#define NLOHMANN_BASIC_JSON_TPL \ + basic_json /*! @brief unnamed namespace with internal helper functions + +This namespace collects some functions that could not be defined inside the +@ref basic_json class. + +@since version 2.1.0 +*/ +namespace detail +{ +//////////////// +// exceptions // +//////////////// + +/*! +@brief general exception of the @ref basic_json class + +This class is an extension of `std::exception` objects with a member @a id for +exception ids. It is used as the base class for all exceptions thrown by the +@ref basic_json class. This class can hence be used as "wildcard" to catch +exceptions. + +Subclasses: +- @ref parse_error for exceptions indicating a parse error +- @ref invalid_iterator for exceptions indicating errors with iterators +- @ref type_error for exceptions indicating executing a member function with + a wrong type +- @ref out_of_range for exceptions indicating access out of the defined range +- @ref other_error for exceptions indicating other library errors + +@internal +@note To have nothrow-copy-constructible exceptions, we internally use + `std::runtime_error` which can cope with arbitrary-length error messages. + Intermediate strings are built with static functions and then passed to + the actual constructor. +@endinternal + +@liveexample{The following code shows how arbitrary library exceptions can be +caught.,exception} + +@since version 3.0.0 +*/ +class exception : public std::exception +{ + public: + /// returns the explanatory string + const char* what() const noexcept override + { + return m.what(); + } + + /// the id of the exception + const int id; + + protected: + exception(int id_, const char* what_arg) : id(id_), m(what_arg) {} + + static std::string name(const std::string& ename, int id) + { + return "[json.exception." + ename + "." + std::to_string(id) + "] "; + } + + private: + /// an exception object as storage for error messages + std::runtime_error m; +}; + +/*! +@brief exception indicating a parse error + +This excpetion is thrown by the library when a parse error occurs. Parse errors +can occur during the deserialization of JSON text, CBOR, MessagePack, as well +as when using JSON Patch. + +Member @a byte holds the byte index of the last read character in the input +file. + +Exceptions have ids 1xx. + +name / id | example message | description +------------------------------ | --------------- | ------------------------- +json.exception.parse_error.101 | parse error at 2: unexpected end of input; expected string literal | This error indicates a syntax error while deserializing a JSON text. The error message describes that an unexpected token (character) was encountered, and the member @a byte indicates the error position. +json.exception.parse_error.102 | parse error at 14: missing or wrong low surrogate | JSON uses the `\uxxxx` format to describe Unicode characters. Code points above above 0xFFFF are split into two `\uxxxx` entries ("surrogate pairs"). This error indicates that the surrogate pair is incomplete or contains an invalid code point. +json.exception.parse_error.103 | parse error: code points above 0x10FFFF are invalid | Unicode supports code points up to 0x10FFFF. Code points above 0x10FFFF are invalid. +json.exception.parse_error.104 | parse error: JSON patch must be an array of objects | [RFC 6902](https://tools.ietf.org/html/rfc6902) requires a JSON Patch document to be a JSON document that represents an array of objects. +json.exception.parse_error.105 | parse error: operation must have string member 'op' | An operation of a JSON Patch document must contain exactly one "op" member, whose value indicates the operation to perform. Its value must be one of "add", "remove", "replace", "move", "copy", or "test"; other values are errors. +json.exception.parse_error.106 | parse error: array index '01' must not begin with '0' | An array index in a JSON Pointer ([RFC 6901](https://tools.ietf.org/html/rfc6901)) may be `0` or any number wihtout a leading `0`. +json.exception.parse_error.107 | parse error: JSON pointer must be empty or begin with '/' - was: 'foo' | A JSON Pointer must be a Unicode string containing a sequence of zero or more reference tokens, each prefixed by a `/` character. +json.exception.parse_error.108 | parse error: escape character '~' must be followed with '0' or '1' | In a JSON Pointer, only `~0` and `~1` are valid escape sequences. +json.exception.parse_error.109 | parse error: array index 'one' is not a number | A JSON Pointer array index must be a number. +json.exception.parse_error.110 | parse error at 1: cannot read 2 bytes from vector | When parsing CBOR or MessagePack, the byte vector ends before the complete value has been read. +json.exception.parse_error.112 | parse error at 1: error reading CBOR; last byte: 0xf8 | Not all types of CBOR or MessagePack are supported. This exception occurs if an unsupported byte was read. +json.exception.parse_error.113 | parse error at 2: expected a CBOR string; last byte: 0x98 | While parsing a map key, a value that is not a string has been read. + +@note For an input with n bytes, 1 is the index of the first character and n+1 + is the index of the terminating null byte or the end of file. This also + holds true when reading a byte vector (CBOR or MessagePack). + +@liveexample{The following code shows how a `parse_error` exception can be +caught.,parse_error} + +@sa @ref exception for the base class of the library exceptions +@sa @ref invalid_iterator for exceptions indicating errors with iterators +@sa @ref type_error for exceptions indicating executing a member function with + a wrong type +@sa @ref out_of_range for exceptions indicating access out of the defined range +@sa @ref other_error for exceptions indicating other library errors + +@since version 3.0.0 +*/ +class parse_error : public exception +{ + public: + /*! + @brief create a parse error exception + @param[in] id the id of the exception + @param[in] byte_ the byte index where the error occurred (or 0 if the + position cannot be determined) + @param[in] what_arg the explanatory string + @return parse_error object + */ + static parse_error create(int id, std::size_t byte_, const std::string& what_arg) + { + std::string w = exception::name("parse_error", id) + "parse error" + + (byte_ != 0 ? (" at " + std::to_string(byte_)) : "") + + ": " + what_arg; + return parse_error(id, byte_, w.c_str()); + } + + /*! + @brief byte index of the parse error + + The byte index of the last read character in the input file. + + @note For an input with n bytes, 1 is the index of the first character and + n+1 is the index of the terminating null byte or the end of file. + This also holds true when reading a byte vector (CBOR or MessagePack). + */ + const std::size_t byte; + + private: + parse_error(int id_, std::size_t byte_, const char* what_arg) + : exception(id_, what_arg), byte(byte_) {} +}; + +/*! +@brief exception indicating errors with iterators + +This exception is thrown if iterators passed to a library function do not match +the expected semantics. + +Exceptions have ids 2xx. + +name / id | example message | description +----------------------------------- | --------------- | ------------------------- +json.exception.invalid_iterator.201 | iterators are not compatible | The iterators passed to constructor @ref basic_json(InputIT first, InputIT last) are not compatible, meaning they do not belong to the same container. Therefore, the range (@a first, @a last) is invalid. +json.exception.invalid_iterator.202 | iterator does not fit current value | In an erase or insert function, the passed iterator @a pos does not belong to the JSON value for which the function was called. It hence does not define a valid position for the deletion/insertion. +json.exception.invalid_iterator.203 | iterators do not fit current value | Either iterator passed to function @ref erase(IteratorType first, IteratorType last) does not belong to the JSON value from which values shall be erased. It hence does not define a valid range to delete values from. +json.exception.invalid_iterator.204 | iterators out of range | When an iterator range for a primitive type (number, boolean, or string) is passed to a constructor or an erase function, this range has to be exactly (@ref begin(), @ref end()), because this is the only way the single stored value is expressed. All other ranges are invalid. +json.exception.invalid_iterator.205 | iterator out of range | When an iterator for a primitive type (number, boolean, or string) is passed to an erase function, the iterator has to be the @ref begin() iterator, because it is the only way to address the stored value. All other iterators are invalid. +json.exception.invalid_iterator.206 | cannot construct with iterators from null | The iterators passed to constructor @ref basic_json(InputIT first, InputIT last) belong to a JSON null value and hence to not define a valid range. +json.exception.invalid_iterator.207 | cannot use key() for non-object iterators | The key() member function can only be used on iterators belonging to a JSON object, because other types do not have a concept of a key. +json.exception.invalid_iterator.208 | cannot use operator[] for object iterators | The operator[] to specify a concrete offset cannot be used on iterators belonging to a JSON object, because JSON objects are unordered. +json.exception.invalid_iterator.209 | cannot use offsets with object iterators | The offset operators (+, -, +=, -=) cannot be used on iterators belonging to a JSON object, because JSON objects are unordered. +json.exception.invalid_iterator.210 | iterators do not fit | The iterator range passed to the insert function are not compatible, meaning they do not belong to the same container. Therefore, the range (@a first, @a last) is invalid. +json.exception.invalid_iterator.211 | passed iterators may not belong to container | The iterator range passed to the insert function must not be a subrange of the container to insert to. +json.exception.invalid_iterator.212 | cannot compare iterators of different containers | When two iterators are compared, they must belong to the same container. +json.exception.invalid_iterator.213 | cannot compare order of object iterators | The order of object iterators cannot be compared, because JSON objects are unordered. +json.exception.invalid_iterator.214 | cannot get value | Cannot get value for iterator: Either the iterator belongs to a null value or it is an iterator to a primitive type (number, boolean, or string), but the iterator is different to @ref begin(). + +@liveexample{The following code shows how an `invalid_iterator` exception can be +caught.,invalid_iterator} + +@sa @ref exception for the base class of the library exceptions +@sa @ref parse_error for exceptions indicating a parse error +@sa @ref type_error for exceptions indicating executing a member function with + a wrong type +@sa @ref out_of_range for exceptions indicating access out of the defined range +@sa @ref other_error for exceptions indicating other library errors + +@since version 3.0.0 +*/ +class invalid_iterator : public exception +{ + public: + static invalid_iterator create(int id, const std::string& what_arg) + { + std::string w = exception::name("invalid_iterator", id) + what_arg; + return invalid_iterator(id, w.c_str()); + } + + private: + invalid_iterator(int id_, const char* what_arg) + : exception(id_, what_arg) {} +}; + +/*! +@brief exception indicating executing a member function with a wrong type + +This exception is thrown in case of a type error; that is, a library function is +executed on a JSON value whose type does not match the expected semantics. + +Exceptions have ids 3xx. + +name / id | example message | description +----------------------------- | --------------- | ------------------------- +json.exception.type_error.301 | cannot create object from initializer list | To create an object from an initializer list, the initializer list must consist only of a list of pairs whose first element is a string. When this constraint is violated, an array is created instead. +json.exception.type_error.302 | type must be object, but is array | During implicit or explicit value conversion, the JSON type must be compatible to the target type. For instance, a JSON string can only be converted into string types, but not into numbers or boolean types. +json.exception.type_error.303 | incompatible ReferenceType for get_ref, actual type is object | To retrieve a reference to a value stored in a @ref basic_json object with @ref get_ref, the type of the reference must match the value type. For instance, for a JSON array, the @a ReferenceType must be @ref array_t&. +json.exception.type_error.304 | cannot use at() with string | The @ref at() member functions can only be executed for certain JSON types. +json.exception.type_error.305 | cannot use operator[] with string | The @ref operator[] member functions can only be executed for certain JSON types. +json.exception.type_error.306 | cannot use value() with string | The @ref value() member functions can only be executed for certain JSON types. +json.exception.type_error.307 | cannot use erase() with string | The @ref erase() member functions can only be executed for certain JSON types. +json.exception.type_error.308 | cannot use push_back() with string | The @ref push_back() and @ref operator+= member functions can only be executed for certain JSON types. +json.exception.type_error.309 | cannot use insert() with | The @ref insert() member functions can only be executed for certain JSON types. +json.exception.type_error.310 | cannot use swap() with number | The @ref swap() member functions can only be executed for certain JSON types. +json.exception.type_error.311 | cannot use emplace_back() with string | The @ref emplace_back() member function can only be executed for certain JSON types. +json.exception.type_error.312 | cannot use update() with string | The @ref update() member functions can only be executed for certain JSON types. +json.exception.type_error.313 | invalid value to unflatten | The @ref unflatten function converts an object whose keys are JSON Pointers back into an arbitrary nested JSON value. The JSON Pointers must not overlap, because then the resulting value would not be well defined. +json.exception.type_error.314 | only objects can be unflattened | The @ref unflatten function only works for an object whose keys are JSON Pointers. +json.exception.type_error.315 | values in object must be primitive | The @ref unflatten function only works for an object whose keys are JSON Pointers and whose values are primitive. + +@liveexample{The following code shows how a `type_error` exception can be +caught.,type_error} + +@sa @ref exception for the base class of the library exceptions +@sa @ref parse_error for exceptions indicating a parse error +@sa @ref invalid_iterator for exceptions indicating errors with iterators +@sa @ref out_of_range for exceptions indicating access out of the defined range +@sa @ref other_error for exceptions indicating other library errors + +@since version 3.0.0 +*/ +class type_error : public exception +{ + public: + static type_error create(int id, const std::string& what_arg) + { + std::string w = exception::name("type_error", id) + what_arg; + return type_error(id, w.c_str()); + } + + private: + type_error(int id_, const char* what_arg) : exception(id_, what_arg) {} +}; + +/*! +@brief exception indicating access out of the defined range + +This exception is thrown in case a library function is called on an input +parameter that exceeds the expected range, for instance in case of array +indices or nonexisting object keys. + +Exceptions have ids 4xx. + +name / id | example message | description +------------------------------- | --------------- | ------------------------- +json.exception.out_of_range.401 | array index 3 is out of range | The provided array index @a i is larger than @a size-1. +json.exception.out_of_range.402 | array index '-' (3) is out of range | The special array index `-` in a JSON Pointer never describes a valid element of the array, but the index past the end. That is, it can only be used to add elements at this position, but not to read it. +json.exception.out_of_range.403 | key 'foo' not found | The provided key was not found in the JSON object. +json.exception.out_of_range.404 | unresolved reference token 'foo' | A reference token in a JSON Pointer could not be resolved. +json.exception.out_of_range.405 | JSON pointer has no parent | The JSON Patch operations 'remove' and 'add' can not be applied to the root element of the JSON value. +json.exception.out_of_range.406 | number overflow parsing '10E1000' | A parsed number could not be stored as without changing it to NaN or INF. + +@liveexample{The following code shows how an `out_of_range` exception can be +caught.,out_of_range} + +@sa @ref exception for the base class of the library exceptions +@sa @ref parse_error for exceptions indicating a parse error +@sa @ref invalid_iterator for exceptions indicating errors with iterators +@sa @ref type_error for exceptions indicating executing a member function with + a wrong type +@sa @ref other_error for exceptions indicating other library errors + +@since version 3.0.0 +*/ +class out_of_range : public exception +{ + public: + static out_of_range create(int id, const std::string& what_arg) + { + std::string w = exception::name("out_of_range", id) + what_arg; + return out_of_range(id, w.c_str()); + } + + private: + out_of_range(int id_, const char* what_arg) : exception(id_, what_arg) {} +}; + +/*! +@brief exception indicating other library errors + +This exception is thrown in case of errors that cannot be classified with the +other exception types. + +Exceptions have ids 5xx. + +name / id | example message | description +------------------------------ | --------------- | ------------------------- +json.exception.other_error.501 | unsuccessful: {"op":"test","path":"/baz", "value":"bar"} | A JSON Patch operation 'test' failed. The unsuccessful operation is also printed. +json.exception.other_error.502 | invalid object size for conversion | Some conversions to user-defined types impose constraints on the object size (e.g. std::pair) + +@sa @ref exception for the base class of the library exceptions +@sa @ref parse_error for exceptions indicating a parse error +@sa @ref invalid_iterator for exceptions indicating errors with iterators +@sa @ref type_error for exceptions indicating executing a member function with + a wrong type +@sa @ref out_of_range for exceptions indicating access out of the defined range + +@liveexample{The following code shows how an `other_error` exception can be +caught.,other_error} + +@since version 3.0.0 +*/ +class other_error : public exception +{ + public: + static other_error create(int id, const std::string& what_arg) + { + std::string w = exception::name("other_error", id) + what_arg; + return other_error(id, w.c_str()); + } + + private: + other_error(int id_, const char* what_arg) : exception(id_, what_arg) {} +}; + + + +/////////////////////////// +// JSON type enumeration // +/////////////////////////// + +/*! +@brief the JSON type enumeration + +This enumeration collects the different JSON types. It is internally used to +distinguish the stored values, and the functions @ref basic_json::is_null(), +@ref basic_json::is_object(), @ref basic_json::is_array(), +@ref basic_json::is_string(), @ref basic_json::is_boolean(), +@ref basic_json::is_number() (with @ref basic_json::is_number_integer(), +@ref basic_json::is_number_unsigned(), and @ref basic_json::is_number_float()), +@ref basic_json::is_discarded(), @ref basic_json::is_primitive(), and +@ref basic_json::is_structured() rely on it. + +@note There are three enumeration entries (number_integer, number_unsigned, and +number_float), because the library distinguishes these three types for numbers: +@ref basic_json::number_unsigned_t is used for unsigned integers, +@ref basic_json::number_integer_t is used for signed integers, and +@ref basic_json::number_float_t is used for floating-point numbers or to +approximate integers which do not fit in the limits of their respective type. + +@sa @ref basic_json::basic_json(const value_t value_type) -- create a JSON +value with the default value for a given type + @since version 1.0.0 */ -namespace +enum class value_t : uint8_t { + null, ///< null value + object, ///< object (unordered set of name/value pairs) + array, ///< array (ordered collection of values) + string, ///< string value + boolean, ///< boolean value + number_integer, ///< number value (signed integer) + number_unsigned, ///< number value (unsigned integer) + number_float, ///< number value (floating-point) + discarded ///< discarded by the the parser callback function +}; + +/*! +@brief comparison operator for JSON types + +Returns an ordering that is similar to Python: +- order: null < boolean < number < object < array < string +- furthermore, each type is not smaller than itself + +@since version 1.0.0 +*/ +inline bool operator<(const value_t lhs, const value_t rhs) noexcept +{ + static constexpr std::array order = {{ + 0, // null + 3, // object + 4, // array + 5, // string + 1, // boolean + 2, // integer + 2, // unsigned + 2, // float + } + }; + + // discarded values are not comparable + return lhs != value_t::discarded and rhs != value_t::discarded and + order[static_cast(lhs)] < order[static_cast(rhs)]; +} + + +///////////// +// helpers // +///////////// + +template struct is_basic_json : std::false_type {}; + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +struct is_basic_json : std::true_type {}; + +// alias templates to reduce boilerplate +template +using enable_if_t = typename std::enable_if::type; + +template +using uncvref_t = typename std::remove_cv::type>::type; + +// implementation of C++14 index_sequence and affiliates +// source: https://stackoverflow.com/a/32223343 +template +struct index_sequence +{ + using type = index_sequence; + using value_type = std::size_t; + static constexpr std::size_t size() noexcept + { + return sizeof...(Ints); + } +}; + +template +struct merge_and_renumber; + +template +struct merge_and_renumber, index_sequence> + : index_sequence < I1..., (sizeof...(I1) + I2)... > + {}; + +template +struct make_index_sequence + : merge_and_renumber < typename make_index_sequence < N / 2 >::type, + typename make_index_sequence < N - N / 2 >::type > +{}; + +template<> struct make_index_sequence<0> : index_sequence<> { }; +template<> struct make_index_sequence<1> : index_sequence<0> { }; + +template +using index_sequence_for = make_index_sequence; + +/* +Implementation of two C++17 constructs: conjunction, negation. This is needed +to avoid evaluating all the traits in a condition + +For example: not std::is_same::value and has_value_type::value +will not compile when T = void (on MSVC at least). Whereas +conjunction>, has_value_type>::value will +stop evaluating if negation<...>::value == false + +Please note that those constructs must be used with caution, since symbols can +become very long quickly (which can slow down compilation and cause MSVC +internal compiler errors). Only use it when you have to (see example ahead). +*/ +template struct conjunction : std::true_type {}; +template struct conjunction : B1 {}; +template +struct conjunction : std::conditional, B1>::type {}; + +template struct negation : std::integral_constant < bool, !B::value > {}; + +// dispatch utility (taken from ranges-v3) +template struct priority_tag : priority_tag < N - 1 > {}; +template<> struct priority_tag<0> {}; + + +////////////////// +// constructors // +////////////////// + +template struct external_constructor; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, typename BasicJsonType::boolean_t b) noexcept + { + j.m_type = value_t::boolean; + j.m_value = b; + j.assert_invariant(); + } +}; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, const typename BasicJsonType::string_t& s) + { + j.m_type = value_t::string; + j.m_value = s; + j.assert_invariant(); + } + + template + static void construct(BasicJsonType& j, typename BasicJsonType::string_t&& s) + { + j.m_type = value_t::string; + j.m_value = std::move(s); + j.assert_invariant(); + } +}; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, typename BasicJsonType::number_float_t val) noexcept + { + j.m_type = value_t::number_float; + j.m_value = val; + j.assert_invariant(); + } +}; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, typename BasicJsonType::number_unsigned_t val) noexcept + { + j.m_type = value_t::number_unsigned; + j.m_value = val; + j.assert_invariant(); + } +}; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, typename BasicJsonType::number_integer_t val) noexcept + { + j.m_type = value_t::number_integer; + j.m_value = val; + j.assert_invariant(); + } +}; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, const typename BasicJsonType::array_t& arr) + { + j.m_type = value_t::array; + j.m_value = arr; + j.assert_invariant(); + } + + template + static void construct(BasicJsonType& j, typename BasicJsonType::array_t&& arr) + { + j.m_type = value_t::array; + j.m_value = std::move(arr); + j.assert_invariant(); + } + + template::value, + int> = 0> + static void construct(BasicJsonType& j, const CompatibleArrayType& arr) + { + using std::begin; + using std::end; + j.m_type = value_t::array; + j.m_value.array = j.template create(begin(arr), end(arr)); + j.assert_invariant(); + } + + template + static void construct(BasicJsonType& j, const std::vector& arr) + { + j.m_type = value_t::array; + j.m_value = value_t::array; + j.m_value.array->reserve(arr.size()); + for (bool x : arr) + { + j.m_value.array->push_back(x); + } + j.assert_invariant(); + } + + template::value, int> = 0> + static void construct(BasicJsonType& j, const std::valarray& arr) + { + using std::begin; + using std::end; + j.m_type = value_t::array; + j.m_value = value_t::array; + j.m_value.array = j.template create(begin(arr), end(arr)); + j.assert_invariant(); + } +}; + +template<> +struct external_constructor +{ + template + static void construct(BasicJsonType& j, const typename BasicJsonType::object_t& obj) + { + j.m_type = value_t::object; + j.m_value = obj; + j.assert_invariant(); + } + + template + static void construct(BasicJsonType& j, typename BasicJsonType::object_t&& obj) + { + j.m_type = value_t::object; + j.m_value = std::move(obj); + j.assert_invariant(); + } + + template::value, int> = 0> + static void construct(BasicJsonType& j, const CompatibleObjectType& obj) + { + using std::begin; + using std::end; + + j.m_type = value_t::object; + j.m_value.object = j.template create(begin(obj), end(obj)); + j.assert_invariant(); + } +}; + + +//////////////////////// +// has_/is_ functions // +//////////////////////// + /*! @brief Helper to determine whether there's a key_type for T. -Thus helper is used to tell associative containers apart from other containers +This helper is used to tell associative containers apart from other containers such as sequence containers. For instance, `std::map` passes the test as it contains a `mapped_type`, whereas `std::vector` fails the test. @sa http://stackoverflow.com/a/7728728/266378 @since version 1.0.0, overworked in version 2.0.6 */ -template -struct has_mapped_type -{ - private: - template - static int detect(U&&); +#define NLOHMANN_JSON_HAS_HELPER(type) \ + template struct has_##type { \ + private: \ + template \ + static int detect(U &&); \ + static void detect(...); \ + public: \ + static constexpr bool value = \ + std::is_integral()))>::value; \ + } - static void detect(...); - public: - static constexpr bool value = - std::is_integral()))>::value; +NLOHMANN_JSON_HAS_HELPER(mapped_type); +NLOHMANN_JSON_HAS_HELPER(key_type); +NLOHMANN_JSON_HAS_HELPER(value_type); +NLOHMANN_JSON_HAS_HELPER(iterator); + +#undef NLOHMANN_JSON_HAS_HELPER + + +template +struct is_compatible_object_type_impl : std::false_type {}; + +template +struct is_compatible_object_type_impl +{ + static constexpr auto value = + std::is_constructible::value and + std::is_constructible::value; }; +template +struct is_compatible_object_type +{ + static auto constexpr value = is_compatible_object_type_impl < + conjunction>, + has_mapped_type, + has_key_type>::value, + typename BasicJsonType::object_t, CompatibleObjectType >::value; +}; + +template +struct is_basic_json_nested_type +{ + static auto constexpr value = std::is_same::value or + std::is_same::value or + std::is_same::value or + std::is_same::value; +}; + +template +struct is_compatible_array_type +{ + static auto constexpr value = + conjunction>, + negation>, + negation>, + negation>, + has_value_type, + has_iterator>::value; +}; + +template +struct is_compatible_integer_type_impl : std::false_type {}; + +template +struct is_compatible_integer_type_impl +{ + // is there an assert somewhere on overflows? + using RealLimits = std::numeric_limits; + using CompatibleLimits = std::numeric_limits; + + static constexpr auto value = + std::is_constructible::value and + CompatibleLimits::is_integer and + RealLimits::is_signed == CompatibleLimits::is_signed; +}; + +template +struct is_compatible_integer_type +{ + static constexpr auto value = + is_compatible_integer_type_impl < + std::is_integral::value and + not std::is_same::value, + RealIntegerType, CompatibleNumberIntegerType > ::value; +}; + + +// trait checking if JSONSerializer::from_json(json const&, udt&) exists +template +struct has_from_json +{ + private: + // also check the return type of from_json + template::from_json( + std::declval(), std::declval()))>::value>> + static int detect(U&&); + static void detect(...); + + public: + static constexpr bool value = std::is_integral>()))>::value; +}; + +// This trait checks if JSONSerializer::from_json(json const&) exists +// this overload is used for non-default-constructible user-defined-types +template +struct has_non_default_from_json +{ + private: + template < + typename U, + typename = enable_if_t::from_json(std::declval()))>::value >> + static int detect(U&&); + static void detect(...); + + public: + static constexpr bool value = std::is_integral>()))>::value; +}; + +// This trait checks if BasicJsonType::json_serializer::to_json exists +template +struct has_to_json +{ + private: + template::to_json( + std::declval(), std::declval()))> + static int detect(U&&); + static void detect(...); + + public: + static constexpr bool value = std::is_integral>()))>::value; +}; + + +///////////// +// to_json // +///////////// + +template::value, int> = 0> +void to_json(BasicJsonType& j, T b) noexcept +{ + external_constructor::construct(j, b); } +template::value, int> = 0> +void to_json(BasicJsonType& j, const CompatibleString& s) +{ + external_constructor::construct(j, s); +} + +template +void to_json(BasicJsonType& j, typename BasicJsonType::string_t&& s) +{ + external_constructor::construct(j, std::move(s)); +} + +template::value, int> = 0> +void to_json(BasicJsonType& j, FloatType val) noexcept +{ + external_constructor::construct(j, static_cast(val)); +} + +template < + typename BasicJsonType, typename CompatibleNumberUnsignedType, + enable_if_t::value, int> = 0 > +void to_json(BasicJsonType& j, CompatibleNumberUnsignedType val) noexcept +{ + external_constructor::construct(j, static_cast(val)); +} + +template < + typename BasicJsonType, typename CompatibleNumberIntegerType, + enable_if_t::value, int> = 0 > +void to_json(BasicJsonType& j, CompatibleNumberIntegerType val) noexcept +{ + external_constructor::construct(j, static_cast(val)); +} + +template::value, int> = 0> +void to_json(BasicJsonType& j, EnumType e) noexcept +{ + using underlying_type = typename std::underlying_type::type; + external_constructor::construct(j, static_cast(e)); +} + +template +void to_json(BasicJsonType& j, const std::vector& e) +{ + external_constructor::construct(j, e); +} + +template < + typename BasicJsonType, typename CompatibleArrayType, + enable_if_t < + is_compatible_array_type::value or + std::is_same::value, + int > = 0 > +void to_json(BasicJsonType& j, const CompatibleArrayType& arr) +{ + external_constructor::construct(j, arr); +} + +template ::value, int> = 0> +void to_json(BasicJsonType& j, std::valarray arr) +{ + external_constructor::construct(j, std::move(arr)); +} + +template +void to_json(BasicJsonType& j, typename BasicJsonType::array_t&& arr) +{ + external_constructor::construct(j, std::move(arr)); +} + +template < + typename BasicJsonType, typename CompatibleObjectType, + enable_if_t::value, + int> = 0 > +void to_json(BasicJsonType& j, const CompatibleObjectType& obj) +{ + external_constructor::construct(j, obj); +} + +template +void to_json(BasicJsonType& j, typename BasicJsonType::object_t&& obj) +{ + external_constructor::construct(j, std::move(obj)); +} + +template::value, + int> = 0> +void to_json(BasicJsonType& j, T (&arr)[N]) +{ + external_constructor::construct(j, arr); +} + +template +void to_json(BasicJsonType& j, const std::pair& p) +{ + j = {p.first, p.second}; +} + +template +void to_json_tuple_impl(BasicJsonType& j, const Tuple& t, index_sequence) +{ + j = {std::get(t)...}; +} + +template +void to_json(BasicJsonType& j, const std::tuple& t) +{ + to_json_tuple_impl(j, t, index_sequence_for {}); +} + +/////////////// +// from_json // +/////////////// + +// overloads for basic_json template parameters +template::value and + not std::is_same::value, + int> = 0> +void get_arithmetic_value(const BasicJsonType& j, ArithmeticType& val) +{ + switch (static_cast(j)) + { + case value_t::number_unsigned: + { + val = static_cast(*j.template get_ptr()); + break; + } + case value_t::number_integer: + { + val = static_cast(*j.template get_ptr()); + break; + } + case value_t::number_float: + { + val = static_cast(*j.template get_ptr()); + break; + } + + default: + JSON_THROW(type_error::create(302, "type must be number, but is " + std::string(j.type_name()))); + } +} + +template +void from_json(const BasicJsonType& j, typename BasicJsonType::boolean_t& b) +{ + if (JSON_UNLIKELY(not j.is_boolean())) + { + JSON_THROW(type_error::create(302, "type must be boolean, but is " + std::string(j.type_name()))); + } + b = *j.template get_ptr(); +} + +template +void from_json(const BasicJsonType& j, typename BasicJsonType::string_t& s) +{ + if (JSON_UNLIKELY(not j.is_string())) + { + JSON_THROW(type_error::create(302, "type must be string, but is " + std::string(j.type_name()))); + } + s = *j.template get_ptr(); +} + +template +void from_json(const BasicJsonType& j, typename BasicJsonType::number_float_t& val) +{ + get_arithmetic_value(j, val); +} + +template +void from_json(const BasicJsonType& j, typename BasicJsonType::number_unsigned_t& val) +{ + get_arithmetic_value(j, val); +} + +template +void from_json(const BasicJsonType& j, typename BasicJsonType::number_integer_t& val) +{ + get_arithmetic_value(j, val); +} + +template::value, int> = 0> +void from_json(const BasicJsonType& j, EnumType& e) +{ + typename std::underlying_type::type val; + get_arithmetic_value(j, val); + e = static_cast(val); +} + +template +void from_json(const BasicJsonType& j, typename BasicJsonType::array_t& arr) +{ + if (JSON_UNLIKELY(not j.is_array())) + { + JSON_THROW(type_error::create(302, "type must be array, but is " + std::string(j.type_name()))); + } + arr = *j.template get_ptr(); +} + +// forward_list doesn't have an insert method +template::value, int> = 0> +void from_json(const BasicJsonType& j, std::forward_list& l) +{ + if (JSON_UNLIKELY(not j.is_array())) + { + JSON_THROW(type_error::create(302, "type must be array, but is " + std::string(j.type_name()))); + } + std::transform(j.rbegin(), j.rend(), + std::front_inserter(l), [](const BasicJsonType & i) + { + return i.template get(); + }); +} + +// valarray doesn't have an insert method +template::value, int> = 0> +void from_json(const BasicJsonType& j, std::valarray& l) +{ + if (JSON_UNLIKELY(not j.is_array())) + { + JSON_THROW(type_error::create(302, "type must be array, but is " + std::string(j.type_name()))); + } + l.resize(j.size()); + for (size_t i = 0; i < j.size(); ++i) + { + l[i] = j[i]; + } +} + +template +void from_json_array_impl(const BasicJsonType& j, CompatibleArrayType& arr, priority_tag<0> /*unused*/) +{ + using std::end; + + std::transform(j.begin(), j.end(), + std::inserter(arr, end(arr)), [](const BasicJsonType & i) + { + // get() returns *this, this won't call a from_json + // method when value_type is BasicJsonType + return i.template get(); + }); +} + +template +auto from_json_array_impl(const BasicJsonType& j, CompatibleArrayType& arr, priority_tag<1> /*unused*/) +-> decltype( + arr.reserve(std::declval()), + void()) +{ + using std::end; + + arr.reserve(j.size()); + std::transform(j.begin(), j.end(), + std::inserter(arr, end(arr)), [](const BasicJsonType & i) + { + // get() returns *this, this won't call a from_json + // method when value_type is BasicJsonType + return i.template get(); + }); +} + +template +void from_json_array_impl(const BasicJsonType& j, std::array& arr, priority_tag<2> /*unused*/) +{ + for (std::size_t i = 0; i < N; ++i) + { + arr[i] = j.at(i).template get(); + } +} + +template::value and + std::is_convertible::value and + not std::is_same::value, int> = 0> +void from_json(const BasicJsonType& j, CompatibleArrayType& arr) +{ + if (JSON_UNLIKELY(not j.is_array())) + { + JSON_THROW(type_error::create(302, "type must be array, but is " + std::string(j.type_name()))); + } + + from_json_array_impl(j, arr, priority_tag<2> {}); +} + +template::value, int> = 0> +void from_json(const BasicJsonType& j, CompatibleObjectType& obj) +{ + if (JSON_UNLIKELY(not j.is_object())) + { + JSON_THROW(type_error::create(302, "type must be object, but is " + std::string(j.type_name()))); + } + + auto inner_object = j.template get_ptr(); + using value_type = typename CompatibleObjectType::value_type; + std::transform( + inner_object->begin(), inner_object->end(), + std::inserter(obj, obj.begin()), + [](typename BasicJsonType::object_t::value_type const & p) + { + return value_type(p.first, p.second.template get()); + }); +} + +// overload for arithmetic types, not chosen for basic_json template arguments +// (BooleanType, etc..); note: Is it really necessary to provide explicit +// overloads for boolean_t etc. in case of a custom BooleanType which is not +// an arithmetic type? +template::value and + not std::is_same::value and + not std::is_same::value and + not std::is_same::value and + not std::is_same::value, + int> = 0> +void from_json(const BasicJsonType& j, ArithmeticType& val) +{ + switch (static_cast(j)) + { + case value_t::number_unsigned: + { + val = static_cast(*j.template get_ptr()); + break; + } + case value_t::number_integer: + { + val = static_cast(*j.template get_ptr()); + break; + } + case value_t::number_float: + { + val = static_cast(*j.template get_ptr()); + break; + } + case value_t::boolean: + { + val = static_cast(*j.template get_ptr()); + break; + } + + default: + JSON_THROW(type_error::create(302, "type must be number, but is " + std::string(j.type_name()))); + } +} + +template +void from_json(const BasicJsonType& j, std::pair& p) +{ + p = {j.at(0).template get(), j.at(1).template get()}; +} + +template +void from_json_tuple_impl(const BasicJsonType& j, Tuple& t, index_sequence) +{ + t = std::make_tuple(j.at(Idx).template get::type>()...); +} + +template +void from_json(const BasicJsonType& j, std::tuple& t) +{ + from_json_tuple_impl(j, t, index_sequence_for {}); +} + +struct to_json_fn +{ + private: + template + auto call(BasicJsonType& j, T&& val, priority_tag<1> /*unused*/) const noexcept(noexcept(to_json(j, std::forward(val)))) + -> decltype(to_json(j, std::forward(val)), void()) + { + return to_json(j, std::forward(val)); + } + + template + void call(BasicJsonType& /*unused*/, T&& /*unused*/, priority_tag<0> /*unused*/) const noexcept + { + static_assert(sizeof(BasicJsonType) == 0, + "could not find to_json() method in T's namespace"); + } + + public: + template + void operator()(BasicJsonType& j, T&& val) const + noexcept(noexcept(std::declval().call(j, std::forward(val), priority_tag<1> {}))) + { + return call(j, std::forward(val), priority_tag<1> {}); + } +}; + +struct from_json_fn +{ + private: + template + auto call(const BasicJsonType& j, T& val, priority_tag<1> /*unused*/) const + noexcept(noexcept(from_json(j, val))) + -> decltype(from_json(j, val), void()) + { + return from_json(j, val); + } + + template + void call(const BasicJsonType& /*unused*/, T& /*unused*/, priority_tag<0> /*unused*/) const noexcept + { + static_assert(sizeof(BasicJsonType) == 0, + "could not find from_json() method in T's namespace"); + } + + public: + template + void operator()(const BasicJsonType& j, T& val) const + noexcept(noexcept(std::declval().call(j, val, priority_tag<1> {}))) + { + return call(j, val, priority_tag<1> {}); + } +}; + +// taken from ranges-v3 +template +struct static_const +{ + static constexpr T value{}; +}; + +template +constexpr T static_const::value; + +//////////////////// +// input adapters // +//////////////////// + +/// abstract input adapter interface +struct input_adapter_protocol +{ + virtual int get_character() = 0; + virtual std::string read(std::size_t offset, std::size_t length) = 0; + virtual ~input_adapter_protocol() = default; +}; + +/// a type to simplify interfaces +using input_adapter_t = std::shared_ptr; + +/// input adapter for cached stream input +template +class cached_input_stream_adapter : public input_adapter_protocol +{ + public: + explicit cached_input_stream_adapter(std::istream& i) + : is(i), start_position(is.tellg()) + { + fill_buffer(); + + // skip byte order mark + if (fill_size >= 3 and buffer[0] == '\xEF' and buffer[1] == '\xBB' and buffer[2] == '\xBF') + { + buffer_pos += 3; + processed_chars += 3; + } + } + + ~cached_input_stream_adapter() override + { + // clear stream flags + is.clear(); + // We initially read a lot of characters into the buffer, and we may + // not have processed all of them. Therefore, we need to "rewind" the + // stream after the last processed char. + is.seekg(start_position); + is.ignore(static_cast(processed_chars)); + // clear stream flags + is.clear(); + } + + int get_character() override + { + // check if refilling is necessary and possible + if (buffer_pos == fill_size and not eof) + { + fill_buffer(); + + // check and remember that filling did not yield new input + if (fill_size == 0) + { + eof = true; + return std::char_traits::eof(); + } + + // the buffer is ready + buffer_pos = 0; + } + + ++processed_chars; + assert(buffer_pos < buffer.size()); + return buffer[buffer_pos++] & 0xFF; + } + + std::string read(std::size_t offset, std::size_t length) override + { + // create buffer + std::string result(length, '\0'); + + // save stream position + const auto current_pos = is.tellg(); + // save stream flags + const auto flags = is.rdstate(); + + // clear stream flags + is.clear(); + // set stream position + is.seekg(static_cast(offset)); + // read bytes + is.read(&result[0], static_cast(length)); + + // reset stream position + is.seekg(current_pos); + // reset stream flags + is.setstate(flags); + + return result; + } + + private: + void fill_buffer() + { + // fill + is.read(buffer.data(), static_cast(buffer.size())); + // store number of bytes in the buffer + fill_size = static_cast(is.gcount()); + } + + /// the associated input stream + std::istream& is; + + /// chars returned via get_character() + std::size_t processed_chars = 0; + /// chars processed in the current buffer + std::size_t buffer_pos = 0; + + /// whether stream reached eof + bool eof = false; + /// how many chars have been copied to the buffer by last (re)fill + std::size_t fill_size = 0; + + /// position of the stream when we started + const std::streampos start_position; + + /// internal buffer + std::array buffer{{}}; +}; + +/// input adapter for buffer input +class input_buffer_adapter : public input_adapter_protocol +{ + public: + input_buffer_adapter(const char* b, const std::size_t l) + : cursor(b), limit(b + l), start(b) + { + // skip byte order mark + if (l >= 3 and b[0] == '\xEF' and b[1] == '\xBB' and b[2] == '\xBF') + { + cursor += 3; + } + } + + // delete because of pointer members + input_buffer_adapter(const input_buffer_adapter&) = delete; + input_buffer_adapter& operator=(input_buffer_adapter&) = delete; + + int get_character() noexcept override + { + if (JSON_LIKELY(cursor < limit)) + { + return *(cursor++) & 0xFF; + } + + return std::char_traits::eof(); + } + + std::string read(std::size_t offset, std::size_t length) override + { + // avoid reading too many characters + const auto max_length = static_cast(limit - start); + return std::string(start + offset, (std::min)(length, max_length - offset)); + } + + private: + /// pointer to the current character + const char* cursor; + /// pointer past the last character + const char* limit; + /// pointer to the first character + const char* start; +}; + +class input_adapter +{ + public: + // native support + + /// input adapter for input stream + input_adapter(std::istream& i) + : ia(std::make_shared>(i)) {} + + /// input adapter for input stream + input_adapter(std::istream&& i) + : ia(std::make_shared>(i)) {} + + /// input adapter for buffer + template::value and + std::is_integral< + typename std::remove_pointer::type>::value and + sizeof(typename std::remove_pointer::type) == 1, + int>::type = 0> + input_adapter(CharT b, std::size_t l) + : ia(std::make_shared(reinterpret_cast(b), l)) {} + + // derived support + + /// input adapter for string literal + template::value and + std::is_integral< + typename std::remove_pointer::type>::value and + sizeof(typename std::remove_pointer::type) == 1, + int>::type = 0> + input_adapter(CharT b) + : input_adapter(reinterpret_cast(b), + std::strlen(reinterpret_cast(b))) {} + + /// input adapter for iterator range with contiguous storage + template::iterator_category, + std::random_access_iterator_tag>::value, + int>::type = 0> + input_adapter(IteratorType first, IteratorType last) + { + // assertion to check that the iterator range is indeed contiguous, + // see http://stackoverflow.com/a/35008842/266378 for more discussion + assert(std::accumulate( + first, last, std::pair(true, 0), + [&first](std::pair res, decltype(*first) val) + { + res.first &= (val == *(std::next(std::addressof(*first), res.second++))); + return res; + }).first); + + // assertion to check that each element is 1 byte long + static_assert( + sizeof(typename std::iterator_traits::value_type) == 1, + "each element in the iterator range must have the size of 1 byte"); + + const auto len = static_cast(std::distance(first, last)); + if (JSON_LIKELY(len > 0)) + { + // there is at least one element: use the address of first + ia = std::make_shared(reinterpret_cast(&(*first)), len); + } + else + { + // the address of first cannot be used: use nullptr + ia = std::make_shared(nullptr, len); + } + } + + /// input adapter for array + template + input_adapter(T (&array)[N]) + : input_adapter(std::begin(array), std::end(array)) {} + + /// input adapter for contiguous container + template < + class ContiguousContainer, + typename std::enable_if < + not std::is_pointer::value and + std::is_base_of()))>::iterator_category>::value, + int >::type = 0 > + input_adapter(const ContiguousContainer& c) + : input_adapter(std::begin(c), std::end(c)) {} + + operator input_adapter_t() + { + return ia; + } + + private: + /// the actual adapter + input_adapter_t ia = nullptr; +}; + +////////////////////// +// lexer and parser // +////////////////////// + +/*! +@brief lexical analysis + +This class organizes the lexical analysis during JSON deserialization. +*/ +template +class lexer +{ + using number_integer_t = typename BasicJsonType::number_integer_t; + using number_unsigned_t = typename BasicJsonType::number_unsigned_t; + using number_float_t = typename BasicJsonType::number_float_t; + + public: + /// token types for the parser + enum class token_type + { + uninitialized, ///< indicating the scanner is uninitialized + literal_true, ///< the `true` literal + literal_false, ///< the `false` literal + literal_null, ///< the `null` literal + value_string, ///< a string -- use get_string() for actual value + value_unsigned, ///< an unsigned integer -- use get_number_unsigned() for actual value + value_integer, ///< a signed integer -- use get_number_integer() for actual value + value_float, ///< an floating point number -- use get_number_float() for actual value + begin_array, ///< the character for array begin `[` + begin_object, ///< the character for object begin `{` + end_array, ///< the character for array end `]` + end_object, ///< the character for object end `}` + name_separator, ///< the name separator `:` + value_separator, ///< the value separator `,` + parse_error, ///< indicating a parse error + end_of_input, ///< indicating the end of the input buffer + literal_or_value ///< a literal or the begin of a value (only for diagnostics) + }; + + /// return name of values of type token_type (only used for errors) + static const char* token_type_name(const token_type t) noexcept + { + switch (t) + { + case token_type::uninitialized: + return ""; + case token_type::literal_true: + return "true literal"; + case token_type::literal_false: + return "false literal"; + case token_type::literal_null: + return "null literal"; + case token_type::value_string: + return "string literal"; + case lexer::token_type::value_unsigned: + case lexer::token_type::value_integer: + case lexer::token_type::value_float: + return "number literal"; + case token_type::begin_array: + return "'['"; + case token_type::begin_object: + return "'{'"; + case token_type::end_array: + return "']'"; + case token_type::end_object: + return "'}'"; + case token_type::name_separator: + return "':'"; + case token_type::value_separator: + return "','"; + case token_type::parse_error: + return ""; + case token_type::end_of_input: + return "end of input"; + case token_type::literal_or_value: + return "'[', '{', or a literal"; + default: // catch non-enum values + return "unknown token"; // LCOV_EXCL_LINE + } + } + + explicit lexer(detail::input_adapter_t adapter) + : ia(std::move(adapter)), decimal_point_char(get_decimal_point()) {} + + // delete because of pointer members + lexer(const lexer&) = delete; + lexer& operator=(lexer&) = delete; + + private: + ///////////////////// + // locales + ///////////////////// + + /// return the locale-dependent decimal point + static char get_decimal_point() noexcept + { + const auto loc = localeconv(); + assert(loc != nullptr); + return (loc->decimal_point == nullptr) ? '.' : loc->decimal_point[0]; + } + + ///////////////////// + // scan functions + ///////////////////// + + /*! + @brief get codepoint from 4 hex characters following `\u` + + For input "\u c1 c2 c3 c4" the codepoint is: + (c1 * 0x1000) + (c2 * 0x0100) + (c3 * 0x0010) + c4 + = (c1 << 12) + (c2 << 8) + (c3 << 4) + (c4 << 0) + + Furthermore, the possible characters '0'..'9', 'A'..'F', and 'a'..'f' + must be converted to the integers 0x0..0x9, 0xA..0xF, 0xA..0xF, resp. The + conversion is done by subtracting the offset (0x30, 0x37, and 0x57) + between the ASCII value of the character and the desired integer value. + + @return codepoint (0x0000..0xFFFF) or -1 in case of an error (e.g. EOF or + non-hex character) + */ + int get_codepoint() + { + // this function only makes sense after reading `\u` + assert(current == 'u'); + int codepoint = 0; + + const auto factors = { 12, 8, 4, 0 }; + for (const auto factor : factors) + { + get(); + + if (current >= '0' and current <= '9') + { + codepoint += ((current - 0x30) << factor); + } + else if (current >= 'A' and current <= 'F') + { + codepoint += ((current - 0x37) << factor); + } + else if (current >= 'a' and current <= 'f') + { + codepoint += ((current - 0x57) << factor); + } + else + { + return -1; + } + } + + assert(0x0000 <= codepoint and codepoint <= 0xFFFF); + return codepoint; + } + + /*! + @brief check if the next byte(s) are inside a given range + + Adds the current byte and, for each passed range, reads a new byte and + checks if it is inside the range. If a violation was detected, set up an + error message and return false. Otherwise, return true. + + @return true if and only if no range violation was detected + */ + bool next_byte_in_range(std::initializer_list ranges) + { + assert(ranges.size() == 2 or ranges.size() == 4 or ranges.size() == 6); + add(current); + + for (auto range = ranges.begin(); range != ranges.end(); ++range) + { + get(); + if (JSON_LIKELY(*range <= current and current <= *(++range))) + { + add(current); + } + else + { + error_message = "invalid string: ill-formed UTF-8 byte"; + return false; + } + } + + return true; + } + + /*! + @brief scan a string literal + + This function scans a string according to Sect. 7 of RFC 7159. While + scanning, bytes are escaped and copied into buffer yytext. Then the + function returns successfully, yytext is null-terminated and yylen + contains the number of bytes in the string. + + @return token_type::value_string if string could be successfully scanned, + token_type::parse_error otherwise + + @note In case of errors, variable error_message contains a textual + description. + */ + token_type scan_string() + { + // reset yytext (ignore opening quote) + reset(); + + // we entered the function by reading an open quote + assert(current == '\"'); + + while (true) + { + // get next character + switch (get()) + { + // end of file while parsing string + case std::char_traits::eof(): + { + error_message = "invalid string: missing closing quote"; + return token_type::parse_error; + } + + // closing quote + case '\"': + { + // terminate yytext + add('\0'); + --yylen; + return token_type::value_string; + } + + // escapes + case '\\': + { + switch (get()) + { + // quotation mark + case '\"': + add('\"'); + break; + // reverse solidus + case '\\': + add('\\'); + break; + // solidus + case '/': + add('/'); + break; + // backspace + case 'b': + add('\b'); + break; + // form feed + case 'f': + add('\f'); + break; + // line feed + case 'n': + add('\n'); + break; + // carriage return + case 'r': + add('\r'); + break; + // tab + case 't': + add('\t'); + break; + + // unicode escapes + case 'u': + { + int codepoint; + const int codepoint1 = get_codepoint(); + + if (JSON_UNLIKELY(codepoint1 == -1)) + { + error_message = "invalid string: '\\u' must be followed by 4 hex digits"; + return token_type::parse_error; + } + + // check if code point is a high surrogate + if (0xD800 <= codepoint1 and codepoint1 <= 0xDBFF) + { + // expect next \uxxxx entry + if (JSON_LIKELY(get() == '\\' and get() == 'u')) + { + const int codepoint2 = get_codepoint(); + + if (JSON_UNLIKELY(codepoint2 == -1)) + { + error_message = "invalid string: '\\u' must be followed by 4 hex digits"; + return token_type::parse_error; + } + + // check if codepoint2 is a low surrogate + if (JSON_LIKELY(0xDC00 <= codepoint2 and codepoint2 <= 0xDFFF)) + { + codepoint = + // high surrogate occupies the most significant 22 bits + (codepoint1 << 10) + // low surrogate occupies the least significant 15 bits + + codepoint2 + // there is still the 0xD800, 0xDC00 and 0x10000 noise + // in the result so we have to subtract with: + // (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00 + - 0x35FDC00; + } + else + { + error_message = "invalid string: surrogate U+DC00..U+DFFF must be followed by U+DC00..U+DFFF"; + return token_type::parse_error; + } + } + else + { + error_message = "invalid string: surrogate U+DC00..U+DFFF must be followed by U+DC00..U+DFFF"; + return token_type::parse_error; + } + } + else + { + if (JSON_UNLIKELY(0xDC00 <= codepoint1 and codepoint1 <= 0xDFFF)) + { + error_message = "invalid string: surrogate U+DC00..U+DFFF must follow U+D800..U+DBFF"; + return token_type::parse_error; + } + + // only work with first code point + codepoint = codepoint1; + } + + // result of the above calculation yields a proper codepoint + assert(0x00 <= codepoint and codepoint <= 0x10FFFF); + + // translate code point to bytes + if (codepoint < 0x80) + { + // 1-byte characters: 0xxxxxxx (ASCII) + add(codepoint); + } + else if (codepoint <= 0x7ff) + { + // 2-byte characters: 110xxxxx 10xxxxxx + add(0xC0 | (codepoint >> 6)); + add(0x80 | (codepoint & 0x3F)); + } + else if (codepoint <= 0xffff) + { + // 3-byte characters: 1110xxxx 10xxxxxx 10xxxxxx + add(0xE0 | (codepoint >> 12)); + add(0x80 | ((codepoint >> 6) & 0x3F)); + add(0x80 | (codepoint & 0x3F)); + } + else + { + // 4-byte characters: 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx + add(0xF0 | (codepoint >> 18)); + add(0x80 | ((codepoint >> 12) & 0x3F)); + add(0x80 | ((codepoint >> 6) & 0x3F)); + add(0x80 | (codepoint & 0x3F)); + } + + break; + } + + // other characters after escape + default: + error_message = "invalid string: forbidden character after backslash"; + return token_type::parse_error; + } + + break; + } + + // invalid control characters + case 0x00: + case 0x01: + case 0x02: + case 0x03: + case 0x04: + case 0x05: + case 0x06: + case 0x07: + case 0x08: + case 0x09: + case 0x0a: + case 0x0b: + case 0x0c: + case 0x0d: + case 0x0e: + case 0x0f: + case 0x10: + case 0x11: + case 0x12: + case 0x13: + case 0x14: + case 0x15: + case 0x16: + case 0x17: + case 0x18: + case 0x19: + case 0x1a: + case 0x1b: + case 0x1c: + case 0x1d: + case 0x1e: + case 0x1f: + { + error_message = "invalid string: control character must be escaped"; + return token_type::parse_error; + } + + // U+0020..U+007F (except U+0022 (quote) and U+005C (backspace)) + case 0x20: + case 0x21: + case 0x23: + case 0x24: + case 0x25: + case 0x26: + case 0x27: + case 0x28: + case 0x29: + case 0x2a: + case 0x2b: + case 0x2c: + case 0x2d: + case 0x2e: + case 0x2f: + case 0x30: + case 0x31: + case 0x32: + case 0x33: + case 0x34: + case 0x35: + case 0x36: + case 0x37: + case 0x38: + case 0x39: + case 0x3a: + case 0x3b: + case 0x3c: + case 0x3d: + case 0x3e: + case 0x3f: + case 0x40: + case 0x41: + case 0x42: + case 0x43: + case 0x44: + case 0x45: + case 0x46: + case 0x47: + case 0x48: + case 0x49: + case 0x4a: + case 0x4b: + case 0x4c: + case 0x4d: + case 0x4e: + case 0x4f: + case 0x50: + case 0x51: + case 0x52: + case 0x53: + case 0x54: + case 0x55: + case 0x56: + case 0x57: + case 0x58: + case 0x59: + case 0x5a: + case 0x5b: + case 0x5d: + case 0x5e: + case 0x5f: + case 0x60: + case 0x61: + case 0x62: + case 0x63: + case 0x64: + case 0x65: + case 0x66: + case 0x67: + case 0x68: + case 0x69: + case 0x6a: + case 0x6b: + case 0x6c: + case 0x6d: + case 0x6e: + case 0x6f: + case 0x70: + case 0x71: + case 0x72: + case 0x73: + case 0x74: + case 0x75: + case 0x76: + case 0x77: + case 0x78: + case 0x79: + case 0x7a: + case 0x7b: + case 0x7c: + case 0x7d: + case 0x7e: + case 0x7f: + { + add(current); + break; + } + + // U+0080..U+07FF: bytes C2..DF 80..BF + case 0xc2: + case 0xc3: + case 0xc4: + case 0xc5: + case 0xc6: + case 0xc7: + case 0xc8: + case 0xc9: + case 0xca: + case 0xcb: + case 0xcc: + case 0xcd: + case 0xce: + case 0xcf: + case 0xd0: + case 0xd1: + case 0xd2: + case 0xd3: + case 0xd4: + case 0xd5: + case 0xd6: + case 0xd7: + case 0xd8: + case 0xd9: + case 0xda: + case 0xdb: + case 0xdc: + case 0xdd: + case 0xde: + case 0xdf: + { + if (JSON_UNLIKELY(not next_byte_in_range({0x80, 0xBF}))) + { + return token_type::parse_error; + } + break; + } + + // U+0800..U+0FFF: bytes E0 A0..BF 80..BF + case 0xe0: + { + if (JSON_UNLIKELY(not (next_byte_in_range({0xA0, 0xBF, 0x80, 0xBF})))) + { + return token_type::parse_error; + } + break; + } + + // U+1000..U+CFFF: bytes E1..EC 80..BF 80..BF + // U+E000..U+FFFF: bytes EE..EF 80..BF 80..BF + case 0xe1: + case 0xe2: + case 0xe3: + case 0xe4: + case 0xe5: + case 0xe6: + case 0xe7: + case 0xe8: + case 0xe9: + case 0xea: + case 0xeb: + case 0xec: + case 0xee: + case 0xef: + { + if (JSON_UNLIKELY(not (next_byte_in_range({0x80, 0xBF, 0x80, 0xBF})))) + { + return token_type::parse_error; + } + break; + } + + // U+D000..U+D7FF: bytes ED 80..9F 80..BF + case 0xed: + { + if (JSON_UNLIKELY(not (next_byte_in_range({0x80, 0x9F, 0x80, 0xBF})))) + { + return token_type::parse_error; + } + break; + } + + // U+10000..U+3FFFF F0 90..BF 80..BF 80..BF + case 0xf0: + { + if (JSON_UNLIKELY(not (next_byte_in_range({0x90, 0xBF, 0x80, 0xBF, 0x80, 0xBF})))) + { + return token_type::parse_error; + } + break; + } + + // U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF + case 0xf1: + case 0xf2: + case 0xf3: + { + if (JSON_UNLIKELY(not (next_byte_in_range({0x80, 0xBF, 0x80, 0xBF, 0x80, 0xBF})))) + { + return token_type::parse_error; + } + break; + } + + // U+100000..U+10FFFF F4 80..8F 80..BF 80..BF + case 0xf4: + { + if (JSON_UNLIKELY(not (next_byte_in_range({0x80, 0x8F, 0x80, 0xBF, 0x80, 0xBF})))) + { + return token_type::parse_error; + } + break; + } + + // remaining bytes (80..C1 and F5..FF) are ill-formed + default: + { + error_message = "invalid string: ill-formed UTF-8 byte"; + return token_type::parse_error; + } + } + } + } + + static void strtof(float& f, const char* str, char** endptr) noexcept + { + f = std::strtof(str, endptr); + } + + static void strtof(double& f, const char* str, char** endptr) noexcept + { + f = std::strtod(str, endptr); + } + + static void strtof(long double& f, const char* str, char** endptr) noexcept + { + f = std::strtold(str, endptr); + } + + /*! + @brief scan a number literal + + This function scans a string according to Sect. 6 of RFC 7159. + + The function is realized with a deterministic finite state machine derived + from the grammar described in RFC 7159. Starting in state "init", the + input is read and used to determined the next state. Only state "done" + accepts the number. State "error" is a trap state to model errors. In the + table below, "anything" means any character but the ones listed before. + + state | 0 | 1-9 | e E | + | - | . | anything + ---------|----------|----------|----------|---------|---------|----------|----------- + init | zero | any1 | [error] | [error] | minus | [error] | [error] + minus | zero | any1 | [error] | [error] | [error] | [error] | [error] + zero | done | done | exponent | done | done | decimal1 | done + any1 | any1 | any1 | exponent | done | done | decimal1 | done + decimal1 | decimal2 | [error] | [error] | [error] | [error] | [error] | [error] + decimal2 | decimal2 | decimal2 | exponent | done | done | done | done + exponent | any2 | any2 | [error] | sign | sign | [error] | [error] + sign | any2 | any2 | [error] | [error] | [error] | [error] | [error] + any2 | any2 | any2 | done | done | done | done | done + + The state machine is realized with one label per state (prefixed with + "scan_number_") and `goto` statements between them. The state machine + contains cycles, but any cycle can be left when EOF is read. Therefore, + the function is guaranteed to terminate. + + During scanning, the read bytes are stored in yytext. This string is + then converted to a signed integer, an unsigned integer, or a + floating-point number. + + @return token_type::value_unsigned, token_type::value_integer, or + token_type::value_float if number could be successfully scanned, + token_type::parse_error otherwise + + @note The scanner is independent of the current locale. Internally, the + locale's decimal point is used instead of `.` to work with the + locale-dependent converters. + */ + token_type scan_number() + { + // reset yytext to store the number's bytes + reset(); + + // the type of the parsed number; initially set to unsigned; will be + // changed if minus sign, decimal point or exponent is read + token_type number_type = token_type::value_unsigned; + + // state (init): we just found out we need to scan a number + switch (current) + { + case '-': + { + add(current); + goto scan_number_minus; + } + + case '0': + { + add(current); + goto scan_number_zero; + } + + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_any1; + } + + default: + { + // all other characters are rejected outside scan_number() + assert(false); // LCOV_EXCL_LINE + } + } + +scan_number_minus: + // state: we just parsed a leading minus sign + number_type = token_type::value_integer; + switch (get()) + { + case '0': + { + add(current); + goto scan_number_zero; + } + + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_any1; + } + + default: + { + error_message = "invalid number; expected digit after '-'"; + return token_type::parse_error; + } + } + +scan_number_zero: + // state: we just parse a zero (maybe with a leading minus sign) + switch (get()) + { + case '.': + { + add(decimal_point_char); + goto scan_number_decimal1; + } + + case 'e': + case 'E': + { + add(current); + goto scan_number_exponent; + } + + default: + goto scan_number_done; + } + +scan_number_any1: + // state: we just parsed a number 0-9 (maybe with a leading minus sign) + switch (get()) + { + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_any1; + } + + case '.': + { + add(decimal_point_char); + goto scan_number_decimal1; + } + + case 'e': + case 'E': + { + add(current); + goto scan_number_exponent; + } + + default: + goto scan_number_done; + } + +scan_number_decimal1: + // state: we just parsed a decimal point + number_type = token_type::value_float; + switch (get()) + { + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_decimal2; + } + + default: + { + error_message = "invalid number; expected digit after '.'"; + return token_type::parse_error; + } + } + +scan_number_decimal2: + // we just parsed at least one number after a decimal point + switch (get()) + { + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_decimal2; + } + + case 'e': + case 'E': + { + add(current); + goto scan_number_exponent; + } + + default: + goto scan_number_done; + } + +scan_number_exponent: + // we just parsed an exponent + number_type = token_type::value_float; + switch (get()) + { + case '+': + case '-': + { + add(current); + goto scan_number_sign; + } + + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_any2; + } + + default: + { + error_message = + "invalid number; expected '+', '-', or digit after exponent"; + return token_type::parse_error; + } + } + +scan_number_sign: + // we just parsed an exponent sign + switch (get()) + { + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_any2; + } + + default: + { + error_message = "invalid number; expected digit after exponent sign"; + return token_type::parse_error; + } + } + +scan_number_any2: + // we just parsed a number after the exponent or exponent sign + switch (get()) + { + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + { + add(current); + goto scan_number_any2; + } + + default: + goto scan_number_done; + } + +scan_number_done: + // unget the character after the number (we only read it to know that + // we are done scanning a number) + --chars_read; + next_unget = true; + + // terminate token + add('\0'); + --yylen; + + char* endptr = nullptr; + errno = 0; + + // try to parse integers first and fall back to floats + if (number_type == token_type::value_unsigned) + { + const auto x = std::strtoull(yytext.data(), &endptr, 10); + + // we checked the number format before + assert(endptr == yytext.data() + yylen); + + if (errno == 0) + { + value_unsigned = static_cast(x); + if (value_unsigned == x) + { + return token_type::value_unsigned; + } + } + } + else if (number_type == token_type::value_integer) + { + const auto x = std::strtoll(yytext.data(), &endptr, 10); + + // we checked the number format before + assert(endptr == yytext.data() + yylen); + + if (errno == 0) + { + value_integer = static_cast(x); + if (value_integer == x) + { + return token_type::value_integer; + } + } + } + + // this code is reached if we parse a floating-point number or if an + // integer conversion above failed + strtof(value_float, yytext.data(), &endptr); + + // we checked the number format before + assert(endptr == yytext.data() + yylen); + + return token_type::value_float; + } + + /*! + @param[in] literal_text the literal text to expect + @param[in] length the length of the passed literal text + @param[in] return_type the token type to return on success + */ + token_type scan_literal(const char* literal_text, const std::size_t length, + token_type return_type) + { + assert(current == literal_text[0]); + for (std::size_t i = 1; i < length; ++i) + { + if (JSON_UNLIKELY(get() != literal_text[i])) + { + error_message = "invalid literal"; + return token_type::parse_error; + } + } + return return_type; + } + + ///////////////////// + // input management + ///////////////////// + + /// reset yytext + void reset() noexcept + { + yylen = 0; + start_pos = chars_read - 1; + } + + /// get a character from the input + int get() + { + ++chars_read; + return next_unget ? (next_unget = false, current) + : (current = ia->get_character()); + } + + /// add a character to yytext + void add(int c) + { + // resize yytext if necessary; this condition is deemed unlikely, + // because we start with a 1024-byte buffer + if (JSON_UNLIKELY((yylen + 1 > yytext.capacity()))) + { + yytext.resize(2 * yytext.capacity(), '\0'); + } + assert(yylen < yytext.size()); + yytext[yylen++] = static_cast(c); + } + + public: + ///////////////////// + // value getters + ///////////////////// + + /// return integer value + constexpr number_integer_t get_number_integer() const noexcept + { + return value_integer; + } + + /// return unsigned integer value + constexpr number_unsigned_t get_number_unsigned() const noexcept + { + return value_unsigned; + } + + /// return floating-point value + constexpr number_float_t get_number_float() const noexcept + { + return value_float; + } + + /// return string value + const std::string get_string() + { + // yytext cannot be returned as char*, because it may contain a null + // byte (parsed as "\u0000") + return std::string(yytext.data(), yylen); + } + + ///////////////////// + // diagnostics + ///////////////////// + + /// return position of last read token + constexpr std::size_t get_position() const noexcept + { + return chars_read; + } + + /// return the last read token (for errors only) + std::string get_token_string() const + { + // get the raw byte sequence of the last token + std::string s = ia->read(start_pos, chars_read - start_pos); + + // escape control characters + std::string result; + for (auto c : s) + { + if (c == '\0' or c == std::char_traits::eof()) + { + // ignore EOF + continue; + } + else if ('\x00' <= c and c <= '\x1f') + { + // escape control characters + std::stringstream ss; + ss << "(c) << ">"; + result += ss.str(); + } + else + { + // add character as is + result.push_back(c); + } + } + + return result; + } + + /// return syntax error message + constexpr const char* get_error_message() const noexcept + { + return error_message; + } + + ///////////////////// + // actual scanner + ///////////////////// + + token_type scan() + { + // read next character and ignore whitespace + do + { + get(); + } + while (current == ' ' or current == '\t' or current == '\n' or current == '\r'); + + switch (current) + { + // structural characters + case '[': + return token_type::begin_array; + case ']': + return token_type::end_array; + case '{': + return token_type::begin_object; + case '}': + return token_type::end_object; + case ':': + return token_type::name_separator; + case ',': + return token_type::value_separator; + + // literals + case 't': + return scan_literal("true", 4, token_type::literal_true); + case 'f': + return scan_literal("false", 5, token_type::literal_false); + case 'n': + return scan_literal("null", 4, token_type::literal_null); + + // string + case '\"': + return scan_string(); + + // number + case '-': + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + return scan_number(); + + // end of input (the null byte is needed when parsing from + // string literals) + case '\0': + case std::char_traits::eof(): + return token_type::end_of_input; + + // error + default: + error_message = "invalid literal"; + return token_type::parse_error; + } + } + + private: + /// input adapter + detail::input_adapter_t ia = nullptr; + + /// the current character + int current = std::char_traits::eof(); + + /// whether get() should return the last character again + bool next_unget = false; + + /// the number of characters read + std::size_t chars_read = 0; + /// the start position of the current token + std::size_t start_pos = 0; + + /// buffer for variable-length tokens (numbers, strings) + std::vector yytext = std::vector(1024, '\0'); + /// current index in yytext + std::size_t yylen = 0; + + /// a description of occurred lexer errors + const char* error_message = ""; + + // number values + number_integer_t value_integer = 0; + number_unsigned_t value_unsigned = 0; + number_float_t value_float = 0; + + /// the decimal point + const char decimal_point_char = '.'; +}; + +/*! +@brief syntax analysis + +This class implements a recursive decent parser. +*/ +template +class parser +{ + using number_integer_t = typename BasicJsonType::number_integer_t; + using number_unsigned_t = typename BasicJsonType::number_unsigned_t; + using number_float_t = typename BasicJsonType::number_float_t; + using lexer_t = lexer; + using token_type = typename lexer_t::token_type; + + public: + enum class parse_event_t : uint8_t + { + /// the parser read `{` and started to process a JSON object + object_start, + /// the parser read `}` and finished processing a JSON object + object_end, + /// the parser read `[` and started to process a JSON array + array_start, + /// the parser read `]` and finished processing a JSON array + array_end, + /// the parser read a key of a value in an object + key, + /// the parser finished reading a JSON value + value + }; + + using parser_callback_t = + std::function; + + /// a parser reading from an input adapter + explicit parser(detail::input_adapter_t adapter, + const parser_callback_t cb = nullptr, + const bool allow_exceptions_ = true) + : callback(cb), m_lexer(adapter), allow_exceptions(allow_exceptions_) + {} + + /*! + @brief public parser interface + + @param[in] strict whether to expect the last token to be EOF + @param[in,out] result parsed JSON value + + @throw parse_error.101 in case of an unexpected token + @throw parse_error.102 if to_unicode fails or surrogate error + @throw parse_error.103 if to_unicode fails + */ + void parse(const bool strict, BasicJsonType& result) + { + // read first token + get_token(); + + parse_internal(true, result); + result.assert_invariant(); + + // in strict mode, input must be completely read + if (strict) + { + get_token(); + expect(token_type::end_of_input); + } + + // in case of an error, return discarded value + if (errored) + { + result = value_t::discarded; + return; + } + + // set top-level value to null if it was discarded by the callback + // function + if (result.is_discarded()) + { + result = nullptr; + } + } + + /*! + @brief public accept interface + + @param[in] strict whether to expect the last token to be EOF + @return whether the input is a proper JSON text + */ + bool accept(const bool strict = true) + { + // read first token + get_token(); + + if (not accept_internal()) + { + return false; + } + + // strict => last token must be EOF + return not strict or (get_token() == token_type::end_of_input); + } + + private: + /*! + @brief the actual parser + @throw parse_error.101 in case of an unexpected token + @throw parse_error.102 if to_unicode fails or surrogate error + @throw parse_error.103 if to_unicode fails + */ + void parse_internal(bool keep, BasicJsonType& result) + { + // never parse after a parse error was detected + assert(not errored); + + // start with a discarded value + if (not result.is_discarded()) + { + result.m_value.destroy(result.m_type); + result.m_type = value_t::discarded; + } + + switch (last_token) + { + case token_type::begin_object: + { + if (keep and (not callback or ((keep = callback(depth++, parse_event_t::object_start, result))))) + { + // explicitly set result to object to cope with {} + result.m_type = value_t::object; + result.m_value = value_t::object; + } + + // read next token + get_token(); + + // closing } -> we are done + if (last_token == token_type::end_object) + { + if (keep and callback and not callback(--depth, parse_event_t::object_end, result)) + { + result.m_value.destroy(result.m_type); + result.m_type = value_t::discarded; + } + break; + } + + // parse values + std::string key; + BasicJsonType value; + while (true) + { + // store key + if (not expect(token_type::value_string)) + { + return; + } + key = m_lexer.get_string(); + + bool keep_tag = false; + if (keep) + { + if (callback) + { + BasicJsonType k(key); + keep_tag = callback(depth, parse_event_t::key, k); + } + else + { + keep_tag = true; + } + } + + // parse separator (:) + get_token(); + if (not expect(token_type::name_separator)) + { + return; + } + + // parse and add value + get_token(); + value.m_value.destroy(value.m_type); + value.m_type = value_t::discarded; + parse_internal(keep, value); + + if (JSON_UNLIKELY(errored)) + { + return; + } + + if (keep and keep_tag and not value.is_discarded()) + { + result.m_value.object->emplace(std::move(key), std::move(value)); + } + + // comma -> next value + get_token(); + if (last_token == token_type::value_separator) + { + get_token(); + continue; + } + + // closing } + if (not expect(token_type::end_object)) + { + return; + } + break; + } + + if (keep and callback and not callback(--depth, parse_event_t::object_end, result)) + { + result.m_value.destroy(result.m_type); + result.m_type = value_t::discarded; + } + break; + } + + case token_type::begin_array: + { + if (keep and (not callback or ((keep = callback(depth++, parse_event_t::array_start, result))))) + { + // explicitly set result to object to cope with [] + result.m_type = value_t::array; + result.m_value = value_t::array; + } + + // read next token + get_token(); + + // closing ] -> we are done + if (last_token == token_type::end_array) + { + if (callback and not callback(--depth, parse_event_t::array_end, result)) + { + result.m_value.destroy(result.m_type); + result.m_type = value_t::discarded; + } + break; + } + + // parse values + BasicJsonType value; + while (true) + { + // parse value + value.m_value.destroy(value.m_type); + value.m_type = value_t::discarded; + parse_internal(keep, value); + + if (JSON_UNLIKELY(errored)) + { + return; + } + + if (keep and not value.is_discarded()) + { + result.m_value.array->push_back(std::move(value)); + } + + // comma -> next value + get_token(); + if (last_token == token_type::value_separator) + { + get_token(); + continue; + } + + // closing ] + if (not expect(token_type::end_array)) + { + return; + } + break; + } + + if (keep and callback and not callback(--depth, parse_event_t::array_end, result)) + { + result.m_value.destroy(result.m_type); + result.m_type = value_t::discarded; + } + break; + } + + case token_type::literal_null: + { + result.m_type = value_t::null; + break; + } + + case token_type::value_string: + { + result.m_type = value_t::string; + result.m_value = m_lexer.get_string(); + break; + } + + case token_type::literal_true: + { + result.m_type = value_t::boolean; + result.m_value = true; + break; + } + + case token_type::literal_false: + { + result.m_type = value_t::boolean; + result.m_value = false; + break; + } + + case token_type::value_unsigned: + { + result.m_type = value_t::number_unsigned; + result.m_value = m_lexer.get_number_unsigned(); + break; + } + + case token_type::value_integer: + { + result.m_type = value_t::number_integer; + result.m_value = m_lexer.get_number_integer(); + break; + } + + case token_type::value_float: + { + result.m_type = value_t::number_float; + result.m_value = m_lexer.get_number_float(); + + // throw in case of infinity or NAN + if (JSON_UNLIKELY(not std::isfinite(result.m_value.number_float))) + { + if (allow_exceptions) + { + JSON_THROW(out_of_range::create(406, "number overflow parsing '" + + m_lexer.get_token_string() + "'")); + } + expect(token_type::uninitialized); + } + break; + } + + case token_type::parse_error: + { + // using "uninitialized" to avoid "expected" message + if (not expect(token_type::uninitialized)) + { + return; + } + break; // LCOV_EXCL_LINE + } + + default: + { + // the last token was unexpected; we expected a value + if (not expect(token_type::literal_or_value)) + { + return; + } + break; // LCOV_EXCL_LINE + } + } + + if (keep and callback and not callback(depth, parse_event_t::value, result)) + { + result.m_type = value_t::discarded; + } + } + + /*! + @brief the acutal acceptor + + @invariant 1. The last token is not yet processed. Therefore, the caller + of this function must make sure a token has been read. + 2. When this function returns, the last token is processed. + That is, the last read character was already considered. + + This invariant makes sure that no token needs to be "unput". + */ + bool accept_internal() + { + switch (last_token) + { + case token_type::begin_object: + { + // read next token + get_token(); + + // closing } -> we are done + if (last_token == token_type::end_object) + { + return true; + } + + // parse values + while (true) + { + // parse key + if (last_token != token_type::value_string) + { + return false; + } + + // parse separator (:) + get_token(); + if (last_token != token_type::name_separator) + { + return false; + } + + // parse value + get_token(); + if (not accept_internal()) + { + return false; + } + + // comma -> next value + get_token(); + if (last_token == token_type::value_separator) + { + get_token(); + continue; + } + + // closing } + return (last_token == token_type::end_object); + } + } + + case token_type::begin_array: + { + // read next token + get_token(); + + // closing ] -> we are done + if (last_token == token_type::end_array) + { + return true; + } + + // parse values + while (true) + { + // parse value + if (not accept_internal()) + { + return false; + } + + // comma -> next value + get_token(); + if (last_token == token_type::value_separator) + { + get_token(); + continue; + } + + // closing ] + return (last_token == token_type::end_array); + } + } + + case token_type::value_float: + { + // reject infinity or NAN + return std::isfinite(m_lexer.get_number_float()); + } + + case token_type::literal_false: + case token_type::literal_null: + case token_type::literal_true: + case token_type::value_integer: + case token_type::value_string: + case token_type::value_unsigned: + return true; + + default: // the last token was unexpected + return false; + } + } + + /// get next token from lexer + token_type get_token() + { + return (last_token = m_lexer.scan()); + } + + /*! + @throw parse_error.101 if expected token did not occur + */ + bool expect(token_type t) + { + if (JSON_UNLIKELY(t != last_token)) + { + errored = true; + expected = t; + if (allow_exceptions) + { + throw_exception(); + } + else + { + return false; + } + } + + return true; + } + + [[noreturn]] void throw_exception() const + { + std::string error_msg = "syntax error - "; + if (last_token == token_type::parse_error) + { + error_msg += std::string(m_lexer.get_error_message()) + "; last read: '" + + m_lexer.get_token_string() + "'"; + } + else + { + error_msg += "unexpected " + std::string(lexer_t::token_type_name(last_token)); + } + + if (expected != token_type::uninitialized) + { + error_msg += "; expected " + std::string(lexer_t::token_type_name(expected)); + } + + JSON_THROW(parse_error::create(101, m_lexer.get_position(), error_msg)); + } + + private: + /// current level of recursion + int depth = 0; + /// callback function + const parser_callback_t callback = nullptr; + /// the type of the last read token + token_type last_token = token_type::uninitialized; + /// the lexer + lexer_t m_lexer; + /// whether a syntax error occurred + bool errored = false; + /// possible reason for the syntax error + token_type expected = token_type::uninitialized; + /// whether to throw exceptions in case of errors + const bool allow_exceptions = true; +}; + +/////////////// +// iterators // +/////////////// + +/*! +@brief an iterator for primitive JSON types + +This class models an iterator for primitive JSON types (boolean, number, +string). It's only purpose is to allow the iterator/const_iterator classes +to "iterate" over primitive values. Internally, the iterator is modeled by +a `difference_type` variable. Value begin_value (`0`) models the begin, +end_value (`1`) models past the end. +*/ +class primitive_iterator_t +{ + public: + using difference_type = std::ptrdiff_t; + + constexpr difference_type get_value() const noexcept + { + return m_it; + } + + /// set iterator to a defined beginning + void set_begin() noexcept + { + m_it = begin_value; + } + + /// set iterator to a defined past the end + void set_end() noexcept + { + m_it = end_value; + } + + /// return whether the iterator can be dereferenced + constexpr bool is_begin() const noexcept + { + return (m_it == begin_value); + } + + /// return whether the iterator is at end + constexpr bool is_end() const noexcept + { + return (m_it == end_value); + } + + friend constexpr bool operator==(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return (lhs.m_it == rhs.m_it); + } + + friend constexpr bool operator!=(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return not(lhs == rhs); + } + + friend constexpr bool operator<(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return lhs.m_it < rhs.m_it; + } + + friend constexpr bool operator<=(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return lhs.m_it <= rhs.m_it; + } + + friend constexpr bool operator>(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return lhs.m_it > rhs.m_it; + } + + friend constexpr bool operator>=(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return lhs.m_it >= rhs.m_it; + } + + primitive_iterator_t operator+(difference_type i) + { + auto result = *this; + result += i; + return result; + } + + friend constexpr difference_type operator-(primitive_iterator_t lhs, primitive_iterator_t rhs) noexcept + { + return lhs.m_it - rhs.m_it; + } + + friend std::ostream& operator<<(std::ostream& os, primitive_iterator_t it) + { + return os << it.m_it; + } + + primitive_iterator_t& operator++() + { + ++m_it; + return *this; + } + + primitive_iterator_t operator++(int) + { + auto result = *this; + m_it++; + return result; + } + + primitive_iterator_t& operator--() + { + --m_it; + return *this; + } + + primitive_iterator_t operator--(int) + { + auto result = *this; + m_it--; + return result; + } + + primitive_iterator_t& operator+=(difference_type n) + { + m_it += n; + return *this; + } + + primitive_iterator_t& operator-=(difference_type n) + { + m_it -= n; + return *this; + } + + private: + static constexpr difference_type begin_value = 0; + static constexpr difference_type end_value = begin_value + 1; + + /// iterator as signed integer type + difference_type m_it = std::numeric_limits::denorm_min(); +}; + +/*! +@brief an iterator value + +@note This structure could easily be a union, but MSVC currently does not allow +unions members with complex constructors, see https://github.com/nlohmann/json/pull/105. +*/ +template struct internal_iterator +{ + /// iterator for JSON objects + typename BasicJsonType::object_t::iterator object_iterator {}; + /// iterator for JSON arrays + typename BasicJsonType::array_t::iterator array_iterator {}; + /// generic iterator for all other types + primitive_iterator_t primitive_iterator {}; +}; + +template class iteration_proxy; + +/*! +@brief a template for a random access iterator for the @ref basic_json class + +This class implements a both iterators (iterator and const_iterator) for the +@ref basic_json class. + +@note An iterator is called *initialized* when a pointer to a JSON value has + been set (e.g., by a constructor or a copy assignment). If the iterator is + default-constructed, it is *uninitialized* and most methods are undefined. + **The library uses assertions to detect calls on uninitialized iterators.** + +@requirement The class satisfies the following concept requirements: +- +[RandomAccessIterator](http://en.cppreference.com/w/cpp/concept/RandomAccessIterator): + The iterator that can be moved to point (forward and backward) to any + element in constant time. + +@since version 1.0.0, simplified in version 2.0.9 +*/ +template +class iter_impl : public std::iterator +{ + /// allow basic_json to access private members + friend iter_impl::value, typename std::remove_const::type, const BasicJsonType>::type>; + friend BasicJsonType; + friend iteration_proxy; + + using object_t = typename BasicJsonType::object_t; + using array_t = typename BasicJsonType::array_t; + // make sure BasicJsonType is basic_json or const basic_json + static_assert(is_basic_json::type>::value, + "iter_impl only accepts (const) basic_json"); + + public: + /// the type of the values when the iterator is dereferenced + using value_type = typename BasicJsonType::value_type; + /// a type to represent differences between iterators + using difference_type = typename BasicJsonType::difference_type; + /// defines a pointer to the type iterated over (value_type) + using pointer = typename std::conditional::value, + typename BasicJsonType::const_pointer, + typename BasicJsonType::pointer>::type; + /// defines a reference to the type iterated over (value_type) + using reference = + typename std::conditional::value, + typename BasicJsonType::const_reference, + typename BasicJsonType::reference>::type; + /// the category of the iterator + using iterator_category = std::bidirectional_iterator_tag; + + /// default constructor + iter_impl() = default; + + /*! + @brief constructor for a given JSON instance + @param[in] object pointer to a JSON object for this iterator + @pre object != nullptr + @post The iterator is initialized; i.e. `m_object != nullptr`. + */ + explicit iter_impl(pointer object) noexcept : m_object(object) + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + m_it.object_iterator = typename object_t::iterator(); + break; + } + + case value_t::array: + { + m_it.array_iterator = typename array_t::iterator(); + break; + } + + default: + { + m_it.primitive_iterator = primitive_iterator_t(); + break; + } + } + } + + /*! + @note The conventional copy constructor and copy assignment are implicitly + defined. Combined with the following converting constructor and + assignment, they support: (1) copy from iterator to iterator, (2) + copy from const iterator to const iterator, and (3) conversion from + iterator to const iterator. However conversion from const iterator + to iterator is not defined. + */ + + /*! + @brief converting constructor + @param[in] other non-const iterator to copy from + @note It is not checked whether @a other is initialized. + */ + iter_impl(const iter_impl::type>& other) noexcept + : m_object(other.m_object), m_it(other.m_it) {} + + /*! + @brief converting assignment + @param[in,out] other non-const iterator to copy from + @return const/non-const iterator + @note It is not checked whether @a other is initialized. + */ + iter_impl& operator=(const iter_impl::type>& other) noexcept + { + m_object = other.m_object; + m_it = other.m_it; + return *this; + } + + private: + /*! + @brief set the iterator to the first value + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + void set_begin() noexcept + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + m_it.object_iterator = m_object->m_value.object->begin(); + break; + } + + case value_t::array: + { + m_it.array_iterator = m_object->m_value.array->begin(); + break; + } + + case value_t::null: + { + // set to end so begin()==end() is true: null is empty + m_it.primitive_iterator.set_end(); + break; + } + + default: + { + m_it.primitive_iterator.set_begin(); + break; + } + } + } + + /*! + @brief set the iterator past the last value + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + void set_end() noexcept + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + m_it.object_iterator = m_object->m_value.object->end(); + break; + } + + case value_t::array: + { + m_it.array_iterator = m_object->m_value.array->end(); + break; + } + + default: + { + m_it.primitive_iterator.set_end(); + break; + } + } + } + + public: + /*! + @brief return a reference to the value pointed to by the iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + reference operator*() const + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + assert(m_it.object_iterator != m_object->m_value.object->end()); + return m_it.object_iterator->second; + } + + case value_t::array: + { + assert(m_it.array_iterator != m_object->m_value.array->end()); + return *m_it.array_iterator; + } + + case value_t::null: + JSON_THROW(invalid_iterator::create(214, "cannot get value")); + + default: + { + if (JSON_LIKELY(m_it.primitive_iterator.is_begin())) + { + return *m_object; + } + + JSON_THROW(invalid_iterator::create(214, "cannot get value")); + } + } + } + + /*! + @brief dereference the iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + pointer operator->() const + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + assert(m_it.object_iterator != m_object->m_value.object->end()); + return &(m_it.object_iterator->second); + } + + case value_t::array: + { + assert(m_it.array_iterator != m_object->m_value.array->end()); + return &*m_it.array_iterator; + } + + default: + { + if (JSON_LIKELY(m_it.primitive_iterator.is_begin())) + { + return m_object; + } + + JSON_THROW(invalid_iterator::create(214, "cannot get value")); + } + } + } + + /*! + @brief post-increment (it++) + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl operator++(int) + { + auto result = *this; + ++(*this); + return result; + } + + /*! + @brief pre-increment (++it) + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl& operator++() + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + std::advance(m_it.object_iterator, 1); + break; + } + + case value_t::array: + { + std::advance(m_it.array_iterator, 1); + break; + } + + default: + { + ++m_it.primitive_iterator; + break; + } + } + + return *this; + } + + /*! + @brief post-decrement (it--) + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl operator--(int) + { + auto result = *this; + --(*this); + return result; + } + + /*! + @brief pre-decrement (--it) + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl& operator--() + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + { + std::advance(m_it.object_iterator, -1); + break; + } + + case value_t::array: + { + std::advance(m_it.array_iterator, -1); + break; + } + + default: + { + --m_it.primitive_iterator; + break; + } + } + + return *this; + } + + /*! + @brief comparison: equal + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + bool operator==(const iter_impl& other) const + { + // if objects are not the same, the comparison is undefined + if (JSON_UNLIKELY(m_object != other.m_object)) + { + JSON_THROW(invalid_iterator::create(212, "cannot compare iterators of different containers")); + } + + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + return (m_it.object_iterator == other.m_it.object_iterator); + + case value_t::array: + return (m_it.array_iterator == other.m_it.array_iterator); + + default: + return (m_it.primitive_iterator == other.m_it.primitive_iterator); + } + } + + /*! + @brief comparison: not equal + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + bool operator!=(const iter_impl& other) const + { + return not operator==(other); + } + + /*! + @brief comparison: smaller + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + bool operator<(const iter_impl& other) const + { + // if objects are not the same, the comparison is undefined + if (JSON_UNLIKELY(m_object != other.m_object)) + { + JSON_THROW(invalid_iterator::create(212, "cannot compare iterators of different containers")); + } + + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + JSON_THROW(invalid_iterator::create(213, "cannot compare order of object iterators")); + + case value_t::array: + return (m_it.array_iterator < other.m_it.array_iterator); + + default: + return (m_it.primitive_iterator < other.m_it.primitive_iterator); + } + } + + /*! + @brief comparison: less than or equal + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + bool operator<=(const iter_impl& other) const + { + return not other.operator < (*this); + } + + /*! + @brief comparison: greater than + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + bool operator>(const iter_impl& other) const + { + return not operator<=(other); + } + + /*! + @brief comparison: greater than or equal + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + bool operator>=(const iter_impl& other) const + { + return not operator<(other); + } + + /*! + @brief add to iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl& operator+=(difference_type i) + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + JSON_THROW(invalid_iterator::create(209, "cannot use offsets with object iterators")); + + case value_t::array: + { + std::advance(m_it.array_iterator, i); + break; + } + + default: + { + m_it.primitive_iterator += i; + break; + } + } + + return *this; + } + + /*! + @brief subtract from iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl& operator-=(difference_type i) + { + return operator+=(-i); + } + + /*! + @brief add to iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl operator+(difference_type i) const + { + auto result = *this; + result += i; + return result; + } + + /*! + @brief addition of distance and iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + friend iter_impl operator+(difference_type i, const iter_impl& it) + { + auto result = it; + result += i; + return result; + } + + /*! + @brief subtract from iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + iter_impl operator-(difference_type i) const + { + auto result = *this; + result -= i; + return result; + } + + /*! + @brief return difference + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + difference_type operator-(const iter_impl& other) const + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + JSON_THROW(invalid_iterator::create(209, "cannot use offsets with object iterators")); + + case value_t::array: + return m_it.array_iterator - other.m_it.array_iterator; + + default: + return m_it.primitive_iterator - other.m_it.primitive_iterator; + } + } + + /*! + @brief access to successor + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + reference operator[](difference_type n) const + { + assert(m_object != nullptr); + + switch (m_object->m_type) + { + case value_t::object: + JSON_THROW(invalid_iterator::create(208, "cannot use operator[] for object iterators")); + + case value_t::array: + return *std::next(m_it.array_iterator, n); + + case value_t::null: + JSON_THROW(invalid_iterator::create(214, "cannot get value")); + + default: + { + if (JSON_LIKELY(m_it.primitive_iterator.get_value() == -n)) + { + return *m_object; + } + + JSON_THROW(invalid_iterator::create(214, "cannot get value")); + } + } + } + + /*! + @brief return the key of an object iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + typename object_t::key_type key() const + { + assert(m_object != nullptr); + + if (JSON_LIKELY(m_object->is_object())) + { + return m_it.object_iterator->first; + } + + JSON_THROW(invalid_iterator::create(207, "cannot use key() for non-object iterators")); + } + + /*! + @brief return the value of an iterator + @pre The iterator is initialized; i.e. `m_object != nullptr`. + */ + reference value() const + { + return operator*(); + } + + private: + /// associated JSON instance + pointer m_object = nullptr; + /// the actual iterator of the associated instance + internal_iterator::type> m_it = {}; +}; + +/// proxy class for the iterator_wrapper functions +template class iteration_proxy +{ + private: + /// helper class for iteration + class iteration_proxy_internal + { + private: + /// the iterator + IteratorType anchor; + /// an index for arrays (used to create key names) + std::size_t array_index = 0; + + public: + explicit iteration_proxy_internal(IteratorType it) noexcept : anchor(it) {} + + /// dereference operator (needed for range-based for) + iteration_proxy_internal& operator*() + { + return *this; + } + + /// increment operator (needed for range-based for) + iteration_proxy_internal& operator++() + { + ++anchor; + ++array_index; + + return *this; + } + + /// inequality operator (needed for range-based for) + bool operator!=(const iteration_proxy_internal& o) const noexcept + { + return anchor != o.anchor; + } + + /// return key of the iterator + std::string key() const + { + assert(anchor.m_object != nullptr); + + switch (anchor.m_object->type()) + { + // use integer array index as key + case value_t::array: + return std::to_string(array_index); + + // use key from the object + case value_t::object: + return anchor.key(); + + // use an empty key for all primitive types + default: + return ""; + } + } + + /// return value of the iterator + typename IteratorType::reference value() const + { + return anchor.value(); + } + }; + + /// the container to iterate + typename IteratorType::reference container; + + public: + /// construct iteration proxy from a container + explicit iteration_proxy(typename IteratorType::reference cont) + : container(cont) {} + + /// return iterator begin (needed for range-based for) + iteration_proxy_internal begin() noexcept + { + return iteration_proxy_internal(container.begin()); + } + + /// return iterator end (needed for range-based for) + iteration_proxy_internal end() noexcept + { + return iteration_proxy_internal(container.end()); + } +}; + +/*! +@brief a template for a reverse iterator class + +@tparam Base the base iterator type to reverse. Valid types are @ref +iterator (to create @ref reverse_iterator) and @ref const_iterator (to +create @ref const_reverse_iterator). + +@requirement The class satisfies the following concept requirements: +- +[RandomAccessIterator](http://en.cppreference.com/w/cpp/concept/RandomAccessIterator): + The iterator that can be moved to point (forward and backward) to any + element in constant time. +- [OutputIterator](http://en.cppreference.com/w/cpp/concept/OutputIterator): + It is possible to write to the pointed-to element (only if @a Base is + @ref iterator). + +@since version 1.0.0 +*/ +template +class json_reverse_iterator : public std::reverse_iterator +{ + public: + using difference_type = std::ptrdiff_t; + /// shortcut to the reverse iterator adaptor + using base_iterator = std::reverse_iterator; + /// the reference type for the pointed-to element + using reference = typename Base::reference; + + /// create reverse iterator from iterator + json_reverse_iterator(const typename base_iterator::iterator_type& it) noexcept + : base_iterator(it) {} + + /// create reverse iterator from base class + json_reverse_iterator(const base_iterator& it) noexcept : base_iterator(it) {} + + /// post-increment (it++) + json_reverse_iterator operator++(int) + { + return static_cast(base_iterator::operator++(1)); + } + + /// pre-increment (++it) + json_reverse_iterator& operator++() + { + return static_cast(base_iterator::operator++()); + } + + /// post-decrement (it--) + json_reverse_iterator operator--(int) + { + return static_cast(base_iterator::operator--(1)); + } + + /// pre-decrement (--it) + json_reverse_iterator& operator--() + { + return static_cast(base_iterator::operator--()); + } + + /// add to iterator + json_reverse_iterator& operator+=(difference_type i) + { + return static_cast(base_iterator::operator+=(i)); + } + + /// add to iterator + json_reverse_iterator operator+(difference_type i) const + { + return static_cast(base_iterator::operator+(i)); + } + + /// subtract from iterator + json_reverse_iterator operator-(difference_type i) const + { + return static_cast(base_iterator::operator-(i)); + } + + /// return difference + difference_type operator-(const json_reverse_iterator& other) const + { + return base_iterator(*this) - base_iterator(other); + } + + /// access to successor + reference operator[](difference_type n) const + { + return *(this->operator+(n)); + } + + /// return the key of an object iterator + auto key() const -> decltype(std::declval().key()) + { + auto it = --this->base(); + return it.key(); + } + + /// return the value of an iterator + reference value() const + { + auto it = --this->base(); + return it.operator * (); + } +}; + +///////////////////// +// output adapters // +///////////////////// + +/// abstract output adapter interface +template struct output_adapter_protocol +{ + virtual void write_character(CharType c) = 0; + virtual void write_characters(const CharType* s, std::size_t length) = 0; + virtual ~output_adapter_protocol() = default; +}; + +/// a type to simplify interfaces +template +using output_adapter_t = std::shared_ptr>; + +/// output adapter for byte vectors +template +class output_vector_adapter : public output_adapter_protocol +{ + public: + explicit output_vector_adapter(std::vector& vec) : v(vec) {} + + void write_character(CharType c) override + { + v.push_back(c); + } + + void write_characters(const CharType* s, std::size_t length) override + { + std::copy(s, s + length, std::back_inserter(v)); + } + + private: + std::vector& v; +}; + +/// output adapter for output streams +template +class output_stream_adapter : public output_adapter_protocol +{ + public: + explicit output_stream_adapter(std::basic_ostream& s) : stream(s) {} + + void write_character(CharType c) override + { + stream.put(c); + } + + void write_characters(const CharType* s, std::size_t length) override + { + stream.write(s, static_cast(length)); + } + + private: + std::basic_ostream& stream; +}; + +/// output adapter for basic_string +template +class output_string_adapter : public output_adapter_protocol +{ + public: + explicit output_string_adapter(std::basic_string& s) : str(s) {} + + void write_character(CharType c) override + { + str.push_back(c); + } + + void write_characters(const CharType* s, std::size_t length) override + { + str.append(s, length); + } + + private: + std::basic_string& str; +}; + +template +class output_adapter +{ + public: + output_adapter(std::vector& vec) + : oa(std::make_shared>(vec)) {} + + output_adapter(std::basic_ostream& s) + : oa(std::make_shared>(s)) {} + + output_adapter(std::basic_string& s) + : oa(std::make_shared>(s)) {} + + operator output_adapter_t() + { + return oa; + } + + private: + output_adapter_t oa = nullptr; +}; + +////////////////////////////// +// binary reader and writer // +////////////////////////////// + +/*! +@brief deserialization of CBOR and MessagePack values +*/ +template +class binary_reader +{ + using number_integer_t = typename BasicJsonType::number_integer_t; + using number_unsigned_t = typename BasicJsonType::number_unsigned_t; + + public: + /*! + @brief create a binary reader + + @param[in] adapter input adapter to read from + */ + explicit binary_reader(input_adapter_t adapter) : ia(std::move(adapter)) + { + assert(ia); + } + + /*! + @brief create a JSON value from CBOR input + + @param[in] strict whether to expect the input to be consumed completed + @return JSON value created from CBOR input + + @throw parse_error.110 if input ended unexpectedly or the end of file was + not reached when @a strict was set to true + @throw parse_error.112 if unsupported byte was read + */ + BasicJsonType parse_cbor(const bool strict) + { + const auto res = parse_cbor_internal(); + if (strict) + { + get(); + check_eof(true); + } + return res; + } + + /*! + @brief create a JSON value from MessagePack input + + @param[in] strict whether to expect the input to be consumed completed + @return JSON value created from MessagePack input + + @throw parse_error.110 if input ended unexpectedly or the end of file was + not reached when @a strict was set to true + @throw parse_error.112 if unsupported byte was read + */ + BasicJsonType parse_msgpack(const bool strict) + { + const auto res = parse_msgpack_internal(); + if (strict) + { + get(); + check_eof(true); + } + return res; + } + + /*! + @brief determine system byte order + + @return true if and only if system's byte order is little endian + + @note from http://stackoverflow.com/a/1001328/266378 + */ + static constexpr bool little_endianess(int num = 1) noexcept + { + return (*reinterpret_cast(&num) == 1); + } + + private: + /*! + @param[in] get_char whether a new character should be retrieved from the + input (true, default) or whether the last read + character should be considered instead + */ + BasicJsonType parse_cbor_internal(const bool get_char = true) + { + switch (get_char ? get() : current) + { + // EOF + case std::char_traits::eof(): + JSON_THROW(parse_error::create(110, chars_read, "unexpected end of input")); + + // Integer 0x00..0x17 (0..23) + case 0x00: + case 0x01: + case 0x02: + case 0x03: + case 0x04: + case 0x05: + case 0x06: + case 0x07: + case 0x08: + case 0x09: + case 0x0a: + case 0x0b: + case 0x0c: + case 0x0d: + case 0x0e: + case 0x0f: + case 0x10: + case 0x11: + case 0x12: + case 0x13: + case 0x14: + case 0x15: + case 0x16: + case 0x17: + return static_cast(current); + + case 0x18: // Unsigned integer (one-byte uint8_t follows) + return get_number(); + + case 0x19: // Unsigned integer (two-byte uint16_t follows) + return get_number(); + + case 0x1a: // Unsigned integer (four-byte uint32_t follows) + return get_number(); + + case 0x1b: // Unsigned integer (eight-byte uint64_t follows) + return get_number(); + + // Negative integer -1-0x00..-1-0x17 (-1..-24) + case 0x20: + case 0x21: + case 0x22: + case 0x23: + case 0x24: + case 0x25: + case 0x26: + case 0x27: + case 0x28: + case 0x29: + case 0x2a: + case 0x2b: + case 0x2c: + case 0x2d: + case 0x2e: + case 0x2f: + case 0x30: + case 0x31: + case 0x32: + case 0x33: + case 0x34: + case 0x35: + case 0x36: + case 0x37: + return static_cast(0x20 - 1 - current); + + case 0x38: // Negative integer (one-byte uint8_t follows) + { + // must be uint8_t ! + return static_cast(-1) - get_number(); + } + + case 0x39: // Negative integer -1-n (two-byte uint16_t follows) + { + return static_cast(-1) - get_number(); + } + + case 0x3a: // Negative integer -1-n (four-byte uint32_t follows) + { + return static_cast(-1) - get_number(); + } + + case 0x3b: // Negative integer -1-n (eight-byte uint64_t follows) + { + return static_cast(-1) - + static_cast(get_number()); + } + + // UTF-8 string (0x00..0x17 bytes follow) + case 0x60: + case 0x61: + case 0x62: + case 0x63: + case 0x64: + case 0x65: + case 0x66: + case 0x67: + case 0x68: + case 0x69: + case 0x6a: + case 0x6b: + case 0x6c: + case 0x6d: + case 0x6e: + case 0x6f: + case 0x70: + case 0x71: + case 0x72: + case 0x73: + case 0x74: + case 0x75: + case 0x76: + case 0x77: + case 0x78: // UTF-8 string (one-byte uint8_t for n follows) + case 0x79: // UTF-8 string (two-byte uint16_t for n follow) + case 0x7a: // UTF-8 string (four-byte uint32_t for n follow) + case 0x7b: // UTF-8 string (eight-byte uint64_t for n follow) + case 0x7f: // UTF-8 string (indefinite length) + { + return get_cbor_string(); + } + + // array (0x00..0x17 data items follow) + case 0x80: + case 0x81: + case 0x82: + case 0x83: + case 0x84: + case 0x85: + case 0x86: + case 0x87: + case 0x88: + case 0x89: + case 0x8a: + case 0x8b: + case 0x8c: + case 0x8d: + case 0x8e: + case 0x8f: + case 0x90: + case 0x91: + case 0x92: + case 0x93: + case 0x94: + case 0x95: + case 0x96: + case 0x97: + { + return get_cbor_array(current & 0x1f); + } + + case 0x98: // array (one-byte uint8_t for n follows) + { + return get_cbor_array(get_number()); + } + + case 0x99: // array (two-byte uint16_t for n follow) + { + return get_cbor_array(get_number()); + } + + case 0x9a: // array (four-byte uint32_t for n follow) + { + return get_cbor_array(get_number()); + } + + case 0x9b: // array (eight-byte uint64_t for n follow) + { + return get_cbor_array(get_number()); + } + + case 0x9f: // array (indefinite length) + { + BasicJsonType result = value_t::array; + while (get() != 0xff) + { + result.push_back(parse_cbor_internal(false)); + } + return result; + } + + // map (0x00..0x17 pairs of data items follow) + case 0xa0: + case 0xa1: + case 0xa2: + case 0xa3: + case 0xa4: + case 0xa5: + case 0xa6: + case 0xa7: + case 0xa8: + case 0xa9: + case 0xaa: + case 0xab: + case 0xac: + case 0xad: + case 0xae: + case 0xaf: + case 0xb0: + case 0xb1: + case 0xb2: + case 0xb3: + case 0xb4: + case 0xb5: + case 0xb6: + case 0xb7: + { + return get_cbor_object(current & 0x1f); + } + + case 0xb8: // map (one-byte uint8_t for n follows) + { + return get_cbor_object(get_number()); + } + + case 0xb9: // map (two-byte uint16_t for n follow) + { + return get_cbor_object(get_number()); + } + + case 0xba: // map (four-byte uint32_t for n follow) + { + return get_cbor_object(get_number()); + } + + case 0xbb: // map (eight-byte uint64_t for n follow) + { + return get_cbor_object(get_number()); + } + + case 0xbf: // map (indefinite length) + { + BasicJsonType result = value_t::object; + while (get() != 0xff) + { + auto key = get_cbor_string(); + result[key] = parse_cbor_internal(); + } + return result; + } + + case 0xf4: // false + { + return false; + } + + case 0xf5: // true + { + return true; + } + + case 0xf6: // null + { + return value_t::null; + } + + case 0xf9: // Half-Precision Float (two-byte IEEE 754) + { + const int byte1 = get(); + check_eof(); + const int byte2 = get(); + check_eof(); + + // code from RFC 7049, Appendix D, Figure 3: + // As half-precision floating-point numbers were only added + // to IEEE 754 in 2008, today's programming platforms often + // still only have limited support for them. It is very + // easy to include at least decoding support for them even + // without such support. An example of a small decoder for + // half-precision floating-point numbers in the C language + // is shown in Fig. 3. + const int half = (byte1 << 8) + byte2; + const int exp = (half >> 10) & 0x1f; + const int mant = half & 0x3ff; + double val; + if (exp == 0) + { + val = std::ldexp(mant, -24); + } + else if (exp != 31) + { + val = std::ldexp(mant + 1024, exp - 25); + } + else + { + val = (mant == 0) ? std::numeric_limits::infinity() + : std::numeric_limits::quiet_NaN(); + } + return (half & 0x8000) != 0 ? -val : val; + } + + case 0xfa: // Single-Precision Float (four-byte IEEE 754) + { + return get_number(); + } + + case 0xfb: // Double-Precision Float (eight-byte IEEE 754) + { + return get_number(); + } + + default: // anything else (0xFF is handled inside the other types) + { + std::stringstream ss; + ss << std::setw(2) << std::setfill('0') << std::hex << current; + JSON_THROW(parse_error::create(112, chars_read, "error reading CBOR; last byte: 0x" + ss.str())); + } + } + } + + BasicJsonType parse_msgpack_internal() + { + switch (get()) + { + // EOF + case std::char_traits::eof(): + JSON_THROW(parse_error::create(110, chars_read, "unexpected end of input")); + + // positive fixint + case 0x00: + case 0x01: + case 0x02: + case 0x03: + case 0x04: + case 0x05: + case 0x06: + case 0x07: + case 0x08: + case 0x09: + case 0x0a: + case 0x0b: + case 0x0c: + case 0x0d: + case 0x0e: + case 0x0f: + case 0x10: + case 0x11: + case 0x12: + case 0x13: + case 0x14: + case 0x15: + case 0x16: + case 0x17: + case 0x18: + case 0x19: + case 0x1a: + case 0x1b: + case 0x1c: + case 0x1d: + case 0x1e: + case 0x1f: + case 0x20: + case 0x21: + case 0x22: + case 0x23: + case 0x24: + case 0x25: + case 0x26: + case 0x27: + case 0x28: + case 0x29: + case 0x2a: + case 0x2b: + case 0x2c: + case 0x2d: + case 0x2e: + case 0x2f: + case 0x30: + case 0x31: + case 0x32: + case 0x33: + case 0x34: + case 0x35: + case 0x36: + case 0x37: + case 0x38: + case 0x39: + case 0x3a: + case 0x3b: + case 0x3c: + case 0x3d: + case 0x3e: + case 0x3f: + case 0x40: + case 0x41: + case 0x42: + case 0x43: + case 0x44: + case 0x45: + case 0x46: + case 0x47: + case 0x48: + case 0x49: + case 0x4a: + case 0x4b: + case 0x4c: + case 0x4d: + case 0x4e: + case 0x4f: + case 0x50: + case 0x51: + case 0x52: + case 0x53: + case 0x54: + case 0x55: + case 0x56: + case 0x57: + case 0x58: + case 0x59: + case 0x5a: + case 0x5b: + case 0x5c: + case 0x5d: + case 0x5e: + case 0x5f: + case 0x60: + case 0x61: + case 0x62: + case 0x63: + case 0x64: + case 0x65: + case 0x66: + case 0x67: + case 0x68: + case 0x69: + case 0x6a: + case 0x6b: + case 0x6c: + case 0x6d: + case 0x6e: + case 0x6f: + case 0x70: + case 0x71: + case 0x72: + case 0x73: + case 0x74: + case 0x75: + case 0x76: + case 0x77: + case 0x78: + case 0x79: + case 0x7a: + case 0x7b: + case 0x7c: + case 0x7d: + case 0x7e: + case 0x7f: + return static_cast(current); + + // fixmap + case 0x80: + case 0x81: + case 0x82: + case 0x83: + case 0x84: + case 0x85: + case 0x86: + case 0x87: + case 0x88: + case 0x89: + case 0x8a: + case 0x8b: + case 0x8c: + case 0x8d: + case 0x8e: + case 0x8f: + { + return get_msgpack_object(current & 0x0f); + } + + // fixarray + case 0x90: + case 0x91: + case 0x92: + case 0x93: + case 0x94: + case 0x95: + case 0x96: + case 0x97: + case 0x98: + case 0x99: + case 0x9a: + case 0x9b: + case 0x9c: + case 0x9d: + case 0x9e: + case 0x9f: + { + return get_msgpack_array(current & 0x0f); + } + + // fixstr + case 0xa0: + case 0xa1: + case 0xa2: + case 0xa3: + case 0xa4: + case 0xa5: + case 0xa6: + case 0xa7: + case 0xa8: + case 0xa9: + case 0xaa: + case 0xab: + case 0xac: + case 0xad: + case 0xae: + case 0xaf: + case 0xb0: + case 0xb1: + case 0xb2: + case 0xb3: + case 0xb4: + case 0xb5: + case 0xb6: + case 0xb7: + case 0xb8: + case 0xb9: + case 0xba: + case 0xbb: + case 0xbc: + case 0xbd: + case 0xbe: + case 0xbf: + return get_msgpack_string(); + + case 0xc0: // nil + return value_t::null; + + case 0xc2: // false + return false; + + case 0xc3: // true + return true; + + case 0xca: // float 32 + return get_number(); + + case 0xcb: // float 64 + return get_number(); + + case 0xcc: // uint 8 + return get_number(); + + case 0xcd: // uint 16 + return get_number(); + + case 0xce: // uint 32 + return get_number(); + + case 0xcf: // uint 64 + return get_number(); + + case 0xd0: // int 8 + return get_number(); + + case 0xd1: // int 16 + return get_number(); + + case 0xd2: // int 32 + return get_number(); + + case 0xd3: // int 64 + return get_number(); + + case 0xd9: // str 8 + case 0xda: // str 16 + case 0xdb: // str 32 + return get_msgpack_string(); + + case 0xdc: // array 16 + { + return get_msgpack_array(get_number()); + } + + case 0xdd: // array 32 + { + return get_msgpack_array(get_number()); + } + + case 0xde: // map 16 + { + return get_msgpack_object(get_number()); + } + + case 0xdf: // map 32 + { + return get_msgpack_object(get_number()); + } + + // positive fixint + case 0xe0: + case 0xe1: + case 0xe2: + case 0xe3: + case 0xe4: + case 0xe5: + case 0xe6: + case 0xe7: + case 0xe8: + case 0xe9: + case 0xea: + case 0xeb: + case 0xec: + case 0xed: + case 0xee: + case 0xef: + case 0xf0: + case 0xf1: + case 0xf2: + case 0xf3: + case 0xf4: + case 0xf5: + case 0xf6: + case 0xf7: + case 0xf8: + case 0xf9: + case 0xfa: + case 0xfb: + case 0xfc: + case 0xfd: + case 0xfe: + case 0xff: + return static_cast(current); + + default: // anything else + { + std::stringstream ss; + ss << std::setw(2) << std::setfill('0') << std::hex << current; + JSON_THROW(parse_error::create(112, chars_read, + "error reading MessagePack; last byte: 0x" + ss.str())); + } + } + } + + /*! + @brief get next character from the input + + This function provides the interface to the used input adapter. It does + not throw in case the input reached EOF, but returns + `std::char_traits::eof()` in that case. + + @return character read from the input + */ + int get() + { + ++chars_read; + return (current = ia->get_character()); + } + + /* + @brief read a number from the input + + @tparam NumberType the type of the number + + @return number of type @a NumberType + + @note This function needs to respect the system's endianess, because + bytes in CBOR and MessagePack are stored in network order (big + endian) and therefore need reordering on little endian systems. + + @throw parse_error.110 if input has less than `sizeof(NumberType)` bytes + */ + template NumberType get_number() + { + // step 1: read input into array with system's byte order + std::array vec; + for (std::size_t i = 0; i < sizeof(NumberType); ++i) + { + get(); + check_eof(); + + // reverse byte order prior to conversion if necessary + if (is_little_endian) + { + vec[sizeof(NumberType) - i - 1] = static_cast(current); + } + else + { + vec[i] = static_cast(current); // LCOV_EXCL_LINE + } + } + + // step 2: convert array into number of type T and return + NumberType result; + std::memcpy(&result, vec.data(), sizeof(NumberType)); + return result; + } + + /*! + @brief create a string by reading characters from the input + + @param[in] len number of bytes to read + + @note We can not reserve @a len bytes for the result, because @a len + may be too large. Usually, @ref check_eof() detects the end of + the input before we run out of string memory. + + @return string created by reading @a len bytes + + @throw parse_error.110 if input has less than @a len bytes + */ + template + std::string get_string(const NumberType len) + { + std::string result; + std::generate_n(std::back_inserter(result), len, [this]() + { + get(); + check_eof(); + return current; + }); + return result; + } + + /*! + @brief reads a CBOR string + + This function first reads starting bytes to determine the expected + string length and then copies this number of bytes into a string. + Additionally, CBOR's strings with indefinite lengths are supported. + + @return string + + @throw parse_error.110 if input ended + @throw parse_error.113 if an unexpected byte is read + */ + std::string get_cbor_string() + { + check_eof(); + + switch (current) + { + // UTF-8 string (0x00..0x17 bytes follow) + case 0x60: + case 0x61: + case 0x62: + case 0x63: + case 0x64: + case 0x65: + case 0x66: + case 0x67: + case 0x68: + case 0x69: + case 0x6a: + case 0x6b: + case 0x6c: + case 0x6d: + case 0x6e: + case 0x6f: + case 0x70: + case 0x71: + case 0x72: + case 0x73: + case 0x74: + case 0x75: + case 0x76: + case 0x77: + { + return get_string(current & 0x1f); + } + + case 0x78: // UTF-8 string (one-byte uint8_t for n follows) + { + return get_string(get_number()); + } + + case 0x79: // UTF-8 string (two-byte uint16_t for n follow) + { + return get_string(get_number()); + } + + case 0x7a: // UTF-8 string (four-byte uint32_t for n follow) + { + return get_string(get_number()); + } + + case 0x7b: // UTF-8 string (eight-byte uint64_t for n follow) + { + return get_string(get_number()); + } + + case 0x7f: // UTF-8 string (indefinite length) + { + std::string result; + while (get() != 0xff) + { + check_eof(); + result.push_back(static_cast(current)); + } + return result; + } + + default: + { + std::stringstream ss; + ss << std::setw(2) << std::setfill('0') << std::hex << current; + JSON_THROW(parse_error::create(113, chars_read, "expected a CBOR string; last byte: 0x" + ss.str())); + } + } + } + + template + BasicJsonType get_cbor_array(const NumberType len) + { + BasicJsonType result = value_t::array; + std::generate_n(std::back_inserter(*result.m_value.array), len, [this]() + { + return parse_cbor_internal(); + }); + return result; + } + + template + BasicJsonType get_cbor_object(const NumberType len) + { + BasicJsonType result = value_t::object; + std::generate_n(std::inserter(*result.m_value.object, + result.m_value.object->end()), + len, [this]() + { + get(); + auto key = get_cbor_string(); + auto val = parse_cbor_internal(); + return std::make_pair(std::move(key), std::move(val)); + }); + return result; + } + + /*! + @brief reads a MessagePack string + + This function first reads starting bytes to determine the expected + string length and then copies this number of bytes into a string. + + @return string + + @throw parse_error.110 if input ended + @throw parse_error.113 if an unexpected byte is read + */ + std::string get_msgpack_string() + { + check_eof(); + + switch (current) + { + // fixstr + case 0xa0: + case 0xa1: + case 0xa2: + case 0xa3: + case 0xa4: + case 0xa5: + case 0xa6: + case 0xa7: + case 0xa8: + case 0xa9: + case 0xaa: + case 0xab: + case 0xac: + case 0xad: + case 0xae: + case 0xaf: + case 0xb0: + case 0xb1: + case 0xb2: + case 0xb3: + case 0xb4: + case 0xb5: + case 0xb6: + case 0xb7: + case 0xb8: + case 0xb9: + case 0xba: + case 0xbb: + case 0xbc: + case 0xbd: + case 0xbe: + case 0xbf: + { + return get_string(current & 0x1f); + } + + case 0xd9: // str 8 + { + return get_string(get_number()); + } + + case 0xda: // str 16 + { + return get_string(get_number()); + } + + case 0xdb: // str 32 + { + return get_string(get_number()); + } + + default: + { + std::stringstream ss; + ss << std::setw(2) << std::setfill('0') << std::hex << current; + JSON_THROW(parse_error::create(113, chars_read, + "expected a MessagePack string; last byte: 0x" + ss.str())); + } + } + } + + template + BasicJsonType get_msgpack_array(const NumberType len) + { + BasicJsonType result = value_t::array; + std::generate_n(std::back_inserter(*result.m_value.array), len, [this]() + { + return parse_msgpack_internal(); + }); + return result; + } + + template + BasicJsonType get_msgpack_object(const NumberType len) + { + BasicJsonType result = value_t::object; + std::generate_n(std::inserter(*result.m_value.object, + result.m_value.object->end()), + len, [this]() + { + get(); + auto key = get_msgpack_string(); + auto val = parse_msgpack_internal(); + return std::make_pair(std::move(key), std::move(val)); + }); + return result; + } + + /*! + @brief check if input ended + @throw parse_error.110 if input ended + */ + void check_eof(const bool expect_eof = false) const + { + if (expect_eof) + { + if (JSON_UNLIKELY(current != std::char_traits::eof())) + { + JSON_THROW(parse_error::create(110, chars_read, "expected end of input")); + } + } + else + { + if (JSON_UNLIKELY(current == std::char_traits::eof())) + { + JSON_THROW(parse_error::create(110, chars_read, "unexpected end of input")); + } + } + } + + private: + /// input adapter + input_adapter_t ia = nullptr; + + /// the current character + int current = std::char_traits::eof(); + + /// the number of characters read + std::size_t chars_read = 0; + + /// whether we can assume little endianess + const bool is_little_endian = little_endianess(); +}; + +/*! +@brief serialization to CBOR and MessagePack values +*/ +template +class binary_writer +{ + public: + /*! + @brief create a binary writer + + @param[in] adapter output adapter to write to + */ + explicit binary_writer(output_adapter_t adapter) : oa(adapter) + { + assert(oa); + } + + /*! + @brief[in] j JSON value to serialize + */ + void write_cbor(const BasicJsonType& j) + { + switch (j.type()) + { + case value_t::null: + { + oa->write_character(static_cast(0xf6)); + break; + } + + case value_t::boolean: + { + oa->write_character(j.m_value.boolean + ? static_cast(0xf5) + : static_cast(0xf4)); + break; + } + + case value_t::number_integer: + { + if (j.m_value.number_integer >= 0) + { + // CBOR does not differentiate between positive signed + // integers and unsigned integers. Therefore, we used the + // code from the value_t::number_unsigned case here. + if (j.m_value.number_integer <= 0x17) + { + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x18)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x19)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x1a)); + write_number(static_cast(j.m_value.number_integer)); + } + else + { + oa->write_character(static_cast(0x1b)); + write_number(static_cast(j.m_value.number_integer)); + } + } + else + { + // The conversions below encode the sign in the first + // byte, and the value is converted to a positive number. + const auto positive_number = -1 - j.m_value.number_integer; + if (j.m_value.number_integer >= -24) + { + write_number(static_cast(0x20 + positive_number)); + } + else if (positive_number <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x38)); + write_number(static_cast(positive_number)); + } + else if (positive_number <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x39)); + write_number(static_cast(positive_number)); + } + else if (positive_number <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x3a)); + write_number(static_cast(positive_number)); + } + else + { + oa->write_character(static_cast(0x3b)); + write_number(static_cast(positive_number)); + } + } + break; + } + + case value_t::number_unsigned: + { + if (j.m_value.number_unsigned <= 0x17) + { + write_number(static_cast(j.m_value.number_unsigned)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x18)); + write_number(static_cast(j.m_value.number_unsigned)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x19)); + write_number(static_cast(j.m_value.number_unsigned)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + oa->write_character(static_cast(0x1a)); + write_number(static_cast(j.m_value.number_unsigned)); + } + else + { + oa->write_character(static_cast(0x1b)); + write_number(static_cast(j.m_value.number_unsigned)); + } + break; + } + + case value_t::number_float: // Double-Precision Float + { + oa->write_character(static_cast(0xfb)); + write_number(j.m_value.number_float); + break; + } + + case value_t::string: + { + // step 1: write control byte and the string length + const auto N = j.m_value.string->size(); + if (N <= 0x17) + { + write_number(static_cast(0x60 + N)); + } + else if (N <= 0xff) + { + oa->write_character(static_cast(0x78)); + write_number(static_cast(N)); + } + else if (N <= 0xffff) + { + oa->write_character(static_cast(0x79)); + write_number(static_cast(N)); + } + else if (N <= 0xffffffff) + { + oa->write_character(static_cast(0x7a)); + write_number(static_cast(N)); + } + // LCOV_EXCL_START + else if (N <= 0xffffffffffffffff) + { + oa->write_character(static_cast(0x7b)); + write_number(static_cast(N)); + } + // LCOV_EXCL_STOP + + // step 2: write the string + oa->write_characters( + reinterpret_cast(j.m_value.string->c_str()), + j.m_value.string->size()); + break; + } + + case value_t::array: + { + // step 1: write control byte and the array size + const auto N = j.m_value.array->size(); + if (N <= 0x17) + { + write_number(static_cast(0x80 + N)); + } + else if (N <= 0xff) + { + oa->write_character(static_cast(0x98)); + write_number(static_cast(N)); + } + else if (N <= 0xffff) + { + oa->write_character(static_cast(0x99)); + write_number(static_cast(N)); + } + else if (N <= 0xffffffff) + { + oa->write_character(static_cast(0x9a)); + write_number(static_cast(N)); + } + // LCOV_EXCL_START + else if (N <= 0xffffffffffffffff) + { + oa->write_character(static_cast(0x9b)); + write_number(static_cast(N)); + } + // LCOV_EXCL_STOP + + // step 2: write each element + for (const auto& el : *j.m_value.array) + { + write_cbor(el); + } + break; + } + + case value_t::object: + { + // step 1: write control byte and the object size + const auto N = j.m_value.object->size(); + if (N <= 0x17) + { + write_number(static_cast(0xa0 + N)); + } + else if (N <= 0xff) + { + oa->write_character(static_cast(0xb8)); + write_number(static_cast(N)); + } + else if (N <= 0xffff) + { + oa->write_character(static_cast(0xb9)); + write_number(static_cast(N)); + } + else if (N <= 0xffffffff) + { + oa->write_character(static_cast(0xba)); + write_number(static_cast(N)); + } + // LCOV_EXCL_START + else if (N <= 0xffffffffffffffff) + { + oa->write_character(static_cast(0xbb)); + write_number(static_cast(N)); + } + // LCOV_EXCL_STOP + + // step 2: write each element + for (const auto& el : *j.m_value.object) + { + write_cbor(el.first); + write_cbor(el.second); + } + break; + } + + default: + break; + } + } + + /*! + @brief[in] j JSON value to serialize + */ + void write_msgpack(const BasicJsonType& j) + { + switch (j.type()) + { + case value_t::null: // nil + { + oa->write_character(static_cast(0xc0)); + break; + } + + case value_t::boolean: // true and false + { + oa->write_character(j.m_value.boolean + ? static_cast(0xc3) + : static_cast(0xc2)); + break; + } + + case value_t::number_integer: + { + if (j.m_value.number_integer >= 0) + { + // MessagePack does not differentiate between positive + // signed integers and unsigned integers. Therefore, we used + // the code from the value_t::number_unsigned case here. + if (j.m_value.number_unsigned < 128) + { + // positive fixnum + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 8 + oa->write_character(static_cast(0xcc)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 16 + oa->write_character(static_cast(0xcd)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 32 + oa->write_character(static_cast(0xce)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 64 + oa->write_character(static_cast(0xcf)); + write_number(static_cast(j.m_value.number_integer)); + } + } + else + { + if (j.m_value.number_integer >= -32) + { + // negative fixnum + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer >= (std::numeric_limits::min)() and + j.m_value.number_integer <= (std::numeric_limits::max)()) + { + // int 8 + oa->write_character(static_cast(0xd0)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer >= (std::numeric_limits::min)() and + j.m_value.number_integer <= (std::numeric_limits::max)()) + { + // int 16 + oa->write_character(static_cast(0xd1)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer >= (std::numeric_limits::min)() and + j.m_value.number_integer <= (std::numeric_limits::max)()) + { + // int 32 + oa->write_character(static_cast(0xd2)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_integer >= (std::numeric_limits::min)() and + j.m_value.number_integer <= (std::numeric_limits::max)()) + { + // int 64 + oa->write_character(static_cast(0xd3)); + write_number(static_cast(j.m_value.number_integer)); + } + } + break; + } + + case value_t::number_unsigned: + { + if (j.m_value.number_unsigned < 128) + { + // positive fixnum + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 8 + oa->write_character(static_cast(0xcc)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 16 + oa->write_character(static_cast(0xcd)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 32 + oa->write_character(static_cast(0xce)); + write_number(static_cast(j.m_value.number_integer)); + } + else if (j.m_value.number_unsigned <= (std::numeric_limits::max)()) + { + // uint 64 + oa->write_character(static_cast(0xcf)); + write_number(static_cast(j.m_value.number_integer)); + } + break; + } + + case value_t::number_float: // float 64 + { + oa->write_character(static_cast(0xcb)); + write_number(j.m_value.number_float); + break; + } + + case value_t::string: + { + // step 1: write control byte and the string length + const auto N = j.m_value.string->size(); + if (N <= 31) + { + // fixstr + write_number(static_cast(0xa0 | N)); + } + else if (N <= 255) + { + // str 8 + oa->write_character(static_cast(0xd9)); + write_number(static_cast(N)); + } + else if (N <= 65535) + { + // str 16 + oa->write_character(static_cast(0xda)); + write_number(static_cast(N)); + } + else if (N <= 4294967295) + { + // str 32 + oa->write_character(static_cast(0xdb)); + write_number(static_cast(N)); + } + + // step 2: write the string + oa->write_characters( + reinterpret_cast(j.m_value.string->c_str()), + j.m_value.string->size()); + break; + } + + case value_t::array: + { + // step 1: write control byte and the array size + const auto N = j.m_value.array->size(); + if (N <= 15) + { + // fixarray + write_number(static_cast(0x90 | N)); + } + else if (N <= 0xffff) + { + // array 16 + oa->write_character(static_cast(0xdc)); + write_number(static_cast(N)); + } + else if (N <= 0xffffffff) + { + // array 32 + oa->write_character(static_cast(0xdd)); + write_number(static_cast(N)); + } + + // step 2: write each element + for (const auto& el : *j.m_value.array) + { + write_msgpack(el); + } + break; + } + + case value_t::object: + { + // step 1: write control byte and the object size + const auto N = j.m_value.object->size(); + if (N <= 15) + { + // fixmap + write_number(static_cast(0x80 | (N & 0xf))); + } + else if (N <= 65535) + { + // map 16 + oa->write_character(static_cast(0xde)); + write_number(static_cast(N)); + } + else if (N <= 4294967295) + { + // map 32 + oa->write_character(static_cast(0xdf)); + write_number(static_cast(N)); + } + + // step 2: write each element + for (const auto& el : *j.m_value.object) + { + write_msgpack(el.first); + write_msgpack(el.second); + } + break; + } + + default: + break; + } + } + + private: + /* + @brief write a number to output input + + @param[in] n number of type @a NumberType + @tparam NumberType the type of the number + + @note This function needs to respect the system's endianess, because bytes + in CBOR and MessagePack are stored in network order (big endian) and + therefore need reordering on little endian systems. + */ + template void write_number(NumberType n) + { + // step 1: write number to array of length NumberType + std::array vec; + std::memcpy(vec.data(), &n, sizeof(NumberType)); + + // step 2: write array to output (with possible reordering) + if (is_little_endian) + { + // reverse byte order prior to conversion if necessary + std::reverse(vec.begin(), vec.end()); + } + + oa->write_characters(vec.data(), sizeof(NumberType)); + } + + private: + /// whether we can assume little endianess + const bool is_little_endian = binary_reader::little_endianess(); + + /// the output + output_adapter_t oa = nullptr; +}; + +/////////////////// +// serialization // +/////////////////// + +template +class serializer +{ + using string_t = typename BasicJsonType::string_t; + using number_float_t = typename BasicJsonType::number_float_t; + using number_integer_t = typename BasicJsonType::number_integer_t; + using number_unsigned_t = typename BasicJsonType::number_unsigned_t; + public: + /*! + @param[in] s output stream to serialize to + @param[in] ichar indentation character to use + */ + serializer(output_adapter_t s, const char ichar) + : o(std::move(s)), loc(std::localeconv()), + thousands_sep(loc->thousands_sep == nullptr ? '\0' : loc->thousands_sep[0]), + decimal_point(loc->decimal_point == nullptr ? '\0' : loc->decimal_point[0]), + indent_char(ichar), indent_string(512, indent_char) {} + + // delete because of pointer members + serializer(const serializer&) = delete; + serializer& operator=(const serializer&) = delete; + + /*! + @brief internal implementation of the serialization function + + This function is called by the public member function dump and organizes + the serialization internally. The indentation level is propagated as + additional parameter. In case of arrays and objects, the function is + called recursively. + + - strings and object keys are escaped using `escape_string()` + - integer numbers are converted implicitly via `operator<<` + - floating-point numbers are converted to a string using `"%g"` format + + @param[in] val value to serialize + @param[in] pretty_print whether the output shall be pretty-printed + @param[in] indent_step the indent level + @param[in] current_indent the current indent level (only used internally) + */ + void dump(const BasicJsonType& val, const bool pretty_print, + const bool ensure_ascii, + const unsigned int indent_step, + const unsigned int current_indent = 0) + { + switch (val.m_type) + { + case value_t::object: + { + if (val.m_value.object->empty()) + { + o->write_characters("{}", 2); + return; + } + + if (pretty_print) + { + o->write_characters("{\n", 2); + + // variable to hold indentation for recursive calls + const auto new_indent = current_indent + indent_step; + if (JSON_UNLIKELY(indent_string.size() < new_indent)) + { + indent_string.resize(indent_string.size() * 2, ' '); + } + + // first n-1 elements + auto i = val.m_value.object->cbegin(); + for (std::size_t cnt = 0; cnt < val.m_value.object->size() - 1; ++cnt, ++i) + { + o->write_characters(indent_string.c_str(), new_indent); + o->write_character('\"'); + dump_escaped(i->first, ensure_ascii); + o->write_characters("\": ", 3); + dump(i->second, true, ensure_ascii, indent_step, new_indent); + o->write_characters(",\n", 2); + } + + // last element + assert(i != val.m_value.object->cend()); + assert(std::next(i) == val.m_value.object->cend()); + o->write_characters(indent_string.c_str(), new_indent); + o->write_character('\"'); + dump_escaped(i->first, ensure_ascii); + o->write_characters("\": ", 3); + dump(i->second, true, ensure_ascii, indent_step, new_indent); + + o->write_character('\n'); + o->write_characters(indent_string.c_str(), current_indent); + o->write_character('}'); + } + else + { + o->write_character('{'); + + // first n-1 elements + auto i = val.m_value.object->cbegin(); + for (std::size_t cnt = 0; cnt < val.m_value.object->size() - 1; ++cnt, ++i) + { + o->write_character('\"'); + dump_escaped(i->first, ensure_ascii); + o->write_characters("\":", 2); + dump(i->second, false, ensure_ascii, indent_step, current_indent); + o->write_character(','); + } + + // last element + assert(i != val.m_value.object->cend()); + assert(std::next(i) == val.m_value.object->cend()); + o->write_character('\"'); + dump_escaped(i->first, ensure_ascii); + o->write_characters("\":", 2); + dump(i->second, false, ensure_ascii, indent_step, current_indent); + + o->write_character('}'); + } + + return; + } + + case value_t::array: + { + if (val.m_value.array->empty()) + { + o->write_characters("[]", 2); + return; + } + + if (pretty_print) + { + o->write_characters("[\n", 2); + + // variable to hold indentation for recursive calls + const auto new_indent = current_indent + indent_step; + if (JSON_UNLIKELY(indent_string.size() < new_indent)) + { + indent_string.resize(indent_string.size() * 2, ' '); + } + + // first n-1 elements + for (auto i = val.m_value.array->cbegin(); + i != val.m_value.array->cend() - 1; ++i) + { + o->write_characters(indent_string.c_str(), new_indent); + dump(*i, true, ensure_ascii, indent_step, new_indent); + o->write_characters(",\n", 2); + } + + // last element + assert(not val.m_value.array->empty()); + o->write_characters(indent_string.c_str(), new_indent); + dump(val.m_value.array->back(), true, ensure_ascii, indent_step, new_indent); + + o->write_character('\n'); + o->write_characters(indent_string.c_str(), current_indent); + o->write_character(']'); + } + else + { + o->write_character('['); + + // first n-1 elements + for (auto i = val.m_value.array->cbegin(); + i != val.m_value.array->cend() - 1; ++i) + { + dump(*i, false, ensure_ascii, indent_step, current_indent); + o->write_character(','); + } + + // last element + assert(not val.m_value.array->empty()); + dump(val.m_value.array->back(), false, ensure_ascii, indent_step, current_indent); + + o->write_character(']'); + } + + return; + } + + case value_t::string: + { + o->write_character('\"'); + dump_escaped(*val.m_value.string, ensure_ascii); + o->write_character('\"'); + return; + } + + case value_t::boolean: + { + if (val.m_value.boolean) + { + o->write_characters("true", 4); + } + else + { + o->write_characters("false", 5); + } + return; + } + + case value_t::number_integer: + { + dump_integer(val.m_value.number_integer); + return; + } + + case value_t::number_unsigned: + { + dump_integer(val.m_value.number_unsigned); + return; + } + + case value_t::number_float: + { + dump_float(val.m_value.number_float); + return; + } + + case value_t::discarded: + { + o->write_characters("", 11); + return; + } + + case value_t::null: + { + o->write_characters("null", 4); + return; + } + } + } + + private: + /*! + @brief returns the number of expected bytes following in UTF-8 string + + @param[in] u the first byte of a UTF-8 string + @return the number of expected bytes following + */ + static constexpr std::size_t bytes_following(const uint8_t u) + { + return ((0 <= u and u <= 127) ? 0 + : ((192 <= u and u <= 223) ? 1 + : ((224 <= u and u <= 239) ? 2 + : ((240 <= u and u <= 247) ? 3 : std::string::npos)))); + } + + /*! + @brief calculates the extra space to escape a JSON string + + @param[in] s the string to escape + @param[in] ensure_ascii whether to escape non-ASCII characters with + \uXXXX sequences + @return the number of characters required to escape string @a s + + @complexity Linear in the length of string @a s. + */ + static std::size_t extra_space(const string_t& s, + const bool ensure_ascii) noexcept + { + std::size_t res = 0; + + for (std::size_t i = 0; i < s.size(); ++i) + { + switch (s[i]) + { + // control characters that can be escaped with a backslash + case '"': + case '\\': + case '\b': + case '\f': + case '\n': + case '\r': + case '\t': + { + // from c (1 byte) to \x (2 bytes) + res += 1; + break; + } + + // control characters that need \uxxxx escaping + case 0x00: + case 0x01: + case 0x02: + case 0x03: + case 0x04: + case 0x05: + case 0x06: + case 0x07: + case 0x0b: + case 0x0e: + case 0x0f: + case 0x10: + case 0x11: + case 0x12: + case 0x13: + case 0x14: + case 0x15: + case 0x16: + case 0x17: + case 0x18: + case 0x19: + case 0x1a: + case 0x1b: + case 0x1c: + case 0x1d: + case 0x1e: + case 0x1f: + { + // from c (1 byte) to \uxxxx (6 bytes) + res += 5; + break; + } + + default: + { + if (ensure_ascii and (s[i] & 0x80 or s[i] == 0x7F)) + { + const auto bytes = bytes_following(static_cast(s[i])); + if (bytes == std::string::npos) + { + // invalid characters are treated as is, so no + // additional space will be used + break; + } + + if (bytes == 3) + { + // codepoints that need 4 bytes (i.e., 3 additional + // bytes) in UTF-8 need a surrogate pair when \u + // escaping is used: from 4 bytes to \uxxxx\uxxxx + // (12 bytes) + res += (12 - bytes - 1); + } + else + { + // from x bytes to \uxxxx (6 bytes) + res += (6 - bytes - 1); + } + + // skip the additional bytes + i += bytes; + } + break; + } + } + } + + return res; + } + + static void escape_codepoint(int codepoint, string_t& result, std::size_t& pos) + { + // expecting a proper codepoint + assert(0x00 <= codepoint and codepoint <= 0x10FFFF); + + // the last written character was the backslash before the 'u' + assert(result[pos] == '\\'); + + // write the 'u' + result[++pos] = 'u'; + + // convert a number 0..15 to its hex representation (0..f) + static const std::array hexify = + { + { + '0', '1', '2', '3', '4', '5', '6', '7', + '8', '9', 'a', 'b', 'c', 'd', 'e', 'f' + } + }; + + if (codepoint < 0x10000) + { + // codepoints U+0000..U+FFFF can be represented as \uxxxx. + result[++pos] = hexify[(codepoint >> 12) & 0x0F]; + result[++pos] = hexify[(codepoint >> 8) & 0x0F]; + result[++pos] = hexify[(codepoint >> 4) & 0x0F]; + result[++pos] = hexify[codepoint & 0x0F]; + } + else + { + // codepoints U+10000..U+10FFFF need a surrogate pair to be + // represented as \uxxxx\uxxxx. + // http://www.unicode.org/faq/utf_bom.html#utf16-4 + codepoint -= 0x10000; + const int high_surrogate = 0xD800 | ((codepoint >> 10) & 0x3FF); + const int low_surrogate = 0xDC00 | (codepoint & 0x3FF); + result[++pos] = hexify[(high_surrogate >> 12) & 0x0F]; + result[++pos] = hexify[(high_surrogate >> 8) & 0x0F]; + result[++pos] = hexify[(high_surrogate >> 4) & 0x0F]; + result[++pos] = hexify[high_surrogate & 0x0F]; + ++pos; // backslash is already in output + result[++pos] = 'u'; + result[++pos] = hexify[(low_surrogate >> 12) & 0x0F]; + result[++pos] = hexify[(low_surrogate >> 8) & 0x0F]; + result[++pos] = hexify[(low_surrogate >> 4) & 0x0F]; + result[++pos] = hexify[low_surrogate & 0x0F]; + } + + ++pos; + } + + /*! + @brief dump escaped string + + Escape a string by replacing certain special characters by a sequence of an + escape character (backslash) and another character and other control + characters by a sequence of "\u" followed by a four-digit hex + representation. The escaped string is written to output stream @a o. + + @param[in] s the string to escape + @param[in] ensure_ascii whether to escape non-ASCII characters with + \uXXXX sequences + + @complexity Linear in the length of string @a s. + */ + void dump_escaped(const string_t& s, const bool ensure_ascii) const + { + const auto space = extra_space(s, ensure_ascii); + if (space == 0) + { + o->write_characters(s.c_str(), s.size()); + return; + } + + // create a result string of necessary size + string_t result(s.size() + space, '\\'); + std::size_t pos = 0; + + for (std::size_t i = 0; i < s.size(); ++i) + { + switch (s[i]) + { + case '"': // quotation mark (0x22) + { + result[pos + 1] = '"'; + pos += 2; + break; + } + + case '\\': // reverse solidus (0x5c) + { + // nothing to change + pos += 2; + break; + } + + case '\b': // backspace (0x08) + { + result[pos + 1] = 'b'; + pos += 2; + break; + } + + case '\f': // formfeed (0x0c) + { + result[pos + 1] = 'f'; + pos += 2; + break; + } + + case '\n': // newline (0x0a) + { + result[pos + 1] = 'n'; + pos += 2; + break; + } + + case '\r': // carriage return (0x0d) + { + result[pos + 1] = 'r'; + pos += 2; + break; + } + + case '\t': // horizontal tab (0x09) + { + result[pos + 1] = 't'; + pos += 2; + break; + } + + default: + { + // escape control characters (0x00..0x1F) or, if + // ensure_ascii parameter is used, non-ASCII characters + if ((0x00 <= s[i] and s[i] <= 0x1F) or + (ensure_ascii and (s[i] & 0x80 or s[i] == 0x7F))) + { + const auto bytes = bytes_following(static_cast(s[i])); + if (bytes == std::string::npos) + { + // copy invalid character as is + result[pos++] = s[i]; + break; + } + + // check that the additional bytes are present + assert(i + bytes < s.size()); + + // to use \uxxxx escaping, we first need to caluclate + // the codepoint from the UTF-8 bytes + int codepoint = 0; + + assert(0 <= bytes and bytes <= 3); + switch (bytes) + { + case 0: + { + codepoint = s[i] & 0xFF; + break; + } + + case 1: + { + codepoint = ((s[i] & 0x3F) << 6) + + (s[i + 1] & 0x7F); + break; + } + + case 2: + { + codepoint = ((s[i] & 0x1F) << 12) + + ((s[i + 1] & 0x7F) << 6) + + (s[i + 2] & 0x7F); + break; + } + + case 3: + { + codepoint = ((s[i] & 0xF) << 18) + + ((s[i + 1] & 0x7F) << 12) + + ((s[i + 2] & 0x7F) << 6) + + (s[i + 3] & 0x7F); + break; + } + + default: + break; // LCOV_EXCL_LINE + } + + escape_codepoint(codepoint, result, pos); + i += bytes; + } + else + { + // all other characters are added as-is + result[pos++] = s[i]; + } + break; + } + } + } + + assert(pos == result.size()); + o->write_characters(result.c_str(), result.size()); + } + + /*! + @brief dump an integer + + Dump a given integer to output stream @a o. Works internally with + @a number_buffer. + + @param[in] x integer number (signed or unsigned) to dump + @tparam NumberType either @a number_integer_t or @a number_unsigned_t + */ + template < + typename NumberType, + detail::enable_if_t::value or + std::is_same::value, + int> = 0 > + void dump_integer(NumberType x) + { + // special case for "0" + if (x == 0) + { + o->write_character('0'); + return; + } + + const bool is_negative = x < 0; + std::size_t i = 0; + + // spare 1 byte for '\0' + while (x != 0 and i < number_buffer.size() - 1) + { + const auto digit = std::labs(static_cast(x % 10)); + number_buffer[i++] = static_cast('0' + digit); + x /= 10; + } + + // make sure the number has been processed completely + assert(x == 0); + + if (is_negative) + { + // make sure there is capacity for the '-' + assert(i < number_buffer.size() - 2); + number_buffer[i++] = '-'; + } + + std::reverse(number_buffer.begin(), number_buffer.begin() + i); + o->write_characters(number_buffer.data(), i); + } + + /*! + @brief dump a floating-point number + + Dump a given floating-point number to output stream @a o. Works internally + with @a number_buffer. + + @param[in] x floating-point number to dump + */ + void dump_float(number_float_t x) + { + // NaN / inf + if (not std::isfinite(x) or std::isnan(x)) + { + o->write_characters("null", 4); + return; + } + + // special case for 0.0 and -0.0 + if (x == 0) + { + if (std::signbit(x)) + { + o->write_characters("-0.0", 4); + } + else + { + o->write_characters("0.0", 3); + } + return; + } + + // get number of digits for a text -> float -> text round-trip + static constexpr auto d = std::numeric_limits::digits10; + + // the actual conversion + std::ptrdiff_t len = snprintf(number_buffer.data(), number_buffer.size(), "%.*g", d, x); + + // negative value indicates an error + assert(len > 0); + // check if buffer was large enough + assert(static_cast(len) < number_buffer.size()); + + // erase thousands separator + if (thousands_sep != '\0') + { + const auto end = std::remove(number_buffer.begin(), + number_buffer.begin() + len, thousands_sep); + std::fill(end, number_buffer.end(), '\0'); + assert((end - number_buffer.begin()) <= len); + len = (end - number_buffer.begin()); + } + + // convert decimal point to '.' + if (decimal_point != '\0' and decimal_point != '.') + { + const auto dec_pos = std::find(number_buffer.begin(), number_buffer.end(), decimal_point); + if (dec_pos != number_buffer.end()) + { + *dec_pos = '.'; + } + } + + o->write_characters(number_buffer.data(), static_cast(len)); + + // determine if need to append ".0" + const bool value_is_int_like = + std::none_of(number_buffer.begin(), number_buffer.begin() + len + 1, + [](char c) + { + return (c == '.' or c == 'e'); + }); + + if (value_is_int_like) + { + o->write_characters(".0", 2); + } + } + + private: + /// the output of the serializer + output_adapter_t o = nullptr; + + /// a (hopefully) large enough character buffer + std::array number_buffer{{}}; + + /// the locale + const std::lconv* loc = nullptr; + /// the locale's thousand separator character + const char thousands_sep = '\0'; + /// the locale's decimal point character + const char decimal_point = '\0'; + + /// the indentation character + const char indent_char; + + /// the indentation string + string_t indent_string; +}; + +template +class json_ref +{ + public: + using value_type = BasicJsonType; + + json_ref(value_type&& value) + : owned_value(std::move(value)), + value_ref(&owned_value), + is_rvalue(true) + {} + + json_ref(const value_type& value) + : value_ref(const_cast(&value)), + is_rvalue(false) + {} + + json_ref(std::initializer_list init) + : owned_value(init), + value_ref(&owned_value), + is_rvalue(true) + {} + + template + json_ref(Args... args) + : owned_value(std::forward(args)...), + value_ref(&owned_value), + is_rvalue(true) + {} + + // class should be movable only + json_ref(json_ref&&) = default; + json_ref(const json_ref&) = delete; + json_ref& operator=(const json_ref&) = delete; + + value_type moved_or_copied() const + { + if (is_rvalue) + { + return std::move(*value_ref); + } + return *value_ref; + } + + value_type const& operator*() const + { + return *static_cast(value_ref); + } + + value_type const* operator->() const + { + return static_cast(value_ref); + } + + private: + mutable value_type owned_value = nullptr; + value_type* value_ref = nullptr; + const bool is_rvalue; +}; + +} // namespace detail + +/// namespace to hold default `to_json` / `from_json` functions +namespace +{ +constexpr const auto& to_json = detail::static_const::value; +constexpr const auto& from_json = detail::static_const::value; +} + + +/*! +@brief default JSONSerializer template argument + +This serializer ignores the template arguments and uses ADL +([argument-dependent lookup](http://en.cppreference.com/w/cpp/language/adl)) +for serialization. +*/ +template +struct adl_serializer +{ + /*! + @brief convert a JSON value to any value type + + This function is usually called by the `get()` function of the + @ref basic_json class (either explicit or via conversion operators). + + @param[in] j JSON value to read from + @param[in,out] val value to write to + */ + template + static void from_json(BasicJsonType&& j, ValueType& val) noexcept( + noexcept(::nlohmann::from_json(std::forward(j), val))) + { + ::nlohmann::from_json(std::forward(j), val); + } + + /*! + @brief convert any value type to a JSON value + + This function is usually called by the constructors of the @ref basic_json + class. + + @param[in,out] j JSON value to write to + @param[in] val value to read from + */ + template + static void to_json(BasicJsonType& j, ValueType&& val) noexcept( + noexcept(::nlohmann::to_json(j, std::forward(val)))) + { + ::nlohmann::to_json(j, std::forward(val)); + } +}; + +/*! +@brief JSON Pointer + +A JSON pointer defines a string syntax for identifying a specific value +within a JSON document. It can be used with functions `at` and +`operator[]`. Furthermore, JSON pointers are the base for JSON patches. + +@sa [RFC 6901](https://tools.ietf.org/html/rfc6901) + +@since version 2.0.0 +*/ +class json_pointer +{ + /// allow basic_json to access private members + NLOHMANN_BASIC_JSON_TPL_DECLARATION + friend class basic_json; + + public: + /*! + @brief create JSON pointer + + Create a JSON pointer according to the syntax described in + [Section 3 of RFC6901](https://tools.ietf.org/html/rfc6901#section-3). + + @param[in] s string representing the JSON pointer; if omitted, the empty + string is assumed which references the whole JSON value + + @throw parse_error.107 if the given JSON pointer @a s is nonempty and + does not begin with a slash (`/`); see example below + + @throw parse_error.108 if a tilde (`~`) in the given JSON pointer @a s + is not followed by `0` (representing `~`) or `1` (representing `/`); + see example below + + @liveexample{The example shows the construction several valid JSON + pointers as well as the exceptional behavior.,json_pointer} + + @since version 2.0.0 + */ + explicit json_pointer(const std::string& s = "") : reference_tokens(split(s)) {} + + /*! + @brief return a string representation of the JSON pointer + + @invariant For each JSON pointer `ptr`, it holds: + @code {.cpp} + ptr == json_pointer(ptr.to_string()); + @endcode + + @return a string representation of the JSON pointer + + @liveexample{The example shows the result of `to_string`., + json_pointer__to_string} + + @since version 2.0.0 + */ + std::string to_string() const noexcept + { + return std::accumulate(reference_tokens.begin(), reference_tokens.end(), + std::string{}, + [](const std::string & a, const std::string & b) + { + return a + "/" + escape(b); + }); + } + + /// @copydoc to_string() + operator std::string() const + { + return to_string(); + } + + private: + /*! + @brief remove and return last reference pointer + @throw out_of_range.405 if JSON pointer has no parent + */ + std::string pop_back() + { + if (JSON_UNLIKELY(is_root())) + { + JSON_THROW(detail::out_of_range::create(405, "JSON pointer has no parent")); + } + + auto last = reference_tokens.back(); + reference_tokens.pop_back(); + return last; + } + + /// return whether pointer points to the root document + bool is_root() const + { + return reference_tokens.empty(); + } + + json_pointer top() const + { + if (JSON_UNLIKELY(is_root())) + { + JSON_THROW(detail::out_of_range::create(405, "JSON pointer has no parent")); + } + + json_pointer result = *this; + result.reference_tokens = {reference_tokens[0]}; + return result; + } + + + /*! + @brief create and return a reference to the pointed to value + + @complexity Linear in the number of reference tokens. + + @throw parse_error.109 if array index is not a number + @throw type_error.313 if value cannot be unflattened + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + NLOHMANN_BASIC_JSON_TPL& get_and_create(NLOHMANN_BASIC_JSON_TPL& j) const; + + /*! + @brief return a reference to the pointed to value + + @note This version does not throw if a value is not present, but tries to + create nested values instead. For instance, calling this function + with pointer `"/this/that"` on a null value is equivalent to calling + `operator[]("this").operator[]("that")` on that value, effectively + changing the null value to an object. + + @param[in] ptr a JSON value + + @return reference to the JSON value pointed to by the JSON pointer + + @complexity Linear in the length of the JSON pointer. + + @throw parse_error.106 if an array index begins with '0' + @throw parse_error.109 if an array index was not a number + @throw out_of_range.404 if the JSON pointer can not be resolved + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + NLOHMANN_BASIC_JSON_TPL& get_unchecked(NLOHMANN_BASIC_JSON_TPL* ptr) const; + + /*! + @throw parse_error.106 if an array index begins with '0' + @throw parse_error.109 if an array index was not a number + @throw out_of_range.402 if the array index '-' is used + @throw out_of_range.404 if the JSON pointer can not be resolved + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + NLOHMANN_BASIC_JSON_TPL& get_checked(NLOHMANN_BASIC_JSON_TPL* ptr) const; + + /*! + @brief return a const reference to the pointed to value + + @param[in] ptr a JSON value + + @return const reference to the JSON value pointed to by the JSON + pointer + + @throw parse_error.106 if an array index begins with '0' + @throw parse_error.109 if an array index was not a number + @throw out_of_range.402 if the array index '-' is used + @throw out_of_range.404 if the JSON pointer can not be resolved + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + const NLOHMANN_BASIC_JSON_TPL& get_unchecked(const NLOHMANN_BASIC_JSON_TPL* ptr) const; + + /*! + @throw parse_error.106 if an array index begins with '0' + @throw parse_error.109 if an array index was not a number + @throw out_of_range.402 if the array index '-' is used + @throw out_of_range.404 if the JSON pointer can not be resolved + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + const NLOHMANN_BASIC_JSON_TPL& get_checked(const NLOHMANN_BASIC_JSON_TPL* ptr) const; + + /*! + @brief split the string input to reference tokens + + @note This function is only called by the json_pointer constructor. + All exceptions below are documented there. + + @throw parse_error.107 if the pointer is not empty or begins with '/' + @throw parse_error.108 if character '~' is not followed by '0' or '1' + */ + static std::vector split(const std::string& reference_string) + { + std::vector result; + + // special case: empty reference string -> no reference tokens + if (reference_string.empty()) + { + return result; + } + + // check if nonempty reference string begins with slash + if (JSON_UNLIKELY(reference_string[0] != '/')) + { + JSON_THROW(detail::parse_error::create(107, 1, + "JSON pointer must be empty or begin with '/' - was: '" + + reference_string + "'")); + } + + // extract the reference tokens: + // - slash: position of the last read slash (or end of string) + // - start: position after the previous slash + for ( + // search for the first slash after the first character + std::size_t slash = reference_string.find_first_of('/', 1), + // set the beginning of the first reference token + start = 1; + // we can stop if start == string::npos+1 = 0 + start != 0; + // set the beginning of the next reference token + // (will eventually be 0 if slash == std::string::npos) + start = slash + 1, + // find next slash + slash = reference_string.find_first_of('/', start)) + { + // use the text between the beginning of the reference token + // (start) and the last slash (slash). + auto reference_token = reference_string.substr(start, slash - start); + + // check reference tokens are properly escaped + for (std::size_t pos = reference_token.find_first_of('~'); + pos != std::string::npos; + pos = reference_token.find_first_of('~', pos + 1)) + { + assert(reference_token[pos] == '~'); + + // ~ must be followed by 0 or 1 + if (JSON_UNLIKELY(pos == reference_token.size() - 1 or + (reference_token[pos + 1] != '0' and + reference_token[pos + 1] != '1'))) + { + JSON_THROW(detail::parse_error::create(108, 0, "escape character '~' must be followed with '0' or '1'")); + } + } + + // finally, store the reference token + unescape(reference_token); + result.push_back(reference_token); + } + + return result; + } + + /*! + @brief replace all occurrences of a substring by another string + + @param[in,out] s the string to manipulate; changed so that all + occurrences of @a f are replaced with @a t + @param[in] f the substring to replace with @a t + @param[in] t the string to replace @a f + + @pre The search string @a f must not be empty. **This precondition is + enforced with an assertion.** + + @since version 2.0.0 + */ + static void replace_substring(std::string& s, const std::string& f, + const std::string& t) + { + assert(not f.empty()); + for (auto pos = s.find(f); // find first occurrence of f + pos != std::string::npos; // make sure f was found + s.replace(pos, f.size(), t), // replace with t, and + pos = s.find(f, pos + t.size())) // find next occurrence of f + {} + } + + /// escape "~"" to "~0" and "/" to "~1" + static std::string escape(std::string s) + { + replace_substring(s, "~", "~0"); + replace_substring(s, "/", "~1"); + return s; + } + + /// unescape "~1" to tilde and "~0" to slash (order is important!) + static void unescape(std::string& s) + { + replace_substring(s, "~1", "/"); + replace_substring(s, "~0", "~"); + } + + /*! + @param[in] reference_string the reference string to the current value + @param[in] value the value to consider + @param[in,out] result the result object to insert values to + + @note Empty objects or arrays are flattened to `null`. + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + static void flatten(const std::string& reference_string, + const NLOHMANN_BASIC_JSON_TPL& value, + NLOHMANN_BASIC_JSON_TPL& result); + + /*! + @param[in] value flattened JSON + + @return unflattened JSON + + @throw parse_error.109 if array index is not a number + @throw type_error.314 if value is not an object + @throw type_error.315 if object values are not primitive + @throw type_error.313 if value cannot be unflattened + */ + NLOHMANN_BASIC_JSON_TPL_DECLARATION + static NLOHMANN_BASIC_JSON_TPL + unflatten(const NLOHMANN_BASIC_JSON_TPL& value); + + friend bool operator==(json_pointer const& lhs, + json_pointer const& rhs) noexcept; + + friend bool operator!=(json_pointer const& lhs, + json_pointer const& rhs) noexcept; + + /// the reference tokens + std::vector reference_tokens; +}; + /*! @brief a class to store JSON values @@ -149,11 +7257,14 @@ default; will be used in @ref number_integer_t) default; will be used in @ref number_float_t) @tparam AllocatorType type of the allocator to use (`std::allocator` by default) +@tparam JSONSerializer the serializer to resolve internal calls to `to_json()` +and `from_json()` (@ref adl_serializer by default) @requirement The class satisfies the following concept requirements: - Basic - [DefaultConstructible](http://en.cppreference.com/w/cpp/concept/DefaultConstructible): - JSON values can be default constructed. The result will be a JSON null value. + JSON values can be default constructed. The result will be a JSON null + value. - [MoveConstructible](http://en.cppreference.com/w/cpp/concept/MoveConstructible): A JSON value can be constructed from an rvalue argument. - [CopyConstructible](http://en.cppreference.com/w/cpp/concept/CopyConstructible): @@ -168,8 +7279,8 @@ default) - [StandardLayoutType](http://en.cppreference.com/w/cpp/concept/StandardLayoutType): JSON values have [standard layout](http://en.cppreference.com/w/cpp/language/data_members#Standard_layout): - All non-static data members are private and standard layout types, the class - has no virtual functions or (virtual) base classes. + All non-static data members are private and standard layout types, the + class has no virtual functions or (virtual) base classes. - Library-wide - [EqualityComparable](http://en.cppreference.com/w/cpp/concept/EqualityComparable): JSON values can be compared with `==`, see @ref @@ -208,29 +7319,77 @@ Format](http://rfc7159.net/rfc7159) @nosubgrouping */ -template < - template class ObjectType = std::map, - template class ArrayType = std::vector, - class StringType = std::string, - class BooleanType = bool, - class NumberIntegerType = std::int64_t, - class NumberUnsignedType = std::uint64_t, - class NumberFloatType = double, - template class AllocatorType = std::allocator - > +NLOHMANN_BASIC_JSON_TPL_DECLARATION class basic_json { private: + template friend struct detail::external_constructor; + friend ::nlohmann::json_pointer; + friend ::nlohmann::detail::parser; + friend ::nlohmann::detail::serializer; + template + friend class ::nlohmann::detail::iter_impl; + template + friend class ::nlohmann::detail::binary_writer; + template + friend class ::nlohmann::detail::binary_reader; + /// workaround type for MSVC - using basic_json_t = basic_json; + using basic_json_t = NLOHMANN_BASIC_JSON_TPL; + + // convenience aliases for types residing in namespace detail; + using lexer = ::nlohmann::detail::lexer; + using parser = ::nlohmann::detail::parser; + + using primitive_iterator_t = ::nlohmann::detail::primitive_iterator_t; + template + using internal_iterator = ::nlohmann::detail::internal_iterator; + template + using iter_impl = ::nlohmann::detail::iter_impl; + template + using iteration_proxy = ::nlohmann::detail::iteration_proxy; + template using json_reverse_iterator = ::nlohmann::detail::json_reverse_iterator; + + template + using output_adapter_t = ::nlohmann::detail::output_adapter_t; + + using binary_reader = ::nlohmann::detail::binary_reader; + template using binary_writer = ::nlohmann::detail::binary_writer; + + using serializer = ::nlohmann::detail::serializer; public: + using value_t = detail::value_t; // forward declarations - template class iter_impl; - template class json_reverse_iterator; - class json_pointer; + using json_pointer = ::nlohmann::json_pointer; + template + using json_serializer = JSONSerializer; + + using initializer_list_t = std::initializer_list>; + + //////////////// + // exceptions // + //////////////// + + /// @name exceptions + /// Classes to implement user-defined exceptions. + /// @{ + + /// @copydoc detail::exception + using exception = detail::exception; + /// @copydoc detail::parse_error + using parse_error = detail::parse_error; + /// @copydoc detail::invalid_iterator + using invalid_iterator = detail::invalid_iterator; + /// @copydoc detail::type_error + using type_error = detail::type_error; + /// @copydoc detail::out_of_range + using out_of_range = detail::out_of_range; + /// @copydoc detail::other_error + using other_error = detail::other_error; + + /// @} + ///////////////////// // container types // @@ -282,6 +7441,84 @@ class basic_json return allocator_type(); } + /*! + @brief returns version information on the library + + This function returns a JSON object with information about the library, + including the version number and information on the platform and compiler. + + @return JSON object holding version information + key | description + ----------- | --------------- + `compiler` | Information on the used compiler. It is an object with the following keys: `c++` (the used C++ standard), `family` (the compiler family; possible values are `clang`, `icc`, `gcc`, `ilecpp`, `msvc`, `pgcpp`, `sunpro`, and `unknown`), and `version` (the compiler version). + `copyright` | The copyright line for the library as string. + `name` | The name of the library as string. + `platform` | The used platform as string. Possible values are `win32`, `linux`, `apple`, `unix`, and `unknown`. + `url` | The URL of the project as string. + `version` | The version of the library. It is an object with the following keys: `major`, `minor`, and `patch` as defined by [Semantic Versioning](http://semver.org), and `string` (the version string). + + @liveexample{The following code shows an example output of the `meta()` + function.,meta} + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + + @complexity Constant. + + @since 2.1.0 + */ + static basic_json meta() + { + basic_json result; + + result["copyright"] = "(C) 2013-2017 Niels Lohmann"; + result["name"] = "JSON for Modern C++"; + result["url"] = "https://github.com/nlohmann/json"; + result["version"] = + { + {"string", "2.1.1"}, {"major", 2}, {"minor", 1}, {"patch", 1} + }; + +#ifdef _WIN32 + result["platform"] = "win32"; +#elif defined __linux__ + result["platform"] = "linux"; +#elif defined __APPLE__ + result["platform"] = "apple"; +#elif defined __unix__ + result["platform"] = "unix"; +#else + result["platform"] = "unknown"; +#endif + +#if defined(__ICC) || defined(__INTEL_COMPILER) + result["compiler"] = {{"family", "icc"}, {"version", __INTEL_COMPILER}}; +#elif defined(__clang__) + result["compiler"] = {{"family", "clang"}, {"version", __clang_version__}}; +#elif defined(__GNUC__) || defined(__GNUG__) + result["compiler"] = {{"family", "gcc"}, {"version", std::to_string(__GNUC__) + "." + std::to_string(__GNUC_MINOR__) + "." + std::to_string(__GNUC_PATCHLEVEL__)}}; +#elif defined(__HP_cc) || defined(__HP_aCC) + result["compiler"] = "hp" +#elif defined(__IBMCPP__) + result["compiler"] = {{"family", "ilecpp"}, {"version", __IBMCPP__}}; +#elif defined(_MSC_VER) + result["compiler"] = {{"family", "msvc"}, {"version", _MSC_VER}}; +#elif defined(__PGI) + result["compiler"] = {{"family", "pgcpp"}, {"version", __PGI}}; +#elif defined(__SUNPRO_CC) + result["compiler"] = {{"family", "sunpro"}, {"version", __SUNPRO_CC}}; +#else + result["compiler"] = {{"family", "unknown"}, {"version", "unknown"}}; +#endif + +#ifdef __cplusplus + result["compiler"]["c++"] = std::to_string(__cplusplus); +#else + result["compiler"]["c++"] = "unknown"; +#endif + return result; + } + /////////////////////////// // JSON value data types // @@ -352,7 +7589,7 @@ class basic_json [RFC 7159](http://rfc7159.net/rfc7159) specifies: > An implementation may set limits on the maximum depth of nesting. - In this class, the object's limit of nesting is not constraint explicitly. + In this class, the object's limit of nesting is not explicitly constrained. However, a maximum depth of nesting may be introduced by the compiler or runtime environment. A theoretical limit can be queried by calling the @ref max_size function of a JSON object. @@ -411,7 +7648,7 @@ class basic_json [RFC 7159](http://rfc7159.net/rfc7159) specifies: > An implementation may set limits on the maximum depth of nesting. - In this class, the array's limit of nesting is not constraint explicitly. + In this class, the array's limit of nesting is not explicitly constrained. However, a maximum depth of nesting may be introduced by the compiler or runtime environment. A theoretical limit can be queried by calling the @ref max_size function of a JSON array. @@ -449,6 +7686,12 @@ class basic_json std::string @endcode + #### Encoding + + Strings are stored in UTF-8 encoding. Therefore, functions like + `std::string::size()` or `std::string::length()` return the number of + bytes in the string rather than the number of characters or glyphs. + #### String comparison [RFC 7159](http://rfc7159.net/rfc7159) states: @@ -713,47 +7956,6 @@ class basic_json /// @} - - /////////////////////////// - // JSON type enumeration // - /////////////////////////// - - /*! - @brief the JSON type enumeration - - This enumeration collects the different JSON types. It is internally used - to distinguish the stored values, and the functions @ref is_null(), @ref - is_object(), @ref is_array(), @ref is_string(), @ref is_boolean(), @ref - is_number() (with @ref is_number_integer(), @ref is_number_unsigned(), and - @ref is_number_float()), @ref is_discarded(), @ref is_primitive(), and - @ref is_structured() rely on it. - - @note There are three enumeration entries (number_integer, - number_unsigned, and number_float), because the library distinguishes - these three types for numbers: @ref number_unsigned_t is used for unsigned - integers, @ref number_integer_t is used for signed integers, and @ref - number_float_t is used for floating-point numbers or to approximate - integers which do not fit in the limits of their respective type. - - @sa @ref basic_json(const value_t value_type) -- create a JSON value with - the default value for a given type - - @since version 1.0.0 - */ - enum class value_t : uint8_t - { - null, ///< null value - object, ///< object (unordered set of name/value pairs) - array, ///< array (ordered collection of values) - string, ///< string value - boolean, ///< boolean value - number_integer, ///< number value (signed integer) - number_unsigned, ///< number value (unsigned integer) - number_float, ///< number value (floating-point) - discarded ///< discarded by the the parser callback function - }; - - private: /// helper for exception-safe object creation @@ -767,7 +7969,7 @@ class basic_json }; std::unique_ptr object(alloc.allocate(1), deleter); alloc.construct(object.get(), std::forward(args)...); - assert(object.get() != nullptr); + assert(object != nullptr); return object.release(); } @@ -880,9 +8082,9 @@ class basic_json default: { - if (t == value_t::null) + if (JSON_UNLIKELY(t == value_t::null)) { - throw std::domain_error("961c151d2e87f2686a955a9be24d316f1362bf21 2.0.10"); // LCOV_EXCL_LINE + JSON_THROW(other_error::create(500, "961c151d2e87f2686a955a9be24d316f1362bf21 2.1.1")); // LCOV_EXCL_LINE } break; } @@ -895,17 +8097,70 @@ class basic_json string = create(value); } + /// constructor for rvalue strings + json_value(string_t&& value) + { + string = create(std::move(value)); + } + /// constructor for objects json_value(const object_t& value) { object = create(value); } + /// constructor for rvalue objects + json_value(object_t&& value) + { + object = create(std::move(value)); + } + /// constructor for arrays json_value(const array_t& value) { array = create(value); } + + /// constructor for rvalue arrays + json_value(array_t&& value) + { + array = create(std::move(value)); + } + + void destroy(value_t t) + { + switch (t) + { + case value_t::object: + { + AllocatorType alloc; + alloc.destroy(object); + alloc.deallocate(object, 1); + break; + } + + case value_t::array: + { + AllocatorType alloc; + alloc.destroy(array); + alloc.deallocate(array, 1); + break; + } + + case value_t::string: + { + AllocatorType alloc; + alloc.destroy(string); + alloc.deallocate(string, 1); + break; + } + + default: + { + break; + } + } + } }; /*! @@ -929,31 +8184,7 @@ class basic_json // JSON parser callback // ////////////////////////// - /*! - @brief JSON callback events - - This enumeration lists the parser events that can trigger calling a - callback function of type @ref parser_callback_t during parsing. - - @image html callback_events.png "Example when certain parse events are triggered" - - @since version 1.0.0 - */ - enum class parse_event_t : uint8_t - { - /// the parser read `{` and started to process a JSON object - object_start, - /// the parser read `}` and finished processing a JSON object - object_end, - /// the parser read `[` and started to process a JSON array - array_start, - /// the parser read `]` and finished processing a JSON array - array_end, - /// the parser read a key of a value in an object - key, - /// the parser finished reading a JSON value - value - }; + using parse_event_t = typename parser::parse_event_t; /*! @brief per-element parser callback type @@ -1007,9 +8238,7 @@ class basic_json @since version 1.0.0 */ - using parser_callback_t = std::function; + using parser_callback_t = typename parser::parser_callback_t; ////////////////// @@ -1036,32 +8265,22 @@ class basic_json object | `{}` array | `[]` - @param[in] value_type the type of the value to create + @param[in] v the type of the value to create @complexity Constant. - @throw std::bad_alloc if allocation for object, array, or string value - fails + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. @liveexample{The following code shows the constructor for different @ref value_t values,basic_json__value_t} - @sa @ref basic_json(std::nullptr_t) -- create a `null` value - @sa @ref basic_json(boolean_t value) -- create a boolean value - @sa @ref basic_json(const string_t&) -- create a string value - @sa @ref basic_json(const object_t&) -- create a object value - @sa @ref basic_json(const array_t&) -- create a array value - @sa @ref basic_json(const number_float_t) -- create a number - (floating-point) value - @sa @ref basic_json(const number_integer_t) -- create a number (integer) - value - @sa @ref basic_json(const number_unsigned_t) -- create a number (unsigned) - value + @sa @ref clear() -- restores the postcondition of this constructor @since version 1.0.0 */ - basic_json(const value_t value_type) - : m_type(value_type), m_value(value_type) + basic_json(const value_t v) + : m_type(v), m_value(v) { assert_invariant(); } @@ -1091,474 +8310,72 @@ class basic_json } /*! - @brief create an object (explicit) + @brief create a JSON value - Create an object JSON value with a given content. + This is a "catch all" constructor for all compatible JSON types; that is, + types for which a `to_json()` method exsits. The constructor forwards the + parameter @a val to that method (to `json_serializer::to_json` method + with `U = uncvref_t`, to be exact). - @param[in] val a value for the object + Template type @a CompatibleType includes, but is not limited to, the + following types: + - **arrays**: @ref array_t and all kinds of compatible containers such as + `std::vector`, `std::deque`, `std::list`, `std::forward_list`, + `std::array`, `std::valarray`, `std::set`, `std::unordered_set`, + `std::multiset`, and `std::unordered_multiset` with a `value_type` from + which a @ref basic_json value can be constructed. + - **objects**: @ref object_t and all kinds of compatible associative + containers such as `std::map`, `std::unordered_map`, `std::multimap`, + and `std::unordered_multimap` with a `key_type` compatible to + @ref string_t and a `value_type` from which a @ref basic_json value can + be constructed. + - **strings**: @ref string_t, string literals, and all compatible string + containers can be used. + - **numbers**: @ref number_integer_t, @ref number_unsigned_t, + @ref number_float_t, and all convertible number types such as `int`, + `size_t`, `int64_t`, `float` or `double` can be used. + - **boolean**: @ref boolean_t / `bool` can be used. - @complexity Linear in the size of the passed @a val. + See the examples below. - @throw std::bad_alloc if allocation for object value fails + @tparam CompatibleType a type such that: + - @a CompatibleType is not derived from `std::istream`, + - @a CompatibleType is not @ref basic_json (to avoid hijacking copy/move + constructors), + - @a CompatibleType is not a @ref basic_json nested type (e.g., + @ref json_pointer, @ref iterator, etc ...) + - @ref @ref json_serializer has a + `to_json(basic_json_t&, CompatibleType&&)` method - @liveexample{The following code shows the constructor with an @ref - object_t parameter.,basic_json__object_t} + @tparam U = `uncvref_t` - @sa @ref basic_json(const CompatibleObjectType&) -- create an object value - from a compatible STL container + @param[in] val the value to be forwarded to the respective constructor - @since version 1.0.0 - */ - basic_json(const object_t& val) - : m_type(value_t::object), m_value(val) - { - assert_invariant(); - } + @complexity Usually linear in the size of the passed @a val, also + depending on the implementation of the called `to_json()` + method. - /*! - @brief create an object (implicit) - - Create an object JSON value with a given content. This constructor allows - any type @a CompatibleObjectType that can be used to construct values of - type @ref object_t. - - @tparam CompatibleObjectType An object type whose `key_type` and - `value_type` is compatible to @ref object_t. Examples include `std::map`, - `std::unordered_map`, `std::multimap`, and `std::unordered_multimap` with - a `key_type` of `std::string`, and a `value_type` from which a @ref - basic_json value can be constructed. - - @param[in] val a value for the object - - @complexity Linear in the size of the passed @a val. - - @throw std::bad_alloc if allocation for object value fails + @exceptionsafety Depends on the called constructor. For types directly + supported by the library (i.e., all types for which no `to_json()` function + was provided), strong guarantee holds: if an exception is thrown, there are + no changes to any JSON value. @liveexample{The following code shows the constructor with several - compatible object type parameters.,basic_json__CompatibleObjectType} + compatible types.,basic_json__CompatibleType} - @sa @ref basic_json(const object_t&) -- create an object value - - @since version 1.0.0 + @since version 2.1.0 */ - template::value and - std::is_constructible::value, int>::type = 0> - basic_json(const CompatibleObjectType& val) - : m_type(value_t::object) - { - using std::begin; - using std::end; - m_value.object = create(begin(val), end(val)); - assert_invariant(); - } - - /*! - @brief create an array (explicit) - - Create an array JSON value with a given content. - - @param[in] val a value for the array - - @complexity Linear in the size of the passed @a val. - - @throw std::bad_alloc if allocation for array value fails - - @liveexample{The following code shows the constructor with an @ref array_t - parameter.,basic_json__array_t} - - @sa @ref basic_json(const CompatibleArrayType&) -- create an array value - from a compatible STL containers - - @since version 1.0.0 - */ - basic_json(const array_t& val) - : m_type(value_t::array), m_value(val) - { - assert_invariant(); - } - - /*! - @brief create an array (implicit) - - Create an array JSON value with a given content. This constructor allows - any type @a CompatibleArrayType that can be used to construct values of - type @ref array_t. - - @tparam CompatibleArrayType An object type whose `value_type` is - compatible to @ref array_t. Examples include `std::vector`, `std::deque`, - `std::list`, `std::forward_list`, `std::array`, `std::set`, - `std::unordered_set`, `std::multiset`, and `unordered_multiset` with a - `value_type` from which a @ref basic_json value can be constructed. - - @param[in] val a value for the array - - @complexity Linear in the size of the passed @a val. - - @throw std::bad_alloc if allocation for array value fails - - @liveexample{The following code shows the constructor with several - compatible array type parameters.,basic_json__CompatibleArrayType} - - @sa @ref basic_json(const array_t&) -- create an array value - - @since version 1.0.0 - */ - template::value and - not std::is_same::value and - not std::is_same::value and - not std::is_same::value and - not std::is_same::value and - not std::is_same::value and - std::is_constructible::value, int>::type = 0> - basic_json(const CompatibleArrayType& val) - : m_type(value_t::array) - { - using std::begin; - using std::end; - m_value.array = create(begin(val), end(val)); - assert_invariant(); - } - - /*! - @brief create a string (explicit) - - Create an string JSON value with a given content. - - @param[in] val a value for the string - - @complexity Linear in the size of the passed @a val. - - @throw std::bad_alloc if allocation for string value fails - - @liveexample{The following code shows the constructor with an @ref - string_t parameter.,basic_json__string_t} - - @sa @ref basic_json(const typename string_t::value_type*) -- create a - string value from a character pointer - @sa @ref basic_json(const CompatibleStringType&) -- create a string value - from a compatible string container - - @since version 1.0.0 - */ - basic_json(const string_t& val) - : m_type(value_t::string), m_value(val) - { - assert_invariant(); - } - - /*! - @brief create a string (explicit) - - Create a string JSON value with a given content. - - @param[in] val a literal value for the string - - @complexity Linear in the size of the passed @a val. - - @throw std::bad_alloc if allocation for string value fails - - @liveexample{The following code shows the constructor with string literal - parameter.,basic_json__string_t_value_type} - - @sa @ref basic_json(const string_t&) -- create a string value - @sa @ref basic_json(const CompatibleStringType&) -- create a string value - from a compatible string container - - @since version 1.0.0 - */ - basic_json(const typename string_t::value_type* val) - : basic_json(string_t(val)) - { - assert_invariant(); - } - - /*! - @brief create a string (implicit) - - Create a string JSON value with a given content. - - @param[in] val a value for the string - - @tparam CompatibleStringType an string type which is compatible to @ref - string_t, for instance `std::string`. - - @complexity Linear in the size of the passed @a val. - - @throw std::bad_alloc if allocation for string value fails - - @liveexample{The following code shows the construction of a string value - from a compatible type.,basic_json__CompatibleStringType} - - @sa @ref basic_json(const string_t&) -- create a string value - @sa @ref basic_json(const typename string_t::value_type*) -- create a - string value from a character pointer - - @since version 1.0.0 - */ - template::value, int>::type = 0> - basic_json(const CompatibleStringType& val) - : basic_json(string_t(val)) - { - assert_invariant(); - } - - /*! - @brief create a boolean (explicit) - - Creates a JSON boolean type from a given value. - - @param[in] val a boolean value to store - - @complexity Constant. - - @liveexample{The example below demonstrates boolean - values.,basic_json__boolean_t} - - @since version 1.0.0 - */ - basic_json(boolean_t val) noexcept - : m_type(value_t::boolean), m_value(val) - { - assert_invariant(); - } - - /*! - @brief create an integer number (explicit) - - Create an integer number JSON value with a given content. - - @tparam T A helper type to remove this function via SFINAE in case @ref - number_integer_t is the same as `int`. In this case, this constructor - would have the same signature as @ref basic_json(const int value). Note - the helper type @a T is not visible in this constructor's interface. - - @param[in] val an integer to create a JSON number from - - @complexity Constant. - - @liveexample{The example below shows the construction of an integer - number value.,basic_json__number_integer_t} - - @sa @ref basic_json(const int) -- create a number value (integer) - @sa @ref basic_json(const CompatibleNumberIntegerType) -- create a number - value (integer) from a compatible number type - - @since version 1.0.0 - */ - template::value) and - std::is_same::value, int>::type = 0> - basic_json(const number_integer_t val) noexcept - : m_type(value_t::number_integer), m_value(val) - { - assert_invariant(); - } - - - /*! - @brief create an integer number from an enum type (explicit) - - Create an integer number JSON value with a given content. - - @param[in] val an integer to create a JSON number from - - @note This constructor allows to pass enums directly to a constructor. As - C++ has no way of specifying the type of an anonymous enum explicitly, we - can only rely on the fact that such values implicitly convert to int. As - int may already be the same type of number_integer_t, we may need to - switch off the constructor @ref basic_json(const number_integer_t). - - @complexity Constant. - - @liveexample{The example below shows the construction of an integer - number value from an anonymous enum.,basic_json__const_int} - - @sa @ref basic_json(const number_integer_t) -- create a number value - (integer) - @sa @ref basic_json(const CompatibleNumberIntegerType) -- create a number - value (integer) from a compatible number type - - @since version 1.0.0 - */ - basic_json(const int val) noexcept - : m_type(value_t::number_integer), - m_value(static_cast(val)) - { - assert_invariant(); - } - - /*! - @brief create an integer number (implicit) - - Create an integer number JSON value with a given content. This constructor - allows any type @a CompatibleNumberIntegerType that can be used to - construct values of type @ref number_integer_t. - - @tparam CompatibleNumberIntegerType An integer type which is compatible to - @ref number_integer_t. Examples include the types `int`, `int32_t`, - `long`, and `short`. - - @param[in] val an integer to create a JSON number from - - @complexity Constant. - - @liveexample{The example below shows the construction of several integer - number values from compatible - types.,basic_json__CompatibleIntegerNumberType} - - @sa @ref basic_json(const number_integer_t) -- create a number value - (integer) - @sa @ref basic_json(const int) -- create a number value (integer) - - @since version 1.0.0 - */ - template::value and - std::numeric_limits::is_integer and - std::numeric_limits::is_signed, - CompatibleNumberIntegerType>::type = 0> - basic_json(const CompatibleNumberIntegerType val) noexcept - : m_type(value_t::number_integer), - m_value(static_cast(val)) - { - assert_invariant(); - } - - /*! - @brief create an unsigned integer number (explicit) - - Create an unsigned integer number JSON value with a given content. - - @tparam T helper type to compare number_unsigned_t and unsigned int (not - visible in) the interface. - - @param[in] val an integer to create a JSON number from - - @complexity Constant. - - @sa @ref basic_json(const CompatibleNumberUnsignedType) -- create a number - value (unsigned integer) from a compatible number type - - @since version 2.0.0 - */ - template::value) and - std::is_same::value, int>::type = 0> - basic_json(const number_unsigned_t val) noexcept - : m_type(value_t::number_unsigned), m_value(val) - { - assert_invariant(); - } - - /*! - @brief create an unsigned number (implicit) - - Create an unsigned number JSON value with a given content. This - constructor allows any type @a CompatibleNumberUnsignedType that can be - used to construct values of type @ref number_unsigned_t. - - @tparam CompatibleNumberUnsignedType An integer type which is compatible - to @ref number_unsigned_t. Examples may include the types `unsigned int`, - `uint32_t`, or `unsigned short`. - - @param[in] val an unsigned integer to create a JSON number from - - @complexity Constant. - - @sa @ref basic_json(const number_unsigned_t) -- create a number value - (unsigned) - - @since version 2.0.0 - */ - template::value and - std::numeric_limits::is_integer and - not std::numeric_limits::is_signed, - CompatibleNumberUnsignedType>::type = 0> - basic_json(const CompatibleNumberUnsignedType val) noexcept - : m_type(value_t::number_unsigned), - m_value(static_cast(val)) - { - assert_invariant(); - } - - /*! - @brief create a floating-point number (explicit) - - Create a floating-point number JSON value with a given content. - - @param[in] val a floating-point value to create a JSON number from - - @note [RFC 7159](http://www.rfc-editor.org/rfc/rfc7159.txt), section 6 - disallows NaN values: - > Numeric values that cannot be represented in the grammar below (such as - > Infinity and NaN) are not permitted. - In case the parameter @a val is not a number, a JSON null value is created - instead. - - @complexity Constant. - - @liveexample{The following example creates several floating-point - values.,basic_json__number_float_t} - - @sa @ref basic_json(const CompatibleNumberFloatType) -- create a number - value (floating-point) from a compatible number type - - @since version 1.0.0 - */ - basic_json(const number_float_t val) noexcept - : m_type(value_t::number_float), m_value(val) - { - // replace infinity and NAN by null - if (not std::isfinite(val)) - { - m_type = value_t::null; - m_value = json_value(); - } - - assert_invariant(); - } - - /*! - @brief create an floating-point number (implicit) - - Create an floating-point number JSON value with a given content. This - constructor allows any type @a CompatibleNumberFloatType that can be used - to construct values of type @ref number_float_t. - - @tparam CompatibleNumberFloatType A floating-point type which is - compatible to @ref number_float_t. Examples may include the types `float` - or `double`. - - @param[in] val a floating-point to create a JSON number from - - @note [RFC 7159](http://www.rfc-editor.org/rfc/rfc7159.txt), section 6 - disallows NaN values: - > Numeric values that cannot be represented in the grammar below (such as - > Infinity and NaN) are not permitted. - In case the parameter @a val is not a number, a JSON null value is - created instead. - - @complexity Constant. - - @liveexample{The example below shows the construction of several - floating-point number values from compatible - types.,basic_json__CompatibleNumberFloatType} - - @sa @ref basic_json(const number_float_t) -- create a number value - (floating-point) - - @since version 1.0.0 - */ - template::value and - std::is_floating_point::value>::type> - basic_json(const CompatibleNumberFloatType val) noexcept - : basic_json(number_float_t(val)) + template, + detail::enable_if_t::value and + not std::is_same::value and + not detail::is_basic_json_nested_type< + basic_json_t, U>::value and + detail::has_to_json::value, + int> = 0> + basic_json(CompatibleType && val) noexcept(noexcept(JSONSerializer::to_json( + std::declval(), std::forward(val)))) { + JSONSerializer::to_json(*this, std::forward(val)); assert_invariant(); } @@ -1581,7 +8398,7 @@ class basic_json 1. The empty initializer list is written as `{}` which is exactly an empty JSON object. - 2. C++ has now way of describing mapped types other than to list a list of + 2. C++ has no way of describing mapped types other than to list a list of pairs. As JSON requires that keys must be of type string, rule 2 is the weakest constraint one can pose on initializer lists to interpret them as an object. @@ -1591,10 +8408,10 @@ class basic_json With the rules described above, the following JSON values cannot be expressed by an initializer list: - - the empty array (`[]`): use @ref array(std::initializer_list) + - the empty array (`[]`): use @ref array(initializer_list_t) with an empty initializer list in this case - arrays whose elements satisfy rule 2: use @ref - array(std::initializer_list) with the same initializer list + array(initializer_list_t) with the same initializer list in this case @note When used without parentheses around an empty initializer list, @ref @@ -1606,41 +8423,46 @@ class basic_json @param[in] type_deduction internal parameter; when set to `true`, the type of the JSON value is deducted from the initializer list @a init; when set to `false`, the type provided via @a manual_type is forced. This mode is - used by the functions @ref array(std::initializer_list) and - @ref object(std::initializer_list). + used by the functions @ref array(initializer_list_t) and + @ref object(initializer_list_t). @param[in] manual_type internal parameter; when @a type_deduction is set to `false`, the created JSON value will use the provided type (only @ref value_t::array and @ref value_t::object are valid); when @a type_deduction is set to `true`, this parameter has no effect - @throw std::domain_error if @a type_deduction is `false`, @a manual_type - is `value_t::object`, but @a init contains an element which is not a pair - whose first element is a string; example: `"cannot create object from - initializer list"` + @throw type_error.301 if @a type_deduction is `false`, @a manual_type is + `value_t::object`, but @a init contains an element which is not a pair + whose first element is a string. In this case, the constructor could not + create an object. If @a type_deduction would have be `true`, an array + would have been created. See @ref object(initializer_list_t) + for an example. @complexity Linear in the size of the initializer list @a init. + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + @liveexample{The example below shows how JSON values are created from initializer lists.,basic_json__list_init_t} - @sa @ref array(std::initializer_list) -- create a JSON array + @sa @ref array(initializer_list_t) -- create a JSON array value from an initializer list - @sa @ref object(std::initializer_list) -- create a JSON object + @sa @ref object(initializer_list_t) -- create a JSON object value from an initializer list @since version 1.0.0 */ - basic_json(std::initializer_list init, + basic_json(initializer_list_t init, bool type_deduction = true, value_t manual_type = value_t::array) { // check if each element is an array with two elements whose first // element is a string bool is_an_object = std::all_of(init.begin(), init.end(), - [](const basic_json & element) + [](const detail::json_ref& element_ref) { - return element.is_array() and element.size() == 2 and element[0].is_string(); + return (element_ref->is_array() and element_ref->size() == 2 and (*element_ref)[0].is_string()); }); // adjust type if type deduction is not wanted @@ -1653,9 +8475,9 @@ class basic_json } // if object is wanted but impossible, throw an exception - if (manual_type == value_t::object and not is_an_object) + if (JSON_UNLIKELY(manual_type == value_t::object and not is_an_object)) { - throw std::domain_error("cannot create object from initializer list"); + JSON_THROW(type_error::create(301, "cannot create object from initializer list")); } } @@ -1665,16 +8487,19 @@ class basic_json m_type = value_t::object; m_value = value_t::object; - std::for_each(init.begin(), init.end(), [this](const basic_json & element) + std::for_each(init.begin(), init.end(), [this](const detail::json_ref& element_ref) { - m_value.object->emplace(*(element[0].m_value.string), element[1]); + auto element = element_ref.moved_or_copied(); + m_value.object->emplace( + std::move(*((*element.m_value.array)[0].m_value.string)), + std::move((*element.m_value.array)[1])); }); } else { // the initializer list describes an array -> create array m_type = value_t::array; - m_value.array = create(init); + m_value.array = create(init.begin(), init.end()); } assert_invariant(); @@ -1689,7 +8514,7 @@ class basic_json @note This function is only needed to express two edge cases that cannot be realized with the initializer list constructor (@ref - basic_json(std::initializer_list, bool, value_t)). These cases + basic_json(initializer_list_t, bool, value_t)). These cases are: 1. creating an array whose elements are all pairs whose first element is a string -- in this case, the initializer list constructor would create an @@ -1704,18 +8529,20 @@ class basic_json @complexity Linear in the size of @a init. + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + @liveexample{The following code shows an example for the `array` function.,array} - @sa @ref basic_json(std::initializer_list, bool, value_t) -- + @sa @ref basic_json(initializer_list_t, bool, value_t) -- create a JSON value from an initializer list - @sa @ref object(std::initializer_list) -- create a JSON object + @sa @ref object(initializer_list_t) -- create a JSON object value from an initializer list @since version 1.0.0 */ - static basic_json array(std::initializer_list init = - std::initializer_list()) + static basic_json array(initializer_list_t init = {}) { return basic_json(init, false, value_t::array); } @@ -1728,34 +8555,37 @@ class basic_json the initializer list is empty, the empty object `{}` is created. @note This function is only added for symmetry reasons. In contrast to the - related function @ref array(std::initializer_list), there are + related function @ref array(initializer_list_t), there are no cases which can only be expressed by this function. That is, any initializer list @a init can also be passed to the initializer list - constructor @ref basic_json(std::initializer_list, bool, - value_t). + constructor @ref basic_json(initializer_list_t, bool, value_t). @param[in] init initializer list to create an object from (optional) @return JSON object value - @throw std::domain_error if @a init is not a pair whose first elements are - strings; thrown by - @ref basic_json(std::initializer_list, bool, value_t) + @throw type_error.301 if @a init is not a list of pairs whose first + elements are strings. In this case, no object can be created. When such a + value is passed to @ref basic_json(initializer_list_t, bool, value_t), + an array would have been created from the passed initializer list @a init. + See example below. @complexity Linear in the size of @a init. + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + @liveexample{The following code shows an example for the `object` function.,object} - @sa @ref basic_json(std::initializer_list, bool, value_t) -- + @sa @ref basic_json(initializer_list_t, bool, value_t) -- create a JSON value from an initializer list - @sa @ref array(std::initializer_list) -- create a JSON array + @sa @ref array(initializer_list_t) -- create a JSON array value from an initializer list @since version 1.0.0 */ - static basic_json object(std::initializer_list init = - std::initializer_list()) + static basic_json object(initializer_list_t init = {}) { return basic_json(init, false, value_t::object); } @@ -1764,14 +8594,18 @@ class basic_json @brief construct an array with count copies of given value Constructs a JSON array value by creating @a cnt copies of a passed value. - In case @a cnt is `0`, an empty array is created. As postcondition, - `std::distance(begin(),end()) == cnt` holds. + In case @a cnt is `0`, an empty array is created. @param[in] cnt the number of JSON copies of @a val to create @param[in] val the JSON value to copy + @post `std::distance(begin(),end()) == cnt` holds. + @complexity Linear in @a cnt. + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + @liveexample{The following code shows examples for the @ref basic_json(size_type\, const basic_json&) constructor.,basic_json__size_type_basic_json} @@ -1790,12 +8624,13 @@ class basic_json Constructs the JSON value with the contents of the range `[first, last)`. The semantics depends on the different types a JSON value can have: - - In case of primitive types (number, boolean, or string), @a first must - be `begin()` and @a last must be `end()`. In this case, the value is - copied. Otherwise, std::out_of_range is thrown. + - In case of a null type, invalid_iterator.206 is thrown. + - In case of other primitive types (number, boolean, or string), @a first + must be `begin()` and @a last must be `end()`. In this case, the value is + copied. Otherwise, invalid_iterator.204 is thrown. - In case of structured types (array, object), the constructor behaves as - similar versions for `std::vector`. - - In case of a null type, std::domain_error is thrown. + similar versions for `std::vector` or `std::map`; that is, a JSON array + or object is constructed from the values in the range. @tparam InputIT an input iterator type (@ref iterator or @ref const_iterator) @@ -1804,19 +8639,36 @@ class basic_json @param[in] last end of the range to copy from (excluded) @pre Iterators @a first and @a last must be initialized. **This - precondition is enforced with an assertion.** + precondition is enforced with an assertion (see warning).** If + assertions are switched off, a violation of this precondition yields + undefined behavior. - @throw std::domain_error if iterators are not compatible; that is, do not - belong to the same JSON value; example: `"iterators are not compatible"` - @throw std::out_of_range if iterators are for a primitive type (number, - boolean, or string) where an out of range error can be detected easily; - example: `"iterators out of range"` - @throw std::bad_alloc if allocation for object, array, or string fails - @throw std::domain_error if called with a null value; example: `"cannot - use construct with iterators from null"` + @pre Range `[first, last)` is valid. Usually, this precondition cannot be + checked efficiently. Only certain edge cases are detected; see the + description of the exceptions below. A violation of this precondition + yields undefined behavior. + + @warning A precondition is enforced with a runtime assertion that will + result in calling `std::abort` if this precondition is not met. + Assertions can be disabled by defining `NDEBUG` at compile time. + See http://en.cppreference.com/w/cpp/error/assert for more + information. + + @throw invalid_iterator.201 if iterators @a first and @a last are not + compatible (i.e., do not belong to the same JSON value). In this case, + the range `[first, last)` is undefined. + @throw invalid_iterator.204 if iterators @a first and @a last belong to a + primitive type (number, boolean, or string), but @a first does not point + to the first element any more. In this case, the range `[first, last)` is + undefined. See example code below. + @throw invalid_iterator.206 if iterators @a first and @a last belong to a + null value. In this case, the range `[first, last)` is undefined. @complexity Linear in distance between @a first and @a last. + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + @liveexample{The example below shows several ways to create JSON values by specifying a subrange with iterators.,basic_json__InputIt_InputIt} @@ -1831,9 +8683,9 @@ class basic_json assert(last.m_object != nullptr); // make sure iterator fits the current value - if (first.m_object != last.m_object) + if (JSON_UNLIKELY(first.m_object != last.m_object)) { - throw std::domain_error("iterators are not compatible"); + JSON_THROW(invalid_iterator::create(201, "iterators are not compatible")); } // copy type from first iterator @@ -1848,17 +8700,16 @@ class basic_json case value_t::number_unsigned: case value_t::string: { - if (not first.m_it.primitive_iterator.is_begin() or not last.m_it.primitive_iterator.is_end()) + if (JSON_UNLIKELY(not first.m_it.primitive_iterator.is_begin() + or not last.m_it.primitive_iterator.is_end())) { - throw std::out_of_range("iterators out of range"); + JSON_THROW(invalid_iterator::create(204, "iterators out of range")); } break; } default: - { break; - } } switch (m_type) @@ -1895,64 +8746,36 @@ class basic_json case value_t::object: { - m_value.object = create(first.m_it.object_iterator, last.m_it.object_iterator); + m_value.object = create(first.m_it.object_iterator, + last.m_it.object_iterator); break; } case value_t::array: { - m_value.array = create(first.m_it.array_iterator, last.m_it.array_iterator); + m_value.array = create(first.m_it.array_iterator, + last.m_it.array_iterator); break; } default: - { - throw std::domain_error("cannot use construct with iterators from " + first.m_object->type_name()); - } + JSON_THROW(invalid_iterator::create(206, "cannot construct with iterators from " + + std::string(first.m_object->type_name()))); } assert_invariant(); } - /*! - @brief construct a JSON value given an input stream - - @param[in,out] i stream to read a serialized JSON value from - @param[in] cb a parser callback function of type @ref parser_callback_t - which is used to control the deserialization by filtering unwanted values - (optional) - - @complexity Linear in the length of the input. The parser is a predictive - LL(1) parser. The complexity can be higher if the parser callback function - @a cb has a super-linear complexity. - - @note A UTF-8 byte order mark is silently ignored. - - @deprecated This constructor is deprecated and will be removed in version - 3.0.0 to unify the interface of the library. Deserialization will be - done by stream operators or by calling one of the `parse` functions, - e.g. @ref parse(std::istream&, const parser_callback_t). That is, calls - like `json j(i);` for an input stream @a i need to be replaced by - `json j = json::parse(i);`. See the example below. - - @liveexample{The example below demonstrates constructing a JSON value from - a `std::stringstream` with and without callback - function.,basic_json__istream} - - @since version 2.0.0, deprecated in version 2.0.3, to be removed in - version 3.0.0 - */ - JSON_DEPRECATED - explicit basic_json(std::istream& i, const parser_callback_t cb = nullptr) - { - *this = parser(i, cb).parse(); - assert_invariant(); - } /////////////////////////////////////// // other constructors and destructor // /////////////////////////////////////// + /// @private + basic_json(const detail::json_ref& ref) + : basic_json(ref.moved_or_copied()) + {} + /*! @brief copy constructor @@ -1960,16 +8783,19 @@ class basic_json @param[in] other the JSON value to copy + @post `*this == other` + @complexity Linear in the size of @a other. + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes to any JSON value. + @requirement This function helps `basic_json` satisfying the [Container](http://en.cppreference.com/w/cpp/concept/Container) requirements: - The complexity is linear. - As postcondition, it holds: `other == basic_json(other)`. - @throw std::bad_alloc if allocation for object, array, or string fails. - @liveexample{The following code shows an example for the copy constructor.,basic_json__basic_json} @@ -2026,9 +8852,7 @@ class basic_json } default: - { break; - } } assert_invariant(); @@ -2043,10 +8867,18 @@ class basic_json @param[in,out] other value to move to this object - @post @a other is a JSON null value + @post `*this` has the same value as @a other before the call. + @post @a other is a JSON null value. @complexity Constant. + @exceptionsafety No-throw guarantee: this constructor never throws + exceptions. + + @requirement This function helps `basic_json` satisfying the + [MoveConstructible](http://en.cppreference.com/w/cpp/concept/MoveConstructible) + requirements. + @liveexample{The code below shows the move constructor explicitly called via std::move.,basic_json__moveconstructor} @@ -2071,7 +8903,7 @@ class basic_json Copy assignment operator. Copies a JSON value via the "copy and swap" strategy: It is expressed in terms of the copy constructor, destructor, - and the swap() member function. + and the `swap()` member function. @param[in] other value to copy from @@ -2125,39 +8957,7 @@ class basic_json ~basic_json() { assert_invariant(); - - switch (m_type) - { - case value_t::object: - { - AllocatorType alloc; - alloc.destroy(m_value.object); - alloc.deallocate(m_value.object, 1); - break; - } - - case value_t::array: - { - AllocatorType alloc; - alloc.destroy(m_value.array); - alloc.deallocate(m_value.array, 1); - break; - } - - case value_t::string: - { - AllocatorType alloc; - alloc.destroy(m_value.string); - alloc.deallocate(m_value.string, 1); - break; - } - - default: - { - // all other types need no specific destructor - break; - } - } + m_value.destroy(m_type); } /// @} @@ -2176,46 +8976,50 @@ class basic_json Serialization function for JSON values. The function tries to mimic Python's `json.dumps()` function, and currently supports its @a indent - parameter. + and @a ensure_ascii parameters. @param[in] indent If indent is nonnegative, then array elements and object members will be pretty-printed with that indent level. An indent level of `0` will only insert newlines. `-1` (the default) selects the most compact representation. + @param[in] indent_char The character to use for indentation if @a indent is + greater than `0`. The default is ` ` (space). + @param[in] ensure_ascii If @a ensure_ascii is true, all non-ASCII characters + in the output are escaped with \uXXXX sequences, and the result consists + of ASCII characters only. @return string containing the serialization of the JSON value @complexity Linear. - @liveexample{The following example shows the effect of different @a indent - parameters to the result of the serialization.,dump} + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. + + @liveexample{The following example shows the effect of different @a indent\, + @a indent_char\, and @a ensure_ascii parameters to the result of the + serialization.,dump} @see https://docs.python.org/2/library/json.html#json.dump - @since version 1.0.0 + @since version 1.0.0; indentation character @a indent_char and option + @a ensure_ascii added in version 3.0.0 */ - string_t dump(const int indent = -1) const + string_t dump(const int indent = -1, const char indent_char = ' ', + const bool ensure_ascii = false) const { - std::stringstream ss; - // fix locale problems - ss.imbue(std::locale::classic()); - - // 6, 15 or 16 digits of precision allows round-trip IEEE 754 - // string->float->string, string->double->string or string->long - // double->string; to be safe, we read this value from - // std::numeric_limits::digits10 - ss.precision(std::numeric_limits::digits10); + string_t result; + serializer s(detail::output_adapter(result), indent_char); if (indent >= 0) { - dump(ss, true, static_cast(indent)); + s.dump(*this, true, ensure_ascii, static_cast(indent)); } else { - dump(ss, false, 0); + s.dump(*this, false, ensure_ascii, 0); } - return ss.str(); + return result; } /*! @@ -2225,6 +9029,17 @@ class basic_json enumeration. @return the type of the JSON value + Value type | return value + ------------------------- | ------------------------- + null | value_t::null + boolean | value_t::boolean + string | value_t::string + number (integer) | value_t::number_integer + number (unsigned integer) | value_t::number_unsigned + number (foating-point) | value_t::number_float + object | value_t::object + array | value_t::array + discarded | value_t::discarded @complexity Constant. @@ -2234,6 +9049,9 @@ class basic_json @liveexample{The following code exemplifies `type()` for all JSON types.,type} + @sa @ref operator value_t() -- return the type of the JSON value (implicit) + @sa @ref type_name() -- return the type as string + @since version 1.0.0 */ constexpr value_t type() const noexcept @@ -2244,8 +9062,8 @@ class basic_json /*! @brief return whether type is primitive - This function returns true iff the JSON type is primitive (string, number, - boolean, or null). + This function returns true if and only if the JSON type is primitive + (string, number, boolean, or null). @return `true` if type is primitive (string, number, boolean, or null), `false` otherwise. @@ -2274,8 +9092,8 @@ class basic_json /*! @brief return whether type is structured - This function returns true iff the JSON type is structured (array or - object). + This function returns true if and only if the JSON type is structured + (array or object). @return `true` if type is structured (array or object), `false` otherwise. @@ -2301,7 +9119,7 @@ class basic_json /*! @brief return whether value is null - This function returns true iff the JSON value is null. + This function returns true if and only if the JSON value is null. @return `true` if type is null, `false` otherwise. @@ -2317,13 +9135,13 @@ class basic_json */ constexpr bool is_null() const noexcept { - return m_type == value_t::null; + return (m_type == value_t::null); } /*! @brief return whether value is a boolean - This function returns true iff the JSON value is a boolean. + This function returns true if and only if the JSON value is a boolean. @return `true` if type is boolean, `false` otherwise. @@ -2339,14 +9157,14 @@ class basic_json */ constexpr bool is_boolean() const noexcept { - return m_type == value_t::boolean; + return (m_type == value_t::boolean); } /*! @brief return whether value is a number - This function returns true iff the JSON value is a number. This includes - both integer and floating-point values. + This function returns true if and only if the JSON value is a number. This + includes both integer (signed and unsigned) and floating-point values. @return `true` if type is number (regardless whether integer, unsigned integer or floating-type), `false` otherwise. @@ -2375,8 +9193,8 @@ class basic_json /*! @brief return whether value is an integer number - This function returns true iff the JSON value is an integer or unsigned - integer number. This excludes floating-point values. + This function returns true if and only if the JSON value is a signed or + unsigned integer number. This excludes floating-point values. @return `true` if type is an integer or unsigned integer number, `false` otherwise. @@ -2398,14 +9216,14 @@ class basic_json */ constexpr bool is_number_integer() const noexcept { - return m_type == value_t::number_integer or m_type == value_t::number_unsigned; + return (m_type == value_t::number_integer or m_type == value_t::number_unsigned); } /*! @brief return whether value is an unsigned integer number - This function returns true iff the JSON value is an unsigned integer - number. This excludes floating-point and (signed) integer values. + This function returns true if and only if the JSON value is an unsigned + integer number. This excludes floating-point and signed integer values. @return `true` if type is an unsigned integer number, `false` otherwise. @@ -2426,14 +9244,14 @@ class basic_json */ constexpr bool is_number_unsigned() const noexcept { - return m_type == value_t::number_unsigned; + return (m_type == value_t::number_unsigned); } /*! @brief return whether value is a floating-point number - This function returns true iff the JSON value is a floating-point number. - This excludes integer and unsigned integer values. + This function returns true if and only if the JSON value is a + floating-point number. This excludes signed and unsigned integer values. @return `true` if type is a floating-point number, `false` otherwise. @@ -2454,13 +9272,13 @@ class basic_json */ constexpr bool is_number_float() const noexcept { - return m_type == value_t::number_float; + return (m_type == value_t::number_float); } /*! @brief return whether value is an object - This function returns true iff the JSON value is an object. + This function returns true if and only if the JSON value is an object. @return `true` if type is object, `false` otherwise. @@ -2476,13 +9294,13 @@ class basic_json */ constexpr bool is_object() const noexcept { - return m_type == value_t::object; + return (m_type == value_t::object); } /*! @brief return whether value is an array - This function returns true iff the JSON value is an array. + This function returns true if and only if the JSON value is an array. @return `true` if type is array, `false` otherwise. @@ -2498,13 +9316,13 @@ class basic_json */ constexpr bool is_array() const noexcept { - return m_type == value_t::array; + return (m_type == value_t::array); } /*! @brief return whether value is a string - This function returns true iff the JSON value is a string. + This function returns true if and only if the JSON value is a string. @return `true` if type is string, `false` otherwise. @@ -2520,14 +9338,14 @@ class basic_json */ constexpr bool is_string() const noexcept { - return m_type == value_t::string; + return (m_type == value_t::string); } /*! @brief return whether value is discarded - This function returns true iff the JSON value was discarded during parsing - with a callback function (see @ref parser_callback_t). + This function returns true if and only if the JSON value was discarded + during parsing with a callback function (see @ref parser_callback_t). @note This function will always be `false` for JSON values after parsing. That is, discarded values can only occur during parsing, but will be @@ -2547,7 +9365,7 @@ class basic_json */ constexpr bool is_discarded() const noexcept { - return m_type == value_t::discarded; + return (m_type == value_t::discarded); } /*! @@ -2566,6 +9384,9 @@ class basic_json @liveexample{The following code exemplifies the @ref value_t operator for all JSON types.,operator__value_t} + @sa @ref type() -- return the type of the JSON value (explicit) + @sa @ref type_name() -- return the type as string + @since version 1.0.0 */ constexpr operator value_t() const noexcept @@ -2580,244 +9401,97 @@ class basic_json // value access // ////////////////// - /// get an object (explicit) - template::value and - std::is_convertible::value, int>::type = 0> - T get_impl(T*) const - { - if (is_object()) - { - return T(m_value.object->begin(), m_value.object->end()); - } - else - { - throw std::domain_error("type must be object, but is " + type_name()); - } - } - - /// get an object (explicit) - object_t get_impl(object_t*) const - { - if (is_object()) - { - return *(m_value.object); - } - else - { - throw std::domain_error("type must be object, but is " + type_name()); - } - } - - /// get an array (explicit) - template::value and - not std::is_same::value and - not std::is_arithmetic::value and - not std::is_convertible::value and - not has_mapped_type::value, int>::type = 0> - T get_impl(T*) const - { - if (is_array()) - { - T to_vector; - std::transform(m_value.array->begin(), m_value.array->end(), - std::inserter(to_vector, to_vector.end()), [](basic_json i) - { - return i.get(); - }); - return to_vector; - } - else - { - throw std::domain_error("type must be array, but is " + type_name()); - } - } - - /// get an array (explicit) - template::value and - not std::is_same::value, int>::type = 0> - std::vector get_impl(std::vector*) const - { - if (is_array()) - { - std::vector to_vector; - to_vector.reserve(m_value.array->size()); - std::transform(m_value.array->begin(), m_value.array->end(), - std::inserter(to_vector, to_vector.end()), [](basic_json i) - { - return i.get(); - }); - return to_vector; - } - else - { - throw std::domain_error("type must be array, but is " + type_name()); - } - } - - /// get an array (explicit) - template::value and - not has_mapped_type::value, int>::type = 0> - T get_impl(T*) const - { - if (is_array()) - { - return T(m_value.array->begin(), m_value.array->end()); - } - else - { - throw std::domain_error("type must be array, but is " + type_name()); - } - } - - /// get an array (explicit) - array_t get_impl(array_t*) const - { - if (is_array()) - { - return *(m_value.array); - } - else - { - throw std::domain_error("type must be array, but is " + type_name()); - } - } - - /// get a string (explicit) - template::value, int>::type = 0> - T get_impl(T*) const - { - if (is_string()) - { - return *m_value.string; - } - else - { - throw std::domain_error("type must be string, but is " + type_name()); - } - } - - /// get a number (explicit) - template::value, int>::type = 0> - T get_impl(T*) const - { - switch (m_type) - { - case value_t::number_integer: - { - return static_cast(m_value.number_integer); - } - - case value_t::number_unsigned: - { - return static_cast(m_value.number_unsigned); - } - - case value_t::number_float: - { - return static_cast(m_value.number_float); - } - - default: - { - throw std::domain_error("type must be number, but is " + type_name()); - } - } - } - /// get a boolean (explicit) - constexpr boolean_t get_impl(boolean_t*) const + boolean_t get_impl(boolean_t* /*unused*/) const { - return is_boolean() - ? m_value.boolean - : throw std::domain_error("type must be boolean, but is " + type_name()); + if (JSON_LIKELY(is_boolean())) + { + return m_value.boolean; + } + + JSON_THROW(type_error::create(302, "type must be boolean, but is " + std::string(type_name()))); } /// get a pointer to the value (object) - object_t* get_impl_ptr(object_t*) noexcept + object_t* get_impl_ptr(object_t* /*unused*/) noexcept { return is_object() ? m_value.object : nullptr; } /// get a pointer to the value (object) - constexpr const object_t* get_impl_ptr(const object_t*) const noexcept + constexpr const object_t* get_impl_ptr(const object_t* /*unused*/) const noexcept { return is_object() ? m_value.object : nullptr; } /// get a pointer to the value (array) - array_t* get_impl_ptr(array_t*) noexcept + array_t* get_impl_ptr(array_t* /*unused*/) noexcept { return is_array() ? m_value.array : nullptr; } /// get a pointer to the value (array) - constexpr const array_t* get_impl_ptr(const array_t*) const noexcept + constexpr const array_t* get_impl_ptr(const array_t* /*unused*/) const noexcept { return is_array() ? m_value.array : nullptr; } /// get a pointer to the value (string) - string_t* get_impl_ptr(string_t*) noexcept + string_t* get_impl_ptr(string_t* /*unused*/) noexcept { return is_string() ? m_value.string : nullptr; } /// get a pointer to the value (string) - constexpr const string_t* get_impl_ptr(const string_t*) const noexcept + constexpr const string_t* get_impl_ptr(const string_t* /*unused*/) const noexcept { return is_string() ? m_value.string : nullptr; } /// get a pointer to the value (boolean) - boolean_t* get_impl_ptr(boolean_t*) noexcept + boolean_t* get_impl_ptr(boolean_t* /*unused*/) noexcept { return is_boolean() ? &m_value.boolean : nullptr; } /// get a pointer to the value (boolean) - constexpr const boolean_t* get_impl_ptr(const boolean_t*) const noexcept + constexpr const boolean_t* get_impl_ptr(const boolean_t* /*unused*/) const noexcept { return is_boolean() ? &m_value.boolean : nullptr; } /// get a pointer to the value (integer number) - number_integer_t* get_impl_ptr(number_integer_t*) noexcept + number_integer_t* get_impl_ptr(number_integer_t* /*unused*/) noexcept { return is_number_integer() ? &m_value.number_integer : nullptr; } /// get a pointer to the value (integer number) - constexpr const number_integer_t* get_impl_ptr(const number_integer_t*) const noexcept + constexpr const number_integer_t* get_impl_ptr(const number_integer_t* /*unused*/) const noexcept { return is_number_integer() ? &m_value.number_integer : nullptr; } /// get a pointer to the value (unsigned number) - number_unsigned_t* get_impl_ptr(number_unsigned_t*) noexcept + number_unsigned_t* get_impl_ptr(number_unsigned_t* /*unused*/) noexcept { return is_number_unsigned() ? &m_value.number_unsigned : nullptr; } /// get a pointer to the value (unsigned number) - constexpr const number_unsigned_t* get_impl_ptr(const number_unsigned_t*) const noexcept + constexpr const number_unsigned_t* get_impl_ptr(const number_unsigned_t* /*unused*/) const noexcept { return is_number_unsigned() ? &m_value.number_unsigned : nullptr; } /// get a pointer to the value (floating-point number) - number_float_t* get_impl_ptr(number_float_t*) noexcept + number_float_t* get_impl_ptr(number_float_t* /*unused*/) noexcept { return is_number_float() ? &m_value.number_float : nullptr; } /// get a pointer to the value (floating-point number) - constexpr const number_float_t* get_impl_ptr(const number_float_t*) const noexcept + constexpr const number_float_t* get_impl_ptr(const number_float_t* /*unused*/) const noexcept { return is_number_float() ? &m_value.number_float : nullptr; } @@ -2825,55 +9499,86 @@ class basic_json /*! @brief helper function to implement get_ref() - This funcion helps to implement get_ref() without code duplication for + This function helps to implement get_ref() without code duplication for const and non-const overloads @tparam ThisType will be deduced as `basic_json` or `const basic_json` - @throw std::domain_error if ReferenceType does not match underlying value + @throw type_error.303 if ReferenceType does not match underlying value type of the current JSON */ template static ReferenceType get_ref_impl(ThisType& obj) { - // helper type - using PointerType = typename std::add_pointer::type; - // delegate the call to get_ptr<>() - auto ptr = obj.template get_ptr(); + auto ptr = obj.template get_ptr::type>(); - if (ptr != nullptr) + if (JSON_LIKELY(ptr != nullptr)) { return *ptr; } - else - { - throw std::domain_error("incompatible ReferenceType for get_ref, actual type is " + - obj.type_name()); - } + + JSON_THROW(type_error::create(303, "incompatible ReferenceType for get_ref, actual type is " + std::string(obj.type_name()))); } public: - /// @name value access /// Direct access to the stored value of a JSON value. /// @{ + /*! + @brief get special-case overload + + This overloads avoids a lot of template boilerplate, it can be seen as the + identity method + + @tparam BasicJsonType == @ref basic_json + + @return a copy of *this + + @complexity Constant. + + @since version 2.1.0 + */ + template < + typename BasicJsonType, + detail::enable_if_t::type, + basic_json_t>::value, + int> = 0 > + basic_json get() const + { + return *this; + } + /*! @brief get a value (explicit) - Explicit type conversion between the JSON value and a compatible value. + Explicit type conversion between the JSON value and a compatible value + which is [CopyConstructible](http://en.cppreference.com/w/cpp/concept/CopyConstructible) + and [DefaultConstructible](http://en.cppreference.com/w/cpp/concept/DefaultConstructible). + The value is converted by calling the @ref json_serializer + `from_json()` method. - @tparam ValueType non-pointer type compatible to the JSON value, for - instance `int` for JSON integer numbers, `bool` for JSON booleans, or - `std::vector` types for JSON arrays + The function is equivalent to executing + @code {.cpp} + ValueType ret; + JSONSerializer::from_json(*this, ret); + return ret; + @endcode - @return copy of the JSON value, converted to type @a ValueType + This overloads is chosen if: + - @a ValueType is not @ref basic_json, + - @ref json_serializer has a `from_json()` method of the form + `void from_json(const basic_json&, ValueType&)`, and + - @ref json_serializer does not have a `from_json()` method of + the form `ValueType from_json(const basic_json&)` - @throw std::domain_error in case passed type @a ValueType is incompatible - to JSON; example: `"type must be object, but is null"` + @tparam ValueTypeCV the provided value type + @tparam ValueType the returned value type - @complexity Linear in the size of the JSON value. + @return copy of the JSON value, converted to @a ValueType + + @throw what @ref json_serializer `from_json()` method throws @liveexample{The example below shows several conversions from JSON values to other types. There a few things to note: (1) Floating-point numbers can @@ -2882,21 +9587,75 @@ class basic_json associative containers such as `std::unordered_map`.,get__ValueType_const} - @internal - The idea of using a casted null pointer to choose the correct - implementation is from . - @endinternal - - @sa @ref operator ValueType() const for implicit conversion - @sa @ref get() for pointer-member access - - @since version 1.0.0 + @since version 2.1.0 */ - template::value, int>::type = 0> - ValueType get() const + template < + typename ValueTypeCV, + typename ValueType = detail::uncvref_t, + detail::enable_if_t < + not std::is_same::value and + detail::has_from_json::value and + not detail::has_non_default_from_json::value, + int > = 0 > + ValueType get() const noexcept(noexcept( + JSONSerializer::from_json(std::declval(), std::declval()))) { - return get_impl(static_cast(nullptr)); + // we cannot static_assert on ValueTypeCV being non-const, because + // there is support for get(), which is why we + // still need the uncvref + static_assert(not std::is_reference::value, + "get() cannot be used with reference types, you might want to use get_ref()"); + static_assert(std::is_default_constructible::value, + "types must be DefaultConstructible when used with get()"); + + ValueType ret; + JSONSerializer::from_json(*this, ret); + return ret; + } + + /*! + @brief get a value (explicit); special case + + Explicit type conversion between the JSON value and a compatible value + which is **not** [CopyConstructible](http://en.cppreference.com/w/cpp/concept/CopyConstructible) + and **not** [DefaultConstructible](http://en.cppreference.com/w/cpp/concept/DefaultConstructible). + The value is converted by calling the @ref json_serializer + `from_json()` method. + + The function is equivalent to executing + @code {.cpp} + return JSONSerializer::from_json(*this); + @endcode + + This overloads is chosen if: + - @a ValueType is not @ref basic_json and + - @ref json_serializer has a `from_json()` method of the form + `ValueType from_json(const basic_json&)` + + @note If @ref json_serializer has both overloads of + `from_json()`, this one is chosen. + + @tparam ValueTypeCV the provided value type + @tparam ValueType the returned value type + + @return copy of the JSON value, converted to @a ValueType + + @throw what @ref json_serializer `from_json()` method throws + + @since version 2.1.0 + */ + template < + typename ValueTypeCV, + typename ValueType = detail::uncvref_t, + detail::enable_if_t::value and + detail::has_non_default_from_json::value, int> = 0 > + ValueType get() const noexcept(noexcept( + JSONSerializer::from_json(std::declval()))) + { + static_assert(not std::is_reference::value, + "get() cannot be used with reference types, you might want to use get_ref()"); + return JSONSerializer::from_json(*this); } /*! @@ -3026,7 +9785,7 @@ class basic_json /*! @brief get a reference value (implicit) - Implict reference access to the internally stored JSON value. No copies + Implicit reference access to the internally stored JSON value. No copies are made. @warning Writing data to the referee of the result yields an undefined @@ -3038,10 +9797,10 @@ class basic_json @return reference to the internally stored JSON value if the requested reference type @a ReferenceType fits to the JSON value; throws - std::domain_error otherwise + type_error.303 otherwise - @throw std::domain_error in case passed type @a ReferenceType is - incompatible with the stored JSON value + @throw type_error.303 in case passed type @a ReferenceType is incompatible + with the stored JSON value; see example below @complexity Constant. @@ -3084,8 +9843,9 @@ class basic_json @return copy of the JSON value, converted to type @a ValueType - @throw std::domain_error in case passed type @a ValueType is incompatible - to JSON, thrown by @ref get() const + @throw type_error.302 in case passed type @a ValueType is incompatible + to the JSON value type (e.g., the JSON value is of type boolean, but a + string is requested); see example below @complexity Linear in the size of the JSON value. @@ -3100,9 +9860,13 @@ class basic_json */ template < typename ValueType, typename std::enable_if < not std::is_pointer::value and + not std::is_same>::value and not std::is_same::value -#ifndef _MSC_VER // Fix for issue #167 operator<< abiguity under VS2015 +#ifndef _MSC_VER // fix for issue #167 operator<< ambiguity under VS2015 and not std::is_same>::value +#endif +#if (defined(__cplusplus) && __cplusplus >= 201703L) || (defined(_MSC_VER) && _MSC_VER >1900 && defined(_HAS_CXX17) && _HAS_CXX17 == 1) // fix for issue #464 + and not std::is_same::value #endif , int >::type = 0 > operator ValueType() const @@ -3132,36 +9896,40 @@ class basic_json @return reference to the element at index @a idx - @throw std::domain_error if the JSON value is not an array; example: - `"cannot use at() with string"` - @throw std::out_of_range if the index @a idx is out of range of the array; - that is, `idx >= size()`; example: `"array index 7 is out of range"` + @throw type_error.304 if the JSON value is not an array; in this case, + calling `at` with an index makes no sense. See example below. + @throw out_of_range.401 if the index @a idx is out of range of the array; + that is, `idx >= size()`. See example below. + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. @complexity Constant. - @liveexample{The example below shows how array elements can be read and - written using `at()`.,at__size_type} - @since version 1.0.0 + + @liveexample{The example below shows how array elements can be read and + written using `at()`. It also demonstrates the different exceptions that + can be thrown.,at__size_type} */ reference at(size_type idx) { // at only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { - try + JSON_TRY { return m_value.array->at(idx); } - catch (std::out_of_range&) + JSON_CATCH (std::out_of_range&) { // create better exception explanation - throw std::out_of_range("array index " + std::to_string(idx) + " is out of range"); + JSON_THROW(out_of_range::create(401, "array index " + std::to_string(idx) + " is out of range")); } } else { - throw std::domain_error("cannot use at() with " + type_name()); + JSON_THROW(type_error::create(304, "cannot use at() with " + std::string(type_name()))); } } @@ -3175,36 +9943,40 @@ class basic_json @return const reference to the element at index @a idx - @throw std::domain_error if the JSON value is not an array; example: - `"cannot use at() with string"` - @throw std::out_of_range if the index @a idx is out of range of the array; - that is, `idx >= size()`; example: `"array index 7 is out of range"` + @throw type_error.304 if the JSON value is not an array; in this case, + calling `at` with an index makes no sense. See example below. + @throw out_of_range.401 if the index @a idx is out of range of the array; + that is, `idx >= size()`. See example below. + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. @complexity Constant. - @liveexample{The example below shows how array elements can be read using - `at()`.,at__size_type_const} - @since version 1.0.0 + + @liveexample{The example below shows how array elements can be read using + `at()`. It also demonstrates the different exceptions that can be thrown., + at__size_type_const} */ const_reference at(size_type idx) const { // at only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { - try + JSON_TRY { return m_value.array->at(idx); } - catch (std::out_of_range&) + JSON_CATCH (std::out_of_range&) { // create better exception explanation - throw std::out_of_range("array index " + std::to_string(idx) + " is out of range"); + JSON_THROW(out_of_range::create(401, "array index " + std::to_string(idx) + " is out of range")); } } else { - throw std::domain_error("cannot use at() with " + type_name()); + JSON_THROW(type_error::create(304, "cannot use at() with " + std::string(type_name()))); } } @@ -3218,40 +9990,44 @@ class basic_json @return reference to the element at key @a key - @throw std::domain_error if the JSON value is not an object; example: - `"cannot use at() with boolean"` - @throw std::out_of_range if the key @a key is is not stored in the object; - that is, `find(key) == end()`; example: `"key "the fast" not found"` + @throw type_error.304 if the JSON value is not an object; in this case, + calling `at` with a key makes no sense. See example below. + @throw out_of_range.403 if the key @a key is is not stored in the object; + that is, `find(key) == end()`. See example below. + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. @complexity Logarithmic in the size of the container. - @liveexample{The example below shows how object elements can be read and - written using `at()`.,at__object_t_key_type} - @sa @ref operator[](const typename object_t::key_type&) for unchecked access by reference @sa @ref value() for access by value with a default value @since version 1.0.0 + + @liveexample{The example below shows how object elements can be read and + written using `at()`. It also demonstrates the different exceptions that + can be thrown.,at__object_t_key_type} */ reference at(const typename object_t::key_type& key) { // at only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { - try + JSON_TRY { return m_value.object->at(key); } - catch (std::out_of_range&) + JSON_CATCH (std::out_of_range&) { // create better exception explanation - throw std::out_of_range("key '" + key + "' not found"); + JSON_THROW(out_of_range::create(403, "key '" + key + "' not found")); } } else { - throw std::domain_error("cannot use at() with " + type_name()); + JSON_THROW(type_error::create(304, "cannot use at() with " + std::string(type_name()))); } } @@ -3265,40 +10041,44 @@ class basic_json @return const reference to the element at key @a key - @throw std::domain_error if the JSON value is not an object; example: - `"cannot use at() with boolean"` - @throw std::out_of_range if the key @a key is is not stored in the object; - that is, `find(key) == end()`; example: `"key "the fast" not found"` + @throw type_error.304 if the JSON value is not an object; in this case, + calling `at` with a key makes no sense. See example below. + @throw out_of_range.403 if the key @a key is is not stored in the object; + that is, `find(key) == end()`. See example below. + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. @complexity Logarithmic in the size of the container. - @liveexample{The example below shows how object elements can be read using - `at()`.,at__object_t_key_type_const} - @sa @ref operator[](const typename object_t::key_type&) for unchecked access by reference @sa @ref value() for access by value with a default value @since version 1.0.0 + + @liveexample{The example below shows how object elements can be read using + `at()`. It also demonstrates the different exceptions that can be thrown., + at__object_t_key_type_const} */ const_reference at(const typename object_t::key_type& key) const { // at only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { - try + JSON_TRY { return m_value.object->at(key); } - catch (std::out_of_range&) + JSON_CATCH (std::out_of_range&) { // create better exception explanation - throw std::out_of_range("key '" + key + "' not found"); + JSON_THROW(out_of_range::create(403, "key '" + key + "' not found")); } } else { - throw std::domain_error("cannot use at() with " + type_name()); + JSON_THROW(type_error::create(304, "cannot use at() with " + std::string(type_name()))); } } @@ -3315,8 +10095,8 @@ class basic_json @return reference to the element at index @a idx - @throw std::domain_error if JSON is not an array or null; example: - `"cannot use operator[] with string"` + @throw type_error.305 if the JSON value is not an array or null; in that + cases, using the [] operator with an index makes no sense. @complexity Constant if @a idx is in the range of the array. Otherwise linear in `idx - size()`. @@ -3338,7 +10118,7 @@ class basic_json } // operator[] only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { // fill up array with null values if given idx is outside range if (idx >= m_value.array->size()) @@ -3350,10 +10130,8 @@ class basic_json return m_value.array->operator[](idx); } - else - { - throw std::domain_error("cannot use operator[] with " + type_name()); - } + + JSON_THROW(type_error::create(305, "cannot use operator[] with " + std::string(type_name()))); } /*! @@ -3365,8 +10143,8 @@ class basic_json @return const reference to the element at index @a idx - @throw std::domain_error if JSON is not an array; example: `"cannot use - operator[] with null"` + @throw type_error.305 if the JSON value is not an array; in that cases, + using the [] operator with an index makes no sense. @complexity Constant. @@ -3378,14 +10156,12 @@ class basic_json const_reference operator[](size_type idx) const { // const operator[] only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { return m_value.array->operator[](idx); } - else - { - throw std::domain_error("cannot use operator[] with " + type_name()); - } + + JSON_THROW(type_error::create(305, "cannot use operator[] with " + std::string(type_name()))); } /*! @@ -3401,8 +10177,8 @@ class basic_json @return reference to the element at key @a key - @throw std::domain_error if JSON is not an object or null; example: - `"cannot use operator[] with string"` + @throw type_error.305 if the JSON value is not an object or null; in that + cases, using the [] operator with a key makes no sense. @complexity Logarithmic in the size of the container. @@ -3426,14 +10202,12 @@ class basic_json } // operator[] only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { return m_value.object->operator[](key); } - else - { - throw std::domain_error("cannot use operator[] with " + type_name()); - } + + JSON_THROW(type_error::create(305, "cannot use operator[] with " + std::string(type_name()))); } /*! @@ -3452,8 +10226,8 @@ class basic_json @pre The element with key @a key must exist. **This precondition is enforced with an assertion.** - @throw std::domain_error if JSON is not an object; example: `"cannot use - operator[] with null"` + @throw type_error.305 if the JSON value is not an object; in that cases, + using the [] operator with a key makes no sense. @complexity Logarithmic in the size of the container. @@ -3469,15 +10243,13 @@ class basic_json const_reference operator[](const typename object_t::key_type& key) const { // const operator[] only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { assert(m_value.object->find(key) != m_value.object->end()); return m_value.object->find(key)->second; } - else - { - throw std::domain_error("cannot use operator[] with " + type_name()); - } + + JSON_THROW(type_error::create(305, "cannot use operator[] with " + std::string(type_name()))); } /*! @@ -3493,76 +10265,8 @@ class basic_json @return reference to the element at key @a key - @throw std::domain_error if JSON is not an object or null; example: - `"cannot use operator[] with string"` - - @complexity Logarithmic in the size of the container. - - @liveexample{The example below shows how object elements can be read and - written using the `[]` operator.,operatorarray__key_type} - - @sa @ref at(const typename object_t::key_type&) for access by reference - with range checking - @sa @ref value() for access by value with a default value - - @since version 1.0.0 - */ - template - reference operator[](T * (&key)[n]) - { - return operator[](static_cast(key)); - } - - /*! - @brief read-only access specified object element - - Returns a const reference to the element at with specified key @a key. No - bounds checking is performed. - - @warning If the element with key @a key does not exist, the behavior is - undefined. - - @note This function is required for compatibility reasons with Clang. - - @param[in] key key of the element to access - - @return const reference to the element at key @a key - - @throw std::domain_error if JSON is not an object; example: `"cannot use - operator[] with null"` - - @complexity Logarithmic in the size of the container. - - @liveexample{The example below shows how object elements can be read using - the `[]` operator.,operatorarray__key_type_const} - - @sa @ref at(const typename object_t::key_type&) for access by reference - with range checking - @sa @ref value() for access by value with a default value - - @since version 1.0.0 - */ - template - const_reference operator[](T * (&key)[n]) const - { - return operator[](static_cast(key)); - } - - /*! - @brief access specified object element - - Returns a reference to the element at with specified key @a key. - - @note If @a key is not found in the object, then it is silently added to - the object and filled with a `null` value to make `key` a valid reference. - In case the value was `null` before, it is converted to an object. - - @param[in] key key of the element to access - - @return reference to the element at key @a key - - @throw std::domain_error if JSON is not an object or null; example: - `"cannot use operator[] with string"` + @throw type_error.305 if the JSON value is not an object or null; in that + cases, using the [] operator with a key makes no sense. @complexity Logarithmic in the size of the container. @@ -3587,14 +10291,12 @@ class basic_json } // at only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { return m_value.object->operator[](key); } - else - { - throw std::domain_error("cannot use operator[] with " + type_name()); - } + + JSON_THROW(type_error::create(305, "cannot use operator[] with " + std::string(type_name()))); } /*! @@ -3613,8 +10315,8 @@ class basic_json @pre The element with key @a key must exist. **This precondition is enforced with an assertion.** - @throw std::domain_error if JSON is not an object; example: `"cannot use - operator[] with null"` + @throw type_error.305 if the JSON value is not an object; in that cases, + using the [] operator with a key makes no sense. @complexity Logarithmic in the size of the container. @@ -3631,15 +10333,13 @@ class basic_json const_reference operator[](T* key) const { // at only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { assert(m_value.object->find(key) != m_value.object->end()); return m_value.object->find(key)->second; } - else - { - throw std::domain_error("cannot use operator[] with " + type_name()); - } + + JSON_THROW(type_error::create(305, "cannot use operator[] with " + std::string(type_name()))); } /*! @@ -3652,7 +10352,7 @@ class basic_json @code {.cpp} try { return at(key); - } catch(std::out_of_range) { + } catch(out_of_range) { return default_value; } @endcode @@ -3675,8 +10375,8 @@ class basic_json @return copy of the element at key @a key or @a default_value if @a key is not found - @throw std::domain_error if JSON is not an object; example: `"cannot use - value() with null"` + @throw type_error.306 if the JSON value is not an objec; in that cases, + using `value()` with a key makes no sense. @complexity Logarithmic in the size of the container. @@ -3692,10 +10392,10 @@ class basic_json */ template::value, int>::type = 0> - ValueType value(const typename object_t::key_type& key, ValueType default_value) const + ValueType value(const typename object_t::key_type& key, const ValueType& default_value) const { // at only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { // if key is found, return value and given default value otherwise const auto it = find(key); @@ -3703,15 +10403,11 @@ class basic_json { return *it; } - else - { - return default_value; - } - } - else - { - throw std::domain_error("cannot use value() with " + type_name()); + + return default_value; } + + JSON_THROW(type_error::create(306, "cannot use value() with " + std::string(type_name()))); } /*! @@ -3733,7 +10429,7 @@ class basic_json @code {.cpp} try { return at(ptr); - } catch(std::out_of_range) { + } catch(out_of_range) { return default_value; } @endcode @@ -3752,8 +10448,8 @@ class basic_json @return copy of the element at key @a key or @a default_value if @a key is not found - @throw std::domain_error if JSON is not an object; example: `"cannot use - value() with null"` + @throw type_error.306 if the JSON value is not an objec; in that cases, + using `value()` with a key makes no sense. @complexity Logarithmic in the size of the container. @@ -3766,25 +10462,23 @@ class basic_json */ template::value, int>::type = 0> - ValueType value(const json_pointer& ptr, ValueType default_value) const + ValueType value(const json_pointer& ptr, const ValueType& default_value) const { // at only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { // if pointer resolves a value, return it or use default value - try + JSON_TRY { return ptr.get_checked(this); } - catch (std::out_of_range&) + JSON_CATCH (out_of_range&) { return default_value; } } - else - { - throw std::domain_error("cannot use value() with " + type_name()); - } + + JSON_THROW(type_error::create(306, "cannot use value() with " + std::string(type_name()))); } /*! @@ -3813,7 +10507,7 @@ class basic_json assertions**). @post The JSON value remains unchanged. - @throw std::out_of_range when called on `null` value + @throw invalid_iterator.214 when called on `null` value @liveexample{The following code shows an example for `front()`.,front} @@ -3856,7 +10550,8 @@ class basic_json assertions**). @post The JSON value remains unchanged. - @throw std::out_of_range when called on `null` value. + @throw invalid_iterator.214 when called on a `null` value. See example + below. @liveexample{The following code shows an example for `back()`.,back} @@ -3900,17 +10595,18 @@ class basic_json @post Invalidates iterators and references at or after the point of the erase, including the `end()` iterator. - @throw std::domain_error if called on a `null` value; example: `"cannot - use erase() with null"` - @throw std::domain_error if called on an iterator which does not belong to - the current JSON value; example: `"iterator does not fit current value"` - @throw std::out_of_range if called on a primitive type with invalid + @throw type_error.307 if called on a `null` value; example: `"cannot use + erase() with null"` + @throw invalid_iterator.202 if called on an iterator which does not belong + to the current JSON value; example: `"iterator does not fit current + value"` + @throw invalid_iterator.205 if called on a primitive type with invalid iterator (i.e., any iterator which is not `begin()`); example: `"iterator out of range"` @complexity The complexity depends on the type: - objects: amortized constant - - arrays: linear in distance between pos and the end of the container + - arrays: linear in distance between @a pos and the end of the container - strings: linear in the length of the string - other types: constant @@ -3933,9 +10629,9 @@ class basic_json IteratorType erase(IteratorType pos) { // make sure iterator fits the current value - if (this != pos.m_object) + if (JSON_UNLIKELY(this != pos.m_object)) { - throw std::domain_error("iterator does not fit current value"); + JSON_THROW(invalid_iterator::create(202, "iterator does not fit current value")); } IteratorType result = end(); @@ -3948,9 +10644,9 @@ class basic_json case value_t::number_unsigned: case value_t::string: { - if (not pos.m_it.primitive_iterator.is_begin()) + if (JSON_UNLIKELY(not pos.m_it.primitive_iterator.is_begin())) { - throw std::out_of_range("iterator out of range"); + JSON_THROW(invalid_iterator::create(205, "iterator out of range")); } if (is_string()) @@ -3979,9 +10675,7 @@ class basic_json } default: - { - throw std::domain_error("cannot use erase() with " + type_name()); - } + JSON_THROW(type_error::create(307, "cannot use erase() with " + std::string(type_name()))); } return result; @@ -4007,11 +10701,11 @@ class basic_json @post Invalidates iterators and references at or after the point of the erase, including the `end()` iterator. - @throw std::domain_error if called on a `null` value; example: `"cannot - use erase() with null"` - @throw std::domain_error if called on iterators which does not belong to - the current JSON value; example: `"iterators do not fit current value"` - @throw std::out_of_range if called on a primitive type with invalid + @throw type_error.307 if called on a `null` value; example: `"cannot use + erase() with null"` + @throw invalid_iterator.203 if called on iterators which does not belong + to the current JSON value; example: `"iterators do not fit current value"` + @throw invalid_iterator.204 if called on a primitive type with invalid iterators (i.e., if `first != begin()` and `last != end()`); example: `"iterators out of range"` @@ -4040,9 +10734,9 @@ class basic_json IteratorType erase(IteratorType first, IteratorType last) { // make sure iterator fits the current value - if (this != first.m_object or this != last.m_object) + if (JSON_UNLIKELY(this != first.m_object or this != last.m_object)) { - throw std::domain_error("iterators do not fit current value"); + JSON_THROW(invalid_iterator::create(203, "iterators do not fit current value")); } IteratorType result = end(); @@ -4055,9 +10749,10 @@ class basic_json case value_t::number_unsigned: case value_t::string: { - if (not first.m_it.primitive_iterator.is_begin() or not last.m_it.primitive_iterator.is_end()) + if (JSON_LIKELY(not first.m_it.primitive_iterator.is_begin() + or not last.m_it.primitive_iterator.is_end())) { - throw std::out_of_range("iterators out of range"); + JSON_THROW(invalid_iterator::create(204, "iterators out of range")); } if (is_string()) @@ -4088,9 +10783,7 @@ class basic_json } default: - { - throw std::domain_error("cannot use erase() with " + type_name()); - } + JSON_THROW(type_error::create(307, "cannot use erase() with " + std::string(type_name()))); } return result; @@ -4110,7 +10803,7 @@ class basic_json @post References and iterators to the erased elements are invalidated. Other references and iterators are not affected. - @throw std::domain_error when called on a type other than JSON object; + @throw type_error.307 when called on a type other than JSON object; example: `"cannot use erase() with null"` @complexity `log(size()) + count(key)` @@ -4128,14 +10821,12 @@ class basic_json size_type erase(const typename object_t::key_type& key) { // this erase only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { return m_value.object->erase(key); } - else - { - throw std::domain_error("cannot use erase() with " + type_name()); - } + + JSON_THROW(type_error::create(307, "cannot use erase() with " + std::string(type_name()))); } /*! @@ -4145,9 +10836,9 @@ class basic_json @param[in] idx index of the element to remove - @throw std::domain_error when called on a type other than JSON array; + @throw type_error.307 when called on a type other than JSON object; example: `"cannot use erase() with null"` - @throw std::out_of_range when `idx >= size()`; example: `"array index 17 + @throw out_of_range.401 when `idx >= size()`; example: `"array index 17 is out of range"` @complexity Linear in distance between @a idx and the end of the container. @@ -4165,18 +10856,18 @@ class basic_json void erase(const size_type idx) { // this erase only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { - if (idx >= size()) + if (JSON_UNLIKELY(idx >= size())) { - throw std::out_of_range("array index " + std::to_string(idx) + " is out of range"); + JSON_THROW(out_of_range::create(401, "array index " + std::to_string(idx) + " is out of range")); } m_value.array->erase(m_value.array->begin() + static_cast(idx)); } else { - throw std::domain_error("cannot use erase() with " + type_name()); + JSON_THROW(type_error::create(307, "cannot use erase() with " + std::string(type_name()))); } } @@ -4550,10 +11241,6 @@ class basic_json return const_reverse_iterator(cbegin()); } - private: - // forward declaration - template class iteration_proxy; - public: /*! @brief wrapper to access iterator member functions in range-based for @@ -4563,6 +11250,8 @@ class basic_json reference to the JSON values is returned, so there is no access to the underlying iterator. + @liveexample{The following code shows how the wrapper is used,iterator_wrapper} + @note The name of this function is not yet final and may change in the future. */ @@ -4590,9 +11279,9 @@ class basic_json /// @{ /*! - @brief checks whether the container is empty + @brief checks whether the container is empty. - Checks if a JSON value has no elements. + Checks if a JSON value has no elements (i.e. whether its @ref size is `0`). @return The return value depends on the different types and is defined as follows: @@ -4605,23 +11294,27 @@ class basic_json object | result of function `object_t::empty()` array | result of function `array_t::empty()` - @note This function does not return whether a string stored as JSON value - is empty - it returns whether the JSON container itself is empty which is - false in the case of a string. + @liveexample{The following code uses `empty()` to check if a JSON + object contains any elements.,empty} @complexity Constant, as long as @ref array_t and @ref object_t satisfy the Container concept; that is, their `empty()` functions have constant complexity. + @iterators No changes. + + @exceptionsafety No-throw guarantee: this function never throws exceptions. + + @note This function does not return whether a string stored as JSON value + is empty - it returns whether the JSON container itself is empty which is + false in the case of a string. + @requirement This function helps `basic_json` satisfying the [Container](http://en.cppreference.com/w/cpp/concept/Container) requirements: - The complexity is constant. - Has the semantics of `begin() == end()`. - @liveexample{The following code uses `empty()` to check if a JSON - object contains any elements.,empty} - @sa @ref size() -- returns the number of elements @since version 1.0.0 @@ -4672,23 +11365,27 @@ class basic_json object | result of function object_t::size() array | result of function array_t::size() - @note This function does not return the length of a string stored as JSON - value - it returns the number of elements in the JSON value which is 1 in - the case of a string. + @liveexample{The following code calls `size()` on the different value + types.,size} @complexity Constant, as long as @ref array_t and @ref object_t satisfy the Container concept; that is, their size() functions have constant complexity. + @iterators No changes. + + @exceptionsafety No-throw guarantee: this function never throws exceptions. + + @note This function does not return the length of a string stored as JSON + value - it returns the number of elements in the JSON value which is 1 in + the case of a string. + @requirement This function helps `basic_json` satisfying the [Container](http://en.cppreference.com/w/cpp/concept/Container) requirements: - The complexity is constant. - Has the semantics of `std::distance(begin(), end())`. - @liveexample{The following code calls `size()` on the different value - types.,size} - @sa @ref empty() -- checks whether the container is empty @sa @ref max_size() -- returns the maximal number of elements @@ -4742,10 +11439,17 @@ class basic_json object | result of function `object_t::max_size()` array | result of function `array_t::max_size()` + @liveexample{The following code calls `max_size()` on the different value + types. Note the output is implementation specific.,max_size} + @complexity Constant, as long as @ref array_t and @ref object_t satisfy the Container concept; that is, their `max_size()` functions have constant complexity. + @iterators No changes. + + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @requirement This function helps `basic_json` satisfying the [Container](http://en.cppreference.com/w/cpp/concept/Container) requirements: @@ -4753,9 +11457,6 @@ class basic_json - Has the semantics of returning `b.size()` where `b` is the largest possible JSON value. - @liveexample{The following code calls `max_size()` on the different value - types. Note the output is implementation specific.,max_size} - @sa @ref size() -- returns the number of elements @since version 1.0.0 @@ -4798,7 +11499,8 @@ class basic_json @brief clears the contents Clears the content of a JSON value and resets it to the default value as - if @ref basic_json(value_t) would have been called: + if @ref basic_json(value_t) would have been called with the current value + type from @ref type(): Value type | initial value ----------- | ------------- @@ -4809,11 +11511,24 @@ class basic_json object | `{}` array | `[]` - @complexity Linear in the size of the JSON value. + @post Has the same effect as calling + @code {.cpp} + *this = basic_json(type()); + @endcode @liveexample{The example below shows the effect of `clear()` to different JSON types.,clear} + @complexity Linear in the size of the JSON value. + + @iterators All iterators, pointers and references related to this container + are invalidated. + + @exceptionsafety No-throw guarantee: this function never throws exceptions. + + @sa @ref basic_json(value_t) -- constructor that creates an object with the + same value than calling `clear()` + @since version 1.0.0 */ void clear() noexcept @@ -4863,9 +11578,7 @@ class basic_json } default: - { break; - } } } @@ -4878,7 +11591,7 @@ class basic_json @param[in] val the value to add to the JSON array - @throw std::domain_error when called on a type other than JSON array or + @throw type_error.308 when called on a type other than JSON array or null; example: `"cannot use push_back() with number"` @complexity Amortized constant. @@ -4892,9 +11605,9 @@ class basic_json void push_back(basic_json&& val) { // push_back only works for null objects or arrays - if (not(is_null() or is_array())) + if (JSON_UNLIKELY(not(is_null() or is_array()))) { - throw std::domain_error("cannot use push_back() with " + type_name()); + JSON_THROW(type_error::create(308, "cannot use push_back() with " + std::string(type_name()))); } // transform null object into an array @@ -4928,9 +11641,9 @@ class basic_json void push_back(const basic_json& val) { // push_back only works for null objects or arrays - if (not(is_null() or is_array())) + if (JSON_UNLIKELY(not(is_null() or is_array()))) { - throw std::domain_error("cannot use push_back() with " + type_name()); + JSON_THROW(type_error::create(308, "cannot use push_back() with " + std::string(type_name()))); } // transform null object into an array @@ -4964,7 +11677,7 @@ class basic_json @param[in] val the value to add to the JSON object - @throw std::domain_error when called on a type other than JSON object or + @throw type_error.308 when called on a type other than JSON object or null; example: `"cannot use push_back() with number"` @complexity Logarithmic in the size of the container, O(log(`size()`)). @@ -4978,9 +11691,9 @@ class basic_json void push_back(const typename object_t::value_type& val) { // push_back only works for null objects or objects - if (not(is_null() or is_object())) + if (JSON_UNLIKELY(not(is_null() or is_object()))) { - throw std::domain_error("cannot use push_back() with " + type_name()); + JSON_THROW(type_error::create(308, "cannot use push_back() with " + std::string(type_name()))); } // transform null object into an object @@ -5018,7 +11731,7 @@ class basic_json @ref push_back(const typename object_t::value_type&). Otherwise, @a init is converted to a JSON value and added using @ref push_back(basic_json&&). - @param init an initializer list + @param[in] init an initializer list @complexity Linear in the size of the initializer list @a init. @@ -5030,12 +11743,13 @@ class basic_json @liveexample{The example shows how initializer lists are treated as objects when possible.,push_back__initializer_list} */ - void push_back(std::initializer_list init) + void push_back(initializer_list_t init) { - if (is_object() and init.size() == 2 and init.begin()->is_string()) + if (is_object() and init.size() == 2 and (*init.begin())->is_string()) { - const string_t key = *init.begin(); - push_back(typename object_t::value_type(key, *(init.begin() + 1))); + basic_json&& key = init.begin()->moved_or_copied(); + push_back(typename object_t::value_type( + std::move(key.get_ref()), (init.begin() + 1)->moved_or_copied())); } else { @@ -5045,9 +11759,9 @@ class basic_json /*! @brief add an object to an object - @copydoc push_back(std::initializer_list) + @copydoc push_back(initializer_list_t) */ - reference operator+=(std::initializer_list init) + reference operator+=(initializer_list_t init) { push_back(init); return *this; @@ -5063,7 +11777,7 @@ class basic_json @param[in] args arguments to forward to a constructor of @ref basic_json @tparam Args compatible types to create a @ref basic_json object - @throw std::domain_error when called on a type other than JSON array or + @throw type_error.311 when called on a type other than JSON array or null; example: `"cannot use emplace_back() with number"` @complexity Amortized constant. @@ -5078,9 +11792,9 @@ class basic_json void emplace_back(Args&& ... args) { // emplace_back only works for null objects or arrays - if (not(is_null() or is_array())) + if (JSON_UNLIKELY(not(is_null() or is_array()))) { - throw std::domain_error("cannot use emplace_back() with " + type_name()); + JSON_THROW(type_error::create(311, "cannot use emplace_back() with " + std::string(type_name()))); } // transform null object into an array @@ -5098,8 +11812,8 @@ class basic_json /*! @brief add an object to an object if key does not exist - Inserts a new element into a JSON object constructed in-place with the given - @a args if there is no element with the key in the container. If the + Inserts a new element into a JSON object constructed in-place with the + given @a args if there is no element with the key in the container. If the function is called on a JSON null value, an empty object is created before appending the value created from @a args. @@ -5110,7 +11824,7 @@ class basic_json already-existing element if no insertion happened, and a bool denoting whether the insertion took place. - @throw std::domain_error when called on a type other than JSON object or + @throw type_error.311 when called on a type other than JSON object or null; example: `"cannot use emplace() with number"` @complexity Logarithmic in the size of the container, O(log(`size()`)). @@ -5126,9 +11840,9 @@ class basic_json std::pair emplace(Args&& ... args) { // emplace only works for null objects or arrays - if (not(is_null() or is_object())) + if (JSON_UNLIKELY(not(is_null() or is_object()))) { - throw std::domain_error("cannot use emplace() with " + type_name()); + JSON_THROW(type_error::create(311, "cannot use emplace() with " + std::string(type_name()))); } // transform null object into an object @@ -5159,13 +11873,13 @@ class basic_json @param[in] val element to insert @return iterator pointing to the inserted @a val. - @throw std::domain_error if called on JSON values other than arrays; + @throw type_error.309 if called on JSON values other than arrays; example: `"cannot use insert() with string"` - @throw std::domain_error if @a pos is not an iterator of *this; example: - `"iterator does not fit current value"` + @throw invalid_iterator.202 if @a pos is not an iterator of *this; + example: `"iterator does not fit current value"` - @complexity Constant plus linear in the distance between pos and end of the - container. + @complexity Constant plus linear in the distance between @a pos and end of + the container. @liveexample{The example shows how `insert()` is used.,insert} @@ -5174,12 +11888,12 @@ class basic_json iterator insert(const_iterator pos, const basic_json& val) { // insert only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { // check if iterator pos fits to this JSON value - if (pos.m_object != this) + if (JSON_UNLIKELY(pos.m_object != this)) { - throw std::domain_error("iterator does not fit current value"); + JSON_THROW(invalid_iterator::create(202, "iterator does not fit current value")); } // insert to array and return iterator @@ -5187,10 +11901,8 @@ class basic_json result.m_it.array_iterator = m_value.array->insert(pos.m_it.array_iterator, val); return result; } - else - { - throw std::domain_error("cannot use insert() with " + type_name()); - } + + JSON_THROW(type_error::create(309, "cannot use insert() with " + std::string(type_name()))); } /*! @@ -5214,10 +11926,10 @@ class basic_json @return iterator pointing to the first element inserted, or @a pos if `cnt==0` - @throw std::domain_error if called on JSON values other than arrays; - example: `"cannot use insert() with string"` - @throw std::domain_error if @a pos is not an iterator of *this; example: - `"iterator does not fit current value"` + @throw type_error.309 if called on JSON values other than arrays; example: + `"cannot use insert() with string"` + @throw invalid_iterator.202 if @a pos is not an iterator of *this; + example: `"iterator does not fit current value"` @complexity Linear in @a cnt plus linear in the distance between @a pos and end of the container. @@ -5229,12 +11941,12 @@ class basic_json iterator insert(const_iterator pos, size_type cnt, const basic_json& val) { // insert only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { // check if iterator pos fits to this JSON value - if (pos.m_object != this) + if (JSON_UNLIKELY(pos.m_object != this)) { - throw std::domain_error("iterator does not fit current value"); + JSON_THROW(invalid_iterator::create(202, "iterator does not fit current value")); } // insert to array and return iterator @@ -5242,10 +11954,8 @@ class basic_json result.m_it.array_iterator = m_value.array->insert(pos.m_it.array_iterator, cnt, val); return result; } - else - { - throw std::domain_error("cannot use insert() with " + type_name()); - } + + JSON_THROW(type_error::create(309, "cannot use insert() with " + std::string(type_name()))); } /*! @@ -5258,13 +11968,13 @@ class basic_json @param[in] first begin of the range of elements to insert @param[in] last end of the range of elements to insert - @throw std::domain_error if called on JSON values other than arrays; - example: `"cannot use insert() with string"` - @throw std::domain_error if @a pos is not an iterator of *this; example: - `"iterator does not fit current value"` - @throw std::domain_error if @a first and @a last do not belong to the same - JSON value; example: `"iterators do not fit"` - @throw std::domain_error if @a first or @a last are iterators into + @throw type_error.309 if called on JSON values other than arrays; example: + `"cannot use insert() with string"` + @throw invalid_iterator.202 if @a pos is not an iterator of *this; + example: `"iterator does not fit current value"` + @throw invalid_iterator.210 if @a first and @a last do not belong to the + same JSON value; example: `"iterators do not fit"` + @throw invalid_iterator.211 if @a first or @a last are iterators into container for which insert is called; example: `"passed iterators may not belong to container"` @@ -5281,26 +11991,26 @@ class basic_json iterator insert(const_iterator pos, const_iterator first, const_iterator last) { // insert only works for arrays - if (not is_array()) + if (JSON_UNLIKELY(not is_array())) { - throw std::domain_error("cannot use insert() with " + type_name()); + JSON_THROW(type_error::create(309, "cannot use insert() with " + std::string(type_name()))); } // check if iterator pos fits to this JSON value - if (pos.m_object != this) + if (JSON_UNLIKELY(pos.m_object != this)) { - throw std::domain_error("iterator does not fit current value"); + JSON_THROW(invalid_iterator::create(202, "iterator does not fit current value")); } // check if range iterators belong to the same JSON object - if (first.m_object != last.m_object) + if (JSON_UNLIKELY(first.m_object != last.m_object)) { - throw std::domain_error("iterators do not fit"); + JSON_THROW(invalid_iterator::create(210, "iterators do not fit")); } - if (first.m_object == this or last.m_object == this) + if (JSON_UNLIKELY(first.m_object == this or last.m_object == this)) { - throw std::domain_error("passed iterators may not belong to container"); + JSON_THROW(invalid_iterator::create(211, "passed iterators may not belong to container")); } // insert to array and return iterator @@ -5321,10 +12031,10 @@ class basic_json the end() iterator @param[in] ilist initializer list to insert the values from - @throw std::domain_error if called on JSON values other than arrays; - example: `"cannot use insert() with string"` - @throw std::domain_error if @a pos is not an iterator of *this; example: - `"iterator does not fit current value"` + @throw type_error.309 if called on JSON values other than arrays; example: + `"cannot use insert() with string"` + @throw invalid_iterator.202 if @a pos is not an iterator of *this; + example: `"iterator does not fit current value"` @return iterator pointing to the first element inserted, or @a pos if `ilist` is empty @@ -5336,26 +12046,177 @@ class basic_json @since version 1.0.0 */ - iterator insert(const_iterator pos, std::initializer_list ilist) + iterator insert(const_iterator pos, initializer_list_t ilist) { // insert only works for arrays - if (not is_array()) + if (JSON_UNLIKELY(not is_array())) { - throw std::domain_error("cannot use insert() with " + type_name()); + JSON_THROW(type_error::create(309, "cannot use insert() with " + std::string(type_name()))); } // check if iterator pos fits to this JSON value - if (pos.m_object != this) + if (JSON_UNLIKELY(pos.m_object != this)) { - throw std::domain_error("iterator does not fit current value"); + JSON_THROW(invalid_iterator::create(202, "iterator does not fit current value")); } // insert to array and return iterator iterator result(this); - result.m_it.array_iterator = m_value.array->insert(pos.m_it.array_iterator, ilist); + result.m_it.array_iterator = m_value.array->insert(pos.m_it.array_iterator, ilist.begin(), ilist.end()); return result; } + /*! + @brief inserts elements + + Inserts elements from range `[first, last)`. + + @param[in] first begin of the range of elements to insert + @param[in] last end of the range of elements to insert + + @throw type_error.309 if called on JSON values other than objects; example: + `"cannot use insert() with string"` + @throw invalid_iterator.202 if iterator @a first or @a last does does not + point to an object; example: `"iterators first and last must point to + objects"` + @throw invalid_iterator.210 if @a first and @a last do not belong to the + same JSON value; example: `"iterators do not fit"` + + @complexity Logarithmic: `O(N*log(size() + N))`, where `N` is the number + of elements to insert. + + @liveexample{The example shows how `insert()` is used.,insert__range_object} + + @since version 3.0.0 + */ + void insert(const_iterator first, const_iterator last) + { + // insert only works for objects + if (JSON_UNLIKELY(not is_object())) + { + JSON_THROW(type_error::create(309, "cannot use insert() with " + std::string(type_name()))); + } + + // check if range iterators belong to the same JSON object + if (JSON_UNLIKELY(first.m_object != last.m_object)) + { + JSON_THROW(invalid_iterator::create(210, "iterators do not fit")); + } + + // passed iterators must belong to objects + if (JSON_UNLIKELY(not first.m_object->is_object() + or not last.m_object->is_object())) + { + JSON_THROW(invalid_iterator::create(202, "iterators first and last must point to objects")); + } + + m_value.object->insert(first.m_it.object_iterator, last.m_it.object_iterator); + } + + /*! + @brief updates a JSON object from another object, overwriting existing keys + + Inserts all values from JSON object @a j and overwrites existing keys. + + @param[in] j JSON object to read values from + + @throw type_error.312 if called on JSON values other than objects; example: + `"cannot use update() with string"` + + @complexity O(N*log(size() + N)), where N is the number of elements to + insert. + + @liveexample{The example shows how `update()` is used.,update} + + @sa https://docs.python.org/3.6/library/stdtypes.html#dict.update + + @since version 3.0.0 + */ + void update(const_reference j) + { + // implicitly convert null value to an empty object + if (is_null()) + { + m_type = value_t::object; + m_value.object = create(); + assert_invariant(); + } + + if (JSON_UNLIKELY(not is_object())) + { + JSON_THROW(type_error::create(312, "cannot use update() with " + std::string(type_name()))); + } + if (JSON_UNLIKELY(not j.is_object())) + { + JSON_THROW(type_error::create(312, "cannot use update() with " + std::string(j.type_name()))); + } + + for (auto it = j.begin(); it != j.end(); ++it) + { + m_value.object->operator[](it.key()) = it.value(); + } + } + + /*! + @brief updates a JSON object from another object, overwriting existing keys + + Inserts all values from from range `[first, last)` and overwrites existing + keys. + + @param[in] first begin of the range of elements to insert + @param[in] last end of the range of elements to insert + + @throw type_error.312 if called on JSON values other than objects; example: + `"cannot use update() with string"` + @throw invalid_iterator.202 if iterator @a first or @a last does does not + point to an object; example: `"iterators first and last must point to + objects"` + @throw invalid_iterator.210 if @a first and @a last do not belong to the + same JSON value; example: `"iterators do not fit"` + + @complexity O(N*log(size() + N)), where N is the number of elements to + insert. + + @liveexample{The example shows how `update()` is used__range.,update} + + @sa https://docs.python.org/3.6/library/stdtypes.html#dict.update + + @since version 3.0.0 + */ + void update(const_iterator first, const_iterator last) + { + // implicitly convert null value to an empty object + if (is_null()) + { + m_type = value_t::object; + m_value.object = create(); + assert_invariant(); + } + + if (JSON_UNLIKELY(not is_object())) + { + JSON_THROW(type_error::create(312, "cannot use update() with " + std::string(type_name()))); + } + + // check if range iterators belong to the same JSON object + if (JSON_UNLIKELY(first.m_object != last.m_object)) + { + JSON_THROW(invalid_iterator::create(210, "iterators do not fit")); + } + + // passed iterators must belong to objects + if (JSON_UNLIKELY(not first.m_object->is_object() + or not first.m_object->is_object())) + { + JSON_THROW(invalid_iterator::create(202, "iterators first and last must point to objects")); + } + + for (auto it = first; it != last; ++it) + { + m_value.object->operator[](it.key()) = it.value(); + } + } + /*! @brief exchanges the values @@ -5395,7 +12256,7 @@ class basic_json @param[in,out] other array to exchange the contents with - @throw std::domain_error when JSON value is not an array; example: `"cannot + @throw type_error.310 when JSON value is not an array; example: `"cannot use swap() with string"` @complexity Constant. @@ -5408,13 +12269,13 @@ class basic_json void swap(array_t& other) { // swap only works for arrays - if (is_array()) + if (JSON_LIKELY(is_array())) { std::swap(*(m_value.array), other); } else { - throw std::domain_error("cannot use swap() with " + type_name()); + JSON_THROW(type_error::create(310, "cannot use swap() with " + std::string(type_name()))); } } @@ -5428,7 +12289,7 @@ class basic_json @param[in,out] other object to exchange the contents with - @throw std::domain_error when JSON value is not an object; example: + @throw type_error.310 when JSON value is not an object; example: `"cannot use swap() with string"` @complexity Constant. @@ -5441,13 +12302,13 @@ class basic_json void swap(object_t& other) { // swap only works for objects - if (is_object()) + if (JSON_LIKELY(is_object())) { std::swap(*(m_value.object), other); } else { - throw std::domain_error("cannot use swap() with " + type_name()); + JSON_THROW(type_error::create(310, "cannot use swap() with " + std::string(type_name()))); } } @@ -5461,7 +12322,7 @@ class basic_json @param[in,out] other string to exchange the contents with - @throw std::domain_error when JSON value is not a string; example: `"cannot + @throw type_error.310 when JSON value is not a string; example: `"cannot use swap() with boolean"` @complexity Constant. @@ -5474,19 +12335,19 @@ class basic_json void swap(string_t& other) { // swap only works for strings - if (is_string()) + if (JSON_LIKELY(is_string())) { std::swap(*(m_value.string), other); } else { - throw std::domain_error("cannot use swap() with " + type_name()); + JSON_THROW(type_error::create(310, "cannot use swap() with " + std::string(type_name()))); } } /// @} - + public: ////////////////////////////////////////// // lexicographical comparison operators // ////////////////////////////////////////// @@ -5494,56 +12355,38 @@ class basic_json /// @name lexicographical comparison operators /// @{ - private: - /*! - @brief comparison operator for JSON types - - Returns an ordering that is similar to Python: - - order: null < boolean < number < object < array < string - - furthermore, each type is not smaller than itself - - @since version 1.0.0 - */ - friend bool operator<(const value_t lhs, const value_t rhs) noexcept - { - static constexpr std::array order = {{ - 0, // null - 3, // object - 4, // array - 5, // string - 1, // boolean - 2, // integer - 2, // unsigned - 2, // float - } - }; - - // discarded values are not comparable - if (lhs == value_t::discarded or rhs == value_t::discarded) - { - return false; - } - - return order[static_cast(lhs)] < order[static_cast(rhs)]; - } - - public: /*! @brief comparison: equal Compares two JSON values for equality according to the following rules: - Two JSON values are equal if (1) they are from the same type and (2) - their stored values are the same. + their stored values are the same according to their respective + `operator==`. - Integer and floating-point numbers are automatically converted before - comparison. Floating-point numbers are compared indirectly: two - floating-point numbers `f1` and `f2` are considered equal if neither - `f1 > f2` nor `f2 > f1` holds. + comparison. Note than two NaN values are always treated as unequal. - Two JSON null values are equal. + @note Floating-point inside JSON values numbers are compared with + `json::number_float_t::operator==` which is `double::operator==` by + default. To compare floating-point while respecting an epsilon, an alternative + [comparison function](https://github.com/mariokonrad/marnav/blob/master/src/marnav/math/floatingpoint.hpp#L34-#L39) + could be used, for instance + @code {.cpp} + template ::value, T>::type> + inline bool is_same(T a, T b, T epsilon = std::numeric_limits::epsilon()) noexcept + { + return std::abs(a - b) <= epsilon; + } + @endcode + + @note NaN values never compare equal to themselves or to other NaN values. + @param[in] lhs first JSON value to consider @param[in] rhs second JSON value to consider @return whether the values @a lhs and @a rhs are equal + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @complexity Linear. @liveexample{The example demonstrates comparing several JSON @@ -5561,66 +12404,56 @@ class basic_json switch (lhs_type) { case value_t::array: - { - return *lhs.m_value.array == *rhs.m_value.array; - } + return (*lhs.m_value.array == *rhs.m_value.array); + case value_t::object: - { - return *lhs.m_value.object == *rhs.m_value.object; - } + return (*lhs.m_value.object == *rhs.m_value.object); + case value_t::null: - { return true; - } + case value_t::string: - { - return *lhs.m_value.string == *rhs.m_value.string; - } + return (*lhs.m_value.string == *rhs.m_value.string); + case value_t::boolean: - { - return lhs.m_value.boolean == rhs.m_value.boolean; - } + return (lhs.m_value.boolean == rhs.m_value.boolean); + case value_t::number_integer: - { - return lhs.m_value.number_integer == rhs.m_value.number_integer; - } + return (lhs.m_value.number_integer == rhs.m_value.number_integer); + case value_t::number_unsigned: - { - return lhs.m_value.number_unsigned == rhs.m_value.number_unsigned; - } + return (lhs.m_value.number_unsigned == rhs.m_value.number_unsigned); + case value_t::number_float: - { - return lhs.m_value.number_float == rhs.m_value.number_float; - } + return (lhs.m_value.number_float == rhs.m_value.number_float); + default: - { return false; - } } } else if (lhs_type == value_t::number_integer and rhs_type == value_t::number_float) { - return static_cast(lhs.m_value.number_integer) == rhs.m_value.number_float; + return (static_cast(lhs.m_value.number_integer) == rhs.m_value.number_float); } else if (lhs_type == value_t::number_float and rhs_type == value_t::number_integer) { - return lhs.m_value.number_float == static_cast(rhs.m_value.number_integer); + return (lhs.m_value.number_float == static_cast(rhs.m_value.number_integer)); } else if (lhs_type == value_t::number_unsigned and rhs_type == value_t::number_float) { - return static_cast(lhs.m_value.number_unsigned) == rhs.m_value.number_float; + return (static_cast(lhs.m_value.number_unsigned) == rhs.m_value.number_float); } else if (lhs_type == value_t::number_float and rhs_type == value_t::number_unsigned) { - return lhs.m_value.number_float == static_cast(rhs.m_value.number_unsigned); + return (lhs.m_value.number_float == static_cast(rhs.m_value.number_unsigned)); } else if (lhs_type == value_t::number_unsigned and rhs_type == value_t::number_integer) { - return static_cast(lhs.m_value.number_unsigned) == rhs.m_value.number_integer; + return (static_cast(lhs.m_value.number_unsigned) == rhs.m_value.number_integer); } else if (lhs_type == value_t::number_integer and rhs_type == value_t::number_unsigned) { - return lhs.m_value.number_integer == static_cast(rhs.m_value.number_unsigned); + return (lhs.m_value.number_integer == static_cast(rhs.m_value.number_unsigned)); } return false; @@ -5628,34 +12461,24 @@ class basic_json /*! @brief comparison: equal - - The functions compares the given JSON value against a null pointer. As the - null pointer can be used to initialize a JSON value to null, a comparison - of JSON value @a v with a null pointer should be equivalent to call - `v.is_null()`. - - @param[in] v JSON value to consider - @return whether @a v is null - - @complexity Constant. - - @liveexample{The example compares several JSON types to the null pointer. - ,operator__equal__nullptr_t} - - @since version 1.0.0 + @copydoc operator==(const_reference, const_reference) */ - friend bool operator==(const_reference v, std::nullptr_t) noexcept + template::value, int>::type = 0> + friend bool operator==(const_reference lhs, const ScalarType rhs) noexcept { - return v.is_null(); + return (lhs == basic_json(rhs)); } /*! @brief comparison: equal - @copydoc operator==(const_reference, std::nullptr_t) + @copydoc operator==(const_reference, const_reference) */ - friend bool operator==(std::nullptr_t, const_reference v) noexcept + template::value, int>::type = 0> + friend bool operator==(const ScalarType lhs, const_reference rhs) noexcept { - return v.is_null(); + return (basic_json(lhs) == rhs); } /*! @@ -5669,6 +12492,8 @@ class basic_json @complexity Linear. + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @liveexample{The example demonstrates comparing several JSON types.,operator__notequal} @@ -5681,34 +12506,24 @@ class basic_json /*! @brief comparison: not equal - - The functions compares the given JSON value against a null pointer. As the - null pointer can be used to initialize a JSON value to null, a comparison - of JSON value @a v with a null pointer should be equivalent to call - `not v.is_null()`. - - @param[in] v JSON value to consider - @return whether @a v is not null - - @complexity Constant. - - @liveexample{The example compares several JSON types to the null pointer. - ,operator__notequal__nullptr_t} - - @since version 1.0.0 + @copydoc operator!=(const_reference, const_reference) */ - friend bool operator!=(const_reference v, std::nullptr_t) noexcept + template::value, int>::type = 0> + friend bool operator!=(const_reference lhs, const ScalarType rhs) noexcept { - return not v.is_null(); + return (lhs != basic_json(rhs)); } /*! @brief comparison: not equal - @copydoc operator!=(const_reference, std::nullptr_t) + @copydoc operator!=(const_reference, const_reference) */ - friend bool operator!=(std::nullptr_t, const_reference v) noexcept + template::value, int>::type = 0> + friend bool operator!=(const ScalarType lhs, const_reference rhs) noexcept { - return not v.is_null(); + return (basic_json(lhs) != rhs); } /*! @@ -5730,6 +12545,8 @@ class basic_json @complexity Linear. + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @liveexample{The example demonstrates comparing several JSON types.,operator__less} @@ -5745,41 +12562,31 @@ class basic_json switch (lhs_type) { case value_t::array: - { - return *lhs.m_value.array < *rhs.m_value.array; - } + return (*lhs.m_value.array) < (*rhs.m_value.array); + case value_t::object: - { return *lhs.m_value.object < *rhs.m_value.object; - } + case value_t::null: - { return false; - } + case value_t::string: - { return *lhs.m_value.string < *rhs.m_value.string; - } + case value_t::boolean: - { return lhs.m_value.boolean < rhs.m_value.boolean; - } + case value_t::number_integer: - { return lhs.m_value.number_integer < rhs.m_value.number_integer; - } + case value_t::number_unsigned: - { return lhs.m_value.number_unsigned < rhs.m_value.number_unsigned; - } + case value_t::number_float: - { return lhs.m_value.number_float < rhs.m_value.number_float; - } + default: - { return false; - } } } else if (lhs_type == value_t::number_integer and rhs_type == value_t::number_float) @@ -5813,6 +12620,28 @@ class basic_json return operator<(lhs_type, rhs_type); } + /*! + @brief comparison: less than + @copydoc operator<(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator<(const_reference lhs, const ScalarType rhs) noexcept + { + return (lhs < basic_json(rhs)); + } + + /*! + @brief comparison: less than + @copydoc operator<(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator<(const ScalarType lhs, const_reference rhs) noexcept + { + return (basic_json(lhs) < rhs); + } + /*! @brief comparison: less than or equal @@ -5825,6 +12654,8 @@ class basic_json @complexity Linear. + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @liveexample{The example demonstrates comparing several JSON types.,operator__greater} @@ -5835,6 +12666,28 @@ class basic_json return not (rhs < lhs); } + /*! + @brief comparison: less than or equal + @copydoc operator<=(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator<=(const_reference lhs, const ScalarType rhs) noexcept + { + return (lhs <= basic_json(rhs)); + } + + /*! + @brief comparison: less than or equal + @copydoc operator<=(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator<=(const ScalarType lhs, const_reference rhs) noexcept + { + return (basic_json(lhs) <= rhs); + } + /*! @brief comparison: greater than @@ -5847,6 +12700,8 @@ class basic_json @complexity Linear. + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @liveexample{The example demonstrates comparing several JSON types.,operator__lessequal} @@ -5857,6 +12712,28 @@ class basic_json return not (lhs <= rhs); } + /*! + @brief comparison: greater than + @copydoc operator>(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator>(const_reference lhs, const ScalarType rhs) noexcept + { + return (lhs > basic_json(rhs)); + } + + /*! + @brief comparison: greater than + @copydoc operator>(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator>(const ScalarType lhs, const_reference rhs) noexcept + { + return (basic_json(lhs) > rhs); + } + /*! @brief comparison: greater than or equal @@ -5869,6 +12746,8 @@ class basic_json @complexity Linear. + @exceptionsafety No-throw guarantee: this function never throws exceptions. + @liveexample{The example demonstrates comparing several JSON types.,operator__greaterequal} @@ -5879,8 +12758,29 @@ class basic_json return not (lhs < rhs); } - /// @} + /*! + @brief comparison: greater than or equal + @copydoc operator>=(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator>=(const_reference lhs, const ScalarType rhs) noexcept + { + return (lhs >= basic_json(rhs)); + } + /*! + @brief comparison: greater than or equal + @copydoc operator>=(const_reference, const_reference) + */ + template::value, int>::type = 0> + friend bool operator>=(const ScalarType lhs, const_reference rhs) noexcept + { + return (basic_json(lhs) >= rhs); + } + + /// @} /////////////////// // serialization // @@ -5893,15 +12793,17 @@ class basic_json @brief serialize to stream Serialize the given JSON value @a j to the output stream @a o. The JSON - value will be serialized using the @ref dump member function. The - indentation of the output can be controlled with the member variable - `width` of the output stream @a o. For instance, using the manipulator - `std::setw(4)` on @a o sets the indentation level to `4` and the - serialization result is the same as calling `dump(4)`. + value will be serialized using the @ref dump member function. - @note During serializaion, the locale and the precision of the output - stream @a o are changed. The original values are restored when the - function returns. + - The indentation of the output can be controlled with the member variable + `width` of the output stream @a o. For instance, using the manipulator + `std::setw(4)` on @a o sets the indentation level to `4` and the + serialization result is the same as calling `dump(4)`. + + - The indentation characrer can be controlled with the member variable + `fill` of the output stream @a o. For instance, the manipulator + `std::setfill('\\t')` sets indentation to use a tab character rather than + the default space character. @param[in,out] o stream to serialize to @param[in] j JSON value to serialize @@ -5913,7 +12815,7 @@ class basic_json @liveexample{The example below shows the serialization with different parameters to `width` to adjust the indentation level.,operator_serialize} - @since version 1.0.0 + @since version 1.0.0; indentaction character added in version 3.0.0 */ friend std::ostream& operator<<(std::ostream& o, const basic_json& j) { @@ -5924,29 +12826,20 @@ class basic_json // reset width to 0 for subsequent calls to this stream o.width(0); - // fix locale problems - const auto old_locale = o.imbue(std::locale::classic()); - // set precision - - // 6, 15 or 16 digits of precision allows round-trip IEEE 754 - // string->float->string, string->double->string or string->long - // double->string; to be safe, we read this value from - // std::numeric_limits::digits10 - const auto old_precision = o.precision(std::numeric_limits::digits10); - // do the actual serialization - j.dump(o, pretty_print, static_cast(indentation)); - - // reset locale and precision - o.imbue(old_locale); - o.precision(old_precision); + serializer s(detail::output_adapter(o), o.fill()); + s.dump(j, pretty_print, false, static_cast(indentation)); return o; } /*! @brief serialize to stream - @copydoc operator<<(std::ostream&, const basic_json&) + @deprecated This stream operator is deprecated and will be removed in a + future version of the library. Please use + @ref operator<<(std::ostream&, const basic_json&) + instead; that is, replace calls like `j >> o;` with `o << j;`. */ + JSON_DEPRECATED friend std::ostream& operator>>(const basic_json& j, std::ostream& o) { return o << j; @@ -5963,21 +12856,47 @@ class basic_json /// @{ /*! - @brief deserialize from an array + @brief deserialize from a compatible input - This function reads from an array of 1-byte values. + This function reads from a compatible input. Examples are: + - an array of 1-byte values + - strings with character/literal type with size of 1 byte + - input streams + - container with contiguous storage of 1-byte values. Compatible container + types include `std::vector`, `std::string`, `std::array`, + `std::valarray`, and `std::initializer_list`. Furthermore, C-style + arrays can be used with `std::begin()`/`std::end()`. User-defined + containers can be used as long as they implement random-access iterators + and a contiguous storage. @pre Each element of the container has a size of 1 byte. Violating this precondition yields undefined behavior. **This precondition is enforced with a static assertion.** - @param[in] array array to read from + @pre The container storage is contiguous. Violating this precondition + yields undefined behavior. **This precondition is enforced with an + assertion.** + @pre Each element of the container has a size of 1 byte. Violating this + precondition yields undefined behavior. **This precondition is enforced + with a static assertion.** + + @warning There is no way to enforce all preconditions at compile-time. If + the function is called with a noncompliant container and with + assertions switched off, the behavior is undefined and will most + likely yield segmentation violation. + + @param[in] i input to read from @param[in] cb a parser callback function of type @ref parser_callback_t which is used to control the deserialization by filtering unwanted values (optional) @return result of the deserialization + @throw parse_error.101 if a parse error occurs; example: `""unexpected end + of input; expected string literal""` + @throw parse_error.102 if to_unicode fails or surrogate error + @throw parse_error.103 if to_unicode fails + @complexity Linear in the length of the input. The parser is a predictive LL(1) parser. The complexity can be higher if the parser callback function @a cb has a super-linear complexity. @@ -5987,90 +12906,46 @@ class basic_json @liveexample{The example below demonstrates the `parse()` function reading from an array.,parse__array__parser_callback_t} - @since version 2.0.3 - */ - template - static basic_json parse(T (&array)[N], - const parser_callback_t cb = nullptr) - { - // delegate the call to the iterator-range parse overload - return parse(std::begin(array), std::end(array), cb); - } - - /*! - @brief deserialize from string literal - - @tparam CharT character/literal type with size of 1 byte - @param[in] s string literal to read a serialized JSON value from - @param[in] cb a parser callback function of type @ref parser_callback_t - which is used to control the deserialization by filtering unwanted values - (optional) - - @return result of the deserialization - - @complexity Linear in the length of the input. The parser is a predictive - LL(1) parser. The complexity can be higher if the parser callback function - @a cb has a super-linear complexity. - - @note A UTF-8 byte order mark is silently ignored. - @note String containers like `std::string` or @ref string_t can be parsed - with @ref parse(const ContiguousContainer&, const parser_callback_t) - @liveexample{The example below demonstrates the `parse()` function with and without callback function.,parse__string__parser_callback_t} - @sa @ref parse(std::istream&, const parser_callback_t) for a version that - reads from an input stream - - @since version 1.0.0 (originally for @ref string_t) - */ - template::value and - std::is_integral::type>::value and - sizeof(typename std::remove_pointer::type) == 1, int>::type = 0> - static basic_json parse(const CharT s, - const parser_callback_t cb = nullptr) - { - return parser(reinterpret_cast(s), cb).parse(); - } - - /*! - @brief deserialize from stream - - @param[in,out] i stream to read a serialized JSON value from - @param[in] cb a parser callback function of type @ref parser_callback_t - which is used to control the deserialization by filtering unwanted values - (optional) - - @return result of the deserialization - - @complexity Linear in the length of the input. The parser is a predictive - LL(1) parser. The complexity can be higher if the parser callback function - @a cb has a super-linear complexity. - - @note A UTF-8 byte order mark is silently ignored. - @liveexample{The example below demonstrates the `parse()` function with and without callback function.,parse__istream__parser_callback_t} - @sa @ref parse(const CharT, const parser_callback_t) for a version - that reads from a string + @liveexample{The example below demonstrates the `parse()` function reading + from a contiguous container.,parse__contiguouscontainer__parser_callback_t} - @since version 1.0.0 + @since version 2.0.3 (contiguous containers) */ - static basic_json parse(std::istream& i, - const parser_callback_t cb = nullptr) + static basic_json parse(detail::input_adapter i, + const parser_callback_t cb = nullptr, + const bool allow_exceptions = true) { - return parser(i, cb).parse(); + basic_json result; + parser(i, cb, allow_exceptions).parse(true, result); + return result; } /*! - @copydoc parse(std::istream&, const parser_callback_t) + @copydoc basic_json parse(detail::input_adapter, const parser_callback_t) */ - static basic_json parse(std::istream&& i, - const parser_callback_t cb = nullptr) + static basic_json parse(detail::input_adapter& i, + const parser_callback_t cb = nullptr, + const bool allow_exceptions = true) { - return parser(i, cb).parse(); + basic_json result; + parser(i, cb, allow_exceptions).parse(true, result); + return result; + } + + static bool accept(detail::input_adapter i) + { + return parser(i).accept(true); + } + + static bool accept(detail::input_adapter& i) + { + return parser(i).accept(true); } /*! @@ -6103,6 +12978,10 @@ class basic_json @return result of the deserialization + @throw parse_error.101 in case of an unexpected token + @throw parse_error.102 if to_unicode fails or surrogate error + @throw parse_error.103 if to_unicode fails + @complexity Linear in the length of the input. The parser is a predictive LL(1) parser. The complexity can be higher if the parser callback function @a cb has a super-linear complexity. @@ -6119,82 +12998,34 @@ class basic_json std::random_access_iterator_tag, typename std::iterator_traits::iterator_category>::value, int>::type = 0> static basic_json parse(IteratorType first, IteratorType last, - const parser_callback_t cb = nullptr) + const parser_callback_t cb = nullptr, + const bool allow_exceptions = true) { - // assertion to check that the iterator range is indeed contiguous, - // see http://stackoverflow.com/a/35008842/266378 for more discussion - assert(std::accumulate(first, last, std::make_pair(true, 0), - [&first](std::pair res, decltype(*first) val) - { - res.first &= (val == *(std::next(std::addressof(*first), res.second++))); - return res; - }).first); + basic_json result; + parser(detail::input_adapter(first, last), cb, allow_exceptions).parse(true, result); + return result; + } - // assertion to check that each element is 1 byte long - static_assert(sizeof(typename std::iterator_traits::value_type) == 1, - "each element in the iterator range must have the size of 1 byte"); - - // if iterator range is empty, create a parser with an empty string - // to generate "unexpected EOF" error message - if (std::distance(first, last) <= 0) - { - return parser("").parse(); - } - - return parser(first, last, cb).parse(); + template::iterator_category>::value, int>::type = 0> + static bool accept(IteratorType first, IteratorType last) + { + return parser(detail::input_adapter(first, last)).accept(true); } /*! - @brief deserialize from a container with contiguous storage - - This function reads from a container with contiguous storage of 1-byte - values. Compatible container types include `std::vector`, `std::string`, - `std::array`, and `std::initializer_list`. User-defined containers can be - used as long as they implement random-access iterators and a contiguous - storage. - - @pre The container storage is contiguous. Violating this precondition - yields undefined behavior. **This precondition is enforced with an - assertion.** - @pre Each element of the container has a size of 1 byte. Violating this - precondition yields undefined behavior. **This precondition is enforced - with a static assertion.** - - @warning There is no way to enforce all preconditions at compile-time. If - the function is called with a noncompliant container and with - assertions switched off, the behavior is undefined and will most - likely yield segmentation violation. - - @tparam ContiguousContainer container type with contiguous storage - @param[in] c container to read from - @param[in] cb a parser callback function of type @ref parser_callback_t - which is used to control the deserialization by filtering unwanted values - (optional) - - @return result of the deserialization - - @complexity Linear in the length of the input. The parser is a predictive - LL(1) parser. The complexity can be higher if the parser callback function - @a cb has a super-linear complexity. - - @note A UTF-8 byte order mark is silently ignored. - - @liveexample{The example below demonstrates the `parse()` function reading - from a contiguous container.,parse__contiguouscontainer__parser_callback_t} - - @since version 2.0.3 + @brief deserialize from stream + @deprecated This stream operator is deprecated and will be removed in a + future version of the library. Please use + @ref operator>>(std::istream&, basic_json&) + instead; that is, replace calls like `j << i;` with `i >> j;`. */ - template::value and - std::is_base_of< - std::random_access_iterator_tag, - typename std::iterator_traits()))>::iterator_category>::value - , int>::type = 0> - static basic_json parse(const ContiguousContainer& c, - const parser_callback_t cb = nullptr) + JSON_DEPRECATED + friend std::istream& operator<<(basic_json& j, std::istream& i) { - // delegate the call to the iterator-range parse overload - return parse(std::begin(c), std::end(c), cb); + return operator>>(i, j); } /*! @@ -6205,7 +13036,9 @@ class basic_json @param[in,out] i input stream to read a serialized JSON value from @param[in,out] j JSON value to write the deserialized input to - @throw std::invalid_argument in case of parse errors + @throw parse_error.101 in case of an unexpected token + @throw parse_error.102 if to_unicode fails or surrogate error + @throw parse_error.103 if to_unicode fails @complexity Linear in the length of the input. The parser is a predictive LL(1) parser. @@ -6220,1516 +13053,14 @@ class basic_json @since version 1.0.0 */ - friend std::istream& operator<<(basic_json& j, std::istream& i) - { - j = parser(i).parse(); - return i; - } - - /*! - @brief deserialize from stream - @copydoc operator<<(basic_json&, std::istream&) - */ friend std::istream& operator>>(std::istream& i, basic_json& j) { - j = parser(i).parse(); + parser(detail::input_adapter(i)).parse(false, j); return i; } /// @} - ////////////////////////////////////////// - // binary serialization/deserialization // - ////////////////////////////////////////// - - /// @name binary serialization/deserialization support - /// @{ - - private: - template - static void add_to_vector(std::vector& vec, size_t bytes, const T number) - { - assert(bytes == 1 or bytes == 2 or bytes == 4 or bytes == 8); - - switch (bytes) - { - case 8: - { - vec.push_back(static_cast((number >> 070) & 0xff)); - vec.push_back(static_cast((number >> 060) & 0xff)); - vec.push_back(static_cast((number >> 050) & 0xff)); - vec.push_back(static_cast((number >> 040) & 0xff)); - // intentional fall-through - } - - case 4: - { - vec.push_back(static_cast((number >> 030) & 0xff)); - vec.push_back(static_cast((number >> 020) & 0xff)); - // intentional fall-through - } - - case 2: - { - vec.push_back(static_cast((number >> 010) & 0xff)); - // intentional fall-through - } - - case 1: - { - vec.push_back(static_cast(number & 0xff)); - break; - } - } - } - - /*! - @brief take sufficient bytes from a vector to fill an integer variable - - In the context of binary serialization formats, we need to read several - bytes from a byte vector and combine them to multi-byte integral data - types. - - @param[in] vec byte vector to read from - @param[in] current_index the position in the vector after which to read - - @return the next sizeof(T) bytes from @a vec, in reverse order as T - - @tparam T the integral return type - - @throw std::out_of_range if there are less than sizeof(T)+1 bytes in the - vector @a vec to read - - In the for loop, the bytes from the vector are copied in reverse order into - the return value. In the figures below, let sizeof(T)=4 and `i` be the loop - variable. - - Precondition: - - vec: | | | a | b | c | d | T: | | | | | - ^ ^ ^ ^ - current_index i ptr sizeof(T) - - Postcondition: - - vec: | | | a | b | c | d | T: | d | c | b | a | - ^ ^ ^ - | i ptr - current_index - - @sa Code adapted from . - */ - template - static T get_from_vector(const std::vector& vec, const size_t current_index) - { - if (current_index + sizeof(T) + 1 > vec.size()) - { - throw std::out_of_range("cannot read " + std::to_string(sizeof(T)) + " bytes from vector"); - } - - T result; - uint8_t* ptr = reinterpret_cast(&result); - for (size_t i = 0; i < sizeof(T); ++i) - { - *ptr++ = vec[current_index + sizeof(T) - i]; - } - return result; - } - - /*! - @brief create a MessagePack serialization of a given JSON value - - This is a straightforward implementation of the MessagePack specification. - - @param[in] j JSON value to serialize - @param[in,out] v byte vector to write the serialization to - - @sa https://github.com/msgpack/msgpack/blob/master/spec.md - */ - static void to_msgpack_internal(const basic_json& j, std::vector& v) - { - switch (j.type()) - { - case value_t::null: - { - // nil - v.push_back(0xc0); - break; - } - - case value_t::boolean: - { - // true and false - v.push_back(j.m_value.boolean ? 0xc3 : 0xc2); - break; - } - - case value_t::number_integer: - { - if (j.m_value.number_integer >= 0) - { - // MessagePack does not differentiate between positive - // signed integers and unsigned integers. Therefore, we used - // the code from the value_t::number_unsigned case here. - if (j.m_value.number_unsigned < 128) - { - // positive fixnum - add_to_vector(v, 1, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT8_MAX) - { - // uint 8 - v.push_back(0xcc); - add_to_vector(v, 1, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT16_MAX) - { - // uint 16 - v.push_back(0xcd); - add_to_vector(v, 2, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT32_MAX) - { - // uint 32 - v.push_back(0xce); - add_to_vector(v, 4, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT64_MAX) - { - // uint 64 - v.push_back(0xcf); - add_to_vector(v, 8, j.m_value.number_unsigned); - } - } - else - { - if (j.m_value.number_integer >= -32) - { - // negative fixnum - add_to_vector(v, 1, j.m_value.number_integer); - } - else if (j.m_value.number_integer >= INT8_MIN and j.m_value.number_integer <= INT8_MAX) - { - // int 8 - v.push_back(0xd0); - add_to_vector(v, 1, j.m_value.number_integer); - } - else if (j.m_value.number_integer >= INT16_MIN and j.m_value.number_integer <= INT16_MAX) - { - // int 16 - v.push_back(0xd1); - add_to_vector(v, 2, j.m_value.number_integer); - } - else if (j.m_value.number_integer >= INT32_MIN and j.m_value.number_integer <= INT32_MAX) - { - // int 32 - v.push_back(0xd2); - add_to_vector(v, 4, j.m_value.number_integer); - } - else if (j.m_value.number_integer >= INT64_MIN and j.m_value.number_integer <= INT64_MAX) - { - // int 64 - v.push_back(0xd3); - add_to_vector(v, 8, j.m_value.number_integer); - } - } - break; - } - - case value_t::number_unsigned: - { - if (j.m_value.number_unsigned < 128) - { - // positive fixnum - add_to_vector(v, 1, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT8_MAX) - { - // uint 8 - v.push_back(0xcc); - add_to_vector(v, 1, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT16_MAX) - { - // uint 16 - v.push_back(0xcd); - add_to_vector(v, 2, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT32_MAX) - { - // uint 32 - v.push_back(0xce); - add_to_vector(v, 4, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= UINT64_MAX) - { - // uint 64 - v.push_back(0xcf); - add_to_vector(v, 8, j.m_value.number_unsigned); - } - break; - } - - case value_t::number_float: - { - // float 64 - v.push_back(0xcb); - const uint8_t* helper = reinterpret_cast(&(j.m_value.number_float)); - for (size_t i = 0; i < 8; ++i) - { - v.push_back(helper[7 - i]); - } - break; - } - - case value_t::string: - { - const auto N = j.m_value.string->size(); - if (N <= 31) - { - // fixstr - v.push_back(static_cast(0xa0 | N)); - } - else if (N <= 255) - { - // str 8 - v.push_back(0xd9); - add_to_vector(v, 1, N); - } - else if (N <= 65535) - { - // str 16 - v.push_back(0xda); - add_to_vector(v, 2, N); - } - else if (N <= 4294967295) - { - // str 32 - v.push_back(0xdb); - add_to_vector(v, 4, N); - } - - // append string - std::copy(j.m_value.string->begin(), j.m_value.string->end(), - std::back_inserter(v)); - break; - } - - case value_t::array: - { - const auto N = j.m_value.array->size(); - if (N <= 15) - { - // fixarray - v.push_back(static_cast(0x90 | N)); - } - else if (N <= 0xffff) - { - // array 16 - v.push_back(0xdc); - add_to_vector(v, 2, N); - } - else if (N <= 0xffffffff) - { - // array 32 - v.push_back(0xdd); - add_to_vector(v, 4, N); - } - - // append each element - for (const auto& el : *j.m_value.array) - { - to_msgpack_internal(el, v); - } - break; - } - - case value_t::object: - { - const auto N = j.m_value.object->size(); - if (N <= 15) - { - // fixmap - v.push_back(static_cast(0x80 | (N & 0xf))); - } - else if (N <= 65535) - { - // map 16 - v.push_back(0xde); - add_to_vector(v, 2, N); - } - else if (N <= 4294967295) - { - // map 32 - v.push_back(0xdf); - add_to_vector(v, 4, N); - } - - // append each element - for (const auto& el : *j.m_value.object) - { - to_msgpack_internal(el.first, v); - to_msgpack_internal(el.second, v); - } - break; - } - - default: - { - break; - } - } - } - - /*! - @brief create a CBOR serialization of a given JSON value - - This is a straightforward implementation of the CBOR specification. - - @param[in] j JSON value to serialize - @param[in,out] v byte vector to write the serialization to - - @sa https://tools.ietf.org/html/rfc7049 - */ - static void to_cbor_internal(const basic_json& j, std::vector& v) - { - switch (j.type()) - { - case value_t::null: - { - v.push_back(0xf6); - break; - } - - case value_t::boolean: - { - v.push_back(j.m_value.boolean ? 0xf5 : 0xf4); - break; - } - - case value_t::number_integer: - { - if (j.m_value.number_integer >= 0) - { - // CBOR does not differentiate between positive signed - // integers and unsigned integers. Therefore, we used the - // code from the value_t::number_unsigned case here. - if (j.m_value.number_integer <= 0x17) - { - add_to_vector(v, 1, j.m_value.number_integer); - } - else if (j.m_value.number_integer <= UINT8_MAX) - { - v.push_back(0x18); - // one-byte uint8_t - add_to_vector(v, 1, j.m_value.number_integer); - } - else if (j.m_value.number_integer <= UINT16_MAX) - { - v.push_back(0x19); - // two-byte uint16_t - add_to_vector(v, 2, j.m_value.number_integer); - } - else if (j.m_value.number_integer <= UINT32_MAX) - { - v.push_back(0x1a); - // four-byte uint32_t - add_to_vector(v, 4, j.m_value.number_integer); - } - else - { - v.push_back(0x1b); - // eight-byte uint64_t - add_to_vector(v, 8, j.m_value.number_integer); - } - } - else - { - // The conversions below encode the sign in the first byte, - // and the value is converted to a positive number. - const auto positive_number = -1 - j.m_value.number_integer; - if (j.m_value.number_integer >= -24) - { - v.push_back(static_cast(0x20 + positive_number)); - } - else if (positive_number <= UINT8_MAX) - { - // int 8 - v.push_back(0x38); - add_to_vector(v, 1, positive_number); - } - else if (positive_number <= UINT16_MAX) - { - // int 16 - v.push_back(0x39); - add_to_vector(v, 2, positive_number); - } - else if (positive_number <= UINT32_MAX) - { - // int 32 - v.push_back(0x3a); - add_to_vector(v, 4, positive_number); - } - else - { - // int 64 - v.push_back(0x3b); - add_to_vector(v, 8, positive_number); - } - } - break; - } - - case value_t::number_unsigned: - { - if (j.m_value.number_unsigned <= 0x17) - { - v.push_back(static_cast(j.m_value.number_unsigned)); - } - else if (j.m_value.number_unsigned <= 0xff) - { - v.push_back(0x18); - // one-byte uint8_t - add_to_vector(v, 1, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= 0xffff) - { - v.push_back(0x19); - // two-byte uint16_t - add_to_vector(v, 2, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= 0xffffffff) - { - v.push_back(0x1a); - // four-byte uint32_t - add_to_vector(v, 4, j.m_value.number_unsigned); - } - else if (j.m_value.number_unsigned <= 0xffffffffffffffff) - { - v.push_back(0x1b); - // eight-byte uint64_t - add_to_vector(v, 8, j.m_value.number_unsigned); - } - break; - } - - case value_t::number_float: - { - // Double-Precision Float - v.push_back(0xfb); - const uint8_t* helper = reinterpret_cast(&(j.m_value.number_float)); - for (size_t i = 0; i < 8; ++i) - { - v.push_back(helper[7 - i]); - } - break; - } - - case value_t::string: - { - const auto N = j.m_value.string->size(); - if (N <= 0x17) - { - v.push_back(0x60 + N); // 1 byte for string + size - } - else if (N <= 0xff) - { - v.push_back(0x78); // one-byte uint8_t for N - add_to_vector(v, 1, N); - } - else if (N <= 0xffff) - { - v.push_back(0x79); // two-byte uint16_t for N - add_to_vector(v, 2, N); - } - else if (N <= 0xffffffff) - { - v.push_back(0x7a); // four-byte uint32_t for N - add_to_vector(v, 4, N); - } - // LCOV_EXCL_START - else if (N <= 0xffffffffffffffff) - { - v.push_back(0x7b); // eight-byte uint64_t for N - add_to_vector(v, 8, N); - } - // LCOV_EXCL_STOP - - // append string - std::copy(j.m_value.string->begin(), j.m_value.string->end(), - std::back_inserter(v)); - break; - } - - case value_t::array: - { - const auto N = j.m_value.array->size(); - if (N <= 0x17) - { - v.push_back(0x80 + N); // 1 byte for array + size - } - else if (N <= 0xff) - { - v.push_back(0x98); // one-byte uint8_t for N - add_to_vector(v, 1, N); - } - else if (N <= 0xffff) - { - v.push_back(0x99); // two-byte uint16_t for N - add_to_vector(v, 2, N); - } - else if (N <= 0xffffffff) - { - v.push_back(0x9a); // four-byte uint32_t for N - add_to_vector(v, 4, N); - } - // LCOV_EXCL_START - else if (N <= 0xffffffffffffffff) - { - v.push_back(0x9b); // eight-byte uint64_t for N - add_to_vector(v, 8, N); - } - // LCOV_EXCL_STOP - - // append each element - for (const auto& el : *j.m_value.array) - { - to_cbor_internal(el, v); - } - break; - } - - case value_t::object: - { - const auto N = j.m_value.object->size(); - if (N <= 0x17) - { - v.push_back(0xa0 + N); // 1 byte for object + size - } - else if (N <= 0xff) - { - v.push_back(0xb8); - add_to_vector(v, 1, N); // one-byte uint8_t for N - } - else if (N <= 0xffff) - { - v.push_back(0xb9); - add_to_vector(v, 2, N); // two-byte uint16_t for N - } - else if (N <= 0xffffffff) - { - v.push_back(0xba); - add_to_vector(v, 4, N); // four-byte uint32_t for N - } - // LCOV_EXCL_START - else if (N <= 0xffffffffffffffff) - { - v.push_back(0xbb); - add_to_vector(v, 8, N); // eight-byte uint64_t for N - } - // LCOV_EXCL_STOP - - // append each element - for (const auto& el : *j.m_value.object) - { - to_cbor_internal(el.first, v); - to_cbor_internal(el.second, v); - } - break; - } - - default: - { - break; - } - } - } - - - /* - @brief checks if given lengths do not exceed the size of a given vector - - To secure the access to the byte vector during CBOR/MessagePack - deserialization, bytes are copied from the vector into buffers. This - function checks if the number of bytes to copy (@a len) does not exceed the - size @s size of the vector. Additionally, an @a offset is given from where - to start reading the bytes. - - This function checks whether reading the bytes is safe; that is, offset is a - valid index in the vector, offset+len - - @param[in] size size of the byte vector - @param[in] len number of bytes to read - @param[in] offset offset where to start reading - - vec: x x x x x X X X X X - ^ ^ ^ - 0 offset len - - @throws out_of_range if `len > v.size()` - */ - static void check_length(const size_t size, const size_t len, const size_t offset) - { - // simple case: requested length is greater than the vector's length - if (len > size or offset > size) - { - throw std::out_of_range("len out of range"); - } - - // second case: adding offset would result in overflow - if ((size > (std::numeric_limits::max() - offset))) - { - throw std::out_of_range("len+offset out of range"); - } - - // last case: reading past the end of the vector - if (len + offset > size) - { - throw std::out_of_range("len+offset out of range"); - } - } - - /*! - @brief create a JSON value from a given MessagePack vector - - @param[in] v MessagePack serialization - @param[in] idx byte index to start reading from @a v - - @return deserialized JSON value - - @throw std::invalid_argument if unsupported features from MessagePack were - used in the given vector @a v or if the input is not valid MessagePack - @throw std::out_of_range if the given vector ends prematurely - - @sa https://github.com/msgpack/msgpack/blob/master/spec.md - */ - static basic_json from_msgpack_internal(const std::vector& v, size_t& idx) - { - // make sure reading 1 byte is safe - check_length(v.size(), 1, idx); - - // store and increment index - const size_t current_idx = idx++; - - if (v[current_idx] <= 0xbf) - { - if (v[current_idx] <= 0x7f) // positive fixint - { - return v[current_idx]; - } - else if (v[current_idx] <= 0x8f) // fixmap - { - basic_json result = value_t::object; - const size_t len = v[current_idx] & 0x0f; - for (size_t i = 0; i < len; ++i) - { - std::string key = from_msgpack_internal(v, idx); - result[key] = from_msgpack_internal(v, idx); - } - return result; - } - else if (v[current_idx] <= 0x9f) // fixarray - { - basic_json result = value_t::array; - const size_t len = v[current_idx] & 0x0f; - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_msgpack_internal(v, idx)); - } - return result; - } - else // fixstr - { - const size_t len = v[current_idx] & 0x1f; - const size_t offset = current_idx + 1; - idx += len; // skip content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - } - else if (v[current_idx] >= 0xe0) // negative fixint - { - return static_cast(v[current_idx]); - } - else - { - switch (v[current_idx]) - { - case 0xc0: // nil - { - return value_t::null; - } - - case 0xc2: // false - { - return false; - } - - case 0xc3: // true - { - return true; - } - - case 0xca: // float 32 - { - // copy bytes in reverse order into the double variable - check_length(v.size(), sizeof(float), 1); - float res; - for (size_t byte = 0; byte < sizeof(float); ++byte) - { - reinterpret_cast(&res)[sizeof(float) - byte - 1] = v[current_idx + 1 + byte]; - } - idx += sizeof(float); // skip content bytes - return res; - } - - case 0xcb: // float 64 - { - // copy bytes in reverse order into the double variable - check_length(v.size(), sizeof(double), 1); - double res; - for (size_t byte = 0; byte < sizeof(double); ++byte) - { - reinterpret_cast(&res)[sizeof(double) - byte - 1] = v[current_idx + 1 + byte]; - } - idx += sizeof(double); // skip content bytes - return res; - } - - case 0xcc: // uint 8 - { - idx += 1; // skip content byte - return get_from_vector(v, current_idx); - } - - case 0xcd: // uint 16 - { - idx += 2; // skip 2 content bytes - return get_from_vector(v, current_idx); - } - - case 0xce: // uint 32 - { - idx += 4; // skip 4 content bytes - return get_from_vector(v, current_idx); - } - - case 0xcf: // uint 64 - { - idx += 8; // skip 8 content bytes - return get_from_vector(v, current_idx); - } - - case 0xd0: // int 8 - { - idx += 1; // skip content byte - return get_from_vector(v, current_idx); - } - - case 0xd1: // int 16 - { - idx += 2; // skip 2 content bytes - return get_from_vector(v, current_idx); - } - - case 0xd2: // int 32 - { - idx += 4; // skip 4 content bytes - return get_from_vector(v, current_idx); - } - - case 0xd3: // int 64 - { - idx += 8; // skip 8 content bytes - return get_from_vector(v, current_idx); - } - - case 0xd9: // str 8 - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 2; - idx += len + 1; // skip size byte + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0xda: // str 16 - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 3; - idx += len + 2; // skip 2 size bytes + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0xdb: // str 32 - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 5; - idx += len + 4; // skip 4 size bytes + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0xdc: // array 16 - { - basic_json result = value_t::array; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 2; // skip 2 size bytes - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_msgpack_internal(v, idx)); - } - return result; - } - - case 0xdd: // array 32 - { - basic_json result = value_t::array; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 4; // skip 4 size bytes - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_msgpack_internal(v, idx)); - } - return result; - } - - case 0xde: // map 16 - { - basic_json result = value_t::object; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 2; // skip 2 size bytes - for (size_t i = 0; i < len; ++i) - { - std::string key = from_msgpack_internal(v, idx); - result[key] = from_msgpack_internal(v, idx); - } - return result; - } - - case 0xdf: // map 32 - { - basic_json result = value_t::object; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 4; // skip 4 size bytes - for (size_t i = 0; i < len; ++i) - { - std::string key = from_msgpack_internal(v, idx); - result[key] = from_msgpack_internal(v, idx); - } - return result; - } - - default: - { - throw std::invalid_argument("error parsing a msgpack @ " + std::to_string(current_idx) + ": " + std::to_string(static_cast(v[current_idx]))); - } - } - } - } - - /*! - @brief create a JSON value from a given CBOR vector - - @param[in] v CBOR serialization - @param[in] idx byte index to start reading from @a v - - @return deserialized JSON value - - @throw std::invalid_argument if unsupported features from CBOR were used in - the given vector @a v or if the input is not valid CBOR - @throw std::out_of_range if the given vector ends prematurely - - @sa https://tools.ietf.org/html/rfc7049 - */ - static basic_json from_cbor_internal(const std::vector& v, size_t& idx) - { - // store and increment index - const size_t current_idx = idx++; - - switch (v.at(current_idx)) - { - // Integer 0x00..0x17 (0..23) - case 0x00: - case 0x01: - case 0x02: - case 0x03: - case 0x04: - case 0x05: - case 0x06: - case 0x07: - case 0x08: - case 0x09: - case 0x0a: - case 0x0b: - case 0x0c: - case 0x0d: - case 0x0e: - case 0x0f: - case 0x10: - case 0x11: - case 0x12: - case 0x13: - case 0x14: - case 0x15: - case 0x16: - case 0x17: - { - return v[current_idx]; - } - - case 0x18: // Unsigned integer (one-byte uint8_t follows) - { - idx += 1; // skip content byte - return get_from_vector(v, current_idx); - } - - case 0x19: // Unsigned integer (two-byte uint16_t follows) - { - idx += 2; // skip 2 content bytes - return get_from_vector(v, current_idx); - } - - case 0x1a: // Unsigned integer (four-byte uint32_t follows) - { - idx += 4; // skip 4 content bytes - return get_from_vector(v, current_idx); - } - - case 0x1b: // Unsigned integer (eight-byte uint64_t follows) - { - idx += 8; // skip 8 content bytes - return get_from_vector(v, current_idx); - } - - // Negative integer -1-0x00..-1-0x17 (-1..-24) - case 0x20: - case 0x21: - case 0x22: - case 0x23: - case 0x24: - case 0x25: - case 0x26: - case 0x27: - case 0x28: - case 0x29: - case 0x2a: - case 0x2b: - case 0x2c: - case 0x2d: - case 0x2e: - case 0x2f: - case 0x30: - case 0x31: - case 0x32: - case 0x33: - case 0x34: - case 0x35: - case 0x36: - case 0x37: - { - return static_cast(0x20 - 1 - v[current_idx]); - } - - case 0x38: // Negative integer (one-byte uint8_t follows) - { - idx += 1; // skip content byte - // must be uint8_t ! - return static_cast(-1) - get_from_vector(v, current_idx); - } - - case 0x39: // Negative integer -1-n (two-byte uint16_t follows) - { - idx += 2; // skip 2 content bytes - return static_cast(-1) - get_from_vector(v, current_idx); - } - - case 0x3a: // Negative integer -1-n (four-byte uint32_t follows) - { - idx += 4; // skip 4 content bytes - return static_cast(-1) - get_from_vector(v, current_idx); - } - - case 0x3b: // Negative integer -1-n (eight-byte uint64_t follows) - { - idx += 8; // skip 8 content bytes - return static_cast(-1) - static_cast(get_from_vector(v, current_idx)); - } - - // UTF-8 string (0x00..0x17 bytes follow) - case 0x60: - case 0x61: - case 0x62: - case 0x63: - case 0x64: - case 0x65: - case 0x66: - case 0x67: - case 0x68: - case 0x69: - case 0x6a: - case 0x6b: - case 0x6c: - case 0x6d: - case 0x6e: - case 0x6f: - case 0x70: - case 0x71: - case 0x72: - case 0x73: - case 0x74: - case 0x75: - case 0x76: - case 0x77: - { - const auto len = static_cast(v[current_idx] - 0x60); - const size_t offset = current_idx + 1; - idx += len; // skip content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0x78: // UTF-8 string (one-byte uint8_t for n follows) - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 2; - idx += len + 1; // skip size byte + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0x79: // UTF-8 string (two-byte uint16_t for n follow) - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 3; - idx += len + 2; // skip 2 size bytes + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0x7a: // UTF-8 string (four-byte uint32_t for n follow) - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 5; - idx += len + 4; // skip 4 size bytes + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0x7b: // UTF-8 string (eight-byte uint64_t for n follow) - { - const auto len = static_cast(get_from_vector(v, current_idx)); - const size_t offset = current_idx + 9; - idx += len + 8; // skip 8 size bytes + content bytes - check_length(v.size(), len, offset); - return std::string(reinterpret_cast(v.data()) + offset, len); - } - - case 0x7f: // UTF-8 string (indefinite length) - { - std::string result; - while (v.at(idx) != 0xff) - { - string_t s = from_cbor_internal(v, idx); - result += s; - } - // skip break byte (0xFF) - idx += 1; - return result; - } - - // array (0x00..0x17 data items follow) - case 0x80: - case 0x81: - case 0x82: - case 0x83: - case 0x84: - case 0x85: - case 0x86: - case 0x87: - case 0x88: - case 0x89: - case 0x8a: - case 0x8b: - case 0x8c: - case 0x8d: - case 0x8e: - case 0x8f: - case 0x90: - case 0x91: - case 0x92: - case 0x93: - case 0x94: - case 0x95: - case 0x96: - case 0x97: - { - basic_json result = value_t::array; - const auto len = static_cast(v[current_idx] - 0x80); - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_cbor_internal(v, idx)); - } - return result; - } - - case 0x98: // array (one-byte uint8_t for n follows) - { - basic_json result = value_t::array; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 1; // skip 1 size byte - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_cbor_internal(v, idx)); - } - return result; - } - - case 0x99: // array (two-byte uint16_t for n follow) - { - basic_json result = value_t::array; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 2; // skip 4 size bytes - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_cbor_internal(v, idx)); - } - return result; - } - - case 0x9a: // array (four-byte uint32_t for n follow) - { - basic_json result = value_t::array; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 4; // skip 4 size bytes - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_cbor_internal(v, idx)); - } - return result; - } - - case 0x9b: // array (eight-byte uint64_t for n follow) - { - basic_json result = value_t::array; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 8; // skip 8 size bytes - for (size_t i = 0; i < len; ++i) - { - result.push_back(from_cbor_internal(v, idx)); - } - return result; - } - - case 0x9f: // array (indefinite length) - { - basic_json result = value_t::array; - while (v.at(idx) != 0xff) - { - result.push_back(from_cbor_internal(v, idx)); - } - // skip break byte (0xFF) - idx += 1; - return result; - } - - // map (0x00..0x17 pairs of data items follow) - case 0xa0: - case 0xa1: - case 0xa2: - case 0xa3: - case 0xa4: - case 0xa5: - case 0xa6: - case 0xa7: - case 0xa8: - case 0xa9: - case 0xaa: - case 0xab: - case 0xac: - case 0xad: - case 0xae: - case 0xaf: - case 0xb0: - case 0xb1: - case 0xb2: - case 0xb3: - case 0xb4: - case 0xb5: - case 0xb6: - case 0xb7: - { - basic_json result = value_t::object; - const auto len = static_cast(v[current_idx] - 0xa0); - for (size_t i = 0; i < len; ++i) - { - std::string key = from_cbor_internal(v, idx); - result[key] = from_cbor_internal(v, idx); - } - return result; - } - - case 0xb8: // map (one-byte uint8_t for n follows) - { - basic_json result = value_t::object; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 1; // skip 1 size byte - for (size_t i = 0; i < len; ++i) - { - std::string key = from_cbor_internal(v, idx); - result[key] = from_cbor_internal(v, idx); - } - return result; - } - - case 0xb9: // map (two-byte uint16_t for n follow) - { - basic_json result = value_t::object; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 2; // skip 2 size bytes - for (size_t i = 0; i < len; ++i) - { - std::string key = from_cbor_internal(v, idx); - result[key] = from_cbor_internal(v, idx); - } - return result; - } - - case 0xba: // map (four-byte uint32_t for n follow) - { - basic_json result = value_t::object; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 4; // skip 4 size bytes - for (size_t i = 0; i < len; ++i) - { - std::string key = from_cbor_internal(v, idx); - result[key] = from_cbor_internal(v, idx); - } - return result; - } - - case 0xbb: // map (eight-byte uint64_t for n follow) - { - basic_json result = value_t::object; - const auto len = static_cast(get_from_vector(v, current_idx)); - idx += 8; // skip 8 size bytes - for (size_t i = 0; i < len; ++i) - { - std::string key = from_cbor_internal(v, idx); - result[key] = from_cbor_internal(v, idx); - } - return result; - } - - case 0xbf: // map (indefinite length) - { - basic_json result = value_t::object; - while (v.at(idx) != 0xff) - { - std::string key = from_cbor_internal(v, idx); - result[key] = from_cbor_internal(v, idx); - } - // skip break byte (0xFF) - idx += 1; - return result; - } - - case 0xf4: // false - { - return false; - } - - case 0xf5: // true - { - return true; - } - - case 0xf6: // null - { - return value_t::null; - } - - case 0xf9: // Half-Precision Float (two-byte IEEE 754) - { - check_length(v.size(), 2, 1); - idx += 2; // skip two content bytes - - // code from RFC 7049, Appendix D, Figure 3: - // As half-precision floating-point numbers were only added to - // IEEE 754 in 2008, today's programming platforms often still - // only have limited support for them. It is very easy to - // include at least decoding support for them even without such - // support. An example of a small decoder for half-precision - // floating-point numbers in the C language is shown in Fig. 3. - const int half = (v[current_idx + 1] << 8) + v[current_idx + 2]; - const int exp = (half >> 10) & 0x1f; - const int mant = half & 0x3ff; - double val; - if (exp == 0) - { - val = std::ldexp(mant, -24); - } - else if (exp != 31) - { - val = std::ldexp(mant + 1024, exp - 25); - } - else - { - val = mant == 0 ? INFINITY : NAN; - } - return half & 0x8000 ? -val : val; - } - - case 0xfa: // Single-Precision Float (four-byte IEEE 754) - { - // copy bytes in reverse order into the float variable - check_length(v.size(), sizeof(float), 1); - float res; - for (size_t byte = 0; byte < sizeof(float); ++byte) - { - reinterpret_cast(&res)[sizeof(float) - byte - 1] = v[current_idx + 1 + byte]; - } - idx += sizeof(float); // skip content bytes - return res; - } - - case 0xfb: // Double-Precision Float (eight-byte IEEE 754) - { - check_length(v.size(), sizeof(double), 1); - // copy bytes in reverse order into the double variable - double res; - for (size_t byte = 0; byte < sizeof(double); ++byte) - { - reinterpret_cast(&res)[sizeof(double) - byte - 1] = v[current_idx + 1 + byte]; - } - idx += sizeof(double); // skip content bytes - return res; - } - - default: // anything else (0xFF is handled inside the other types) - { - throw std::invalid_argument("error parsing a CBOR @ " + std::to_string(current_idx) + ": " + std::to_string(static_cast(v[current_idx]))); - } - } - } - - public: - /*! - @brief create a MessagePack serialization of a given JSON value - - Serializes a given JSON value @a j to a byte vector using the MessagePack - serialization format. MessagePack is a binary serialization format which - aims to be more compact than JSON itself, yet more efficient to parse. - - @param[in] j JSON value to serialize - @return MessagePack serialization as byte vector - - @complexity Linear in the size of the JSON value @a j. - - @liveexample{The example shows the serialization of a JSON value to a byte - vector in MessagePack format.,to_msgpack} - - @sa http://msgpack.org - @sa @ref from_msgpack(const std::vector&) for the analogous - deserialization - @sa @ref to_cbor(const basic_json& for the related CBOR format - */ - static std::vector to_msgpack(const basic_json& j) - { - std::vector result; - to_msgpack_internal(j, result); - return result; - } - - /*! - @brief create a JSON value from a byte vector in MessagePack format - - Deserializes a given byte vector @a v to a JSON value using the MessagePack - serialization format. - - @param[in] v a byte vector in MessagePack format - @return deserialized JSON value - - @throw std::invalid_argument if unsupported features from MessagePack were - used in the given vector @a v or if the input is not valid MessagePack - @throw std::out_of_range if the given vector ends prematurely - - @complexity Linear in the size of the byte vector @a v. - - @liveexample{The example shows the deserialization of a byte vector in - MessagePack format to a JSON value.,from_msgpack} - - @sa http://msgpack.org - @sa @ref to_msgpack(const basic_json&) for the analogous serialization - @sa @ref from_cbor(const std::vector&) for the related CBOR format - */ - static basic_json from_msgpack(const std::vector& v) - { - size_t i = 0; - return from_msgpack_internal(v, i); - } - - /*! - @brief create a MessagePack serialization of a given JSON value - - Serializes a given JSON value @a j to a byte vector using the CBOR (Concise - Binary Object Representation) serialization format. CBOR is a binary - serialization format which aims to be more compact than JSON itself, yet - more efficient to parse. - - @param[in] j JSON value to serialize - @return MessagePack serialization as byte vector - - @complexity Linear in the size of the JSON value @a j. - - @liveexample{The example shows the serialization of a JSON value to a byte - vector in CBOR format.,to_cbor} - - @sa http://cbor.io - @sa @ref from_cbor(const std::vector&) for the analogous - deserialization - @sa @ref to_msgpack(const basic_json& for the related MessagePack format - */ - static std::vector to_cbor(const basic_json& j) - { - std::vector result; - to_cbor_internal(j, result); - return result; - } - - /*! - @brief create a JSON value from a byte vector in CBOR format - - Deserializes a given byte vector @a v to a JSON value using the CBOR - (Concise Binary Object Representation) serialization format. - - @param[in] v a byte vector in CBOR format - @return deserialized JSON value - - @throw std::invalid_argument if unsupported features from CBOR were used in - the given vector @a v or if the input is not valid MessagePack - @throw std::out_of_range if the given vector ends prematurely - - @complexity Linear in the size of the byte vector @a v. - - @liveexample{The example shows the deserialization of a byte vector in CBOR - format to a JSON value.,from_cbor} - - @sa http://cbor.io - @sa @ref to_cbor(const basic_json&) for the analogous serialization - @sa @ref from_msgpack(const std::vector&) for the related - MessagePack format - */ - static basic_json from_cbor(const std::vector& v) - { - size_t i = 0; - return from_cbor_internal(v, i); - } - - /// @} - - private: /////////////////////////// // convenience functions // /////////////////////////// @@ -7740,353 +13071,54 @@ class basic_json Returns the type name as string to be used in error messages - usually to indicate that a function was called on a wrong JSON type. - @return basically a string representation of a the @a m_type member + @return a string representation of a the @a m_type member: + Value type | return value + ----------- | ------------- + null | `"null"` + boolean | `"boolean"` + string | `"string"` + number | `"number"` (for all number types) + object | `"object"` + array | `"array"` + discarded | `"discarded"` + + @exceptionsafety No-throw guarantee: this function never throws exceptions. @complexity Constant. - @since version 1.0.0 + @liveexample{The following code exemplifies `type_name()` for all JSON + types.,type_name} + + @sa @ref type() -- return the type of the JSON value + @sa @ref operator value_t() -- return the type of the JSON value (implicit) + + @since version 1.0.0, public since 2.1.0, `const char*` and `noexcept` + since 3.0.0 */ - std::string type_name() const + const char* type_name() const noexcept { - switch (m_type) { - case value_t::null: - return "null"; - case value_t::object: - return "object"; - case value_t::array: - return "array"; - case value_t::string: - return "string"; - case value_t::boolean: - return "boolean"; - case value_t::discarded: - return "discarded"; - default: - return "number"; - } - } - - /*! - @brief calculates the extra space to escape a JSON string - - @param[in] s the string to escape - @return the number of characters required to escape string @a s - - @complexity Linear in the length of string @a s. - */ - static std::size_t extra_space(const string_t& s) noexcept - { - return std::accumulate(s.begin(), s.end(), size_t{}, - [](size_t res, typename string_t::value_type c) - { - switch (c) + switch (m_type) { - case '"': - case '\\': - case '\b': - case '\f': - case '\n': - case '\r': - case '\t': - { - // from c (1 byte) to \x (2 bytes) - return res + 1; - } - + case value_t::null: + return "null"; + case value_t::object: + return "object"; + case value_t::array: + return "array"; + case value_t::string: + return "string"; + case value_t::boolean: + return "boolean"; + case value_t::discarded: + return "discarded"; default: - { - if (c >= 0x00 and c <= 0x1f) - { - // from c (1 byte) to \uxxxx (6 bytes) - return res + 5; - } - else - { - return res; - } - } - } - }); - } - - /*! - @brief escape a string - - Escape a string by replacing certain special characters by a sequence of - an escape character (backslash) and another character and other control - characters by a sequence of "\u" followed by a four-digit hex - representation. - - @param[in] s the string to escape - @return the escaped string - - @complexity Linear in the length of string @a s. - */ - static string_t escape_string(const string_t& s) - { - const auto space = extra_space(s); - if (space == 0) - { - return s; - } - - // create a result string of necessary size - string_t result(s.size() + space, '\\'); - std::size_t pos = 0; - - for (const auto& c : s) - { - switch (c) - { - // quotation mark (0x22) - case '"': - { - result[pos + 1] = '"'; - pos += 2; - break; - } - - // reverse solidus (0x5c) - case '\\': - { - // nothing to change - pos += 2; - break; - } - - // backspace (0x08) - case '\b': - { - result[pos + 1] = 'b'; - pos += 2; - break; - } - - // formfeed (0x0c) - case '\f': - { - result[pos + 1] = 'f'; - pos += 2; - break; - } - - // newline (0x0a) - case '\n': - { - result[pos + 1] = 'n'; - pos += 2; - break; - } - - // carriage return (0x0d) - case '\r': - { - result[pos + 1] = 'r'; - pos += 2; - break; - } - - // horizontal tab (0x09) - case '\t': - { - result[pos + 1] = 't'; - pos += 2; - break; - } - - default: - { - if (c >= 0x00 and c <= 0x1f) - { - // convert a number 0..15 to its hex representation - // (0..f) - static const char hexify[16] = - { - '0', '1', '2', '3', '4', '5', '6', '7', - '8', '9', 'a', 'b', 'c', 'd', 'e', 'f' - }; - - // print character c as \uxxxx - for (const char m : - { 'u', '0', '0', hexify[c >> 4], hexify[c & 0x0f] - }) - { - result[++pos] = m; - } - - ++pos; - } - else - { - // all other characters are added as-is - result[pos++] = c; - } - break; - } - } - } - - return result; - } - - /*! - @brief internal implementation of the serialization function - - This function is called by the public member function dump and organizes - the serialization internally. The indentation level is propagated as - additional parameter. In case of arrays and objects, the function is - called recursively. Note that - - - strings and object keys are escaped using `escape_string()` - - integer numbers are converted implicitly via `operator<<` - - floating-point numbers are converted to a string using `"%g"` format - - @param[out] o stream to write to - @param[in] pretty_print whether the output shall be pretty-printed - @param[in] indent_step the indent level - @param[in] current_indent the current indent level (only used internally) - */ - void dump(std::ostream& o, - const bool pretty_print, - const unsigned int indent_step, - const unsigned int current_indent = 0) const - { - // variable to hold indentation for recursive calls - unsigned int new_indent = current_indent; - - switch (m_type) - { - case value_t::object: - { - if (m_value.object->empty()) - { - o << "{}"; - return; - } - - o << "{"; - - // increase indentation - if (pretty_print) - { - new_indent += indent_step; - o << "\n"; - } - - for (auto i = m_value.object->cbegin(); i != m_value.object->cend(); ++i) - { - if (i != m_value.object->cbegin()) - { - o << (pretty_print ? ",\n" : ","); - } - o << string_t(new_indent, ' ') << "\"" - << escape_string(i->first) << "\":" - << (pretty_print ? " " : ""); - i->second.dump(o, pretty_print, indent_step, new_indent); - } - - // decrease indentation - if (pretty_print) - { - new_indent -= indent_step; - o << "\n"; - } - - o << string_t(new_indent, ' ') + "}"; - return; - } - - case value_t::array: - { - if (m_value.array->empty()) - { - o << "[]"; - return; - } - - o << "["; - - // increase indentation - if (pretty_print) - { - new_indent += indent_step; - o << "\n"; - } - - for (auto i = m_value.array->cbegin(); i != m_value.array->cend(); ++i) - { - if (i != m_value.array->cbegin()) - { - o << (pretty_print ? ",\n" : ","); - } - o << string_t(new_indent, ' '); - i->dump(o, pretty_print, indent_step, new_indent); - } - - // decrease indentation - if (pretty_print) - { - new_indent -= indent_step; - o << "\n"; - } - - o << string_t(new_indent, ' ') << "]"; - return; - } - - case value_t::string: - { - o << string_t("\"") << escape_string(*m_value.string) << "\""; - return; - } - - case value_t::boolean: - { - o << (m_value.boolean ? "true" : "false"); - return; - } - - case value_t::number_integer: - { - o << m_value.number_integer; - return; - } - - case value_t::number_unsigned: - { - o << m_value.number_unsigned; - return; - } - - case value_t::number_float: - { - if (m_value.number_float == 0) - { - // special case for zero to get "0.0"/"-0.0" - o << (std::signbit(m_value.number_float) ? "-0.0" : "0.0"); - } - else - { - o << m_value.number_float; - } - return; - } - - case value_t::discarded: - { - o << ""; - return; - } - - case value_t::null: - { - o << "null"; - return; + return "number"; } } } + private: ////////////////////// // member variables // @@ -8098,3426 +13130,407 @@ class basic_json /// the value of the current element json_value m_value = {}; + ////////////////////////////////////////// + // binary serialization/deserialization // + ////////////////////////////////////////// - private: - /////////////// - // iterators // - /////////////// - - /*! - @brief an iterator for primitive JSON types - - This class models an iterator for primitive JSON types (boolean, number, - string). It's only purpose is to allow the iterator/const_iterator classes - to "iterate" over primitive values. Internally, the iterator is modeled by - a `difference_type` variable. Value begin_value (`0`) models the begin, - end_value (`1`) models past the end. - */ - class primitive_iterator_t - { - public: - /// set iterator to a defined beginning - void set_begin() noexcept - { - m_it = begin_value; - } - - /// set iterator to a defined past the end - void set_end() noexcept - { - m_it = end_value; - } - - /// return whether the iterator can be dereferenced - constexpr bool is_begin() const noexcept - { - return (m_it == begin_value); - } - - /// return whether the iterator is at end - constexpr bool is_end() const noexcept - { - return (m_it == end_value); - } - - /// return reference to the value to change and compare - operator difference_type& () noexcept - { - return m_it; - } - - /// return value to compare - constexpr operator difference_type () const noexcept - { - return m_it; - } - - private: - static constexpr difference_type begin_value = 0; - static constexpr difference_type end_value = begin_value + 1; - - /// iterator as signed integer type - difference_type m_it = std::numeric_limits::denorm_min(); - }; - - /*! - @brief an iterator value - - @note This structure could easily be a union, but MSVC currently does not - allow unions members with complex constructors, see - https://github.com/nlohmann/json/pull/105. - */ - struct internal_iterator - { - /// iterator for JSON objects - typename object_t::iterator object_iterator; - /// iterator for JSON arrays - typename array_t::iterator array_iterator; - /// generic iterator for all other types - primitive_iterator_t primitive_iterator; - - /// create an uninitialized internal_iterator - internal_iterator() noexcept - : object_iterator(), array_iterator(), primitive_iterator() - {} - }; - - /// proxy class for the iterator_wrapper functions - template - class iteration_proxy - { - private: - /// helper class for iteration - class iteration_proxy_internal - { - private: - /// the iterator - IteratorType anchor; - /// an index for arrays (used to create key names) - size_t array_index = 0; - - public: - explicit iteration_proxy_internal(IteratorType it) noexcept - : anchor(it) - {} - - /// dereference operator (needed for range-based for) - iteration_proxy_internal& operator*() - { - return *this; - } - - /// increment operator (needed for range-based for) - iteration_proxy_internal& operator++() - { - ++anchor; - ++array_index; - - return *this; - } - - /// inequality operator (needed for range-based for) - bool operator!= (const iteration_proxy_internal& o) const - { - return anchor != o.anchor; - } - - /// return key of the iterator - typename basic_json::string_t key() const - { - assert(anchor.m_object != nullptr); - - switch (anchor.m_object->type()) - { - // use integer array index as key - case value_t::array: - { - return std::to_string(array_index); - } - - // use key from the object - case value_t::object: - { - return anchor.key(); - } - - // use an empty key for all primitive types - default: - { - return ""; - } - } - } - - /// return value of the iterator - typename IteratorType::reference value() const - { - return anchor.value(); - } - }; - - /// the container to iterate - typename IteratorType::reference container; - - public: - /// construct iteration proxy from a container - explicit iteration_proxy(typename IteratorType::reference cont) - : container(cont) - {} - - /// return iterator begin (needed for range-based for) - iteration_proxy_internal begin() noexcept - { - return iteration_proxy_internal(container.begin()); - } - - /// return iterator end (needed for range-based for) - iteration_proxy_internal end() noexcept - { - return iteration_proxy_internal(container.end()); - } - }; + /// @name binary serialization/deserialization support + /// @{ public: /*! - @brief a template for a random access iterator for the @ref basic_json class + @brief create a CBOR serialization of a given JSON value - This class implements a both iterators (iterator and const_iterator) for the - @ref basic_json class. + Serializes a given JSON value @a j to a byte vector using the CBOR (Concise + Binary Object Representation) serialization format. CBOR is a binary + serialization format which aims to be more compact than JSON itself, yet + more efficient to parse. - @note An iterator is called *initialized* when a pointer to a JSON value - has been set (e.g., by a constructor or a copy assignment). If the - iterator is default-constructed, it is *uninitialized* and most - methods are undefined. **The library uses assertions to detect calls - on uninitialized iterators.** + The library uses the following mapping from JSON values types to + CBOR types according to the CBOR specification (RFC 7049): - @requirement The class satisfies the following concept requirements: - - [RandomAccessIterator](http://en.cppreference.com/w/cpp/concept/RandomAccessIterator): - The iterator that can be moved to point (forward and backward) to any - element in constant time. + JSON value type | value/range | CBOR type | first byte + --------------- | ------------------------------------------ | ---------------------------------- | --------------- + null | `null` | Null | 0xf6 + boolean | `true` | True | 0xf5 + boolean | `false` | False | 0xf4 + number_integer | -9223372036854775808..-2147483649 | Negative integer (8 bytes follow) | 0x3b + number_integer | -2147483648..-32769 | Negative integer (4 bytes follow) | 0x3a + number_integer | -32768..-129 | Negative integer (2 bytes follow) | 0x39 + number_integer | -128..-25 | Negative integer (1 byte follow) | 0x38 + number_integer | -24..-1 | Negative integer | 0x20..0x37 + number_integer | 0..23 | Integer | 0x00..0x17 + number_integer | 24..255 | Unsigned integer (1 byte follow) | 0x18 + number_integer | 256..65535 | Unsigned integer (2 bytes follow) | 0x19 + number_integer | 65536..4294967295 | Unsigned integer (4 bytes follow) | 0x1a + number_integer | 4294967296..18446744073709551615 | Unsigned integer (8 bytes follow) | 0x1b + number_unsigned | 0..23 | Integer | 0x00..0x17 + number_unsigned | 24..255 | Unsigned integer (1 byte follow) | 0x18 + number_unsigned | 256..65535 | Unsigned integer (2 bytes follow) | 0x19 + number_unsigned | 65536..4294967295 | Unsigned integer (4 bytes follow) | 0x1a + number_unsigned | 4294967296..18446744073709551615 | Unsigned integer (8 bytes follow) | 0x1b + number_float | *any value* | Double-Precision Float | 0xfb + string | *length*: 0..23 | UTF-8 string | 0x60..0x77 + string | *length*: 23..255 | UTF-8 string (1 byte follow) | 0x78 + string | *length*: 256..65535 | UTF-8 string (2 bytes follow) | 0x79 + string | *length*: 65536..4294967295 | UTF-8 string (4 bytes follow) | 0x7a + string | *length*: 4294967296..18446744073709551615 | UTF-8 string (8 bytes follow) | 0x7b + array | *size*: 0..23 | array | 0x80..0x97 + array | *size*: 23..255 | array (1 byte follow) | 0x98 + array | *size*: 256..65535 | array (2 bytes follow) | 0x99 + array | *size*: 65536..4294967295 | array (4 bytes follow) | 0x9a + array | *size*: 4294967296..18446744073709551615 | array (8 bytes follow) | 0x9b + object | *size*: 0..23 | map | 0xa0..0xb7 + object | *size*: 23..255 | map (1 byte follow) | 0xb8 + object | *size*: 256..65535 | map (2 bytes follow) | 0xb9 + object | *size*: 65536..4294967295 | map (4 bytes follow) | 0xba + object | *size*: 4294967296..18446744073709551615 | map (8 bytes follow) | 0xbb - @since version 1.0.0, simplified in version 2.0.9 + @note The mapping is **complete** in the sense that any JSON value type + can be converted to a CBOR value. + + @note If NaN or Infinity are stored inside a JSON number, they are + serialized properly. This behavior differs from the @ref dump() + function which serializes NaN or Infinity to `null`. + + @note The following CBOR types are not used in the conversion: + - byte strings (0x40..0x5f) + - UTF-8 strings terminated by "break" (0x7f) + - arrays terminated by "break" (0x9f) + - maps terminated by "break" (0xbf) + - date/time (0xc0..0xc1) + - bignum (0xc2..0xc3) + - decimal fraction (0xc4) + - bigfloat (0xc5) + - tagged items (0xc6..0xd4, 0xd8..0xdb) + - expected conversions (0xd5..0xd7) + - simple values (0xe0..0xf3, 0xf8) + - undefined (0xf7) + - half and single-precision floats (0xf9-0xfa) + - break (0xff) + + @param[in] j JSON value to serialize + @return MessagePack serialization as byte vector + + @complexity Linear in the size of the JSON value @a j. + + @liveexample{The example shows the serialization of a JSON value to a byte + vector in CBOR format.,to_cbor} + + @sa http://cbor.io + @sa @ref from_cbor(const std::vector&, const size_t) for the + analogous deserialization + @sa @ref to_msgpack(const basic_json&) for the related MessagePack format + + @since version 2.0.9 */ - template - class iter_impl : public std::iterator + static std::vector to_cbor(const basic_json& j) { - /// allow basic_json to access private members - friend class basic_json; - - // make sure U is basic_json or const basic_json - static_assert(std::is_same::value - or std::is_same::value, - "iter_impl only accepts (const) basic_json"); - - public: - /// the type of the values when the iterator is dereferenced - using value_type = typename basic_json::value_type; - /// a type to represent differences between iterators - using difference_type = typename basic_json::difference_type; - /// defines a pointer to the type iterated over (value_type) - using pointer = typename std::conditional::value, - typename basic_json::const_pointer, - typename basic_json::pointer>::type; - /// defines a reference to the type iterated over (value_type) - using reference = typename std::conditional::value, - typename basic_json::const_reference, - typename basic_json::reference>::type; - /// the category of the iterator - using iterator_category = std::bidirectional_iterator_tag; - - /// default constructor - iter_impl() = default; - - /*! - @brief constructor for a given JSON instance - @param[in] object pointer to a JSON object for this iterator - @pre object != nullptr - @post The iterator is initialized; i.e. `m_object != nullptr`. - */ - explicit iter_impl(pointer object) noexcept - : m_object(object) - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - m_it.object_iterator = typename object_t::iterator(); - break; - } - - case basic_json::value_t::array: - { - m_it.array_iterator = typename array_t::iterator(); - break; - } - - default: - { - m_it.primitive_iterator = primitive_iterator_t(); - break; - } - } - } - - /* - Use operator `const_iterator` instead of `const_iterator(const iterator& - other) noexcept` to avoid two class definitions for @ref iterator and - @ref const_iterator. - - This function is only called if this class is an @ref iterator. If this - class is a @ref const_iterator this function is not called. - */ - operator const_iterator() const - { - const_iterator ret; - - if (m_object) - { - ret.m_object = m_object; - ret.m_it = m_it; - } - - return ret; - } - - /*! - @brief copy constructor - @param[in] other iterator to copy from - @note It is not checked whether @a other is initialized. - */ - iter_impl(const iter_impl& other) noexcept - : m_object(other.m_object), m_it(other.m_it) - {} - - /*! - @brief copy assignment - @param[in,out] other iterator to copy from - @note It is not checked whether @a other is initialized. - */ - iter_impl& operator=(iter_impl other) noexcept( - std::is_nothrow_move_constructible::value and - std::is_nothrow_move_assignable::value and - std::is_nothrow_move_constructible::value and - std::is_nothrow_move_assignable::value - ) - { - std::swap(m_object, other.m_object); - std::swap(m_it, other.m_it); - return *this; - } - - private: - /*! - @brief set the iterator to the first value - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - void set_begin() noexcept - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - m_it.object_iterator = m_object->m_value.object->begin(); - break; - } - - case basic_json::value_t::array: - { - m_it.array_iterator = m_object->m_value.array->begin(); - break; - } - - case basic_json::value_t::null: - { - // set to end so begin()==end() is true: null is empty - m_it.primitive_iterator.set_end(); - break; - } - - default: - { - m_it.primitive_iterator.set_begin(); - break; - } - } - } - - /*! - @brief set the iterator past the last value - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - void set_end() noexcept - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - m_it.object_iterator = m_object->m_value.object->end(); - break; - } - - case basic_json::value_t::array: - { - m_it.array_iterator = m_object->m_value.array->end(); - break; - } - - default: - { - m_it.primitive_iterator.set_end(); - break; - } - } - } - - public: - /*! - @brief return a reference to the value pointed to by the iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - reference operator*() const - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - assert(m_it.object_iterator != m_object->m_value.object->end()); - return m_it.object_iterator->second; - } - - case basic_json::value_t::array: - { - assert(m_it.array_iterator != m_object->m_value.array->end()); - return *m_it.array_iterator; - } - - case basic_json::value_t::null: - { - throw std::out_of_range("cannot get value"); - } - - default: - { - if (m_it.primitive_iterator.is_begin()) - { - return *m_object; - } - else - { - throw std::out_of_range("cannot get value"); - } - } - } - } - - /*! - @brief dereference the iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - pointer operator->() const - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - assert(m_it.object_iterator != m_object->m_value.object->end()); - return &(m_it.object_iterator->second); - } - - case basic_json::value_t::array: - { - assert(m_it.array_iterator != m_object->m_value.array->end()); - return &*m_it.array_iterator; - } - - default: - { - if (m_it.primitive_iterator.is_begin()) - { - return m_object; - } - else - { - throw std::out_of_range("cannot get value"); - } - } - } - } - - /*! - @brief post-increment (it++) - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl operator++(int) - { - auto result = *this; - ++(*this); - return result; - } - - /*! - @brief pre-increment (++it) - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl& operator++() - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - std::advance(m_it.object_iterator, 1); - break; - } - - case basic_json::value_t::array: - { - std::advance(m_it.array_iterator, 1); - break; - } - - default: - { - ++m_it.primitive_iterator; - break; - } - } - - return *this; - } - - /*! - @brief post-decrement (it--) - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl operator--(int) - { - auto result = *this; - --(*this); - return result; - } - - /*! - @brief pre-decrement (--it) - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl& operator--() - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - std::advance(m_it.object_iterator, -1); - break; - } - - case basic_json::value_t::array: - { - std::advance(m_it.array_iterator, -1); - break; - } - - default: - { - --m_it.primitive_iterator; - break; - } - } - - return *this; - } - - /*! - @brief comparison: equal - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - bool operator==(const iter_impl& other) const - { - // if objects are not the same, the comparison is undefined - if (m_object != other.m_object) - { - throw std::domain_error("cannot compare iterators of different containers"); - } - - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - return (m_it.object_iterator == other.m_it.object_iterator); - } - - case basic_json::value_t::array: - { - return (m_it.array_iterator == other.m_it.array_iterator); - } - - default: - { - return (m_it.primitive_iterator == other.m_it.primitive_iterator); - } - } - } - - /*! - @brief comparison: not equal - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - bool operator!=(const iter_impl& other) const - { - return not operator==(other); - } - - /*! - @brief comparison: smaller - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - bool operator<(const iter_impl& other) const - { - // if objects are not the same, the comparison is undefined - if (m_object != other.m_object) - { - throw std::domain_error("cannot compare iterators of different containers"); - } - - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - throw std::domain_error("cannot compare order of object iterators"); - } - - case basic_json::value_t::array: - { - return (m_it.array_iterator < other.m_it.array_iterator); - } - - default: - { - return (m_it.primitive_iterator < other.m_it.primitive_iterator); - } - } - } - - /*! - @brief comparison: less than or equal - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - bool operator<=(const iter_impl& other) const - { - return not other.operator < (*this); - } - - /*! - @brief comparison: greater than - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - bool operator>(const iter_impl& other) const - { - return not operator<=(other); - } - - /*! - @brief comparison: greater than or equal - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - bool operator>=(const iter_impl& other) const - { - return not operator<(other); - } - - /*! - @brief add to iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl& operator+=(difference_type i) - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - throw std::domain_error("cannot use offsets with object iterators"); - } - - case basic_json::value_t::array: - { - std::advance(m_it.array_iterator, i); - break; - } - - default: - { - m_it.primitive_iterator += i; - break; - } - } - - return *this; - } - - /*! - @brief subtract from iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl& operator-=(difference_type i) - { - return operator+=(-i); - } - - /*! - @brief add to iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl operator+(difference_type i) - { - auto result = *this; - result += i; - return result; - } - - /*! - @brief subtract from iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - iter_impl operator-(difference_type i) - { - auto result = *this; - result -= i; - return result; - } - - /*! - @brief return difference - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - difference_type operator-(const iter_impl& other) const - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - throw std::domain_error("cannot use offsets with object iterators"); - } - - case basic_json::value_t::array: - { - return m_it.array_iterator - other.m_it.array_iterator; - } - - default: - { - return m_it.primitive_iterator - other.m_it.primitive_iterator; - } - } - } - - /*! - @brief access to successor - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - reference operator[](difference_type n) const - { - assert(m_object != nullptr); - - switch (m_object->m_type) - { - case basic_json::value_t::object: - { - throw std::domain_error("cannot use operator[] for object iterators"); - } - - case basic_json::value_t::array: - { - return *std::next(m_it.array_iterator, n); - } - - case basic_json::value_t::null: - { - throw std::out_of_range("cannot get value"); - } - - default: - { - if (m_it.primitive_iterator == -n) - { - return *m_object; - } - else - { - throw std::out_of_range("cannot get value"); - } - } - } - } - - /*! - @brief return the key of an object iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - typename object_t::key_type key() const - { - assert(m_object != nullptr); - - if (m_object->is_object()) - { - return m_it.object_iterator->first; - } - else - { - throw std::domain_error("cannot use key() for non-object iterators"); - } - } - - /*! - @brief return the value of an iterator - @pre The iterator is initialized; i.e. `m_object != nullptr`. - */ - reference value() const - { - return operator*(); - } - - private: - /// associated JSON instance - pointer m_object = nullptr; - /// the actual iterator of the associated instance - internal_iterator m_it = internal_iterator(); - }; + std::vector result; + to_cbor(j, result); + return result; + } + + static void to_cbor(const basic_json& j, detail::output_adapter o) + { + binary_writer(o).write_cbor(j); + } + + static void to_cbor(const basic_json& j, detail::output_adapter o) + { + binary_writer(o).write_cbor(j); + } /*! - @brief a template for a reverse iterator class + @brief create a MessagePack serialization of a given JSON value - @tparam Base the base iterator type to reverse. Valid types are @ref - iterator (to create @ref reverse_iterator) and @ref const_iterator (to - create @ref const_reverse_iterator). + Serializes a given JSON value @a j to a byte vector using the MessagePack + serialization format. MessagePack is a binary serialization format which + aims to be more compact than JSON itself, yet more efficient to parse. - @requirement The class satisfies the following concept requirements: - - [RandomAccessIterator](http://en.cppreference.com/w/cpp/concept/RandomAccessIterator): - The iterator that can be moved to point (forward and backward) to any - element in constant time. - - [OutputIterator](http://en.cppreference.com/w/cpp/concept/OutputIterator): - It is possible to write to the pointed-to element (only if @a Base is - @ref iterator). + The library uses the following mapping from JSON values types to + MessagePack types according to the MessagePack specification: - @since version 1.0.0 + JSON value type | value/range | MessagePack type | first byte + --------------- | --------------------------------- | ---------------- | ---------- + null | `null` | nil | 0xc0 + boolean | `true` | true | 0xc3 + boolean | `false` | false | 0xc2 + number_integer | -9223372036854775808..-2147483649 | int64 | 0xd3 + number_integer | -2147483648..-32769 | int32 | 0xd2 + number_integer | -32768..-129 | int16 | 0xd1 + number_integer | -128..-33 | int8 | 0xd0 + number_integer | -32..-1 | negative fixint | 0xe0..0xff + number_integer | 0..127 | positive fixint | 0x00..0x7f + number_integer | 128..255 | uint 8 | 0xcc + number_integer | 256..65535 | uint 16 | 0xcd + number_integer | 65536..4294967295 | uint 32 | 0xce + number_integer | 4294967296..18446744073709551615 | uint 64 | 0xcf + number_unsigned | 0..127 | positive fixint | 0x00..0x7f + number_unsigned | 128..255 | uint 8 | 0xcc + number_unsigned | 256..65535 | uint 16 | 0xcd + number_unsigned | 65536..4294967295 | uint 32 | 0xce + number_unsigned | 4294967296..18446744073709551615 | uint 64 | 0xcf + number_float | *any value* | float 64 | 0xcb + string | *length*: 0..31 | fixstr | 0xa0..0xbf + string | *length*: 32..255 | str 8 | 0xd9 + string | *length*: 256..65535 | str 16 | 0xda + string | *length*: 65536..4294967295 | str 32 | 0xdb + array | *size*: 0..15 | fixarray | 0x90..0x9f + array | *size*: 16..65535 | array 16 | 0xdc + array | *size*: 65536..4294967295 | array 32 | 0xdd + object | *size*: 0..15 | fix map | 0x80..0x8f + object | *size*: 16..65535 | map 16 | 0xde + object | *size*: 65536..4294967295 | map 32 | 0xdf + + @note The mapping is **complete** in the sense that any JSON value type + can be converted to a MessagePack value. + + @note The following values can **not** be converted to a MessagePack value: + - strings with more than 4294967295 bytes + - arrays with more than 4294967295 elements + - objects with more than 4294967295 elements + + @note The following MessagePack types are not used in the conversion: + - bin 8 - bin 32 (0xc4..0xc6) + - ext 8 - ext 32 (0xc7..0xc9) + - float 32 (0xca) + - fixext 1 - fixext 16 (0xd4..0xd8) + + @note Any MessagePack output created @ref to_msgpack can be successfully + parsed by @ref from_msgpack. + + @note If NaN or Infinity are stored inside a JSON number, they are + serialized properly. This behavior differs from the @ref dump() + function which serializes NaN or Infinity to `null`. + + @param[in] j JSON value to serialize + @return MessagePack serialization as byte vector + + @complexity Linear in the size of the JSON value @a j. + + @liveexample{The example shows the serialization of a JSON value to a byte + vector in MessagePack format.,to_msgpack} + + @sa http://msgpack.org + @sa @ref from_msgpack(const std::vector&, const size_t) for the + analogous deserialization + @sa @ref to_cbor(const basic_json& for the related CBOR format + + @since version 2.0.9 */ - template - class json_reverse_iterator : public std::reverse_iterator + static std::vector to_msgpack(const basic_json& j) { - public: - /// shortcut to the reverse iterator adaptor - using base_iterator = std::reverse_iterator; - /// the reference type for the pointed-to element - using reference = typename Base::reference; + std::vector result; + to_msgpack(j, result); + return result; + } - /// create reverse iterator from iterator - json_reverse_iterator(const typename base_iterator::iterator_type& it) noexcept - : base_iterator(it) - {} + static void to_msgpack(const basic_json& j, detail::output_adapter o) + { + binary_writer(o).write_msgpack(j); + } - /// create reverse iterator from base class - json_reverse_iterator(const base_iterator& it) noexcept - : base_iterator(it) - {} - - /// post-increment (it++) - json_reverse_iterator operator++(int) - { - return base_iterator::operator++(1); - } - - /// pre-increment (++it) - json_reverse_iterator& operator++() - { - base_iterator::operator++(); - return *this; - } - - /// post-decrement (it--) - json_reverse_iterator operator--(int) - { - return base_iterator::operator--(1); - } - - /// pre-decrement (--it) - json_reverse_iterator& operator--() - { - base_iterator::operator--(); - return *this; - } - - /// add to iterator - json_reverse_iterator& operator+=(difference_type i) - { - base_iterator::operator+=(i); - return *this; - } - - /// add to iterator - json_reverse_iterator operator+(difference_type i) const - { - auto result = *this; - result += i; - return result; - } - - /// subtract from iterator - json_reverse_iterator operator-(difference_type i) const - { - auto result = *this; - result -= i; - return result; - } - - /// return difference - difference_type operator-(const json_reverse_iterator& other) const - { - return this->base() - other.base(); - } - - /// access to successor - reference operator[](difference_type n) const - { - return *(this->operator+(n)); - } - - /// return the key of an object iterator - typename object_t::key_type key() const - { - auto it = --this->base(); - return it.key(); - } - - /// return the value of an iterator - reference value() const - { - auto it = --this->base(); - return it.operator * (); - } - }; - - - private: - ////////////////////// - // lexer and parser // - ////////////////////// + static void to_msgpack(const basic_json& j, detail::output_adapter o) + { + binary_writer(o).write_msgpack(j); + } /*! - @brief lexical analysis + @brief create a JSON value from an input in CBOR format - This class organizes the lexical analysis during JSON deserialization. The - core of it is a scanner generated by [re2c](http://re2c.org) that - processes a buffer and recognizes tokens according to RFC 7159. + Deserializes a given input @a i to a JSON value using the CBOR (Concise + Binary Object Representation) serialization format. + + The library maps CBOR types to JSON value types as follows: + + CBOR type | JSON value type | first byte + ---------------------- | --------------- | ---------- + Integer | number_unsigned | 0x00..0x17 + Unsigned integer | number_unsigned | 0x18 + Unsigned integer | number_unsigned | 0x19 + Unsigned integer | number_unsigned | 0x1a + Unsigned integer | number_unsigned | 0x1b + Negative integer | number_integer | 0x20..0x37 + Negative integer | number_integer | 0x38 + Negative integer | number_integer | 0x39 + Negative integer | number_integer | 0x3a + Negative integer | number_integer | 0x3b + Negative integer | number_integer | 0x40..0x57 + UTF-8 string | string | 0x60..0x77 + UTF-8 string | string | 0x78 + UTF-8 string | string | 0x79 + UTF-8 string | string | 0x7a + UTF-8 string | string | 0x7b + UTF-8 string | string | 0x7f + array | array | 0x80..0x97 + array | array | 0x98 + array | array | 0x99 + array | array | 0x9a + array | array | 0x9b + array | array | 0x9f + map | object | 0xa0..0xb7 + map | object | 0xb8 + map | object | 0xb9 + map | object | 0xba + map | object | 0xbb + map | object | 0xbf + False | `false` | 0xf4 + True | `true` | 0xf5 + Nill | `null` | 0xf6 + Half-Precision Float | number_float | 0xf9 + Single-Precision Float | number_float | 0xfa + Double-Precision Float | number_float | 0xfb + + @warning The mapping is **incomplete** in the sense that not all CBOR + types can be converted to a JSON value. The following CBOR types + are not supported and will yield parse errors (parse_error.112): + - byte strings (0x40..0x5f) + - date/time (0xc0..0xc1) + - bignum (0xc2..0xc3) + - decimal fraction (0xc4) + - bigfloat (0xc5) + - tagged items (0xc6..0xd4, 0xd8..0xdb) + - expected conversions (0xd5..0xd7) + - simple values (0xe0..0xf3, 0xf8) + - undefined (0xf7) + + @warning CBOR allows map keys of any type, whereas JSON only allows + strings as keys in object values. Therefore, CBOR maps with keys + other than UTF-8 strings are rejected (parse_error.113). + + @note Any CBOR output created @ref to_cbor can be successfully parsed by + @ref from_cbor. + + @param[in] i an input in CBOR format convertible to an input adapter + @param[in] strict whether to expect the input to be consumed until EOF + (true by default) + @return deserialized JSON value + + @throw parse_error.110 if the given input ends prematurely or the end of + file was not reached when @a strict was set to true + @throw parse_error.112 if unsupported features from CBOR were + used in the given input @a v or if the input is not valid CBOR + @throw parse_error.113 if a string was expected as map key, but not found + + @complexity Linear in the size of the input @a i. + + @liveexample{The example shows the deserialization of a byte vector in CBOR + format to a JSON value.,from_cbor} + + @sa http://cbor.io + @sa @ref to_cbor(const basic_json&) for the analogous serialization + @sa @ref from_msgpack(detail::input_adapter, const bool) for the + related MessagePack format + + @since version 2.0.9; parameter @a start_index since 2.1.1; changed to + consume input adapters, removed start_index parameter, and added + @a strict parameter since 3.0.0 */ - class lexer + static basic_json from_cbor(detail::input_adapter i, + const bool strict = true) { - public: - /// token types for the parser - enum class token_type - { - uninitialized, ///< indicating the scanner is uninitialized - literal_true, ///< the `true` literal - literal_false, ///< the `false` literal - literal_null, ///< the `null` literal - value_string, ///< a string -- use get_string() for actual value - value_number, ///< a number -- use get_number() for actual value - begin_array, ///< the character for array begin `[` - begin_object, ///< the character for object begin `{` - end_array, ///< the character for array end `]` - end_object, ///< the character for object end `}` - name_separator, ///< the name separator `:` - value_separator, ///< the value separator `,` - parse_error, ///< indicating a parse error - end_of_input ///< indicating the end of the input buffer - }; - - /// the char type to use in the lexer - using lexer_char_t = unsigned char; - - /// a lexer from a buffer with given length - lexer(const lexer_char_t* buff, const size_t len) noexcept - : m_content(buff) - { - assert(m_content != nullptr); - m_start = m_cursor = m_content; - m_limit = m_content + len; - } - - /// a lexer from an input stream - explicit lexer(std::istream& s) - : m_stream(&s), m_line_buffer() - { - // immediately abort if stream is erroneous - if (s.fail()) - { - throw std::invalid_argument("stream error"); - } - - // fill buffer - fill_line_buffer(); - - // skip UTF-8 byte-order mark - if (m_line_buffer.size() >= 3 and m_line_buffer.substr(0, 3) == "\xEF\xBB\xBF") - { - m_line_buffer[0] = ' '; - m_line_buffer[1] = ' '; - m_line_buffer[2] = ' '; - } - } - - // switch off unwanted functions (due to pointer members) - lexer() = delete; - lexer(const lexer&) = delete; - lexer operator=(const lexer&) = delete; - - /*! - @brief create a string from one or two Unicode code points - - There are two cases: (1) @a codepoint1 is in the Basic Multilingual - Plane (U+0000 through U+FFFF) and @a codepoint2 is 0, or (2) - @a codepoint1 and @a codepoint2 are a UTF-16 surrogate pair to - represent a code point above U+FFFF. - - @param[in] codepoint1 the code point (can be high surrogate) - @param[in] codepoint2 the code point (can be low surrogate or 0) - - @return string representation of the code point; the length of the - result string is between 1 and 4 characters. - - @throw std::out_of_range if code point is > 0x10ffff; example: `"code - points above 0x10FFFF are invalid"` - @throw std::invalid_argument if the low surrogate is invalid; example: - `""missing or wrong low surrogate""` - - @complexity Constant. - - @see - */ - static string_t to_unicode(const std::size_t codepoint1, - const std::size_t codepoint2 = 0) - { - // calculate the code point from the given code points - std::size_t codepoint = codepoint1; - - // check if codepoint1 is a high surrogate - if (codepoint1 >= 0xD800 and codepoint1 <= 0xDBFF) - { - // check if codepoint2 is a low surrogate - if (codepoint2 >= 0xDC00 and codepoint2 <= 0xDFFF) - { - codepoint = - // high surrogate occupies the most significant 22 bits - (codepoint1 << 10) - // low surrogate occupies the least significant 15 bits - + codepoint2 - // there is still the 0xD800, 0xDC00 and 0x10000 noise - // in the result so we have to subtract with: - // (0xD800 << 10) + DC00 - 0x10000 = 0x35FDC00 - - 0x35FDC00; - } - else - { - throw std::invalid_argument("missing or wrong low surrogate"); - } - } - - string_t result; - - if (codepoint < 0x80) - { - // 1-byte characters: 0xxxxxxx (ASCII) - result.append(1, static_cast(codepoint)); - } - else if (codepoint <= 0x7ff) - { - // 2-byte characters: 110xxxxx 10xxxxxx - result.append(1, static_cast(0xC0 | ((codepoint >> 6) & 0x1F))); - result.append(1, static_cast(0x80 | (codepoint & 0x3F))); - } - else if (codepoint <= 0xffff) - { - // 3-byte characters: 1110xxxx 10xxxxxx 10xxxxxx - result.append(1, static_cast(0xE0 | ((codepoint >> 12) & 0x0F))); - result.append(1, static_cast(0x80 | ((codepoint >> 6) & 0x3F))); - result.append(1, static_cast(0x80 | (codepoint & 0x3F))); - } - else if (codepoint <= 0x10ffff) - { - // 4-byte characters: 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx - result.append(1, static_cast(0xF0 | ((codepoint >> 18) & 0x07))); - result.append(1, static_cast(0x80 | ((codepoint >> 12) & 0x3F))); - result.append(1, static_cast(0x80 | ((codepoint >> 6) & 0x3F))); - result.append(1, static_cast(0x80 | (codepoint & 0x3F))); - } - else - { - throw std::out_of_range("code points above 0x10FFFF are invalid"); - } - - return result; - } - - /// return name of values of type token_type (only used for errors) - static std::string token_type_name(const token_type t) - { - switch (t) - { - case token_type::uninitialized: - return ""; - case token_type::literal_true: - return "true literal"; - case token_type::literal_false: - return "false literal"; - case token_type::literal_null: - return "null literal"; - case token_type::value_string: - return "string literal"; - case token_type::value_number: - return "number literal"; - case token_type::begin_array: - return "'['"; - case token_type::begin_object: - return "'{'"; - case token_type::end_array: - return "']'"; - case token_type::end_object: - return "'}'"; - case token_type::name_separator: - return "':'"; - case token_type::value_separator: - return "','"; - case token_type::parse_error: - return ""; - case token_type::end_of_input: - return "end of input"; - default: - { - // catch non-enum values - return "unknown token"; // LCOV_EXCL_LINE - } - } - } - - /*! - This function implements a scanner for JSON. It is specified using - regular expressions that try to follow RFC 7159 as close as possible. - These regular expressions are then translated into a minimized - deterministic finite automaton (DFA) by the tool - [re2c](http://re2c.org). As a result, the translated code for this - function consists of a large block of code with `goto` jumps. - - @return the class of the next token read from the buffer - - @complexity Linear in the length of the input.\n - - Proposition: The loop below will always terminate for finite input.\n - - Proof (by contradiction): Assume a finite input. To loop forever, the - loop must never hit code with a `break` statement. The only code - snippets without a `break` statement are the continue statements for - whitespace and byte-order-marks. To loop forever, the input must be an - infinite sequence of whitespace or byte-order-marks. This contradicts - the assumption of finite input, q.e.d. - */ - token_type scan() - { - while (true) - { - // pointer for backtracking information - m_marker = nullptr; - - // remember the begin of the token - m_start = m_cursor; - assert(m_start != nullptr); - - - { - lexer_char_t yych; - unsigned int yyaccept = 0; - static const unsigned char yybm[] = - { - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 32, 32, 0, 0, 32, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 160, 128, 0, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 192, 192, 192, 192, 192, 192, 192, 192, - 192, 192, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 0, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 128, 128, 128, 128, 128, 128, 128, 128, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - 0, 0, 0, 0, 0, 0, 0, 0, - }; - if ((m_limit - m_cursor) < 5) - { - fill_line_buffer(5); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yybm[0 + yych] & 32) - { - goto basic_json_parser_6; - } - if (yych <= '[') - { - if (yych <= '-') - { - if (yych <= '"') - { - if (yych <= 0x00) - { - goto basic_json_parser_2; - } - if (yych <= '!') - { - goto basic_json_parser_4; - } - goto basic_json_parser_9; - } - else - { - if (yych <= '+') - { - goto basic_json_parser_4; - } - if (yych <= ',') - { - goto basic_json_parser_10; - } - goto basic_json_parser_12; - } - } - else - { - if (yych <= '9') - { - if (yych <= '/') - { - goto basic_json_parser_4; - } - if (yych <= '0') - { - goto basic_json_parser_13; - } - goto basic_json_parser_15; - } - else - { - if (yych <= ':') - { - goto basic_json_parser_17; - } - if (yych <= 'Z') - { - goto basic_json_parser_4; - } - goto basic_json_parser_19; - } - } - } - else - { - if (yych <= 'n') - { - if (yych <= 'e') - { - if (yych == ']') - { - goto basic_json_parser_21; - } - goto basic_json_parser_4; - } - else - { - if (yych <= 'f') - { - goto basic_json_parser_23; - } - if (yych <= 'm') - { - goto basic_json_parser_4; - } - goto basic_json_parser_24; - } - } - else - { - if (yych <= 'z') - { - if (yych == 't') - { - goto basic_json_parser_25; - } - goto basic_json_parser_4; - } - else - { - if (yych <= '{') - { - goto basic_json_parser_26; - } - if (yych == '}') - { - goto basic_json_parser_28; - } - goto basic_json_parser_4; - } - } - } -basic_json_parser_2: - ++m_cursor; - { - last_token_type = token_type::end_of_input; - break; - } -basic_json_parser_4: - ++m_cursor; -basic_json_parser_5: - { - last_token_type = token_type::parse_error; - break; - } -basic_json_parser_6: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yybm[0 + yych] & 32) - { - goto basic_json_parser_6; - } - { - continue; - } -basic_json_parser_9: - yyaccept = 0; - yych = *(m_marker = ++m_cursor); - if (yych <= 0x1F) - { - goto basic_json_parser_5; - } - if (yych <= 0x7F) - { - goto basic_json_parser_31; - } - if (yych <= 0xC1) - { - goto basic_json_parser_5; - } - if (yych <= 0xF4) - { - goto basic_json_parser_31; - } - goto basic_json_parser_5; -basic_json_parser_10: - ++m_cursor; - { - last_token_type = token_type::value_separator; - break; - } -basic_json_parser_12: - yych = *++m_cursor; - if (yych <= '/') - { - goto basic_json_parser_5; - } - if (yych <= '0') - { - goto basic_json_parser_13; - } - if (yych <= '9') - { - goto basic_json_parser_15; - } - goto basic_json_parser_5; -basic_json_parser_13: - yyaccept = 1; - yych = *(m_marker = ++m_cursor); - if (yych <= 'D') - { - if (yych == '.') - { - goto basic_json_parser_43; - } - } - else - { - if (yych <= 'E') - { - goto basic_json_parser_44; - } - if (yych == 'e') - { - goto basic_json_parser_44; - } - } -basic_json_parser_14: - { - last_token_type = token_type::value_number; - break; - } -basic_json_parser_15: - yyaccept = 1; - m_marker = ++m_cursor; - if ((m_limit - m_cursor) < 3) - { - fill_line_buffer(3); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yybm[0 + yych] & 64) - { - goto basic_json_parser_15; - } - if (yych <= 'D') - { - if (yych == '.') - { - goto basic_json_parser_43; - } - goto basic_json_parser_14; - } - else - { - if (yych <= 'E') - { - goto basic_json_parser_44; - } - if (yych == 'e') - { - goto basic_json_parser_44; - } - goto basic_json_parser_14; - } -basic_json_parser_17: - ++m_cursor; - { - last_token_type = token_type::name_separator; - break; - } -basic_json_parser_19: - ++m_cursor; - { - last_token_type = token_type::begin_array; - break; - } -basic_json_parser_21: - ++m_cursor; - { - last_token_type = token_type::end_array; - break; - } -basic_json_parser_23: - yyaccept = 0; - yych = *(m_marker = ++m_cursor); - if (yych == 'a') - { - goto basic_json_parser_45; - } - goto basic_json_parser_5; -basic_json_parser_24: - yyaccept = 0; - yych = *(m_marker = ++m_cursor); - if (yych == 'u') - { - goto basic_json_parser_46; - } - goto basic_json_parser_5; -basic_json_parser_25: - yyaccept = 0; - yych = *(m_marker = ++m_cursor); - if (yych == 'r') - { - goto basic_json_parser_47; - } - goto basic_json_parser_5; -basic_json_parser_26: - ++m_cursor; - { - last_token_type = token_type::begin_object; - break; - } -basic_json_parser_28: - ++m_cursor; - { - last_token_type = token_type::end_object; - break; - } -basic_json_parser_30: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; -basic_json_parser_31: - if (yybm[0 + yych] & 128) - { - goto basic_json_parser_30; - } - if (yych <= 0xE0) - { - if (yych <= '\\') - { - if (yych <= 0x1F) - { - goto basic_json_parser_32; - } - if (yych <= '"') - { - goto basic_json_parser_33; - } - goto basic_json_parser_35; - } - else - { - if (yych <= 0xC1) - { - goto basic_json_parser_32; - } - if (yych <= 0xDF) - { - goto basic_json_parser_36; - } - goto basic_json_parser_37; - } - } - else - { - if (yych <= 0xEF) - { - if (yych == 0xED) - { - goto basic_json_parser_39; - } - goto basic_json_parser_38; - } - else - { - if (yych <= 0xF0) - { - goto basic_json_parser_40; - } - if (yych <= 0xF3) - { - goto basic_json_parser_41; - } - if (yych <= 0xF4) - { - goto basic_json_parser_42; - } - } - } -basic_json_parser_32: - m_cursor = m_marker; - if (yyaccept == 0) - { - goto basic_json_parser_5; - } - else - { - goto basic_json_parser_14; - } -basic_json_parser_33: - ++m_cursor; - { - last_token_type = token_type::value_string; - break; - } -basic_json_parser_35: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 'e') - { - if (yych <= '/') - { - if (yych == '"') - { - goto basic_json_parser_30; - } - if (yych <= '.') - { - goto basic_json_parser_32; - } - goto basic_json_parser_30; - } - else - { - if (yych <= '\\') - { - if (yych <= '[') - { - goto basic_json_parser_32; - } - goto basic_json_parser_30; - } - else - { - if (yych == 'b') - { - goto basic_json_parser_30; - } - goto basic_json_parser_32; - } - } - } - else - { - if (yych <= 'q') - { - if (yych <= 'f') - { - goto basic_json_parser_30; - } - if (yych == 'n') - { - goto basic_json_parser_30; - } - goto basic_json_parser_32; - } - else - { - if (yych <= 's') - { - if (yych <= 'r') - { - goto basic_json_parser_30; - } - goto basic_json_parser_32; - } - else - { - if (yych <= 't') - { - goto basic_json_parser_30; - } - if (yych <= 'u') - { - goto basic_json_parser_48; - } - goto basic_json_parser_32; - } - } - } -basic_json_parser_36: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x7F) - { - goto basic_json_parser_32; - } - if (yych <= 0xBF) - { - goto basic_json_parser_30; - } - goto basic_json_parser_32; -basic_json_parser_37: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x9F) - { - goto basic_json_parser_32; - } - if (yych <= 0xBF) - { - goto basic_json_parser_36; - } - goto basic_json_parser_32; -basic_json_parser_38: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x7F) - { - goto basic_json_parser_32; - } - if (yych <= 0xBF) - { - goto basic_json_parser_36; - } - goto basic_json_parser_32; -basic_json_parser_39: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x7F) - { - goto basic_json_parser_32; - } - if (yych <= 0x9F) - { - goto basic_json_parser_36; - } - goto basic_json_parser_32; -basic_json_parser_40: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x8F) - { - goto basic_json_parser_32; - } - if (yych <= 0xBF) - { - goto basic_json_parser_38; - } - goto basic_json_parser_32; -basic_json_parser_41: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x7F) - { - goto basic_json_parser_32; - } - if (yych <= 0xBF) - { - goto basic_json_parser_38; - } - goto basic_json_parser_32; -basic_json_parser_42: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 0x7F) - { - goto basic_json_parser_32; - } - if (yych <= 0x8F) - { - goto basic_json_parser_38; - } - goto basic_json_parser_32; -basic_json_parser_43: - yych = *++m_cursor; - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych <= '9') - { - goto basic_json_parser_49; - } - goto basic_json_parser_32; -basic_json_parser_44: - yych = *++m_cursor; - if (yych <= ',') - { - if (yych == '+') - { - goto basic_json_parser_51; - } - goto basic_json_parser_32; - } - else - { - if (yych <= '-') - { - goto basic_json_parser_51; - } - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych <= '9') - { - goto basic_json_parser_52; - } - goto basic_json_parser_32; - } -basic_json_parser_45: - yych = *++m_cursor; - if (yych == 'l') - { - goto basic_json_parser_54; - } - goto basic_json_parser_32; -basic_json_parser_46: - yych = *++m_cursor; - if (yych == 'l') - { - goto basic_json_parser_55; - } - goto basic_json_parser_32; -basic_json_parser_47: - yych = *++m_cursor; - if (yych == 'u') - { - goto basic_json_parser_56; - } - goto basic_json_parser_32; -basic_json_parser_48: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= '@') - { - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych <= '9') - { - goto basic_json_parser_57; - } - goto basic_json_parser_32; - } - else - { - if (yych <= 'F') - { - goto basic_json_parser_57; - } - if (yych <= '`') - { - goto basic_json_parser_32; - } - if (yych <= 'f') - { - goto basic_json_parser_57; - } - goto basic_json_parser_32; - } -basic_json_parser_49: - yyaccept = 1; - m_marker = ++m_cursor; - if ((m_limit - m_cursor) < 3) - { - fill_line_buffer(3); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= 'D') - { - if (yych <= '/') - { - goto basic_json_parser_14; - } - if (yych <= '9') - { - goto basic_json_parser_49; - } - goto basic_json_parser_14; - } - else - { - if (yych <= 'E') - { - goto basic_json_parser_44; - } - if (yych == 'e') - { - goto basic_json_parser_44; - } - goto basic_json_parser_14; - } -basic_json_parser_51: - yych = *++m_cursor; - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych >= ':') - { - goto basic_json_parser_32; - } -basic_json_parser_52: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= '/') - { - goto basic_json_parser_14; - } - if (yych <= '9') - { - goto basic_json_parser_52; - } - goto basic_json_parser_14; -basic_json_parser_54: - yych = *++m_cursor; - if (yych == 's') - { - goto basic_json_parser_58; - } - goto basic_json_parser_32; -basic_json_parser_55: - yych = *++m_cursor; - if (yych == 'l') - { - goto basic_json_parser_59; - } - goto basic_json_parser_32; -basic_json_parser_56: - yych = *++m_cursor; - if (yych == 'e') - { - goto basic_json_parser_61; - } - goto basic_json_parser_32; -basic_json_parser_57: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= '@') - { - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych <= '9') - { - goto basic_json_parser_63; - } - goto basic_json_parser_32; - } - else - { - if (yych <= 'F') - { - goto basic_json_parser_63; - } - if (yych <= '`') - { - goto basic_json_parser_32; - } - if (yych <= 'f') - { - goto basic_json_parser_63; - } - goto basic_json_parser_32; - } -basic_json_parser_58: - yych = *++m_cursor; - if (yych == 'e') - { - goto basic_json_parser_64; - } - goto basic_json_parser_32; -basic_json_parser_59: - ++m_cursor; - { - last_token_type = token_type::literal_null; - break; - } -basic_json_parser_61: - ++m_cursor; - { - last_token_type = token_type::literal_true; - break; - } -basic_json_parser_63: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= '@') - { - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych <= '9') - { - goto basic_json_parser_66; - } - goto basic_json_parser_32; - } - else - { - if (yych <= 'F') - { - goto basic_json_parser_66; - } - if (yych <= '`') - { - goto basic_json_parser_32; - } - if (yych <= 'f') - { - goto basic_json_parser_66; - } - goto basic_json_parser_32; - } -basic_json_parser_64: - ++m_cursor; - { - last_token_type = token_type::literal_false; - break; - } -basic_json_parser_66: - ++m_cursor; - if (m_limit <= m_cursor) - { - fill_line_buffer(1); // LCOV_EXCL_LINE - } - yych = *m_cursor; - if (yych <= '@') - { - if (yych <= '/') - { - goto basic_json_parser_32; - } - if (yych <= '9') - { - goto basic_json_parser_30; - } - goto basic_json_parser_32; - } - else - { - if (yych <= 'F') - { - goto basic_json_parser_30; - } - if (yych <= '`') - { - goto basic_json_parser_32; - } - if (yych <= 'f') - { - goto basic_json_parser_30; - } - goto basic_json_parser_32; - } - } - - } - - return last_token_type; - } - - /*! - @brief append data from the stream to the line buffer - - This function is called by the scan() function when the end of the - buffer (`m_limit`) is reached and the `m_cursor` pointer cannot be - incremented without leaving the limits of the line buffer. Note re2c - decides when to call this function. - - If the lexer reads from contiguous storage, there is no trailing null - byte. Therefore, this function must make sure to add these padding - null bytes. - - If the lexer reads from an input stream, this function reads the next - line of the input. - - @pre - p p p p p p u u u u u x . . . . . . - ^ ^ ^ ^ - m_content m_start | m_limit - m_cursor - - @post - u u u u u x x x x x x x . . . . . . - ^ ^ ^ - | m_cursor m_limit - m_start - m_content - */ - void fill_line_buffer(size_t n = 0) - { - // if line buffer is used, m_content points to its data - assert(m_line_buffer.empty() - or m_content == reinterpret_cast(m_line_buffer.data())); - - // if line buffer is used, m_limit is set past the end of its data - assert(m_line_buffer.empty() - or m_limit == m_content + m_line_buffer.size()); - - // pointer relationships - assert(m_content <= m_start); - assert(m_start <= m_cursor); - assert(m_cursor <= m_limit); - assert(m_marker == nullptr or m_marker <= m_limit); - - // number of processed characters (p) - const size_t num_processed_chars = static_cast(m_start - m_content); - // offset for m_marker wrt. to m_start - const auto offset_marker = (m_marker == nullptr) ? 0 : m_marker - m_start; - // number of unprocessed characters (u) - const auto offset_cursor = m_cursor - m_start; - - // no stream is used or end of file is reached - if (m_stream == nullptr or m_stream->eof()) - { - // m_start may or may not be pointing into m_line_buffer at - // this point. We trust the standand library to do the right - // thing. See http://stackoverflow.com/q/28142011/266378 - m_line_buffer.assign(m_start, m_limit); - - // append n characters to make sure that there is sufficient - // space between m_cursor and m_limit - m_line_buffer.append(1, '\x00'); - if (n > 0) - { - m_line_buffer.append(n - 1, '\x01'); - } - } - else - { - // delete processed characters from line buffer - m_line_buffer.erase(0, num_processed_chars); - // read next line from input stream - m_line_buffer_tmp.clear(); - std::getline(*m_stream, m_line_buffer_tmp, '\n'); - - // add line with newline symbol to the line buffer - m_line_buffer += m_line_buffer_tmp; - m_line_buffer.push_back('\n'); - } - - // set pointers - m_content = reinterpret_cast(m_line_buffer.data()); - assert(m_content != nullptr); - m_start = m_content; - m_marker = m_start + offset_marker; - m_cursor = m_start + offset_cursor; - m_limit = m_start + m_line_buffer.size(); - } - - /// return string representation of last read token - string_t get_token_string() const - { - assert(m_start != nullptr); - return string_t(reinterpret_cast(m_start), - static_cast(m_cursor - m_start)); - } - - /*! - @brief return string value for string tokens - - The function iterates the characters between the opening and closing - quotes of the string value. The complete string is the range - [m_start,m_cursor). Consequently, we iterate from m_start+1 to - m_cursor-1. - - We differentiate two cases: - - 1. Escaped characters. In this case, a new character is constructed - according to the nature of the escape. Some escapes create new - characters (e.g., `"\\n"` is replaced by `"\n"`), some are copied - as is (e.g., `"\\\\"`). Furthermore, Unicode escapes of the shape - `"\\uxxxx"` need special care. In this case, to_unicode takes care - of the construction of the values. - 2. Unescaped characters are copied as is. - - @pre `m_cursor - m_start >= 2`, meaning the length of the last token - is at least 2 bytes which is trivially true for any string (which - consists of at least two quotes). - - " c1 c2 c3 ... " - ^ ^ - m_start m_cursor - - @complexity Linear in the length of the string.\n - - Lemma: The loop body will always terminate.\n - - Proof (by contradiction): Assume the loop body does not terminate. As - the loop body does not contain another loop, one of the called - functions must never return. The called functions are `std::strtoul` - and to_unicode. Neither function can loop forever, so the loop body - will never loop forever which contradicts the assumption that the loop - body does not terminate, q.e.d.\n - - Lemma: The loop condition for the for loop is eventually false.\n - - Proof (by contradiction): Assume the loop does not terminate. Due to - the above lemma, this can only be due to a tautological loop - condition; that is, the loop condition i < m_cursor - 1 must always be - true. Let x be the change of i for any loop iteration. Then - m_start + 1 + x < m_cursor - 1 must hold to loop indefinitely. This - can be rephrased to m_cursor - m_start - 2 > x. With the - precondition, we x <= 0, meaning that the loop condition holds - indefinitly if i is always decreased. However, observe that the value - of i is strictly increasing with each iteration, as it is incremented - by 1 in the iteration expression and never decremented inside the loop - body. Hence, the loop condition will eventually be false which - contradicts the assumption that the loop condition is a tautology, - q.e.d. - - @return string value of current token without opening and closing - quotes - @throw std::out_of_range if to_unicode fails - */ - string_t get_string() const - { - assert(m_cursor - m_start >= 2); - - string_t result; - result.reserve(static_cast(m_cursor - m_start - 2)); - - // iterate the result between the quotes - for (const lexer_char_t* i = m_start + 1; i < m_cursor - 1; ++i) - { - // find next escape character - auto e = std::find(i, m_cursor - 1, '\\'); - if (e != i) - { - // see https://github.com/nlohmann/json/issues/365#issuecomment-262874705 - for (auto k = i; k < e; k++) - { - result.push_back(static_cast(*k)); - } - i = e - 1; // -1 because of ++i - } - else - { - // processing escaped character - // read next character - ++i; - - switch (*i) - { - // the default escapes - case 't': - { - result += "\t"; - break; - } - case 'b': - { - result += "\b"; - break; - } - case 'f': - { - result += "\f"; - break; - } - case 'n': - { - result += "\n"; - break; - } - case 'r': - { - result += "\r"; - break; - } - case '\\': - { - result += "\\"; - break; - } - case '/': - { - result += "/"; - break; - } - case '"': - { - result += "\""; - break; - } - - // unicode - case 'u': - { - // get code xxxx from uxxxx - auto codepoint = std::strtoul(std::string(reinterpret_cast(i + 1), - 4).c_str(), nullptr, 16); - - // check if codepoint is a high surrogate - if (codepoint >= 0xD800 and codepoint <= 0xDBFF) - { - // make sure there is a subsequent unicode - if ((i + 6 >= m_limit) or * (i + 5) != '\\' or * (i + 6) != 'u') - { - throw std::invalid_argument("missing low surrogate"); - } - - // get code yyyy from uxxxx\uyyyy - auto codepoint2 = std::strtoul(std::string(reinterpret_cast - (i + 7), 4).c_str(), nullptr, 16); - result += to_unicode(codepoint, codepoint2); - // skip the next 10 characters (xxxx\uyyyy) - i += 10; - } - else if (codepoint >= 0xDC00 and codepoint <= 0xDFFF) - { - // we found a lone low surrogate - throw std::invalid_argument("missing high surrogate"); - } - else - { - // add unicode character(s) - result += to_unicode(codepoint); - // skip the next four characters (xxxx) - i += 4; - } - break; - } - } - } - } - - return result; - } - - /*! - @brief parse floating point number - - This function (and its overloads) serves to select the most approprate - standard floating point number parsing function based on the type - supplied via the first parameter. Set this to @a - static_cast(nullptr). - - @param[in,out] endptr recieves a pointer to the first character after - the number - - @return the floating point number - */ - long double str_to_float_t(long double* /* type */, char** endptr) const - { - return std::strtold(reinterpret_cast(m_start), endptr); - } - - /*! - @brief parse floating point number - - This function (and its overloads) serves to select the most approprate - standard floating point number parsing function based on the type - supplied via the first parameter. Set this to @a - static_cast(nullptr). - - @param[in,out] endptr recieves a pointer to the first character after - the number - - @return the floating point number - */ - double str_to_float_t(double* /* type */, char** endptr) const - { - return std::strtod(reinterpret_cast(m_start), endptr); - } - - /*! - @brief parse floating point number - - This function (and its overloads) serves to select the most approprate - standard floating point number parsing function based on the type - supplied via the first parameter. Set this to @a - static_cast(nullptr). - - @param[in,out] endptr recieves a pointer to the first character after - the number - - @return the floating point number - */ - float str_to_float_t(float* /* type */, char** endptr) const - { - return std::strtof(reinterpret_cast(m_start), endptr); - } - - /*! - @brief return number value for number tokens - - This function translates the last token into the most appropriate - number type (either integer, unsigned integer or floating point), - which is passed back to the caller via the result parameter. - - This function parses the integer component up to the radix point or - exponent while collecting information about the 'floating point - representation', which it stores in the result parameter. If there is - no radix point or exponent, and the number can fit into a @ref - number_integer_t or @ref number_unsigned_t then it sets the result - parameter accordingly. - - If the number is a floating point number the number is then parsed - using @a std:strtod (or @a std:strtof or @a std::strtold). - - @param[out] result @ref basic_json object to receive the number, or - NAN if the conversion read past the current token. The latter case - needs to be treated by the caller function. - */ - void get_number(basic_json& result) const - { - assert(m_start != nullptr); - - const lexer::lexer_char_t* curptr = m_start; - - // accumulate the integer conversion result (unsigned for now) - number_unsigned_t value = 0; - - // maximum absolute value of the relevant integer type - number_unsigned_t max; - - // temporarily store the type to avoid unecessary bitfield access - value_t type; - - // look for sign - if (*curptr == '-') - { - type = value_t::number_integer; - max = static_cast((std::numeric_limits::max)()) + 1; - curptr++; - } - else - { - type = value_t::number_unsigned; - max = static_cast((std::numeric_limits::max)()); - } - - // count the significant figures - for (; curptr < m_cursor; curptr++) - { - // quickly skip tests if a digit - if (*curptr < '0' || *curptr > '9') - { - if (*curptr == '.') - { - // don't count '.' but change to float - type = value_t::number_float; - continue; - } - // assume exponent (if not then will fail parse): change to - // float, stop counting and record exponent details - type = value_t::number_float; - break; - } - - // skip if definitely not an integer - if (type != value_t::number_float) - { - auto digit = static_cast(*curptr - '0'); - - // overflow if value * 10 + digit > max, move terms around - // to avoid overflow in intermediate values - if (value > (max - digit) / 10) - { - // overflow - type = value_t::number_float; - } - else - { - // no overflow - value = value * 10 + digit; - } - } - } - - // save the value (if not a float) - if (type == value_t::number_unsigned) - { - result.m_value.number_unsigned = value; - } - else if (type == value_t::number_integer) - { - // invariant: if we parsed a '-', the absolute value is between - // 0 (we allow -0) and max == -INT64_MIN - assert(value >= 0); - assert(value <= max); - - if (value == max) - { - // we cannot simply negate value (== max == -INT64_MIN), - // see https://github.com/nlohmann/json/issues/389 - result.m_value.number_integer = static_cast(INT64_MIN); - } - else - { - // all other values can be negated safely - result.m_value.number_integer = -static_cast(value); - } - } - else - { - // parse with strtod - result.m_value.number_float = str_to_float_t(static_cast(nullptr), NULL); - - // replace infinity and NAN by null - if (not std::isfinite(result.m_value.number_float)) - { - type = value_t::null; - result.m_value = basic_json::json_value(); - } - } - - // save the type - result.m_type = type; - } - - private: - /// optional input stream - std::istream* m_stream = nullptr; - /// line buffer buffer for m_stream - string_t m_line_buffer {}; - /// used for filling m_line_buffer - string_t m_line_buffer_tmp {}; - /// the buffer pointer - const lexer_char_t* m_content = nullptr; - /// pointer to the beginning of the current symbol - const lexer_char_t* m_start = nullptr; - /// pointer for backtracking information - const lexer_char_t* m_marker = nullptr; - /// pointer to the current symbol - const lexer_char_t* m_cursor = nullptr; - /// pointer to the end of the buffer - const lexer_char_t* m_limit = nullptr; - /// the last token type - token_type last_token_type = token_type::end_of_input; - }; + return binary_reader(i).parse_cbor(strict); + } /*! - @brief syntax analysis - - This class implements a recursive decent parser. + @copydoc from_cbor(detail::input_adapter, const bool) */ - class parser + template::value, int> = 0> + static basic_json from_cbor(A1 && a1, A2 && a2, const bool strict = true) { - public: - /// a parser reading from a string literal - parser(const char* buff, const parser_callback_t cb = nullptr) - : callback(cb), - m_lexer(reinterpret_cast(buff), std::strlen(buff)) - {} + return binary_reader(detail::input_adapter(std::forward(a1), std::forward(a2))).parse_cbor(strict); + } - /// a parser reading from an input stream - parser(std::istream& is, const parser_callback_t cb = nullptr) - : callback(cb), m_lexer(is) - {} - - /// a parser reading from an iterator range with contiguous storage - template::iterator_category, std::random_access_iterator_tag>::value - , int>::type - = 0> - parser(IteratorType first, IteratorType last, const parser_callback_t cb = nullptr) - : callback(cb), - m_lexer(reinterpret_cast(&(*first)), - static_cast(std::distance(first, last))) - {} - - /// public parser interface - basic_json parse() - { - // read first token - get_token(); - - basic_json result = parse_internal(true); - result.assert_invariant(); - - expect(lexer::token_type::end_of_input); - - // return parser result and replace it with null in case the - // top-level value was discarded by the callback function - return result.is_discarded() ? basic_json() : std::move(result); - } - - private: - /// the actual parser - basic_json parse_internal(bool keep) - { - auto result = basic_json(value_t::discarded); - - switch (last_token) - { - case lexer::token_type::begin_object: - { - if (keep and (not callback - or ((keep = callback(depth++, parse_event_t::object_start, result)) != 0))) - { - // explicitly set result to object to cope with {} - result.m_type = value_t::object; - result.m_value = value_t::object; - } - - // read next token - get_token(); - - // closing } -> we are done - if (last_token == lexer::token_type::end_object) - { - get_token(); - if (keep and callback and not callback(--depth, parse_event_t::object_end, result)) - { - result = basic_json(value_t::discarded); - } - return result; - } - - // no comma is expected here - unexpect(lexer::token_type::value_separator); - - // otherwise: parse key-value pairs - do - { - // ugly, but could be fixed with loop reorganization - if (last_token == lexer::token_type::value_separator) - { - get_token(); - } - - // store key - expect(lexer::token_type::value_string); - const auto key = m_lexer.get_string(); - - bool keep_tag = false; - if (keep) - { - if (callback) - { - basic_json k(key); - keep_tag = callback(depth, parse_event_t::key, k); - } - else - { - keep_tag = true; - } - } - - // parse separator (:) - get_token(); - expect(lexer::token_type::name_separator); - - // parse and add value - get_token(); - auto value = parse_internal(keep); - if (keep and keep_tag and not value.is_discarded()) - { - result[key] = std::move(value); - } - } - while (last_token == lexer::token_type::value_separator); - - // closing } - expect(lexer::token_type::end_object); - get_token(); - if (keep and callback and not callback(--depth, parse_event_t::object_end, result)) - { - result = basic_json(value_t::discarded); - } - - return result; - } - - case lexer::token_type::begin_array: - { - if (keep and (not callback - or ((keep = callback(depth++, parse_event_t::array_start, result)) != 0))) - { - // explicitly set result to object to cope with [] - result.m_type = value_t::array; - result.m_value = value_t::array; - } - - // read next token - get_token(); - - // closing ] -> we are done - if (last_token == lexer::token_type::end_array) - { - get_token(); - if (callback and not callback(--depth, parse_event_t::array_end, result)) - { - result = basic_json(value_t::discarded); - } - return result; - } - - // no comma is expected here - unexpect(lexer::token_type::value_separator); - - // otherwise: parse values - do - { - // ugly, but could be fixed with loop reorganization - if (last_token == lexer::token_type::value_separator) - { - get_token(); - } - - // parse value - auto value = parse_internal(keep); - if (keep and not value.is_discarded()) - { - result.push_back(std::move(value)); - } - } - while (last_token == lexer::token_type::value_separator); - - // closing ] - expect(lexer::token_type::end_array); - get_token(); - if (keep and callback and not callback(--depth, parse_event_t::array_end, result)) - { - result = basic_json(value_t::discarded); - } - - return result; - } - - case lexer::token_type::literal_null: - { - get_token(); - result.m_type = value_t::null; - break; - } - - case lexer::token_type::value_string: - { - const auto s = m_lexer.get_string(); - get_token(); - result = basic_json(s); - break; - } - - case lexer::token_type::literal_true: - { - get_token(); - result.m_type = value_t::boolean; - result.m_value = true; - break; - } - - case lexer::token_type::literal_false: - { - get_token(); - result.m_type = value_t::boolean; - result.m_value = false; - break; - } - - case lexer::token_type::value_number: - { - m_lexer.get_number(result); - get_token(); - break; - } - - default: - { - // the last token was unexpected - unexpect(last_token); - } - } - - if (keep and callback and not callback(depth, parse_event_t::value, result)) - { - result = basic_json(value_t::discarded); - } - return result; - } - - /// get next token from lexer - typename lexer::token_type get_token() - { - last_token = m_lexer.scan(); - return last_token; - } - - void expect(typename lexer::token_type t) const - { - if (t != last_token) - { - std::string error_msg = "parse error - unexpected "; - error_msg += (last_token == lexer::token_type::parse_error ? ("'" + m_lexer.get_token_string() + - "'") : - lexer::token_type_name(last_token)); - error_msg += "; expected " + lexer::token_type_name(t); - throw std::invalid_argument(error_msg); - } - } - - void unexpect(typename lexer::token_type t) const - { - if (t == last_token) - { - std::string error_msg = "parse error - unexpected "; - error_msg += (last_token == lexer::token_type::parse_error ? ("'" + m_lexer.get_token_string() + - "'") : - lexer::token_type_name(last_token)); - throw std::invalid_argument(error_msg); - } - } - - private: - /// current level of recursion - int depth = 0; - /// callback function - const parser_callback_t callback = nullptr; - /// the type of the last read token - typename lexer::token_type last_token = lexer::token_type::uninitialized; - /// the lexer - lexer m_lexer; - }; - - public: /*! - @brief JSON Pointer + @brief create a JSON value from an input in MessagePack format - A JSON pointer defines a string syntax for identifying a specific value - within a JSON document. It can be used with functions `at` and - `operator[]`. Furthermore, JSON pointers are the base for JSON patches. + Deserializes a given input @a i to a JSON value using the MessagePack + serialization format. - @sa [RFC 6901](https://tools.ietf.org/html/rfc6901) + The library maps MessagePack types to JSON value types as follows: - @since version 2.0.0 + MessagePack type | JSON value type | first byte + ---------------- | --------------- | ---------- + positive fixint | number_unsigned | 0x00..0x7f + fixmap | object | 0x80..0x8f + fixarray | array | 0x90..0x9f + fixstr | string | 0xa0..0xbf + nil | `null` | 0xc0 + false | `false` | 0xc2 + true | `true` | 0xc3 + float 32 | number_float | 0xca + float 64 | number_float | 0xcb + uint 8 | number_unsigned | 0xcc + uint 16 | number_unsigned | 0xcd + uint 32 | number_unsigned | 0xce + uint 64 | number_unsigned | 0xcf + int 8 | number_integer | 0xd0 + int 16 | number_integer | 0xd1 + int 32 | number_integer | 0xd2 + int 64 | number_integer | 0xd3 + str 8 | string | 0xd9 + str 16 | string | 0xda + str 32 | string | 0xdb + array 16 | array | 0xdc + array 32 | array | 0xdd + map 16 | object | 0xde + map 32 | object | 0xdf + negative fixint | number_integer | 0xe0-0xff + + @warning The mapping is **incomplete** in the sense that not all + MessagePack types can be converted to a JSON value. The following + MessagePack types are not supported and will yield parse errors: + - bin 8 - bin 32 (0xc4..0xc6) + - ext 8 - ext 32 (0xc7..0xc9) + - fixext 1 - fixext 16 (0xd4..0xd8) + + @note Any MessagePack output created @ref to_msgpack can be successfully + parsed by @ref from_msgpack. + + @param[in] i an input in MessagePack format convertible to an input + adapter + @param[in] strict whether to expect the input to be consumed until EOF + (true by default) + + @throw parse_error.110 if the given input ends prematurely or the end of + file was not reached when @a strict was set to true + @throw parse_error.112 if unsupported features from MessagePack were + used in the given input @a i or if the input is not valid MessagePack + @throw parse_error.113 if a string was expected as map key, but not found + + @complexity Linear in the size of the input @a i. + + @liveexample{The example shows the deserialization of a byte vector in + MessagePack format to a JSON value.,from_msgpack} + + @sa http://msgpack.org + @sa @ref to_msgpack(const basic_json&) for the analogous serialization + @sa @ref from_cbor(detail::input_adapter, const bool) for the related CBOR + format + + @since version 2.0.9; parameter @a start_index since 2.1.1; changed to + consume input adapters, removed start_index parameter, and added + @a strict parameter since 3.0.0 */ - class json_pointer + static basic_json from_msgpack(detail::input_adapter i, + const bool strict = true) { - /// allow basic_json to access private members - friend class basic_json; - - public: - /*! - @brief create JSON pointer - - Create a JSON pointer according to the syntax described in - [Section 3 of RFC6901](https://tools.ietf.org/html/rfc6901#section-3). - - @param[in] s string representing the JSON pointer; if omitted, the - empty string is assumed which references the whole JSON - value - - @throw std::domain_error if reference token is nonempty and does not - begin with a slash (`/`); example: `"JSON pointer must be empty or - begin with /"` - @throw std::domain_error if a tilde (`~`) is not followed by `0` - (representing `~`) or `1` (representing `/`); example: `"escape error: - ~ must be followed with 0 or 1"` - - @liveexample{The example shows the construction several valid JSON - pointers as well as the exceptional behavior.,json_pointer} - - @since version 2.0.0 - */ - explicit json_pointer(const std::string& s = "") - : reference_tokens(split(s)) - {} - - /*! - @brief return a string representation of the JSON pointer - - @invariant For each JSON pointer `ptr`, it holds: - @code {.cpp} - ptr == json_pointer(ptr.to_string()); - @endcode - - @return a string representation of the JSON pointer - - @liveexample{The example shows the result of `to_string`., - json_pointer__to_string} - - @since version 2.0.0 - */ - std::string to_string() const noexcept - { - return std::accumulate(reference_tokens.begin(), - reference_tokens.end(), std::string{}, - [](const std::string & a, const std::string & b) - { - return a + "/" + escape(b); - }); - } - - /// @copydoc to_string() - operator std::string() const - { - return to_string(); - } - - private: - /// remove and return last reference pointer - std::string pop_back() - { - if (is_root()) - { - throw std::domain_error("JSON pointer has no parent"); - } - - auto last = reference_tokens.back(); - reference_tokens.pop_back(); - return last; - } - - /// return whether pointer points to the root document - bool is_root() const - { - return reference_tokens.empty(); - } - - json_pointer top() const - { - if (is_root()) - { - throw std::domain_error("JSON pointer has no parent"); - } - - json_pointer result = *this; - result.reference_tokens = {reference_tokens[0]}; - return result; - } - - /*! - @brief create and return a reference to the pointed to value - - @complexity Linear in the number of reference tokens. - */ - reference get_and_create(reference j) const - { - pointer result = &j; - - // in case no reference tokens exist, return a reference to the - // JSON value j which will be overwritten by a primitive value - for (const auto& reference_token : reference_tokens) - { - switch (result->m_type) - { - case value_t::null: - { - if (reference_token == "0") - { - // start a new array if reference token is 0 - result = &result->operator[](0); - } - else - { - // start a new object otherwise - result = &result->operator[](reference_token); - } - break; - } - - case value_t::object: - { - // create an entry in the object - result = &result->operator[](reference_token); - break; - } - - case value_t::array: - { - // create an entry in the array - result = &result->operator[](static_cast(std::stoi(reference_token))); - break; - } - - /* - The following code is only reached if there exists a - reference token _and_ the current value is primitive. In - this case, we have an error situation, because primitive - values may only occur as single value; that is, with an - empty list of reference tokens. - */ - default: - { - throw std::domain_error("invalid value to unflatten"); - } - } - } - - return *result; - } - - /*! - @brief return a reference to the pointed to value - - @note This version does not throw if a value is not present, but tries - to create nested values instead. For instance, calling this function - with pointer `"/this/that"` on a null value is equivalent to calling - `operator[]("this").operator[]("that")` on that value, effectively - changing the null value to an object. - - @param[in] ptr a JSON value - - @return reference to the JSON value pointed to by the JSON pointer - - @complexity Linear in the length of the JSON pointer. - - @throw std::out_of_range if the JSON pointer can not be resolved - @throw std::domain_error if an array index begins with '0' - @throw std::invalid_argument if an array index was not a number - */ - reference get_unchecked(pointer ptr) const - { - for (const auto& reference_token : reference_tokens) - { - // convert null values to arrays or objects before continuing - if (ptr->m_type == value_t::null) - { - // check if reference token is a number - const bool nums = std::all_of(reference_token.begin(), - reference_token.end(), - [](const char x) - { - return std::isdigit(x); - }); - - // change value to array for numbers or "-" or to object - // otherwise - if (nums or reference_token == "-") - { - *ptr = value_t::array; - } - else - { - *ptr = value_t::object; - } - } - - switch (ptr->m_type) - { - case value_t::object: - { - // use unchecked object access - ptr = &ptr->operator[](reference_token); - break; - } - - case value_t::array: - { - // error condition (cf. RFC 6901, Sect. 4) - if (reference_token.size() > 1 and reference_token[0] == '0') - { - throw std::domain_error("array index must not begin with '0'"); - } - - if (reference_token == "-") - { - // explicityly treat "-" as index beyond the end - ptr = &ptr->operator[](ptr->m_value.array->size()); - } - else - { - // convert array index to number; unchecked access - ptr = &ptr->operator[](static_cast(std::stoi(reference_token))); - } - break; - } - - default: - { - throw std::out_of_range("unresolved reference token '" + reference_token + "'"); - } - } - } - - return *ptr; - } - - reference get_checked(pointer ptr) const - { - for (const auto& reference_token : reference_tokens) - { - switch (ptr->m_type) - { - case value_t::object: - { - // note: at performs range check - ptr = &ptr->at(reference_token); - break; - } - - case value_t::array: - { - if (reference_token == "-") - { - // "-" always fails the range check - throw std::out_of_range("array index '-' (" + - std::to_string(ptr->m_value.array->size()) + - ") is out of range"); - } - - // error condition (cf. RFC 6901, Sect. 4) - if (reference_token.size() > 1 and reference_token[0] == '0') - { - throw std::domain_error("array index must not begin with '0'"); - } - - // note: at performs range check - ptr = &ptr->at(static_cast(std::stoi(reference_token))); - break; - } - - default: - { - throw std::out_of_range("unresolved reference token '" + reference_token + "'"); - } - } - } - - return *ptr; - } - - /*! - @brief return a const reference to the pointed to value - - @param[in] ptr a JSON value - - @return const reference to the JSON value pointed to by the JSON - pointer - */ - const_reference get_unchecked(const_pointer ptr) const - { - for (const auto& reference_token : reference_tokens) - { - switch (ptr->m_type) - { - case value_t::object: - { - // use unchecked object access - ptr = &ptr->operator[](reference_token); - break; - } - - case value_t::array: - { - if (reference_token == "-") - { - // "-" cannot be used for const access - throw std::out_of_range("array index '-' (" + - std::to_string(ptr->m_value.array->size()) + - ") is out of range"); - } - - // error condition (cf. RFC 6901, Sect. 4) - if (reference_token.size() > 1 and reference_token[0] == '0') - { - throw std::domain_error("array index must not begin with '0'"); - } - - // use unchecked array access - ptr = &ptr->operator[](static_cast(std::stoi(reference_token))); - break; - } - - default: - { - throw std::out_of_range("unresolved reference token '" + reference_token + "'"); - } - } - } - - return *ptr; - } - - const_reference get_checked(const_pointer ptr) const - { - for (const auto& reference_token : reference_tokens) - { - switch (ptr->m_type) - { - case value_t::object: - { - // note: at performs range check - ptr = &ptr->at(reference_token); - break; - } - - case value_t::array: - { - if (reference_token == "-") - { - // "-" always fails the range check - throw std::out_of_range("array index '-' (" + - std::to_string(ptr->m_value.array->size()) + - ") is out of range"); - } - - // error condition (cf. RFC 6901, Sect. 4) - if (reference_token.size() > 1 and reference_token[0] == '0') - { - throw std::domain_error("array index must not begin with '0'"); - } - - // note: at performs range check - ptr = &ptr->at(static_cast(std::stoi(reference_token))); - break; - } - - default: - { - throw std::out_of_range("unresolved reference token '" + reference_token + "'"); - } - } - } - - return *ptr; - } - - /// split the string input to reference tokens - static std::vector split(const std::string& reference_string) - { - std::vector result; - - // special case: empty reference string -> no reference tokens - if (reference_string.empty()) - { - return result; - } - - // check if nonempty reference string begins with slash - if (reference_string[0] != '/') - { - throw std::domain_error("JSON pointer must be empty or begin with '/'"); - } - - // extract the reference tokens: - // - slash: position of the last read slash (or end of string) - // - start: position after the previous slash - for ( - // search for the first slash after the first character - size_t slash = reference_string.find_first_of("/", 1), - // set the beginning of the first reference token - start = 1; - // we can stop if start == string::npos+1 = 0 - start != 0; - // set the beginning of the next reference token - // (will eventually be 0 if slash == std::string::npos) - start = slash + 1, - // find next slash - slash = reference_string.find_first_of("/", start)) - { - // use the text between the beginning of the reference token - // (start) and the last slash (slash). - auto reference_token = reference_string.substr(start, slash - start); - - // check reference tokens are properly escaped - for (size_t pos = reference_token.find_first_of("~"); - pos != std::string::npos; - pos = reference_token.find_first_of("~", pos + 1)) - { - assert(reference_token[pos] == '~'); - - // ~ must be followed by 0 or 1 - if (pos == reference_token.size() - 1 or - (reference_token[pos + 1] != '0' and - reference_token[pos + 1] != '1')) - { - throw std::domain_error("escape error: '~' must be followed with '0' or '1'"); - } - } - - // finally, store the reference token - unescape(reference_token); - result.push_back(reference_token); - } - - return result; - } - - private: - /*! - @brief replace all occurrences of a substring by another string - - @param[in,out] s the string to manipulate; changed so that all - occurrences of @a f are replaced with @a t - @param[in] f the substring to replace with @a t - @param[in] t the string to replace @a f - - @pre The search string @a f must not be empty. - - @since version 2.0.0 - */ - static void replace_substring(std::string& s, - const std::string& f, - const std::string& t) - { - assert(not f.empty()); - - for ( - size_t pos = s.find(f); // find first occurrence of f - pos != std::string::npos; // make sure f was found - s.replace(pos, f.size(), t), // replace with t - pos = s.find(f, pos + t.size()) // find next occurrence of f - ); - } - - /// escape tilde and slash - static std::string escape(std::string s) - { - // escape "~"" to "~0" and "/" to "~1" - replace_substring(s, "~", "~0"); - replace_substring(s, "/", "~1"); - return s; - } - - /// unescape tilde and slash - static void unescape(std::string& s) - { - // first transform any occurrence of the sequence '~1' to '/' - replace_substring(s, "~1", "/"); - // then transform any occurrence of the sequence '~0' to '~' - replace_substring(s, "~0", "~"); - } - - /*! - @param[in] reference_string the reference string to the current value - @param[in] value the value to consider - @param[in,out] result the result object to insert values to - - @note Empty objects or arrays are flattened to `null`. - */ - static void flatten(const std::string& reference_string, - const basic_json& value, - basic_json& result) - { - switch (value.m_type) - { - case value_t::array: - { - if (value.m_value.array->empty()) - { - // flatten empty array as null - result[reference_string] = nullptr; - } - else - { - // iterate array and use index as reference string - for (size_t i = 0; i < value.m_value.array->size(); ++i) - { - flatten(reference_string + "/" + std::to_string(i), - value.m_value.array->operator[](i), result); - } - } - break; - } - - case value_t::object: - { - if (value.m_value.object->empty()) - { - // flatten empty object as null - result[reference_string] = nullptr; - } - else - { - // iterate object and use keys as reference string - for (const auto& element : *value.m_value.object) - { - flatten(reference_string + "/" + escape(element.first), - element.second, result); - } - } - break; - } - - default: - { - // add primitive value with its reference string - result[reference_string] = value; - break; - } - } - } - - /*! - @param[in] value flattened JSON - - @return unflattened JSON - */ - static basic_json unflatten(const basic_json& value) - { - if (not value.is_object()) - { - throw std::domain_error("only objects can be unflattened"); - } - - basic_json result; - - // iterate the JSON object values - for (const auto& element : *value.m_value.object) - { - if (not element.second.is_primitive()) - { - throw std::domain_error("values in object must be primitive"); - } - - // assign value to reference pointed to by JSON pointer; Note - // that if the JSON pointer is "" (i.e., points to the whole - // value), function get_and_create returns a reference to - // result itself. An assignment will then create a primitive - // value. - json_pointer(element.first).get_and_create(result) = element.second; - } - - return result; - } - - private: - /// the reference tokens - std::vector reference_tokens {}; - }; + return binary_reader(i).parse_msgpack(strict); + } + + /*! + @copydoc from_msgpack(detail::input_adapter, const bool) + */ + template::value, int> = 0> + static basic_json from_msgpack(A1 && a1, A2 && a2, const bool strict = true) + { + return binary_reader(detail::input_adapter(std::forward(a1), std::forward(a2))).parse_msgpack(strict); + } + + /// @} ////////////////////////// // JSON Pointer support // @@ -11551,9 +13564,9 @@ basic_json_parser_66: @complexity Constant. - @throw std::out_of_range if the JSON pointer can not be resolved - @throw std::domain_error if an array index begins with '0' - @throw std::invalid_argument if an array index was not a number + @throw parse_error.106 if an array index begins with '0' + @throw parse_error.109 if an array index was not a number + @throw out_of_range.404 if the JSON pointer can not be resolved @liveexample{The behavior is shown in the example.,operatorjson_pointer} @@ -11578,9 +13591,10 @@ basic_json_parser_66: @complexity Constant. - @throw std::out_of_range if the JSON pointer can not be resolved - @throw std::domain_error if an array index begins with '0' - @throw std::invalid_argument if an array index was not a number + @throw parse_error.106 if an array index begins with '0' + @throw parse_error.109 if an array index was not a number + @throw out_of_range.402 if the array index '-' is used + @throw out_of_range.404 if the JSON pointer can not be resolved @liveexample{The behavior is shown in the example.,operatorjson_pointer_const} @@ -11601,15 +13615,30 @@ basic_json_parser_66: @return reference to the element pointed to by @a ptr + @throw parse_error.106 if an array index in the passed JSON pointer @a ptr + begins with '0'. See example below. + + @throw parse_error.109 if an array index in the passed JSON pointer @a ptr + is not a number. See example below. + + @throw out_of_range.401 if an array index in the passed JSON pointer @a ptr + is out of range. See example below. + + @throw out_of_range.402 if the array index '-' is used in the passed JSON + pointer @a ptr. As `at` provides checked access (and no elements are + implicitly inserted), the index '-' is always invalid. See example below. + + @throw out_of_range.404 if the JSON pointer @a ptr can not be resolved. + See example below. + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. + @complexity Constant. - @throw std::out_of_range if the JSON pointer can not be resolved - @throw std::domain_error if an array index begins with '0' - @throw std::invalid_argument if an array index was not a number + @since version 2.0.0 @liveexample{The behavior is shown in the example.,at_json_pointer} - - @since version 2.0.0 */ reference at(const json_pointer& ptr) { @@ -11626,15 +13655,30 @@ basic_json_parser_66: @return reference to the element pointed to by @a ptr + @throw parse_error.106 if an array index in the passed JSON pointer @a ptr + begins with '0'. See example below. + + @throw parse_error.109 if an array index in the passed JSON pointer @a ptr + is not a number. See example below. + + @throw out_of_range.401 if an array index in the passed JSON pointer @a ptr + is out of range. See example below. + + @throw out_of_range.402 if the array index '-' is used in the passed JSON + pointer @a ptr. As `at` provides checked access (and no elements are + implicitly inserted), the index '-' is always invalid. See example below. + + @throw out_of_range.404 if the JSON pointer @a ptr can not be resolved. + See example below. + + @exceptionsafety Strong guarantee: if an exception is thrown, there are no + changes in the JSON value. + @complexity Constant. - @throw std::out_of_range if the JSON pointer can not be resolved - @throw std::domain_error if an array index begins with '0' - @throw std::invalid_argument if an array index was not a number + @since version 2.0.0 @liveexample{The behavior is shown in the example.,at_json_pointer_const} - - @since version 2.0.0 */ const_reference at(const json_pointer& ptr) const { @@ -11649,7 +13693,7 @@ basic_json_parser_66: primitive. The original JSON value can be restored using the @ref unflatten() function. - @return an object that maps JSON pointers to primitve values + @return an object that maps JSON pointers to primitive values @note Empty objects and arrays are flattened to `null` and will not be reconstructed correctly by the @ref unflatten() function. @@ -11690,6 +13734,9 @@ basic_json_parser_66: @complexity Linear in the size the JSON value. + @throw type_error.314 if value is not an object + @throw type_error.315 if object values are not primitive + @liveexample{The following code shows how a flattened JSON object is unflattened into the original nested JSON object.,unflatten} @@ -11716,7 +13763,7 @@ basic_json_parser_66: [JSON Patch](http://jsonpatch.com) defines a JSON document structure for expressing a sequence of operations to apply to a JSON) document. With - this funcion, a JSON Patch is applied to the current JSON value by + this function, a JSON Patch is applied to the current JSON value by executing all operations from the patch. @param[in] json_patch JSON patch document @@ -11727,12 +13774,23 @@ basic_json_parser_66: any case, the original value is not changed: the patch is applied to a copy of the value. - @throw std::out_of_range if a JSON pointer inside the patch could not - be resolved successfully in the current JSON value; example: `"key baz - not found"` - @throw invalid_argument if the JSON patch is malformed (e.g., mandatory + @throw parse_error.104 if the JSON patch does not consist of an array of + objects + + @throw parse_error.105 if the JSON patch is malformed (e.g., mandatory attributes are missing); example: `"operation add must have member path"` + @throw out_of_range.401 if an array index is out of range. + + @throw out_of_range.403 if a JSON pointer inside the patch could not be + resolved successfully in the current JSON value; example: `"key baz not + found"` + + @throw out_of_range.405 if JSON pointer has no parent ("add", "remove", + "move") + + @throw other_error.501 if "test" operation was unsuccessful + @complexity Linear in the size of the JSON value and the length of the JSON patch. As usually only a fraction of the JSON value is affected by the patch, the complexity can usually be neglected. @@ -11755,7 +13813,7 @@ basic_json_parser_66: // the valid JSON Patch operations enum class patch_operations {add, remove, replace, move, copy, test, invalid}; - const auto get_op = [](const std::string op) + const auto get_op = [](const std::string & op) { if (op == "add") { @@ -11826,10 +13884,10 @@ basic_json_parser_66: else { const auto idx = std::stoi(last_path); - if (static_cast(idx) > parent.size()) + if (JSON_UNLIKELY(static_cast(idx) > parent.size())) { // avoid undefined behavior - throw std::out_of_range("array index " + std::to_string(idx) + " is out of range"); + JSON_THROW(out_of_range::create(401, "array index " + std::to_string(idx) + " is out of range")); } else { @@ -11861,13 +13919,13 @@ basic_json_parser_66: { // perform range check auto it = parent.find(last_path); - if (it != parent.end()) + if (JSON_LIKELY(it != parent.end())) { parent.erase(it); } else { - throw std::out_of_range("key '" + last_path + "' not found"); + JSON_THROW(out_of_range::create(403, "key '" + last_path + "' not found")); } } else if (parent.is_array()) @@ -11877,14 +13935,13 @@ basic_json_parser_66: } }; - // type check - if (not json_patch.is_array()) + // type check: top level value must be an array + if (JSON_UNLIKELY(not json_patch.is_array())) { - // a JSON patch must be an array of objects - throw std::invalid_argument("JSON patch must be an array of objects"); + JSON_THROW(parse_error::create(104, 0, "JSON patch must be an array of objects")); } - // iterate and apply th eoperations + // iterate and apply the operations for (const auto& val : json_patch) { // wrapper to get a value for an operation @@ -11899,25 +13956,25 @@ basic_json_parser_66: const auto error_msg = (op == "op") ? "operation" : "operation '" + op + "'"; // check if desired value is present - if (it == val.m_value.object->end()) + if (JSON_UNLIKELY(it == val.m_value.object->end())) { - throw std::invalid_argument(error_msg + " must have member '" + member + "'"); + JSON_THROW(parse_error::create(105, 0, error_msg + " must have member '" + member + "'")); } // check if result is of type string - if (string_type and not it->second.is_string()) + if (JSON_UNLIKELY(string_type and not it->second.is_string())) { - throw std::invalid_argument(error_msg + " must have string member '" + member + "'"); + JSON_THROW(parse_error::create(105, 0, error_msg + " must have string member '" + member + "'")); } // no error: return value return it->second; }; - // type check - if (not val.is_object()) + // type check: every element of the array must be an object + if (JSON_UNLIKELY(not val.is_object())) { - throw std::invalid_argument("JSON patch must be an array of objects"); + JSON_THROW(parse_error::create(104, 0, "JSON patch must be an array of objects")); } // collect mandatory members @@ -11965,7 +14022,7 @@ basic_json_parser_66: case patch_operations::copy: { - const std::string from_path = get_value("copy", "from", true);; + const std::string from_path = get_value("copy", "from", true); const json_pointer from_ptr(from_path); // the "from" location must exist - use at() @@ -11976,21 +14033,21 @@ basic_json_parser_66: case patch_operations::test: { bool success = false; - try + JSON_TRY { // check if "value" matches the one at "path" // the "path" location must exist - use at() success = (result.at(ptr) == get_value("test", "value", false)); } - catch (std::out_of_range&) + JSON_CATCH (out_of_range&) { // ignore out of range errors: success remains false } // throw an exception if test fails - if (not success) + if (JSON_UNLIKELY(not success)) { - throw std::domain_error("unsuccessful: " + val.dump()); + JSON_THROW(other_error::create(501, "unsuccessful: " + val.dump())); } break; @@ -12000,7 +14057,7 @@ basic_json_parser_66: { // op must be "add", "remove", "replace", "move", "copy", or // "test" - throw std::invalid_argument("operation value '" + op + "' is invalid"); + JSON_THROW(parse_error::create(105, 0, "operation value '" + op + "' is invalid")); } } } @@ -12023,8 +14080,8 @@ basic_json_parser_66: @note Currently, only `remove`, `add`, and `replace` operations are generated. - @param[in] source JSON value to copare from - @param[in] target JSON value to copare against + @param[in] source JSON value to compare from + @param[in] target JSON value to compare against @param[in] path helper value to create JSON pointers @return a JSON patch to convert the @a source to @a target @@ -12040,8 +14097,7 @@ basic_json_parser_66: @since version 2.0.0 */ - static basic_json diff(const basic_json& source, - const basic_json& target, + static basic_json diff(const basic_json& source, const basic_json& target, const std::string& path = "") { // the patch @@ -12058,9 +14114,7 @@ basic_json_parser_66: // different types: replace value result.push_back( { - {"op", "replace"}, - {"path", path}, - {"value", target} + {"op", "replace"}, {"path", path}, {"value", target} }); } else @@ -12070,7 +14124,7 @@ basic_json_parser_66: case value_t::array: { // first pass: traverse common elements - size_t i = 0; + std::size_t i = 0; while (i < source.size() and i < target.size()) { // recursive call to compare array values at index i @@ -12130,8 +14184,7 @@ basic_json_parser_66: // found a key that is not in o -> remove it result.push_back(object( { - {"op", "remove"}, - {"path", path + "/" + key} + {"op", "remove"}, {"path", path + "/" + key} })); } } @@ -12145,8 +14198,7 @@ basic_json_parser_66: const auto key = json_pointer::escape(it.key()); result.push_back( { - {"op", "add"}, - {"path", path + "/" + key}, + {"op", "add"}, {"path", path + "/" + key}, {"value", it.value()} }); } @@ -12160,9 +14212,7 @@ basic_json_parser_66: // both primitive type: replace value result.push_back( { - {"op", "replace"}, - {"path", path}, - {"value", target} + {"op", "replace"}, {"path", path}, {"value", target} }); break; } @@ -12175,7 +14225,6 @@ basic_json_parser_66: /// @} }; - ///////////// // presets // ///////////// @@ -12189,8 +14238,402 @@ uses the standard template types. @since version 1.0.0 */ using json = basic_json<>; + +////////////////// +// json_pointer // +////////////////// + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +NLOHMANN_BASIC_JSON_TPL& +json_pointer::get_and_create(NLOHMANN_BASIC_JSON_TPL& j) const +{ + using size_type = typename NLOHMANN_BASIC_JSON_TPL::size_type; + auto result = &j; + + // in case no reference tokens exist, return a reference to the JSON value + // j which will be overwritten by a primitive value + for (const auto& reference_token : reference_tokens) + { + switch (result->m_type) + { + case detail::value_t::null: + { + if (reference_token == "0") + { + // start a new array if reference token is 0 + result = &result->operator[](0); + } + else + { + // start a new object otherwise + result = &result->operator[](reference_token); + } + break; + } + + case detail::value_t::object: + { + // create an entry in the object + result = &result->operator[](reference_token); + break; + } + + case detail::value_t::array: + { + // create an entry in the array + JSON_TRY + { + result = &result->operator[](static_cast(std::stoi(reference_token))); + } + JSON_CATCH(std::invalid_argument&) + { + JSON_THROW(detail::parse_error::create(109, 0, "array index '" + reference_token + "' is not a number")); + } + break; + } + + /* + The following code is only reached if there exists a reference + token _and_ the current value is primitive. In this case, we have + an error situation, because primitive values may only occur as + single value; that is, with an empty list of reference tokens. + */ + default: + JSON_THROW(detail::type_error::create(313, "invalid value to unflatten")); + } + } + + return *result; } +NLOHMANN_BASIC_JSON_TPL_DECLARATION +NLOHMANN_BASIC_JSON_TPL& +json_pointer::get_unchecked(NLOHMANN_BASIC_JSON_TPL* ptr) const +{ + using size_type = typename NLOHMANN_BASIC_JSON_TPL::size_type; + for (const auto& reference_token : reference_tokens) + { + // convert null values to arrays or objects before continuing + if (ptr->m_type == detail::value_t::null) + { + // check if reference token is a number + const bool nums = + std::all_of(reference_token.begin(), reference_token.end(), + [](const char x) + { + return (x >= '0' and x <= '9'); + }); + + // change value to array for numbers or "-" or to object otherwise + *ptr = (nums or reference_token == "-") + ? detail::value_t::array + : detail::value_t::object; + } + + switch (ptr->m_type) + { + case detail::value_t::object: + { + // use unchecked object access + ptr = &ptr->operator[](reference_token); + break; + } + + case detail::value_t::array: + { + // error condition (cf. RFC 6901, Sect. 4) + if (JSON_UNLIKELY(reference_token.size() > 1 and reference_token[0] == '0')) + { + JSON_THROW(detail::parse_error::create(106, 0, + "array index '" + reference_token + + "' must not begin with '0'")); + } + + if (reference_token == "-") + { + // explicitly treat "-" as index beyond the end + ptr = &ptr->operator[](ptr->m_value.array->size()); + } + else + { + // convert array index to number; unchecked access + JSON_TRY + { + ptr = &ptr->operator[]( + static_cast(std::stoi(reference_token))); + } + JSON_CATCH(std::invalid_argument&) + { + JSON_THROW(detail::parse_error::create(109, 0, "array index '" + reference_token + "' is not a number")); + } + } + break; + } + + default: + JSON_THROW(detail::out_of_range::create(404, "unresolved reference token '" + reference_token + "'")); + } + } + + return *ptr; +} + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +NLOHMANN_BASIC_JSON_TPL& +json_pointer::get_checked(NLOHMANN_BASIC_JSON_TPL* ptr) const +{ + using size_type = typename NLOHMANN_BASIC_JSON_TPL::size_type; + for (const auto& reference_token : reference_tokens) + { + switch (ptr->m_type) + { + case detail::value_t::object: + { + // note: at performs range check + ptr = &ptr->at(reference_token); + break; + } + + case detail::value_t::array: + { + if (JSON_UNLIKELY(reference_token == "-")) + { + // "-" always fails the range check + JSON_THROW(detail::out_of_range::create(402, + "array index '-' (" + std::to_string(ptr->m_value.array->size()) + + ") is out of range")); + } + + // error condition (cf. RFC 6901, Sect. 4) + if (JSON_UNLIKELY(reference_token.size() > 1 and reference_token[0] == '0')) + { + JSON_THROW(detail::parse_error::create(106, 0, + "array index '" + reference_token + + "' must not begin with '0'")); + } + + // note: at performs range check + JSON_TRY + { + ptr = &ptr->at(static_cast(std::stoi(reference_token))); + } + JSON_CATCH(std::invalid_argument&) + { + JSON_THROW(detail::parse_error::create(109, 0, "array index '" + reference_token + "' is not a number")); + } + break; + } + + default: + JSON_THROW(detail::out_of_range::create(404, "unresolved reference token '" + reference_token + "'")); + } + } + + return *ptr; +} + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +const NLOHMANN_BASIC_JSON_TPL& +json_pointer::get_unchecked(const NLOHMANN_BASIC_JSON_TPL* ptr) const +{ + using size_type = typename NLOHMANN_BASIC_JSON_TPL::size_type; + for (const auto& reference_token : reference_tokens) + { + switch (ptr->m_type) + { + case detail::value_t::object: + { + // use unchecked object access + ptr = &ptr->operator[](reference_token); + break; + } + + case detail::value_t::array: + { + if (JSON_UNLIKELY(reference_token == "-")) + { + // "-" cannot be used for const access + JSON_THROW(detail::out_of_range::create(402, + "array index '-' (" + std::to_string(ptr->m_value.array->size()) + + ") is out of range")); + } + + // error condition (cf. RFC 6901, Sect. 4) + if (JSON_UNLIKELY(reference_token.size() > 1 and reference_token[0] == '0')) + { + JSON_THROW(detail::parse_error::create(106, 0, + "array index '" + reference_token + + "' must not begin with '0'")); + } + + // use unchecked array access + JSON_TRY + { + ptr = &ptr->operator[]( + static_cast(std::stoi(reference_token))); + } + JSON_CATCH(std::invalid_argument&) + { + JSON_THROW(detail::parse_error::create(109, 0, "array index '" + reference_token + "' is not a number")); + } + break; + } + + default: + JSON_THROW(detail::out_of_range::create(404, "unresolved reference token '" + reference_token + "'")); + } + } + + return *ptr; +} + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +const NLOHMANN_BASIC_JSON_TPL& +json_pointer::get_checked(const NLOHMANN_BASIC_JSON_TPL* ptr) const +{ + using size_type = typename NLOHMANN_BASIC_JSON_TPL::size_type; + for (const auto& reference_token : reference_tokens) + { + switch (ptr->m_type) + { + case detail::value_t::object: + { + // note: at performs range check + ptr = &ptr->at(reference_token); + break; + } + + case detail::value_t::array: + { + if (JSON_UNLIKELY(reference_token == "-")) + { + // "-" always fails the range check + JSON_THROW(detail::out_of_range::create(402, + "array index '-' (" + std::to_string(ptr->m_value.array->size()) + + ") is out of range")); + } + + // error condition (cf. RFC 6901, Sect. 4) + if (JSON_UNLIKELY(reference_token.size() > 1 and reference_token[0] == '0')) + { + JSON_THROW(detail::parse_error::create(106, 0, + "array index '" + reference_token + + "' must not begin with '0'")); + } + + // note: at performs range check + JSON_TRY + { + ptr = &ptr->at(static_cast(std::stoi(reference_token))); + } + JSON_CATCH(std::invalid_argument&) + { + JSON_THROW(detail::parse_error::create(109, 0, "array index '" + reference_token + "' is not a number")); + } + break; + } + + default: + JSON_THROW(detail::out_of_range::create(404, "unresolved reference token '" + reference_token + "'")); + } + } + + return *ptr; +} + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +void json_pointer::flatten(const std::string& reference_string, + const NLOHMANN_BASIC_JSON_TPL& value, + NLOHMANN_BASIC_JSON_TPL& result) +{ + switch (value.m_type) + { + case detail::value_t::array: + { + if (value.m_value.array->empty()) + { + // flatten empty array as null + result[reference_string] = nullptr; + } + else + { + // iterate array and use index as reference string + for (std::size_t i = 0; i < value.m_value.array->size(); ++i) + { + flatten(reference_string + "/" + std::to_string(i), + value.m_value.array->operator[](i), result); + } + } + break; + } + + case detail::value_t::object: + { + if (value.m_value.object->empty()) + { + // flatten empty object as null + result[reference_string] = nullptr; + } + else + { + // iterate object and use keys as reference string + for (const auto& element : *value.m_value.object) + { + flatten(reference_string + "/" + escape(element.first), element.second, result); + } + } + break; + } + + default: + { + // add primitive value with its reference string + result[reference_string] = value; + break; + } + } +} + +NLOHMANN_BASIC_JSON_TPL_DECLARATION +NLOHMANN_BASIC_JSON_TPL +json_pointer::unflatten(const NLOHMANN_BASIC_JSON_TPL& value) +{ + if (JSON_UNLIKELY(not value.is_object())) + { + JSON_THROW(detail::type_error::create(314, "only objects can be unflattened")); + } + + NLOHMANN_BASIC_JSON_TPL result; + + // iterate the JSON object values + for (const auto& element : *value.m_value.object) + { + if (JSON_UNLIKELY(not element.second.is_primitive())) + { + JSON_THROW(detail::type_error::create(315, "values in object must be primitive")); + } + + // assign value to reference pointed to by JSON pointer; Note that if + // the JSON pointer is "" (i.e., points to the whole value), function + // get_and_create returns a reference to result itself. An assignment + // will then create a primitive value. + json_pointer(element.first).get_and_create(result) = element.second; + } + + return result; +} + +inline bool operator==(json_pointer const& lhs, json_pointer const& rhs) noexcept +{ + return (lhs.reference_tokens == rhs.reference_tokens); +} + +inline bool operator!=(json_pointer const& lhs, json_pointer const& rhs) noexcept +{ + return not (lhs == rhs); +} +} // namespace nlohmann + /////////////////////// // nonmember support // @@ -12230,7 +14673,25 @@ struct hash return h(j.dump()); } }; -} + +/// specialization for std::less +/// @note: do not remove the space after '<', +/// see https://github.com/nlohmann/json/pull/679 +template<> +struct less< ::nlohmann::detail::value_t> +{ + /*! + @brief compare two value_t enum values + @since version 3.0.0 + */ + bool operator()(nlohmann::detail::value_t lhs, + nlohmann::detail::value_t rhs) const noexcept + { + return nlohmann::detail::operator<(lhs, rhs); + } +}; + +} // namespace std /*! @brief user-defined string literal for JSON values @@ -12272,5 +14733,18 @@ inline nlohmann::json::json_pointer operator "" _json_pointer(const char* s, std #if defined(__clang__) || defined(__GNUC__) || defined(__GNUG__) #pragma GCC diagnostic pop #endif +#if defined(__clang__) + #pragma GCC diagnostic pop +#endif + +// clean up +#undef JSON_CATCH +#undef JSON_THROW +#undef JSON_TRY +#undef JSON_LIKELY +#undef JSON_UNLIKELY +#undef JSON_DEPRECATED +#undef NLOHMANN_BASIC_JSON_TPL_DECLARATION +#undef NLOHMANN_BASIC_JSON_TPL #endif diff --git a/lib/lattice/Lattice_reduction.h b/lib/lattice/Lattice_reduction.h index db012c8c..8a3fbece 100644 --- a/lib/lattice/Lattice_reduction.h +++ b/lib/lattice/Lattice_reduction.h @@ -544,7 +544,6 @@ static void sliceInnerProductMatrix( Eigen::MatrixXcd &mat, const Lattice for(int i=0;i &out, const Lattice &in){ merge(out._odata[out_oidx], ptrs, 0); } } + +//////////////////////////////////////////////////////////////////////////////// +// Communicate between grids +//////////////////////////////////////////////////////////////////////////////// +// +// All to all plan +// +// Subvolume on fine grid is v. Vectors a,b,c,d +// +/////////////////////////////////////////////////////////////////////////////////////////////////////////// +// SIMPLEST CASE: +/////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Mesh of nodes (2) ; subdivide to 1 subdivisions +// +// Lex ord: +// N0 va0 vb0 N1 va1 vb1 +// +// For each dimension do an all to all +// +// full AllToAll(0) +// N0 va0 va1 N1 vb0 vb1 +// +// REARRANGE +// N0 va01 N1 vb01 +// +// Must also rearrange data to get into the NEW lex order of grid at each stage. Some kind of "insert/extract". +// NB: Easiest to programme if keep in lex order. +// +/////////////////////////////////////////////////////////////////////////////////////////////////////////// +// SIMPLE CASE: +/////////////////////////////////////////////////////////////////////////////////////////////////////////// +// +// Mesh of nodes (2x2) ; subdivide to 1x1 subdivisions +// +// Lex ord: +// N0 va0 vb0 vc0 vd0 N1 va1 vb1 vc1 vd1 +// N2 va2 vb2 vc2 vd2 N3 va3 vb3 vc3 vd3 +// +// Ratio = full[dim] / split[dim] +// +// For each dimension do an all to all; get Nvec -> Nvec / ratio +// Ldim -> Ldim * ratio +// LocalVol -> LocalVol * ratio +// full AllToAll(0) +// N0 va0 vb0 va1 vb1 N1 vc0 vd0 vc1 vd1 +// N2 va2 vb2 va3 vb3 N3 vc2 vd2 vc3 vd3 +// +// REARRANGE +// N0 va01 vb01 N1 vc01 vd01 +// N2 va23 vb23 N3 vc23 vd23 +// +// full AllToAll(1) // Not what is wanted. FIXME +// N0 va01 va23 N1 vc01 vc23 +// N2 vb01 vb23 N3 vd01 vd23 +// +// REARRANGE +// N0 va0123 N1 vc0123 +// N2 vb0123 N3 vd0123 +// +// Must also rearrange data to get into the NEW lex order of grid at each stage. Some kind of "insert/extract". +// NB: Easiest to programme if keep in lex order. +// +///////////////////////////////////////////////////////// +template +void Grid_split(std::vector > & full,Lattice & split) +{ + typedef typename Vobj::scalar_object Sobj; + + int full_vecs = full.size(); + + assert(full_vecs>=1); + + GridBase * full_grid = full[0]._grid; + GridBase *split_grid = split._grid; + + int ndim = full_grid->_ndimension; + int full_nproc = full_grid->_Nprocessors; + int split_nproc =split_grid->_Nprocessors; + + //////////////////////////////// + // Checkerboard management + //////////////////////////////// + int cb = full[0].checkerboard; + split.checkerboard = cb; + + ////////////////////////////// + // Checks + ////////////////////////////// + assert(full_grid->_ndimension==split_grid->_ndimension); + for(int n=0;n_gdimensions[d]==split._grid->_gdimensions[d]); + assert(full[n]._grid->_fdimensions[d]==split._grid->_fdimensions[d]); + } + } + + int nvector =full_nproc/split_nproc; + assert(nvector*split_nproc==full_nproc); + assert(nvector == full_vecs); + + std::vector ratio(ndim); + for(int d=0;d_processors[d]/ split_grid->_processors[d]; + } + + int lsites = full_grid->lSites(); + Integer sz = lsites * nvector; + std::vector tmpdata(sz); + std::vector alldata(sz); + std::vector scalardata(lsites); + for(int v=0;v ldims = full_grid->_ldimensions; + std::vector lcoor(ndim); + + for(int d=0;dAllToAll(d,alldata,tmpdata); + + ////////////////////////////////////////// + //Local volume for this dimension is expanded by ratio of processor extents + // Number of vectors is decreased by same factor + // Rearrange to lexico for bigger volume + ////////////////////////////////////////// + nvec /= ratio[d]; + auto rdims = ldims; rdims[d] *= ratio[d]; + auto rsites= lsites*ratio[d]; + for(int v=0;v_processors[d] > 1 ) { + tmpdata = alldata; + split_grid->AllToAll(d,tmpdata,alldata); + } + } + } + + vectorizeFromLexOrdArray(alldata,split); +} + +template +void Grid_split(Lattice &full,Lattice & split) +{ + int nvector = full._grid->_Nprocessors / split._grid->_Nprocessors; + std::vector > full_v(nvector,full._grid); + for(int n=0;n +void Grid_unsplit(std::vector > & full,Lattice & split) +{ + typedef typename Vobj::scalar_object Sobj; + + int full_vecs = full.size(); + + assert(full_vecs>=1); + + GridBase * full_grid = full[0]._grid; + GridBase *split_grid = split._grid; + + int ndim = full_grid->_ndimension; + int full_nproc = full_grid->_Nprocessors; + int split_nproc =split_grid->_Nprocessors; + + //////////////////////////////// + // Checkerboard management + //////////////////////////////// + int cb = full[0].checkerboard; + split.checkerboard = cb; + + ////////////////////////////// + // Checks + ////////////////////////////// + assert(full_grid->_ndimension==split_grid->_ndimension); + for(int n=0;n_gdimensions[d]==split._grid->_gdimensions[d]); + assert(full[n]._grid->_fdimensions[d]==split._grid->_fdimensions[d]); + } + } + + int nvector =full_nproc/split_nproc; + assert(nvector*split_nproc==full_nproc); + assert(nvector == full_vecs); + + std::vector ratio(ndim); + for(int d=0;d_processors[d]/ split_grid->_processors[d]; + } + + int lsites = full_grid->lSites(); + Integer sz = lsites * nvector; + std::vector tmpdata(sz); + std::vector alldata(sz); + std::vector scalardata(lsites); + + unvectorizeToLexOrdArray(alldata,split); + + ///////////////////////////////////////////////////////////////// + // Start from split grid and work towards full grid + ///////////////////////////////////////////////////////////////// + std::vector lcoor(ndim); + std::vector rcoor(ndim); + + int nvec = 1; + lsites = split_grid->lSites(); + std::vector ldims = split_grid->_ldimensions; + + for(int d=ndim-1;d>=0;d--){ + + if ( ratio[d] != 1 ) { + + if ( split_grid->_processors[d] > 1 ) { + tmpdata = alldata; + split_grid->AllToAll(d,tmpdata,alldata); + } + + ////////////////////////////////////////// + //Local volume for this dimension is expanded by ratio of processor extents + // Number of vectors is decreased by same factor + // Rearrange to lexico for bigger volume + ////////////////////////////////////////// + auto rsites= lsites/ratio[d]; + auto rdims = ldims; rdims[d]/=ratio[d]; + + for(int v=0;v smaller local volume + // lsite, lcoor --> bigger original (single node?) volume + // For loop over each site within smaller subvol + for(int rsite=0;rsiteAllToAll(d,tmpdata,alldata); + } + } + + lsites = full_grid->lSites(); + for(int v=0;v(); - // std::cout << " Lorentz N/S/V/M : " << _LorentzN<<" "<<_LorentzScalar<<"/"<<_LorentzVector<<"/"<<_LorentzMatrix<_gsites; + + // std::cout << "R sizeof(sobj)= " <_gsites< munge; - BinaryIO::readLatticeObject< sobj, sobj >(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb); + BinaryIO::readLatticeObject< vobj, sobj >(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb); ///////////////////////////////////////////// // Insist checksum is next record ///////////////////////////////////////////// - readLimeObject(scidacChecksum_,std::string("scidacChecksum"),record_name); + readLimeObject(scidacChecksum_,std::string("scidacChecksum"),std::string(SCIDAC_CHECKSUM)); ///////////////////////////////////////////// // Verify checksums @@ -242,11 +252,19 @@ class GridLimeReader : public BinaryIO { // should this be a do while; can we miss a first record?? while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { + // std::cout << GridLogMessage<< " readLimeObject seeking "<< record_name <<" found record :" < xmlc(nbytes+1,'\0'); limeReaderReadData((void *)&xmlc[0], &nbytes, LimeR); + + // std::cout << GridLogMessage<< " readLimeObject matches XML " << &xmlc[0] <(record_name.c_str()), nbytes); + assert(h!= NULL); err=limeWriteRecordHeader(h, LimeW); assert(err>=0); err=limeWriteRecordData(&xmlstring[0], &nbytes, LimeW); assert(err>=0); err=limeWriterCloseRecord(LimeW); assert(err>=0); limeDestroyHeader(h); + // std::cout << " File offset is now"<_gsites; createLimeRecordHeader(record_name, 0, 0, PayloadSize); + + // std::cout << "W sizeof(sobj)" <_gsites<(); BinarySimpleMunger munge; BinaryIO::writeLatticeObject(field, filename, munge, offset, format,nersc_csum,scidac_csuma,scidac_csumb); @@ -354,7 +383,7 @@ class GridLimeWriter : public BinaryIO { checksum.suma= streama.str(); checksum.sumb= streamb.str(); std::cout << GridLogMessage<<" writing scidac checksums "< + template void writeScidacFieldRecord(Lattice &field,userRecord _userRecord) { - typedef typename vobj::scalar_object sobj; - uint64_t nbytes; GridBase * grid = field._grid; //////////////////////////////////////// @@ -397,6 +424,66 @@ class ScidacWriter : public GridLimeWriter { } }; + +class ScidacReader : public GridLimeReader { + public: + + template + void readScidacFileRecord(GridBase *grid,SerialisableUserFile &_userFile) + { + scidacFile _scidacFile(grid); + readLimeObject(_scidacFile,_scidacFile.SerialisableClassName(),std::string(SCIDAC_PRIVATE_FILE_XML)); + readLimeObject(_userFile,_userFile.SerialisableClassName(),std::string(SCIDAC_FILE_XML)); + } + //////////////////////////////////////////////// + // Write generic lattice field in scidac format + //////////////////////////////////////////////// + template + void readScidacFieldRecord(Lattice &field,userRecord &_userRecord) + { + typedef typename vobj::scalar_object sobj; + GridBase * grid = field._grid; + + //////////////////////////////////////// + // fill the Grid header + //////////////////////////////////////// + FieldMetaData header; + scidacRecord _scidacRecord; + scidacFile _scidacFile; + + ////////////////////////////////////////////// + // Fill the Lime file record by record + ////////////////////////////////////////////// + readLimeObject(header ,std::string("FieldMetaData"),std::string(GRID_FORMAT)); // Open message + readLimeObject(_userRecord,_userRecord.SerialisableClassName(),std::string(SCIDAC_RECORD_XML)); + readLimeObject(_scidacRecord,_scidacRecord.SerialisableClassName(),std::string(SCIDAC_PRIVATE_RECORD_XML)); + readLimeLatticeBinaryObject(field,std::string(ILDG_BINARY_DATA)); + } + void skipPastBinaryRecord(void) { + std::string rec_name(ILDG_BINARY_DATA); + while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { + if ( !strncmp(limeReaderType(LimeR), rec_name.c_str(),strlen(rec_name.c_str()) ) ) { + skipPastObjectRecord(std::string(SCIDAC_CHECKSUM)); + return; + } + } + } + void skipPastObjectRecord(std::string rec_name) { + while ( limeReaderNextRecord(LimeR) == LIME_SUCCESS ) { + if ( !strncmp(limeReaderType(LimeR), rec_name.c_str(),strlen(rec_name.c_str()) ) ) { + return; + } + } + } + void skipScidacFieldRecord() { + skipPastObjectRecord(std::string(GRID_FORMAT)); + skipPastObjectRecord(std::string(SCIDAC_RECORD_XML)); + skipPastObjectRecord(std::string(SCIDAC_PRIVATE_RECORD_XML)); + skipPastBinaryRecord(); + } +}; + + class IldgWriter : public ScidacWriter { public: @@ -425,8 +512,6 @@ class IldgWriter : public ScidacWriter { typedef iLorentzColourMatrix vobj; typedef typename vobj::scalar_object sobj; - uint64_t nbytes; - //////////////////////////////////////// // fill the Grid header //////////////////////////////////////// diff --git a/lib/parallelIO/IldgIOtypes.h b/lib/parallelIO/IldgIOtypes.h index c3a5321c..5b397e14 100644 --- a/lib/parallelIO/IldgIOtypes.h +++ b/lib/parallelIO/IldgIOtypes.h @@ -64,6 +64,11 @@ namespace Grid { // file compatability, so should be correct to assume the undocumented but defacto file structure. ///////////////////////////////////////////////////////////////////////////////// +struct emptyUserRecord : Serializable { + GRID_SERIALIZABLE_CLASS_MEMBERS(emptyUserRecord,int,dummy); + emptyUserRecord() { dummy=0; }; +}; + //////////////////////// // Scidac private file xml // 1.1416 16 16 32 0 diff --git a/lib/parallelIO/MetaData.h b/lib/parallelIO/MetaData.h index 6d45d0a5..ccc8b18f 100644 --- a/lib/parallelIO/MetaData.h +++ b/lib/parallelIO/MetaData.h @@ -85,6 +85,9 @@ namespace Grid { nd=4; dimension.resize(4); boundary.resize(4); + scidac_checksuma=0; + scidac_checksumb=0; + checksum=0; } }; @@ -104,6 +107,7 @@ namespace Grid { header.nd = nd; header.dimension.resize(nd); header.boundary.resize(nd); + header.data_start = 0; for(int d=0;d_fdimensions[d]; } diff --git a/lib/qcd/action/fermion/AbstractEOFAFermion.h b/lib/qcd/action/fermion/AbstractEOFAFermion.h new file mode 100644 index 00000000..15faa401 --- /dev/null +++ b/lib/qcd/action/fermion/AbstractEOFAFermion.h @@ -0,0 +1,100 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/AbstractEOFAFermion.h + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: David Murphy + +This program is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2 of the License, or +(at your option) any later version. + +This program is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU General Public License for more details. + +You should have received a copy of the GNU General Public License along +with this program; if not, write to the Free Software Foundation, Inc., +51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + +See the full license in the file "LICENSE" in the top level distribution directory +*************************************************************************************/ +/* END LEGAL */ +#ifndef GRID_QCD_ABSTRACT_EOFA_FERMION_H +#define GRID_QCD_ABSTRACT_EOFA_FERMION_H + +#include + +namespace Grid { +namespace QCD { + + // DJM: Abstract base class for EOFA fermion types. + // Defines layout of additional EOFA-specific parameters and operators. + // Use to construct EOFA pseudofermion actions that are agnostic to + // Shamir / Mobius / etc., and ensure that no one can construct EOFA + // pseudofermion action with non-EOFA fermion type. + template + class AbstractEOFAFermion : public CayleyFermion5D { + public: + INHERIT_IMPL_TYPES(Impl); + + public: + // Fermion operator: D(mq1) + shift*\gamma_{5}*R_{5}*\Delta_{\pm}(mq2,mq3)*P_{\pm} + RealD mq1; + RealD mq2; + RealD mq3; + RealD shift; + int pm; + + RealD alpha; // Mobius scale + RealD k; // EOFA normalization constant + + virtual void Instantiatable(void) = 0; + + // EOFA-specific operations + // Force user to implement in derived classes + virtual void Omega (const FermionField& in, FermionField& out, int sign, int dag) = 0; + virtual void Dtilde (const FermionField& in, FermionField& out) = 0; + virtual void DtildeInv(const FermionField& in, FermionField& out) = 0; + + // Implement derivatives in base class: + // for EOFA both DWF and Mobius just need d(Dw)/dU + virtual void MDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag){ + this->DhopDeriv(mat, U, V, dag); + }; + virtual void MoeDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag){ + this->DhopDerivOE(mat, U, V, dag); + }; + virtual void MeoDeriv(GaugeField& mat, const FermionField& U, const FermionField& V, int dag){ + this->DhopDerivEO(mat, U, V, dag); + }; + + // Recompute 5D coefficients for different value of shift constant + // (needed for heatbath loop over poles) + virtual void RefreshShiftCoefficients(RealD new_shift) = 0; + + // Constructors + AbstractEOFAFermion(GaugeField& _Umu, GridCartesian& FiveDimGrid, GridRedBlackCartesian& FiveDimRedBlackGrid, + GridCartesian& FourDimGrid, GridRedBlackCartesian& FourDimRedBlackGrid, + RealD _mq1, RealD _mq2, RealD _mq3, RealD _shift, int _pm, + RealD _M5, RealD _b, RealD _c, const ImplParams& p=ImplParams()) + : CayleyFermion5D(_Umu, FiveDimGrid, FiveDimRedBlackGrid, FourDimGrid, FourDimRedBlackGrid, + _mq1, _M5, p), mq1(_mq1), mq2(_mq2), mq3(_mq3), shift(_shift), pm(_pm) + { + int Ls = this->Ls; + this->alpha = _b + _c; + this->k = this->alpha * (_mq3-_mq2) * std::pow(this->alpha+1.0,2*Ls) / + ( std::pow(this->alpha+1.0,Ls) + _mq2*std::pow(this->alpha-1.0,Ls) ) / + ( std::pow(this->alpha+1.0,Ls) + _mq3*std::pow(this->alpha-1.0,Ls) ); + }; + }; +}} + +#endif diff --git a/lib/qcd/action/fermion/CayleyFermion5D.cc b/lib/qcd/action/fermion/CayleyFermion5D.cc index 838b1c3d..eace6484 100644 --- a/lib/qcd/action/fermion/CayleyFermion5D.cc +++ b/lib/qcd/action/fermion/CayleyFermion5D.cc @@ -77,7 +77,6 @@ void CayleyFermion5D::DminusDag(const FermionField &psi, FermionField &chi } } - template void CayleyFermion5D::CayleyReport(void) { this->Report(); @@ -119,7 +118,6 @@ template void CayleyFermion5D::CayleyZeroCounters(void) MooeeInvTime=0; } - template void CayleyFermion5D::M5D (const FermionField &psi, FermionField &chi) { diff --git a/lib/qcd/action/fermion/CayleyFermion5D.h b/lib/qcd/action/fermion/CayleyFermion5D.h index cce13e12..ef75235a 100644 --- a/lib/qcd/action/fermion/CayleyFermion5D.h +++ b/lib/qcd/action/fermion/CayleyFermion5D.h @@ -1,6 +1,6 @@ /************************************************************************************* - Grid physics library, www.github.com/paboyle/Grid + Grid physics library, www.github.com/paboyle/Grid Source file: ./lib/qcd/action/fermion/CayleyFermion5D.h @@ -35,24 +35,24 @@ namespace Grid { namespace QCD { - template struct switcheroo { - static inline int iscomplex() { return 0; } + template struct switcheroo { + static inline int iscomplex() { return 0; } template static inline vec mult(vec a, vec b) { return real_mult(a,b); } }; - template<> struct switcheroo { - static inline int iscomplex() { return 1; } + template<> struct switcheroo { + static inline int iscomplex() { return 1; } template static inline vec mult(vec a, vec b) { return a*b; } }; - template<> struct switcheroo { - static inline int iscomplex() { return 1; } + template<> struct switcheroo { + static inline int iscomplex() { return 1; } template static inline vec mult(vec a, vec b) { return a*b; @@ -90,14 +90,14 @@ namespace Grid { // Instantiate different versions depending on Impl ///////////////////////////////////////////////////// void M5D(const FermionField &psi, - const FermionField &phi, + const FermionField &phi, FermionField &chi, std::vector &lower, std::vector &diag, std::vector &upper); void M5Ddag(const FermionField &psi, - const FermionField &phi, + const FermionField &phi, FermionField &chi, std::vector &lower, std::vector &diag, @@ -125,7 +125,7 @@ namespace Grid { // Efficient support for multigrid coarsening virtual void Mdir (const FermionField &in, FermionField &out,int dir,int disp); - + void Meooe5D (const FermionField &in, FermionField &out); void MeooeDag5D (const FermionField &in, FermionField &out); @@ -133,23 +133,23 @@ namespace Grid { RealD mass; // Cayley form Moebius (tanh and zolotarev) - std::vector omega; + std::vector omega; std::vector bs; // S dependent coeffs - std::vector cs; - std::vector as; + std::vector cs; + std::vector as; // For preconditioning Cayley form - std::vector bee; - std::vector cee; - std::vector aee; - std::vector beo; - std::vector ceo; - std::vector aeo; + std::vector bee; + std::vector cee; + std::vector aee; + std::vector beo; + std::vector ceo; + std::vector aeo; // LDU factorisation of the eeoo matrix - std::vector lee; - std::vector leem; - std::vector uee; - std::vector ueem; - std::vector dee; + std::vector lee; + std::vector leem; + std::vector uee; + std::vector ueem; + std::vector dee; // Matrices of 5d ee inverse params Vector > MatpInv; @@ -165,7 +165,7 @@ namespace Grid { GridRedBlackCartesian &FourDimRedBlackGrid, RealD _mass,RealD _M5,const ImplParams &p= ImplParams()); - + void CayleyReport(void); void CayleyZeroCounters(void); @@ -179,9 +179,9 @@ namespace Grid { double MooeeInvTime; protected: - void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c); - void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c); - void SetCoefficientsInternal(RealD zolo_hi,std::vector & gamma,RealD b,RealD c); + virtual void SetCoefficientsZolotarev(RealD zolohi,Approx::zolotarev_data *zdata,RealD b,RealD c); + virtual void SetCoefficientsTanh(Approx::zolotarev_data *zdata,RealD b,RealD c); + virtual void SetCoefficientsInternal(RealD zolo_hi,std::vector & gamma,RealD b,RealD c); }; } diff --git a/lib/qcd/action/fermion/DomainWallEOFAFermion.cc b/lib/qcd/action/fermion/DomainWallEOFAFermion.cc new file mode 100644 index 00000000..37ab5fa6 --- /dev/null +++ b/lib/qcd/action/fermion/DomainWallEOFAFermion.cc @@ -0,0 +1,438 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include +#include + +namespace Grid { +namespace QCD { + + template + DomainWallEOFAFermion::DomainWallEOFAFermion( + GaugeField &_Umu, + GridCartesian &FiveDimGrid, + GridRedBlackCartesian &FiveDimRedBlackGrid, + GridCartesian &FourDimGrid, + GridRedBlackCartesian &FourDimRedBlackGrid, + RealD _mq1, RealD _mq2, RealD _mq3, + RealD _shift, int _pm, RealD _M5, const ImplParams &p) : + AbstractEOFAFermion(_Umu, FiveDimGrid, FiveDimRedBlackGrid, + FourDimGrid, FourDimRedBlackGrid, _mq1, _mq2, _mq3, + _shift, _pm, _M5, 1.0, 0.0, p) + { + RealD eps = 1.0; + Approx::zolotarev_data *zdata = Approx::higham(eps,this->Ls); + assert(zdata->n == this->Ls); + + std::cout << GridLogMessage << "DomainWallEOFAFermion with Ls=" << this->Ls << std::endl; + this->SetCoefficientsTanh(zdata, 1.0, 0.0); + + Approx::zolotarev_free(zdata); + } + + /*************************************************************** + * Additional EOFA operators only called outside the inverter. + * Since speed is not essential, simple axpby-style + * implementations should be fine. + ***************************************************************/ + template + void DomainWallEOFAFermion::Omega(const FermionField& psi, FermionField& Din, int sign, int dag) + { + int Ls = this->Ls; + + Din = zero; + if((sign == 1) && (dag == 0)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, Ls-1, 0); } + else if((sign == -1) && (dag == 0)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, 0, 0); } + else if((sign == 1 ) && (dag == 1)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, 0, Ls-1); } + else if((sign == -1) && (dag == 1)){ axpby_ssp(Din, 0.0, psi, 1.0, psi, 0, 0); } + } + + // This is just the identity for DWF + template + void DomainWallEOFAFermion::Dtilde(const FermionField& psi, FermionField& chi){ chi = psi; } + + // This is just the identity for DWF + template + void DomainWallEOFAFermion::DtildeInv(const FermionField& psi, FermionField& chi){ chi = psi; } + + /*****************************************************************************************************/ + + template + RealD DomainWallEOFAFermion::M(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + FermionField Din(psi._grid); + + this->Meooe5D(psi, Din); + this->DW(Din, chi, DaggerNo); + axpby(chi, 1.0, 1.0, chi, psi); + this->M5D(psi, chi); + return(norm2(chi)); + } + + template + RealD DomainWallEOFAFermion::Mdag(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + FermionField Din(psi._grid); + + this->DW(psi, Din, DaggerYes); + this->MeooeDag5D(Din, chi); + this->M5Ddag(psi, chi); + axpby(chi, 1.0, 1.0, chi, psi); + return(norm2(chi)); + } + + /******************************************************************** + * Performance critical fermion operators called inside the inverter + ********************************************************************/ + + template + void DomainWallEOFAFermion::M5D(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + int pm = this->pm; + RealD shift = this->shift; + RealD mq1 = this->mq1; + RealD mq2 = this->mq2; + RealD mq3 = this->mq3; + + // coefficients for shift operator ( = shift*\gamma_{5}*R_{5}*\Delta_{\pm}(mq2,mq3)*P_{\pm} ) + Coeff_t shiftp(0.0), shiftm(0.0); + if(shift != 0.0){ + if(pm == 1){ shiftp = shift*(mq3-mq2); } + else{ shiftm = -shift*(mq3-mq2); } + } + + std::vector diag(Ls,1.0); + std::vector upper(Ls,-1.0); upper[Ls-1] = mq1 + shiftm; + std::vector lower(Ls,-1.0); lower[0] = mq1 + shiftp; + + #if(0) + std::cout << GridLogMessage << "DomainWallEOFAFermion::M5D(FF&,FF&):" << std::endl; + for(int i=0; i::iscomplex()) { + sp[l] = PplusMat (l*istride+s1*ostride,s2); + sm[l] = PminusMat(l*istride+s1*ostride,s2); + } else { + // if real + scalar_type tmp; + tmp = PplusMat (l*istride+s1*ostride,s2); + sp[l] = scalar_type(tmp.real(),tmp.real()); + tmp = PminusMat(l*istride+s1*ostride,s2); + sm[l] = scalar_type(tmp.real(),tmp.real()); + } + } + Matp[LLs*s2+s1] = Vp; + Matm[LLs*s2+s1] = Vm; + }} + } + + FermOpTemplateInstantiate(DomainWallEOFAFermion); + GparityFermOpTemplateInstantiate(DomainWallEOFAFermion); + +}} diff --git a/lib/qcd/action/fermion/DomainWallEOFAFermion.h b/lib/qcd/action/fermion/DomainWallEOFAFermion.h new file mode 100644 index 00000000..5362cda8 --- /dev/null +++ b/lib/qcd/action/fermion/DomainWallEOFAFermion.h @@ -0,0 +1,115 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermion.h + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: David Murphy + +This program is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2 of the License, or +(at your option) any later version. + +This program is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU General Public License for more details. + +You should have received a copy of the GNU General Public License along +with this program; if not, write to the Free Software Foundation, Inc., +51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + +See the full license in the file "LICENSE" in the top level distribution directory +*************************************************************************************/ +/* END LEGAL */ +#ifndef GRID_QCD_DOMAIN_WALL_EOFA_FERMION_H +#define GRID_QCD_DOMAIN_WALL_EOFA_FERMION_H + +#include + +namespace Grid { +namespace QCD { + + template + class DomainWallEOFAFermion : public AbstractEOFAFermion + { + public: + INHERIT_IMPL_TYPES(Impl); + + public: + // Modified (0,Ls-1) and (Ls-1,0) elements of Mooee + // for red-black preconditioned Shamir EOFA + Coeff_t dm; + Coeff_t dp; + + virtual void Instantiatable(void) {}; + + // EOFA-specific operations + virtual void Omega (const FermionField& in, FermionField& out, int sign, int dag); + virtual void Dtilde (const FermionField& in, FermionField& out); + virtual void DtildeInv (const FermionField& in, FermionField& out); + + // override multiply + virtual RealD M (const FermionField& in, FermionField& out); + virtual RealD Mdag (const FermionField& in, FermionField& out); + + // half checkerboard operations + virtual void Mooee (const FermionField& in, FermionField& out); + virtual void MooeeDag (const FermionField& in, FermionField& out); + virtual void MooeeInv (const FermionField& in, FermionField& out); + virtual void MooeeInvDag(const FermionField& in, FermionField& out); + + virtual void M5D (const FermionField& psi, FermionField& chi); + virtual void M5Ddag (const FermionField& psi, FermionField& chi); + + ///////////////////////////////////////////////////// + // Instantiate different versions depending on Impl + ///////////////////////////////////////////////////// + void M5D(const FermionField& psi, const FermionField& phi, FermionField& chi, + std::vector& lower, std::vector& diag, std::vector& upper); + + void M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi, + std::vector& lower, std::vector& diag, std::vector& upper); + + void MooeeInternal(const FermionField& in, FermionField& out, int dag, int inv); + + void MooeeInternalCompute(int dag, int inv, Vector>& Matp, Vector>& Matm); + + void MooeeInternalAsm(const FermionField& in, FermionField& out, int LLs, int site, + Vector>& Matp, Vector>& Matm); + + void MooeeInternalZAsm(const FermionField& in, FermionField& out, int LLs, int site, + Vector>& Matp, Vector>& Matm); + + virtual void RefreshShiftCoefficients(RealD new_shift); + + // Constructors + DomainWallEOFAFermion(GaugeField& _Umu, GridCartesian& FiveDimGrid, GridRedBlackCartesian& FiveDimRedBlackGrid, + GridCartesian& FourDimGrid, GridRedBlackCartesian& FourDimRedBlackGrid, + RealD _mq1, RealD _mq2, RealD _mq3, RealD _shift, int pm, + RealD _M5, const ImplParams& p=ImplParams()); + + protected: + void SetCoefficientsInternal(RealD zolo_hi, std::vector& gamma, RealD b, RealD c); + }; +}} + +#define INSTANTIATE_DPERP_DWF_EOFA(A)\ +template void DomainWallEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, FermionField& chi, \ + std::vector& lower, std::vector& diag, std::vector& upper); \ +template void DomainWallEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi, \ + std::vector& lower, std::vector& diag, std::vector& upper); \ +template void DomainWallEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi); \ +template void DomainWallEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi); + +#undef DOMAIN_WALL_EOFA_DPERP_DENSE +#define DOMAIN_WALL_EOFA_DPERP_CACHE +#undef DOMAIN_WALL_EOFA_DPERP_LINALG +#define DOMAIN_WALL_EOFA_DPERP_VEC + +#endif diff --git a/lib/qcd/action/fermion/DomainWallEOFAFermioncache.cc b/lib/qcd/action/fermion/DomainWallEOFAFermioncache.cc new file mode 100644 index 00000000..0b214d31 --- /dev/null +++ b/lib/qcd/action/fermion/DomainWallEOFAFermioncache.cc @@ -0,0 +1,248 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermioncache.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include + +namespace Grid { +namespace QCD { + + // FIXME -- make a version of these routines with site loop outermost for cache reuse. + + // Pminus fowards + // Pplus backwards.. + template + void DomainWallEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + int Ls = this->Ls; + GridBase* grid = psi._grid; + + assert(phi.checkerboard == psi.checkerboard); + chi.checkerboard = psi.checkerboard; + // Flops = 6.0*(Nc*Ns) *Ls*vol + this->M5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ // adds Ls + for(int s=0; sM5Dtime += usecond(); + } + + template + void DomainWallEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + int Ls = this->Ls; + GridBase* grid = psi._grid; + assert(phi.checkerboard == psi.checkerboard); + chi.checkerboard=psi.checkerboard; + + // Flops = 6.0*(Nc*Ns) *Ls*vol + this->M5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ // adds Ls + auto tmp = psi._odata[0]; + for(int s=0; sM5Dtime += usecond(); + } + + template + void DomainWallEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + GridBase* grid = psi._grid; + int Ls = this->Ls; + + chi.checkerboard = psi.checkerboard; + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ // adds Ls + + auto tmp1 = psi._odata[0]; + auto tmp2 = psi._odata[0]; + + // flops = 12*2*Ls + 12*2*Ls + 3*12*Ls + 12*2*Ls = 12*Ls * (9) = 108*Ls flops + // Apply (L^{\prime})^{-1} + chi[ss] = psi[ss]; // chi[0]=psi[0] + for(int s=1; slee[s-1]*tmp1; + } + + // L_m^{-1} + for(int s=0; sleem[s]*tmp1; + } + + // U_m^{-1} D^{-1} + for(int s=0; sdee[s])*chi[ss+s] - (this->ueem[s]/this->dee[Ls])*tmp1; + } + spProj5m(tmp2, chi[ss+Ls-1]); + chi[ss+Ls-1] = (1.0/this->dee[Ls])*tmp1 + (1.0/this->dee[Ls-1])*tmp2; + + // Apply U^{-1} + for(int s=Ls-2; s>=0; s--){ + spProj5m(tmp1, chi[ss+s+1]); + chi[ss+s] = chi[ss+s] - this->uee[s]*tmp1; + } + } + + this->MooeeInvTime += usecond(); + } + + template + void DomainWallEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + GridBase* grid = psi._grid; + int Ls = this->Ls; + + assert(psi.checkerboard == psi.checkerboard); + chi.checkerboard = psi.checkerboard; + + std::vector ueec(Ls); + std::vector deec(Ls+1); + std::vector leec(Ls); + std::vector ueemc(Ls); + std::vector leemc(Ls); + + for(int s=0; suee[s]); + deec[s] = conjugate(this->dee[s]); + leec[s] = conjugate(this->lee[s]); + ueemc[s] = conjugate(this->ueem[s]); + leemc[s] = conjugate(this->leem[s]); + } + deec[Ls] = conjugate(this->dee[Ls]); + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ // adds Ls + + auto tmp1 = psi._odata[0]; + auto tmp2 = psi._odata[0]; + + // Apply (U^{\prime})^{-dagger} + chi[ss] = psi[ss]; + for(int s=1; s=0; s--){ + spProj5p(tmp1, chi[ss+s+1]); + chi[ss+s] = chi[ss+s] - leec[s]*tmp1; + } + } + + this->MooeeInvTime += usecond(); + } + + #ifdef DOMAIN_WALL_EOFA_DPERP_CACHE + + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplF); + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplD); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplF); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplD); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplF); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplD); + + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplDF); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplDF); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplDF); + + #endif + +}} diff --git a/lib/qcd/action/fermion/DomainWallEOFAFermiondense.cc b/lib/qcd/action/fermion/DomainWallEOFAFermiondense.cc new file mode 100644 index 00000000..c27074d9 --- /dev/null +++ b/lib/qcd/action/fermion/DomainWallEOFAFermiondense.cc @@ -0,0 +1,159 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermiondense.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include +#include + +namespace Grid { +namespace QCD { + + /* + * Dense matrix versions of routines + */ + template + void DomainWallEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerYes, InverseYes); + } + + template + void DomainWallEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerNo, InverseYes); + } + + template + void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv) + { + int Ls = this->Ls; + int LLs = psi._grid->_rdimensions[0]; + int vol = psi._grid->oSites()/LLs; + + chi.checkerboard = psi.checkerboard; + + assert(Ls==LLs); + + Eigen::MatrixXd Pplus = Eigen::MatrixXd::Zero(Ls,Ls); + Eigen::MatrixXd Pminus = Eigen::MatrixXd::Zero(Ls,Ls); + + for(int s=0;sbee[s]; + Pminus(s,s) = this->bee[s]; + } + + for(int s=0; scee[s]; + } + + for(int s=0; scee[s+1]; + } + + Pplus (0,Ls-1) = this->dp; + Pminus(Ls-1,0) = this->dm; + + Eigen::MatrixXd PplusMat ; + Eigen::MatrixXd PminusMat; + + if(inv) { + PplusMat = Pplus.inverse(); + PminusMat = Pminus.inverse(); + } else { + PplusMat = Pplus; + PminusMat = Pminus; + } + + if(dag){ + PplusMat.adjointInPlace(); + PminusMat.adjointInPlace(); + } + + // For the non-vectorised s-direction this is simple + + for(auto site=0; site::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplDF); + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplDF); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplDF); + + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + #endif + +}} diff --git a/lib/qcd/action/fermion/DomainWallEOFAFermionssp.cc b/lib/qcd/action/fermion/DomainWallEOFAFermionssp.cc new file mode 100644 index 00000000..80a4bf09 --- /dev/null +++ b/lib/qcd/action/fermion/DomainWallEOFAFermionssp.cc @@ -0,0 +1,168 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermionssp.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include + +namespace Grid { +namespace QCD { + + // FIXME -- make a version of these routines with site loop outermost for cache reuse. + // Pminus fowards + // Pplus backwards + template + void DomainWallEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; s + void DomainWallEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; s + void DomainWallEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + Coeff_t one(1.0); + Coeff_t czero(0.0); + chi.checkerboard = psi.checkerboard; + int Ls = this->Ls; + + FermionField tmp(psi._grid); + + // Apply (L^{\prime})^{-1} + axpby_ssp(chi, one, psi, czero, psi, 0, 0); // chi[0]=psi[0] + for(int s=1; slee[s-1], chi, s, s-1);// recursion Psi[s] -lee P_+ chi[s-1] + } + + // L_m^{-1} + for(int s=0; sleem[s], chi, Ls-1, s); + } + + // U_m^{-1} D^{-1} + for(int s=0; sdee[s], chi, -this->ueem[s]/this->dee[Ls], chi, s, Ls-1); + } + axpby_ssp_pminus(tmp, czero, chi, one/this->dee[Ls-1], chi, Ls-1, Ls-1); + axpby_ssp_pplus(chi, one, tmp, one/this->dee[Ls], chi, Ls-1, Ls-1); + + // Apply U^{-1} + for(int s=Ls-2; s>=0; s--){ + axpby_ssp_pminus(chi, one, chi, -this->uee[s], chi, s, s+1); // chi[Ls] + } + } + + template + void DomainWallEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + Coeff_t one(1.0); + Coeff_t czero(0.0); + chi.checkerboard = psi.checkerboard; + int Ls = this->Ls; + + FermionField tmp(psi._grid); + + // Apply (U^{\prime})^{-dagger} + axpby_ssp(chi, one, psi, czero, psi, 0, 0); // chi[0]=psi[0] + for(int s=1; suee[s-1]), chi, s, s-1); + } + + // U_m^{-\dagger} + for(int s=0; sueem[s]), chi, Ls-1, s); + } + + // L_m^{-\dagger} D^{-dagger} + for(int s=0; sdee[s]), chi, -conjugate(this->leem[s]/this->dee[Ls-1]), chi, s, Ls-1); + } + axpby_ssp_pminus(tmp, czero, chi, one/conjugate(this->dee[Ls-1]), chi, Ls-1, Ls-1); + axpby_ssp_pplus(chi, one, tmp, one/conjugate(this->dee[Ls]), chi, Ls-1, Ls-1); + + // Apply L^{-dagger} + for(int s=Ls-2; s>=0; s--){ + axpby_ssp_pplus(chi, one, chi, -conjugate(this->lee[s]), chi, s, s+1); // chi[Ls] + } + } + + #ifdef DOMAIN_WALL_EOFA_DPERP_LINALG + + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplF); + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplD); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplF); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplD); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplF); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplD); + + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(WilsonImplDF); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(GparityWilsonImplDF); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplFH); + INSTANTIATE_DPERP_DWF_EOFA(ZWilsonImplDF); + + #endif + +}} diff --git a/lib/qcd/action/fermion/DomainWallEOFAFermionvec.cc b/lib/qcd/action/fermion/DomainWallEOFAFermionvec.cc new file mode 100644 index 00000000..81ce448c --- /dev/null +++ b/lib/qcd/action/fermion/DomainWallEOFAFermionvec.cc @@ -0,0 +1,605 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/DomainWallEOFAFermionvec.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include + +namespace Grid { +namespace QCD { + + /* + * Dense matrix versions of routines + */ + template + void DomainWallEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerYes, InverseYes); + } + + template + void DomainWallEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerNo, InverseYes); + } + + template + void DomainWallEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + GridBase* grid = psi._grid; + int Ls = this->Ls; + int LLs = grid->_rdimensions[0]; + const int nsimd = Simd::Nsimd(); + + Vector > u(LLs); + Vector > l(LLs); + Vector > d(LLs); + + assert(Ls/LLs == nsimd); + assert(phi.checkerboard == psi.checkerboard); + + chi.checkerboard = psi.checkerboard; + + // just directly address via type pun + typedef typename Simd::scalar_type scalar_type; + scalar_type* u_p = (scalar_type*) &u[0]; + scalar_type* l_p = (scalar_type*) &l[0]; + scalar_type* d_p = (scalar_type*) &d[0]; + + for(int o=0;oM5Dcalls++; + this->M5Dtime -= usecond(); + + assert(Nc == 3); + + parallel_for(int ss=0; ssoSites(); ss+=LLs){ // adds LLs + + #if 0 + + alignas(64) SiteHalfSpinor hp; + alignas(64) SiteHalfSpinor hm; + alignas(64) SiteSpinor fp; + alignas(64) SiteSpinor fm; + + for(int v=0; v= v){ rotate(hm, hm, nsimd-1); } + + hp = 0.5*hp; + hm = 0.5*hm; + + spRecon5m(fp, hp); + spRecon5p(fm, hm); + + chi[ss+v] = d[v]*phi[ss+v]; + chi[ss+v] = chi[ss+v] + u[v]*fp; + chi[ss+v] = chi[ss+v] + l[v]*fm; + + } + + #else + + for(int v=0; v(hp_00.v); + hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v); + hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v); + hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v); + hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v); + hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v); + } + + if(vm >= v){ + hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v); + hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v); + hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v); + hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v); + hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v); + hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v); + } + + // Can force these to real arithmetic and save 2x. + Simd p_00 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(l[v]()()(), hm_00); + Simd p_01 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(l[v]()()(), hm_01); + Simd p_02 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(l[v]()()(), hm_02); + Simd p_10 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(l[v]()()(), hm_10); + Simd p_11 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(l[v]()()(), hm_11); + Simd p_12 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(l[v]()()(), hm_12); + Simd p_20 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(u[v]()()(), hp_00); + Simd p_21 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(u[v]()()(), hp_01); + Simd p_22 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(u[v]()()(), hp_02); + Simd p_30 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(u[v]()()(), hp_10); + Simd p_31 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(u[v]()()(), hp_11); + Simd p_32 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(u[v]()()(), hp_12); + + vstream(chi[ss+v]()(0)(0), p_00); + vstream(chi[ss+v]()(0)(1), p_01); + vstream(chi[ss+v]()(0)(2), p_02); + vstream(chi[ss+v]()(1)(0), p_10); + vstream(chi[ss+v]()(1)(1), p_11); + vstream(chi[ss+v]()(1)(2), p_12); + vstream(chi[ss+v]()(2)(0), p_20); + vstream(chi[ss+v]()(2)(1), p_21); + vstream(chi[ss+v]()(2)(2), p_22); + vstream(chi[ss+v]()(3)(0), p_30); + vstream(chi[ss+v]()(3)(1), p_31); + vstream(chi[ss+v]()(3)(2), p_32); + } + + #endif + } + + this->M5Dtime += usecond(); + } + + template + void DomainWallEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + GridBase* grid = psi._grid; + int Ls = this->Ls; + int LLs = grid->_rdimensions[0]; + int nsimd = Simd::Nsimd(); + + Vector > u(LLs); + Vector > l(LLs); + Vector > d(LLs); + + assert(Ls/LLs == nsimd); + assert(phi.checkerboard == psi.checkerboard); + + chi.checkerboard = psi.checkerboard; + + // just directly address via type pun + typedef typename Simd::scalar_type scalar_type; + scalar_type* u_p = (scalar_type*) &u[0]; + scalar_type* l_p = (scalar_type*) &l[0]; + scalar_type* d_p = (scalar_type*) &d[0]; + + for(int o=0; oM5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=LLs){ // adds LLs + + #if 0 + + alignas(64) SiteHalfSpinor hp; + alignas(64) SiteHalfSpinor hm; + alignas(64) SiteSpinor fp; + alignas(64) SiteSpinor fm; + + for(int v=0; v= v){ rotate(hm, hm, nsimd-1); } + + hp = hp*0.5; + hm = hm*0.5; + spRecon5p(fp, hp); + spRecon5m(fm, hm); + + chi[ss+v] = d[v]*phi[ss+v]+u[v]*fp; + chi[ss+v] = chi[ss+v] +l[v]*fm; + } + + #else + + for(int v=0; v(hp_00.v); + hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v); + hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v); + hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v); + hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v); + hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v); + } + + if(vm >= v){ + hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v); + hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v); + hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v); + hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v); + hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v); + hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v); + } + + Simd p_00 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(u[v]()()(), hp_00); + Simd p_01 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(u[v]()()(), hp_01); + Simd p_02 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(u[v]()()(), hp_02); + Simd p_10 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(u[v]()()(), hp_10); + Simd p_11 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(u[v]()()(), hp_11); + Simd p_12 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(u[v]()()(), hp_12); + Simd p_20 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(l[v]()()(), hm_00); + Simd p_21 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(l[v]()()(), hm_01); + Simd p_22 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(l[v]()()(), hm_02); + Simd p_30 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(l[v]()()(), hm_10); + Simd p_31 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(l[v]()()(), hm_11); + Simd p_32 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(l[v]()()(), hm_12); + + vstream(chi[ss+v]()(0)(0), p_00); + vstream(chi[ss+v]()(0)(1), p_01); + vstream(chi[ss+v]()(0)(2), p_02); + vstream(chi[ss+v]()(1)(0), p_10); + vstream(chi[ss+v]()(1)(1), p_11); + vstream(chi[ss+v]()(1)(2), p_12); + vstream(chi[ss+v]()(2)(0), p_20); + vstream(chi[ss+v]()(2)(1), p_21); + vstream(chi[ss+v]()(2)(2), p_22); + vstream(chi[ss+v]()(3)(0), p_30); + vstream(chi[ss+v]()(3)(1), p_31); + vstream(chi[ss+v]()(3)(2), p_32); + } + #endif + + } + + this->M5Dtime += usecond(); + } + + #ifdef AVX512 + #include + #include + #include + #endif + + template + void DomainWallEOFAFermion::MooeeInternalAsm(const FermionField& psi, FermionField& chi, + int LLs, int site, Vector >& Matp, Vector >& Matm) + { + #ifndef AVX512 + { + SiteHalfSpinor BcastP; + SiteHalfSpinor BcastM; + SiteHalfSpinor SiteChiP; + SiteHalfSpinor SiteChiM; + + // Ls*Ls * 2 * 12 * vol flops + for(int s1=0; s1); + for(int s1=0; s1 + void DomainWallEOFAFermion::MooeeInternalZAsm(const FermionField& psi, FermionField& chi, + int LLs, int site, Vector >& Matp, Vector >& Matm) + { + std::cout << "Error: zMobius not implemented for EOFA" << std::endl; + exit(-1); + }; + + template + void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv) + { + int Ls = this->Ls; + int LLs = psi._grid->_rdimensions[0]; + int vol = psi._grid->oSites()/LLs; + + chi.checkerboard = psi.checkerboard; + + Vector > Matp; + Vector > Matm; + Vector > *_Matp; + Vector > *_Matm; + + // MooeeInternalCompute(dag,inv,Matp,Matm); + if(inv && dag){ + _Matp = &this->MatpInvDag; + _Matm = &this->MatmInvDag; + } + + if(inv && (!dag)){ + _Matp = &this->MatpInv; + _Matm = &this->MatmInv; + } + + if(!inv){ + MooeeInternalCompute(dag, inv, Matp, Matm); + _Matp = &Matp; + _Matm = &Matm; + } + + assert(_Matp->size() == Ls*LLs); + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + if(switcheroo::iscomplex()){ + parallel_for(auto site=0; siteMooeeInvTime += usecond(); + } + + #ifdef DOMAIN_WALL_EOFA_DPERP_VEC + + INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplD); + INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplF); + INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplD); + INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplF); + + INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplDF); + INSTANTIATE_DPERP_DWF_EOFA(DomainWallVec5dImplFH); + INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplDF); + INSTANTIATE_DPERP_DWF_EOFA(ZDomainWallVec5dImplFH); + + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void DomainWallEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + #endif + +}} diff --git a/lib/qcd/action/fermion/Fermion.h b/lib/qcd/action/fermion/Fermion.h index 0f803f44..ad2f383d 100644 --- a/lib/qcd/action/fermion/Fermion.h +++ b/lib/qcd/action/fermion/Fermion.h @@ -1,6 +1,6 @@ /************************************************************************************* - Grid physics library, www.github.com/paboyle/Grid + Grid physics library, www.github.com/paboyle/Grid Source file: ./lib/qcd/action/fermion/Fermion_base_aggregate.h @@ -38,6 +38,8 @@ Author: Peter Boyle // - ContinuedFractionFermion5D.cc // - WilsonFermion.cc // - WilsonKernels.cc +// - DomainWallEOFAFermion.cc +// - MobiusEOFAFermion.cc // // The explicit instantiation is only avoidable if we move this source to headers and end up with include/parse/recompile // for EVERY .cc file. This define centralises the list and restores global push of impl cases @@ -55,8 +57,9 @@ Author: Peter Boyle #include #include // Cayley types #include -#include +#include #include +#include #include #include #include @@ -113,6 +116,14 @@ typedef DomainWallFermion DomainWallFermionRL; typedef DomainWallFermion DomainWallFermionFH; typedef DomainWallFermion DomainWallFermionDF; +typedef DomainWallEOFAFermion DomainWallEOFAFermionR; +typedef DomainWallEOFAFermion DomainWallEOFAFermionF; +typedef DomainWallEOFAFermion DomainWallEOFAFermionD; + +typedef DomainWallEOFAFermion DomainWallEOFAFermionRL; +typedef DomainWallEOFAFermion DomainWallEOFAFermionFH; +typedef DomainWallEOFAFermion DomainWallEOFAFermionDF; + typedef MobiusFermion MobiusFermionR; typedef MobiusFermion MobiusFermionF; typedef MobiusFermion MobiusFermionD; @@ -121,6 +132,14 @@ typedef MobiusFermion MobiusFermionRL; typedef MobiusFermion MobiusFermionFH; typedef MobiusFermion MobiusFermionDF; +typedef MobiusEOFAFermion MobiusEOFAFermionR; +typedef MobiusEOFAFermion MobiusEOFAFermionF; +typedef MobiusEOFAFermion MobiusEOFAFermionD; + +typedef MobiusEOFAFermion MobiusEOFAFermionRL; +typedef MobiusEOFAFermion MobiusEOFAFermionFH; +typedef MobiusEOFAFermion MobiusEOFAFermionDF; + typedef ZMobiusFermion ZMobiusFermionR; typedef ZMobiusFermion ZMobiusFermionF; typedef ZMobiusFermion ZMobiusFermionD; @@ -129,7 +148,7 @@ typedef ZMobiusFermion ZMobiusFermionRL; typedef ZMobiusFermion ZMobiusFermionFH; typedef ZMobiusFermion ZMobiusFermionDF; -// Ls vectorised +// Ls vectorised typedef DomainWallFermion DomainWallFermionVec5dR; typedef DomainWallFermion DomainWallFermionVec5dF; typedef DomainWallFermion DomainWallFermionVec5dD; @@ -138,6 +157,14 @@ typedef DomainWallFermion DomainWallFermionVec5dRL; typedef DomainWallFermion DomainWallFermionVec5dFH; typedef DomainWallFermion DomainWallFermionVec5dDF; +typedef DomainWallEOFAFermion DomainWallEOFAFermionVec5dR; +typedef DomainWallEOFAFermion DomainWallEOFAFermionVec5dF; +typedef DomainWallEOFAFermion DomainWallEOFAFermionVec5dD; + +typedef DomainWallEOFAFermion DomainWallEOFAFermionVec5dRL; +typedef DomainWallEOFAFermion DomainWallEOFAFermionVec5dFH; +typedef DomainWallEOFAFermion DomainWallEOFAFermionVec5dDF; + typedef MobiusFermion MobiusFermionVec5dR; typedef MobiusFermion MobiusFermionVec5dF; typedef MobiusFermion MobiusFermionVec5dD; @@ -146,6 +173,14 @@ typedef MobiusFermion MobiusFermionVec5dRL; typedef MobiusFermion MobiusFermionVec5dFH; typedef MobiusFermion MobiusFermionVec5dDF; +typedef MobiusEOFAFermion MobiusEOFAFermionVec5dR; +typedef MobiusEOFAFermion MobiusEOFAFermionVec5dF; +typedef MobiusEOFAFermion MobiusEOFAFermionVec5dD; + +typedef MobiusEOFAFermion MobiusEOFAFermionVec5dRL; +typedef MobiusEOFAFermion MobiusEOFAFermionVec5dFH; +typedef MobiusEOFAFermion MobiusEOFAFermionVec5dDF; + typedef ZMobiusFermion ZMobiusFermionVec5dR; typedef ZMobiusFermion ZMobiusFermionVec5dF; typedef ZMobiusFermion ZMobiusFermionVec5dD; @@ -206,6 +241,14 @@ typedef DomainWallFermion GparityDomainWallFermionRL; typedef DomainWallFermion GparityDomainWallFermionFH; typedef DomainWallFermion GparityDomainWallFermionDF; +typedef DomainWallEOFAFermion GparityDomainWallEOFAFermionR; +typedef DomainWallEOFAFermion GparityDomainWallEOFAFermionF; +typedef DomainWallEOFAFermion GparityDomainWallEOFAFermionD; + +typedef DomainWallEOFAFermion GparityDomainWallEOFAFermionRL; +typedef DomainWallEOFAFermion GparityDomainWallEOFAFermionFH; +typedef DomainWallEOFAFermion GparityDomainWallEOFAFermionDF; + typedef WilsonTMFermion GparityWilsonTMFermionR; typedef WilsonTMFermion GparityWilsonTMFermionF; typedef WilsonTMFermion GparityWilsonTMFermionD; @@ -222,6 +265,14 @@ typedef MobiusFermion GparityMobiusFermionRL; typedef MobiusFermion GparityMobiusFermionFH; typedef MobiusFermion GparityMobiusFermionDF; +typedef MobiusEOFAFermion GparityMobiusEOFAFermionR; +typedef MobiusEOFAFermion GparityMobiusEOFAFermionF; +typedef MobiusEOFAFermion GparityMobiusEOFAFermionD; + +typedef MobiusEOFAFermion GparityMobiusEOFAFermionRL; +typedef MobiusEOFAFermion GparityMobiusEOFAFermionFH; +typedef MobiusEOFAFermion GparityMobiusEOFAFermionDF; + typedef ImprovedStaggeredFermion ImprovedStaggeredFermionR; typedef ImprovedStaggeredFermion ImprovedStaggeredFermionF; typedef ImprovedStaggeredFermion ImprovedStaggeredFermionD; diff --git a/lib/qcd/action/fermion/FermionOperatorImpl.h b/lib/qcd/action/fermion/FermionOperatorImpl.h index caa89d05..a9510222 100644 --- a/lib/qcd/action/fermion/FermionOperatorImpl.h +++ b/lib/qcd/action/fermion/FermionOperatorImpl.h @@ -557,12 +557,16 @@ class GparityWilsonImpl : public ConjugateGaugeImpl + inline void loadLinkElement(Simd ®, ref &memory) { + reg = memory; } inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu) diff --git a/lib/qcd/action/fermion/MobiusEOFAFermion.cc b/lib/qcd/action/fermion/MobiusEOFAFermion.cc new file mode 100644 index 00000000..0344afbf --- /dev/null +++ b/lib/qcd/action/fermion/MobiusEOFAFermion.cc @@ -0,0 +1,502 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include +#include + +namespace Grid { +namespace QCD { + + template + MobiusEOFAFermion::MobiusEOFAFermion( + GaugeField &_Umu, + GridCartesian &FiveDimGrid, + GridRedBlackCartesian &FiveDimRedBlackGrid, + GridCartesian &FourDimGrid, + GridRedBlackCartesian &FourDimRedBlackGrid, + RealD _mq1, RealD _mq2, RealD _mq3, + RealD _shift, int _pm, RealD _M5, + RealD _b, RealD _c, const ImplParams &p) : + AbstractEOFAFermion(_Umu, FiveDimGrid, FiveDimRedBlackGrid, + FourDimGrid, FourDimRedBlackGrid, _mq1, _mq2, _mq3, + _shift, _pm, _M5, _b, _c, p) + { + int Ls = this->Ls; + + RealD eps = 1.0; + Approx::zolotarev_data *zdata = Approx::higham(eps, this->Ls); + assert(zdata->n == this->Ls); + + std::cout << GridLogMessage << "MobiusEOFAFermion (b=" << _b << + ",c=" << _c << ") with Ls=" << Ls << std::endl; + this->SetCoefficientsTanh(zdata, _b, _c); + std::cout << GridLogMessage << "EOFA parameters: (mq1=" << _mq1 << + ",mq2=" << _mq2 << ",mq3=" << _mq3 << ",shift=" << _shift << + ",pm=" << _pm << ")" << std::endl; + + Approx::zolotarev_free(zdata); + + if(_shift != 0.0){ + SetCoefficientsPrecondShiftOps(); + } else { + Mooee_shift.resize(Ls, 0.0); + MooeeInv_shift_lc.resize(Ls, 0.0); + MooeeInv_shift_norm.resize(Ls, 0.0); + MooeeInvDag_shift_lc.resize(Ls, 0.0); + MooeeInvDag_shift_norm.resize(Ls, 0.0); + } + } + + /**************************************************************** + * Additional EOFA operators only called outside the inverter. + * Since speed is not essential, simple axpby-style + * implementations should be fine. + ***************************************************************/ + template + void MobiusEOFAFermion::Omega(const FermionField& psi, FermionField& Din, int sign, int dag) + { + int Ls = this->Ls; + RealD alpha = this->alpha; + + Din = zero; + if((sign == 1) && (dag == 0)) { // \Omega_{+} + for(int s=0; s + void MobiusEOFAFermion::Dtilde(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + RealD b = 0.5 * ( 1.0 + this->alpha ); + RealD c = 0.5 * ( 1.0 - this->alpha ); + RealD mq1 = this->mq1; + + for(int s=0; s + void MobiusEOFAFermion::DtildeInv(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + RealD m = this->mq1; + RealD c = 0.5 * this->alpha; + RealD d = 0.5; + + RealD DtInv_p(0.0), DtInv_m(0.0); + RealD N = std::pow(c+d,Ls) + m*std::pow(c-d,Ls); + FermionField tmp(this->FermionGrid()); + + for(int s=0; s sp) ? 0.0 : std::pow(-1.0,sp-s) * std::pow(c-d,sp-s) / std::pow(c+d,sp-s+1); + + if(sp == 0){ + axpby_ssp_pplus (tmp, 0.0, tmp, DtInv_p, psi, s, sp); + axpby_ssp_pminus(tmp, 0.0, tmp, DtInv_m, psi, s, sp); + } else { + axpby_ssp_pplus (tmp, 1.0, tmp, DtInv_p, psi, s, sp); + axpby_ssp_pminus(tmp, 1.0, tmp, DtInv_m, psi, s, sp); + } + + }} + } + + /*****************************************************************************************************/ + + template + RealD MobiusEOFAFermion::M(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + FermionField Din(psi._grid); + + this->Meooe5D(psi, Din); + this->DW(Din, chi, DaggerNo); + axpby(chi, 1.0, 1.0, chi, psi); + this->M5D(psi, chi); + return(norm2(chi)); + } + + template + RealD MobiusEOFAFermion::Mdag(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + FermionField Din(psi._grid); + + this->DW(psi, Din, DaggerYes); + this->MeooeDag5D(Din, chi); + this->M5Ddag(psi, chi); + axpby(chi, 1.0, 1.0, chi, psi); + return(norm2(chi)); + } + + /******************************************************************** + * Performance critical fermion operators called inside the inverter + ********************************************************************/ + + template + void MobiusEOFAFermion::M5D(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + std::vector diag(Ls,1.0); + std::vector upper(Ls,-1.0); upper[Ls-1] = this->mq1; + std::vector lower(Ls,-1.0); lower[0] = this->mq1; + + // no shift term + if(this->shift == 0.0){ this->M5D(psi, chi, chi, lower, diag, upper); } + + // fused M + shift operation + else{ this->M5D_shift(psi, chi, chi, lower, diag, upper, Mooee_shift); } + } + + template + void MobiusEOFAFermion::M5Ddag(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + std::vector diag(Ls,1.0); + std::vector upper(Ls,-1.0); upper[Ls-1] = this->mq1; + std::vector lower(Ls,-1.0); lower[0] = this->mq1; + + // no shift term + if(this->shift == 0.0){ this->M5Ddag(psi, chi, chi, lower, diag, upper); } + + // fused M + shift operation + else{ this->M5Ddag_shift(psi, chi, chi, lower, diag, upper, Mooee_shift); } + } + + // half checkerboard operations + template + void MobiusEOFAFermion::Mooee(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + // coefficients of Mooee + std::vector diag = this->bee; + std::vector upper(Ls); + std::vector lower(Ls); + for(int s=0; scee[s]; + lower[s] = -this->cee[s]; + } + upper[Ls-1] *= -this->mq1; + lower[0] *= -this->mq1; + + // no shift term + if(this->shift == 0.0){ this->M5D(psi, psi, chi, lower, diag, upper); } + + // fused M + shift operation + else { this->M5D_shift(psi, psi, chi, lower, diag, upper, Mooee_shift); } + } + + template + void MobiusEOFAFermion::MooeeDag(const FermionField& psi, FermionField& chi) + { + int Ls = this->Ls; + + // coefficients of MooeeDag + std::vector diag = this->bee; + std::vector upper(Ls); + std::vector lower(Ls); + for(int s=0; scee[s+1]; + lower[s] = this->mq1*this->cee[Ls-1]; + } else if(s==(Ls-1)) { + upper[s] = this->mq1*this->cee[0]; + lower[s] = -this->cee[s-1]; + } else { + upper[s] = -this->cee[s+1]; + lower[s] = -this->cee[s-1]; + } + } + + // no shift term + if(this->shift == 0.0){ this->M5Ddag(psi, psi, chi, lower, diag, upper); } + + // fused M + shift operation + else{ this->M5Ddag_shift(psi, psi, chi, lower, diag, upper, Mooee_shift); } + } + + /****************************************************************************************/ + + // Computes coefficients for applying Cayley preconditioned shift operators + // (Mooee + \Delta) --> Mooee_shift + // (Mooee + \Delta)^{-1} --> MooeeInv_shift_lc, MooeeInv_shift_norm + // (Mooee + \Delta)^{-dag} --> MooeeInvDag_shift_lc, MooeeInvDag_shift_norm + // For the latter two cases, the operation takes the form + // [ (Mooee + \Delta)^{-1} \psi ]_{i} = Mooee_{ij} \psi_{j} + + // ( MooeeInv_shift_norm )_{i} ( \sum_{j} [ MooeeInv_shift_lc ]_{j} P_{pm} \psi_{j} ) + template + void MobiusEOFAFermion::SetCoefficientsPrecondShiftOps() + { + int Ls = this->Ls; + int pm = this->pm; + RealD alpha = this->alpha; + RealD k = this->k; + RealD mq1 = this->mq1; + RealD shift = this->shift; + + // Initialize + Mooee_shift.resize(Ls); + MooeeInv_shift_lc.resize(Ls); + MooeeInv_shift_norm.resize(Ls); + MooeeInvDag_shift_lc.resize(Ls); + MooeeInvDag_shift_norm.resize(Ls); + + // Construct Mooee_shift + int idx(0); + Coeff_t N = ( (pm == 1) ? 1.0 : -1.0 ) * (2.0*shift*k) * + ( std::pow(alpha+1.0,Ls) + mq1*std::pow(alpha-1.0,Ls) ); + for(int s=0; s d = Mooee_shift; + std::vector u(Ls,0.0); + std::vector y(Ls,0.0); + std::vector q(Ls,0.0); + if(pm == 1){ u[0] = 1.0; } + else{ u[Ls-1] = 1.0; } + + // Tridiagonal matrix algorithm + Sherman-Morrison formula + // + // We solve + // ( Mooee' + u \otimes v ) MooeeInvDag_shift_lc = Mooee_shift + // where Mooee' is the tridiagonal part of Mooee_{+}, and + // u = (1,0,...,0) and v = (0,...,0,mq1*cee[0]) are chosen + // so that the outer-product u \otimes v gives the (0,Ls-1) + // entry of Mooee_{+}. + // + // We do this as two solves: Mooee'*y = d and Mooee'*q = u, + // and then construct the solution to the original system + // MooeeInvDag_shift_lc = y - / ( 1 + ) q + if(pm == 1){ + for(int s=1; scee[s] / this->bee[s-1]; + d[s] -= m*d[s-1]; + u[s] -= m*u[s-1]; + } + } + y[Ls-1] = d[Ls-1] / this->bee[Ls-1]; + q[Ls-1] = u[Ls-1] / this->bee[Ls-1]; + for(int s=Ls-2; s>=0; --s){ + if(pm == 1){ + y[s] = d[s] / this->bee[s]; + q[s] = u[s] / this->bee[s]; + } else { + y[s] = ( d[s] + this->cee[s]*y[s+1] ) / this->bee[s]; + q[s] = ( u[s] + this->cee[s]*q[s+1] ) / this->bee[s]; + } + } + + // Construct MooeeInvDag_shift_lc + for(int s=0; scee[0]*y[Ls-1] / + (1.0+mq1*this->cee[0]*q[Ls-1]) * q[s]; + } else { + MooeeInvDag_shift_lc[s] = y[s] - mq1*this->cee[Ls-1]*y[0] / + (1.0+mq1*this->cee[Ls-1]*q[0]) * q[s]; + } + } + + // Compute remaining coefficients + N = (pm == 1) ? (1.0 + MooeeInvDag_shift_lc[Ls-1]) : (1.0 + MooeeInvDag_shift_lc[0]); + for(int s=0; sbee[s],s) * std::pow(this->cee[s],Ls-1-s); } + else{ MooeeInv_shift_lc[s] = std::pow(this->bee[s],Ls-1-s) * std::pow(this->cee[s],s); } + + // MooeeInv_shift_norm + MooeeInv_shift_norm[s] = -MooeeInvDag_shift_lc[s] / + ( std::pow(this->bee[s],Ls) + mq1*std::pow(this->cee[s],Ls) ) / N; + + // MooeeInvDag_shift_norm + if(pm == 1){ MooeeInvDag_shift_norm[s] = -std::pow(this->bee[s],s) * std::pow(this->cee[s],Ls-1-s) / + ( std::pow(this->bee[s],Ls) + mq1*std::pow(this->cee[s],Ls) ) / N; } + else{ MooeeInvDag_shift_norm[s] = -std::pow(this->bee[s],Ls-1-s) * std::pow(this->cee[s],s) / + ( std::pow(this->bee[s],Ls) + mq1*std::pow(this->cee[s],Ls) ) / N; } + } + } + } + + // Recompute coefficients for a different value of shift constant + template + void MobiusEOFAFermion::RefreshShiftCoefficients(RealD new_shift) + { + this->shift = new_shift; + if(new_shift != 0.0){ + SetCoefficientsPrecondShiftOps(); + } else { + int Ls = this->Ls; + Mooee_shift.resize(Ls,0.0); + MooeeInv_shift_lc.resize(Ls,0.0); + MooeeInv_shift_norm.resize(Ls,0.0); + MooeeInvDag_shift_lc.resize(Ls,0.0); + MooeeInvDag_shift_norm.resize(Ls,0.0); + } + } + + template + void MobiusEOFAFermion::MooeeInternalCompute(int dag, int inv, + Vector >& Matp, Vector >& Matm) + { + int Ls = this->Ls; + + GridBase* grid = this->FermionRedBlackGrid(); + int LLs = grid->_rdimensions[0]; + + if(LLs == Ls){ return; } // Not vectorised in 5th direction + + Eigen::MatrixXcd Pplus = Eigen::MatrixXcd::Zero(Ls,Ls); + Eigen::MatrixXcd Pminus = Eigen::MatrixXcd::Zero(Ls,Ls); + + for(int s=0; sbee[s]; + Pminus(s,s) = this->bee[s]; + } + + for(int s=0; scee[s]; + Pplus(s+1,s) = -this->cee[s+1]; + } + + Pplus (0,Ls-1) = this->mq1*this->cee[0]; + Pminus(Ls-1,0) = this->mq1*this->cee[Ls-1]; + + if(this->shift != 0.0){ + RealD c = 0.5 * this->alpha; + RealD d = 0.5; + RealD N = this->shift * this->k * ( std::pow(c+d,Ls) + this->mq1*std::pow(c-d,Ls) ); + if(this->pm == 1) { + for(int s=0; s::iscomplex()) { + sp[l] = PplusMat (l*istride+s1*ostride,s2); + sm[l] = PminusMat(l*istride+s1*ostride,s2); + } else { + // if real + scalar_type tmp; + tmp = PplusMat (l*istride+s1*ostride,s2); + sp[l] = scalar_type(tmp.real(),tmp.real()); + tmp = PminusMat(l*istride+s1*ostride,s2); + sm[l] = scalar_type(tmp.real(),tmp.real()); + } + } + Matp[LLs*s2+s1] = Vp; + Matm[LLs*s2+s1] = Vm; + }} + } + + FermOpTemplateInstantiate(MobiusEOFAFermion); + GparityFermOpTemplateInstantiate(MobiusEOFAFermion); + +}} diff --git a/lib/qcd/action/fermion/MobiusEOFAFermion.h b/lib/qcd/action/fermion/MobiusEOFAFermion.h new file mode 100644 index 00000000..519b49e7 --- /dev/null +++ b/lib/qcd/action/fermion/MobiusEOFAFermion.h @@ -0,0 +1,133 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/MobiusEOFAFermion.h + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: David Murphy + +This program is free software; you can redistribute it and/or modify +it under the terms of the GNU General Public License as published by +the Free Software Foundation; either version 2 of the License, or +(at your option) any later version. + +This program is distributed in the hope that it will be useful, +but WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +GNU General Public License for more details. + +You should have received a copy of the GNU General Public License along +with this program; if not, write to the Free Software Foundation, Inc., +51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. + +See the full license in the file "LICENSE" in the top level distribution directory +*************************************************************************************/ +/* END LEGAL */ +#ifndef GRID_QCD_MOBIUS_EOFA_FERMION_H +#define GRID_QCD_MOBIUS_EOFA_FERMION_H + +#include + +namespace Grid { +namespace QCD { + + template + class MobiusEOFAFermion : public AbstractEOFAFermion + { + public: + INHERIT_IMPL_TYPES(Impl); + + public: + // Shift operator coefficients for red-black preconditioned Mobius EOFA + std::vector Mooee_shift; + std::vector MooeeInv_shift_lc; + std::vector MooeeInv_shift_norm; + std::vector MooeeInvDag_shift_lc; + std::vector MooeeInvDag_shift_norm; + + virtual void Instantiatable(void) {}; + + // EOFA-specific operations + virtual void Omega (const FermionField& in, FermionField& out, int sign, int dag); + virtual void Dtilde (const FermionField& in, FermionField& out); + virtual void DtildeInv (const FermionField& in, FermionField& out); + + // override multiply + virtual RealD M (const FermionField& in, FermionField& out); + virtual RealD Mdag (const FermionField& in, FermionField& out); + + // half checkerboard operations + virtual void Mooee (const FermionField& in, FermionField& out); + virtual void MooeeDag (const FermionField& in, FermionField& out); + virtual void MooeeInv (const FermionField& in, FermionField& out); + virtual void MooeeInv_shift (const FermionField& in, FermionField& out); + virtual void MooeeInvDag (const FermionField& in, FermionField& out); + virtual void MooeeInvDag_shift(const FermionField& in, FermionField& out); + + virtual void M5D (const FermionField& psi, FermionField& chi); + virtual void M5Ddag (const FermionField& psi, FermionField& chi); + + ///////////////////////////////////////////////////// + // Instantiate different versions depending on Impl + ///////////////////////////////////////////////////// + void M5D(const FermionField& psi, const FermionField& phi, FermionField& chi, + std::vector& lower, std::vector& diag, std::vector& upper); + + void M5D_shift(const FermionField& psi, const FermionField& phi, FermionField& chi, + std::vector& lower, std::vector& diag, std::vector& upper, + std::vector& shift_coeffs); + + void M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi, + std::vector& lower, std::vector& diag, std::vector& upper); + + void M5Ddag_shift(const FermionField& psi, const FermionField& phi, FermionField& chi, + std::vector& lower, std::vector& diag, std::vector& upper, + std::vector& shift_coeffs); + + void MooeeInternal(const FermionField& in, FermionField& out, int dag, int inv); + + void MooeeInternalCompute(int dag, int inv, Vector>& Matp, Vector>& Matm); + + void MooeeInternalAsm(const FermionField& in, FermionField& out, int LLs, int site, + Vector>& Matp, Vector>& Matm); + + void MooeeInternalZAsm(const FermionField& in, FermionField& out, int LLs, int site, + Vector>& Matp, Vector>& Matm); + + virtual void RefreshShiftCoefficients(RealD new_shift); + + // Constructors + MobiusEOFAFermion(GaugeField& _Umu, GridCartesian& FiveDimGrid, GridRedBlackCartesian& FiveDimRedBlackGrid, + GridCartesian& FourDimGrid, GridRedBlackCartesian& FourDimRedBlackGrid, + RealD _mq1, RealD _mq2, RealD _mq3, RealD _shift, int pm, + RealD _M5, RealD _b, RealD _c, const ImplParams& p=ImplParams()); + + protected: + void SetCoefficientsPrecondShiftOps(void); + }; +}} + +#define INSTANTIATE_DPERP_MOBIUS_EOFA(A)\ +template void MobiusEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, FermionField& chi, \ + std::vector& lower, std::vector& diag, std::vector& upper); \ +template void MobiusEOFAFermion::M5D_shift(const FermionField& psi, const FermionField& phi, FermionField& chi, \ + std::vector& lower, std::vector& diag, std::vector& upper, std::vector& shift_coeffs); \ +template void MobiusEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, FermionField& chi, \ + std::vector& lower, std::vector& diag, std::vector& upper); \ +template void MobiusEOFAFermion::M5Ddag_shift(const FermionField& psi, const FermionField& phi, FermionField& chi, \ + std::vector& lower, std::vector& diag, std::vector& upper, std::vector& shift_coeffs); \ +template void MobiusEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi); \ +template void MobiusEOFAFermion::MooeeInv_shift(const FermionField& psi, FermionField& chi); \ +template void MobiusEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi); \ +template void MobiusEOFAFermion::MooeeInvDag_shift(const FermionField& psi, FermionField& chi); + +#undef MOBIUS_EOFA_DPERP_DENSE +#define MOBIUS_EOFA_DPERP_CACHE +#undef MOBIUS_EOFA_DPERP_LINALG +#define MOBIUS_EOFA_DPERP_VEC + +#endif diff --git a/lib/qcd/action/fermion/MobiusEOFAFermioncache.cc b/lib/qcd/action/fermion/MobiusEOFAFermioncache.cc new file mode 100644 index 00000000..420f6390 --- /dev/null +++ b/lib/qcd/action/fermion/MobiusEOFAFermioncache.cc @@ -0,0 +1,429 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/MobiusEOFAFermioncache.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include + +namespace Grid { +namespace QCD { + + // FIXME -- make a version of these routines with site loop outermost for cache reuse. + + template + void MobiusEOFAFermion::M5D(const FermionField &psi, const FermionField &phi, FermionField &chi, + std::vector &lower, std::vector &diag, std::vector &upper) + { + int Ls = this->Ls; + GridBase *grid = psi._grid; + + assert(phi.checkerboard == psi.checkerboard); + chi.checkerboard = psi.checkerboard; + + // Flops = 6.0*(Nc*Ns) *Ls*vol + this->M5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + for(int s=0; sM5Dtime += usecond(); + } + + template + void MobiusEOFAFermion::M5D_shift(const FermionField &psi, const FermionField &phi, FermionField &chi, + std::vector &lower, std::vector &diag, std::vector &upper, + std::vector &shift_coeffs) + { + int Ls = this->Ls; + int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator + GridBase *grid = psi._grid; + + assert(phi.checkerboard == psi.checkerboard); + chi.checkerboard = psi.checkerboard; + + // Flops = 6.0*(Nc*Ns) *Ls*vol + this->M5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + for(int s=0; spm == 1){ spProj5p(tmp, psi._odata[ss+shift_s]); } + else{ spProj5m(tmp, psi._odata[ss+shift_s]); } + chi[ss+s] = chi[ss+s] + shift_coeffs[s]*tmp; + } + } + + this->M5Dtime += usecond(); + } + + template + void MobiusEOFAFermion::M5Ddag(const FermionField &psi, const FermionField &phi, FermionField &chi, + std::vector &lower, std::vector &diag, std::vector &upper) + { + int Ls = this->Ls; + GridBase *grid = psi._grid; + + assert(phi.checkerboard == psi.checkerboard); + chi.checkerboard = psi.checkerboard; + + // Flops = 6.0*(Nc*Ns) *Ls*vol + this->M5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + auto tmp = psi._odata[0]; + for(int s=0; sM5Dtime += usecond(); + } + + template + void MobiusEOFAFermion::M5Ddag_shift(const FermionField &psi, const FermionField &phi, FermionField &chi, + std::vector &lower, std::vector &diag, std::vector &upper, + std::vector &shift_coeffs) + { + int Ls = this->Ls; + int shift_s = (this->pm == 1) ? (Ls-1) : 0; // s-component modified by shift operator + GridBase *grid = psi._grid; + + assert(phi.checkerboard == psi.checkerboard); + chi.checkerboard = psi.checkerboard; + + // Flops = 6.0*(Nc*Ns) *Ls*vol + this->M5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + chi[ss+Ls-1] = zero; + auto tmp = psi._odata[0]; + for(int s=0; spm == 1){ spProj5p(tmp, psi._odata[ss+s]); } + else{ spProj5m(tmp, psi._odata[ss+s]); } + chi[ss+shift_s] = chi[ss+shift_s] + shift_coeffs[s]*tmp; + } + } + + this->M5Dtime += usecond(); + } + + template + void MobiusEOFAFermion::MooeeInv(const FermionField &psi, FermionField &chi) + { + if(this->shift != 0.0){ MooeeInv_shift(psi,chi); return; } + + GridBase *grid = psi._grid; + int Ls = this->Ls; + + chi.checkerboard = psi.checkerboard; + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + + auto tmp = psi._odata[0]; + + // Apply (L^{\prime})^{-1} + chi[ss] = psi[ss]; // chi[0]=psi[0] + for(int s=1; slee[s-1]*tmp; + } + + // L_m^{-1} + for(int s=0; sleem[s]*tmp; + } + + // U_m^{-1} D^{-1} + for(int s=0; sdee[s])*chi[ss+s] - (this->ueem[s]/this->dee[Ls-1])*tmp; + } + chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1]; + + // Apply U^{-1} + for(int s=Ls-2; s>=0; s--){ + spProj5m(tmp, chi[ss+s+1]); + chi[ss+s] = chi[ss+s] - this->uee[s]*tmp; + } + } + + this->MooeeInvTime += usecond(); + } + + template + void MobiusEOFAFermion::MooeeInv_shift(const FermionField &psi, FermionField &chi) + { + GridBase *grid = psi._grid; + int Ls = this->Ls; + + chi.checkerboard = psi.checkerboard; + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + + auto tmp1 = psi._odata[0]; + auto tmp2 = psi._odata[0]; + auto tmp2_spProj = psi._odata[0]; + + // Apply (L^{\prime})^{-1} and accumulate MooeeInv_shift_lc[j]*psi[j] in tmp2 + chi[ss] = psi[ss]; // chi[0]=psi[0] + tmp2 = MooeeInv_shift_lc[0]*psi[ss]; + for(int s=1; slee[s-1]*tmp1; + tmp2 = tmp2 + MooeeInv_shift_lc[s]*psi[ss+s]; + } + if(this->pm == 1){ spProj5p(tmp2_spProj, tmp2);} + else{ spProj5m(tmp2_spProj, tmp2); } + + // L_m^{-1} + for(int s=0; sleem[s]*tmp1; + } + + // U_m^{-1} D^{-1} + for(int s=0; sdee[s])*chi[ss+s] - (this->ueem[s]/this->dee[Ls-1])*tmp1; + } + // chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1] + MooeeInv_shift_norm[Ls-1]*tmp2_spProj; + chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1]; + spProj5m(tmp1, chi[ss+Ls-1]); + chi[ss+Ls-1] = chi[ss+Ls-1] + MooeeInv_shift_norm[Ls-1]*tmp2_spProj; + + // Apply U^{-1} and add shift term + for(int s=Ls-2; s>=0; s--){ + chi[ss+s] = chi[ss+s] - this->uee[s]*tmp1; + spProj5m(tmp1, chi[ss+s]); + chi[ss+s] = chi[ss+s] + MooeeInv_shift_norm[s]*tmp2_spProj; + } + } + + this->MooeeInvTime += usecond(); + } + + template + void MobiusEOFAFermion::MooeeInvDag(const FermionField &psi, FermionField &chi) + { + if(this->shift != 0.0){ MooeeInvDag_shift(psi,chi); return; } + + GridBase *grid = psi._grid; + int Ls = this->Ls; + + chi.checkerboard = psi.checkerboard; + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + + auto tmp = psi._odata[0]; + + // Apply (U^{\prime})^{-dag} + chi[ss] = psi[ss]; + for(int s=1; suee[s-1]*tmp; + } + + // U_m^{-\dag} + for(int s=0; sueem[s]*tmp; + } + + // L_m^{-\dag} D^{-dag} + for(int s=0; sdee[s])*chi[ss+s] - (this->leem[s]/this->dee[Ls-1])*tmp; + } + chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1]; + + // Apply L^{-dag} + for(int s=Ls-2; s>=0; s--){ + spProj5p(tmp, chi[ss+s+1]); + chi[ss+s] = chi[ss+s] - this->lee[s]*tmp; + } + } + + this->MooeeInvTime += usecond(); + } + + template + void MobiusEOFAFermion::MooeeInvDag_shift(const FermionField &psi, FermionField &chi) + { + GridBase *grid = psi._grid; + int Ls = this->Ls; + + chi.checkerboard = psi.checkerboard; + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=Ls){ + + auto tmp1 = psi._odata[0]; + auto tmp2 = psi._odata[0]; + auto tmp2_spProj = psi._odata[0]; + + // Apply (U^{\prime})^{-dag} and accumulate MooeeInvDag_shift_lc[j]*psi[j] in tmp2 + chi[ss] = psi[ss]; + tmp2 = MooeeInvDag_shift_lc[0]*psi[ss]; + for(int s=1; suee[s-1]*tmp1; + tmp2 = tmp2 + MooeeInvDag_shift_lc[s]*psi[ss+s]; + } + if(this->pm == 1){ spProj5p(tmp2_spProj, tmp2);} + else{ spProj5m(tmp2_spProj, tmp2); } + + // U_m^{-\dag} + for(int s=0; sueem[s]*tmp1; + } + + // L_m^{-\dag} D^{-dag} + for(int s=0; sdee[s])*chi[ss+s] - (this->leem[s]/this->dee[Ls-1])*tmp1; + } + chi[ss+Ls-1] = (1.0/this->dee[Ls-1])*chi[ss+Ls-1]; + spProj5p(tmp1, chi[ss+Ls-1]); + chi[ss+Ls-1] = chi[ss+Ls-1] + MooeeInvDag_shift_norm[Ls-1]*tmp2_spProj; + + // Apply L^{-dag} + for(int s=Ls-2; s>=0; s--){ + chi[ss+s] = chi[ss+s] - this->lee[s]*tmp1; + spProj5p(tmp1, chi[ss+s]); + chi[ss+s] = chi[ss+s] + MooeeInvDag_shift_norm[s]*tmp2_spProj; + } + } + + this->MooeeInvTime += usecond(); + } + + #ifdef MOBIUS_EOFA_DPERP_CACHE + + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplD); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplD); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplD); + + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplDF); + + #endif + +}} diff --git a/lib/qcd/action/fermion/MobiusEOFAFermiondense.cc b/lib/qcd/action/fermion/MobiusEOFAFermiondense.cc new file mode 100644 index 00000000..d66b8cd9 --- /dev/null +++ b/lib/qcd/action/fermion/MobiusEOFAFermiondense.cc @@ -0,0 +1,184 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/MobiusEOFAFermiondense.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include +#include + +namespace Grid { +namespace QCD { + + /* + * Dense matrix versions of routines + */ + template + void MobiusEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerNo, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInv_shift(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerNo, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerYes, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInvDag_shift(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerYes, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv) + { + int Ls = this->Ls; + int LLs = psi._grid->_rdimensions[0]; + int vol = psi._grid->oSites()/LLs; + + int pm = this->pm; + RealD shift = this->shift; + RealD alpha = this->alpha; + RealD k = this->k; + RealD mq1 = this->mq1; + + chi.checkerboard = psi.checkerboard; + + assert(Ls==LLs); + + Eigen::MatrixXd Pplus = Eigen::MatrixXd::Zero(Ls,Ls); + Eigen::MatrixXd Pminus = Eigen::MatrixXd::Zero(Ls,Ls); + + for(int s=0;sbee[s]; + Pminus(s,s) = this->bee[s]; + } + + for(int s=0; scee[s]; + } + + for(int s=0; scee[s+1]; + } + Pplus (0,Ls-1) = mq1*this->cee[0]; + Pminus(Ls-1,0) = mq1*this->cee[Ls-1]; + + if(shift != 0.0){ + Coeff_t N = 2.0 * ( std::pow(alpha+1.0,Ls) + mq1*std::pow(alpha-1.0,Ls) ); + for(int s=0; s::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplDF); + + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + #endif + +}} diff --git a/lib/qcd/action/fermion/MobiusEOFAFermionssp.cc b/lib/qcd/action/fermion/MobiusEOFAFermionssp.cc new file mode 100644 index 00000000..c86bb995 --- /dev/null +++ b/lib/qcd/action/fermion/MobiusEOFAFermionssp.cc @@ -0,0 +1,290 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/MobiusEOFAFermionssp.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include + +namespace Grid { +namespace QCD { + + // FIXME -- make a version of these routines with site loop outermost for cache reuse. + // Pminus fowards + // Pplus backwards + template + void MobiusEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; s + void MobiusEOFAFermion::M5D_shift(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper, + std::vector& shift_coeffs) + { + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; spm == 1){ axpby_ssp_pplus(chi, one, chi, shift_coeffs[s], psi, s, Ls-1); } + else{ axpby_ssp_pminus(chi, one, chi, shift_coeffs[s], psi, s, 0); } + } + } + + template + void MobiusEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; s + void MobiusEOFAFermion::M5Ddag_shift(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper, + std::vector& shift_coeffs) + { + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; spm == 1){ axpby_ssp_pplus(chi, one, chi, shift_coeffs[s], psi, Ls-1, s); } + else{ axpby_ssp_pminus(chi, one, chi, shift_coeffs[s], psi, 0, s); } + } + } + + template + void MobiusEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + if(this->shift != 0.0){ MooeeInv_shift(psi,chi); return; } + + Coeff_t one(1.0); + Coeff_t czero(0.0); + chi.checkerboard = psi.checkerboard; + int Ls = this->Ls; + + // Apply (L^{\prime})^{-1} + axpby_ssp(chi, one, psi, czero, psi, 0, 0); // chi[0]=psi[0] + for(int s=1; slee[s-1], chi, s, s-1);// recursion Psi[s] -lee P_+ chi[s-1] + } + + // L_m^{-1} + for(int s=0; sleem[s], chi, Ls-1, s); + } + + // U_m^{-1} D^{-1} + for(int s=0; sdee[s], chi, -this->ueem[s]/this->dee[Ls-1], chi, s, Ls-1); + } + axpby_ssp(chi, one/this->dee[Ls-1], chi, czero, chi, Ls-1, Ls-1); + + // Apply U^{-1} + for(int s=Ls-2; s>=0; s--){ + axpby_ssp_pminus(chi, one, chi, -this->uee[s], chi, s, s+1); // chi[Ls] + } + } + + template + void MobiusEOFAFermion::MooeeInv_shift(const FermionField& psi, FermionField& chi) + { + Coeff_t one(1.0); + Coeff_t czero(0.0); + chi.checkerboard = psi.checkerboard; + int Ls = this->Ls; + + FermionField tmp(psi._grid); + + // Apply (L^{\prime})^{-1} + axpby_ssp(chi, one, psi, czero, psi, 0, 0); // chi[0]=psi[0] + axpby_ssp(tmp, czero, tmp, this->MooeeInv_shift_lc[0], psi, 0, 0); + for(int s=1; slee[s-1], chi, s, s-1);// recursion Psi[s] -lee P_+ chi[s-1] + axpby_ssp(tmp, one, tmp, this->MooeeInv_shift_lc[s], psi, 0, s); + } + + // L_m^{-1} + for(int s=0; sleem[s], chi, Ls-1, s); + } + + // U_m^{-1} D^{-1} + for(int s=0; sdee[s], chi, -this->ueem[s]/this->dee[Ls-1], chi, s, Ls-1); + } + axpby_ssp(chi, one/this->dee[Ls-1], chi, czero, chi, Ls-1, Ls-1); + + // Apply U^{-1} and add shift term + if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInv_shift_norm[Ls-1], tmp, Ls-1, 0); } + else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInv_shift_norm[Ls-1], tmp, Ls-1, 0); } + for(int s=Ls-2; s>=0; s--){ + axpby_ssp_pminus(chi, one, chi, -this->uee[s], chi, s, s+1); // chi[Ls] + if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInv_shift_norm[s], tmp, s, 0); } + else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInv_shift_norm[s], tmp, s, 0); } + } + } + + template + void MobiusEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + if(this->shift != 0.0){ MooeeInvDag_shift(psi,chi); return; } + + Coeff_t one(1.0); + Coeff_t czero(0.0); + chi.checkerboard = psi.checkerboard; + int Ls = this->Ls; + + // Apply (U^{\prime})^{-dagger} + axpby_ssp(chi, one, psi, czero, psi, 0, 0); // chi[0]=psi[0] + for(int s=1; suee[s-1]), chi, s, s-1); + } + + // U_m^{-\dagger} + for(int s=0; sueem[s]), chi, Ls-1, s); + } + + // L_m^{-\dagger} D^{-dagger} + for(int s=0; sdee[s]), chi, -conjugate(this->leem[s]/this->dee[Ls-1]), chi, s, Ls-1); + } + axpby_ssp(chi, one/conjugate(this->dee[Ls-1]), chi, czero, chi, Ls-1, Ls-1); + + // Apply L^{-dagger} + for(int s=Ls-2; s>=0; s--){ + axpby_ssp_pplus(chi, one, chi, -conjugate(this->lee[s]), chi, s, s+1); // chi[Ls] + } + } + + template + void MobiusEOFAFermion::MooeeInvDag_shift(const FermionField& psi, FermionField& chi) + { + Coeff_t one(1.0); + Coeff_t czero(0.0); + chi.checkerboard = psi.checkerboard; + int Ls = this->Ls; + + FermionField tmp(psi._grid); + + // Apply (U^{\prime})^{-dagger} and accumulate (MooeeInvDag_shift_lc)_{j} \psi_{j} in tmp[0] + axpby_ssp(chi, one, psi, czero, psi, 0, 0); // chi[0]=psi[0] + axpby_ssp(tmp, czero, tmp, this->MooeeInvDag_shift_lc[0], psi, 0, 0); + for(int s=1; suee[s-1]), chi, s, s-1); + axpby_ssp(tmp, one, tmp, this->MooeeInvDag_shift_lc[s], psi, 0, s); + } + + // U_m^{-\dagger} + for(int s=0; sueem[s]), chi, Ls-1, s); + } + + // L_m^{-\dagger} D^{-dagger} + for(int s=0; sdee[s]), chi, -conjugate(this->leem[s]/this->dee[Ls-1]), chi, s, Ls-1); + } + axpby_ssp(chi, one/conjugate(this->dee[Ls-1]), chi, czero, chi, Ls-1, Ls-1); + + // Apply L^{-dagger} and add shift + if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInvDag_shift_norm[Ls-1], tmp, Ls-1, 0); } + else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInvDag_shift_norm[Ls-1], tmp, Ls-1, 0); } + for(int s=Ls-2; s>=0; s--){ + axpby_ssp_pplus(chi, one, chi, -conjugate(this->lee[s]), chi, s, s+1); // chi[Ls] + if(this->pm == 1){ axpby_ssp_pplus(chi, one, chi, this->MooeeInvDag_shift_norm[s], tmp, s, 0); } + else{ axpby_ssp_pminus(chi, one, chi, this->MooeeInvDag_shift_norm[s], tmp, s, 0); } + } + } + + #ifdef MOBIUS_EOFA_DPERP_LINALG + + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplD); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplD); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplD); + + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(WilsonImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(GparityWilsonImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZWilsonImplDF); + + #endif + +}} diff --git a/lib/qcd/action/fermion/MobiusEOFAFermionvec.cc b/lib/qcd/action/fermion/MobiusEOFAFermionvec.cc new file mode 100644 index 00000000..c4eaf0f3 --- /dev/null +++ b/lib/qcd/action/fermion/MobiusEOFAFermionvec.cc @@ -0,0 +1,983 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/fermion/MobiusEOFAFermionvec.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: Peter Boyle +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 +#include + +namespace Grid { +namespace QCD { + + /* + * Dense matrix versions of routines + */ + template + void MobiusEOFAFermion::MooeeInv(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerNo, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInv_shift(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerNo, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInvDag(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerYes, InverseYes); + } + + template + void MobiusEOFAFermion::MooeeInvDag_shift(const FermionField& psi, FermionField& chi) + { + this->MooeeInternal(psi, chi, DaggerYes, InverseYes); + } + + template + void MobiusEOFAFermion::M5D(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + GridBase* grid = psi._grid; + int Ls = this->Ls; + int LLs = grid->_rdimensions[0]; + const int nsimd = Simd::Nsimd(); + + Vector> u(LLs); + Vector> l(LLs); + Vector> d(LLs); + + assert(Ls/LLs == nsimd); + assert(phi.checkerboard == psi.checkerboard); + + chi.checkerboard = psi.checkerboard; + + // just directly address via type pun + typedef typename Simd::scalar_type scalar_type; + scalar_type* u_p = (scalar_type*) &u[0]; + scalar_type* l_p = (scalar_type*) &l[0]; + scalar_type* d_p = (scalar_type*) &d[0]; + + for(int o=0; oM5Dcalls++; + this->M5Dtime -= usecond(); + + assert(Nc == 3); + + parallel_for(int ss=0; ssoSites(); ss+=LLs){ // adds LLs + + #if 0 + + alignas(64) SiteHalfSpinor hp; + alignas(64) SiteHalfSpinor hm; + alignas(64) SiteSpinor fp; + alignas(64) SiteSpinor fm; + + for(int v=0; v= v){ rotate(hm, hm, nsimd-1); } + + hp = 0.5*hp; + hm = 0.5*hm; + + spRecon5m(fp, hp); + spRecon5p(fm, hm); + + chi[ss+v] = d[v]*phi[ss+v]; + chi[ss+v] = chi[ss+v] + u[v]*fp; + chi[ss+v] = chi[ss+v] + l[v]*fm; + + } + + #else + + for(int v=0; v(hp_00.v); + hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v); + hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v); + hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v); + hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v); + hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v); + } + + if(vm >= v){ + hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v); + hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v); + hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v); + hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v); + hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v); + hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v); + } + + // Can force these to real arithmetic and save 2x. + Simd p_00 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(l[v]()()(), hm_00); + Simd p_01 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(l[v]()()(), hm_01); + Simd p_02 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(l[v]()()(), hm_02); + Simd p_10 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(l[v]()()(), hm_10); + Simd p_11 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(l[v]()()(), hm_11); + Simd p_12 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(l[v]()()(), hm_12); + Simd p_20 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(u[v]()()(), hp_00); + Simd p_21 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(u[v]()()(), hp_01); + Simd p_22 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(u[v]()()(), hp_02); + Simd p_30 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(u[v]()()(), hp_10); + Simd p_31 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(u[v]()()(), hp_11); + Simd p_32 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(u[v]()()(), hp_12); + + vstream(chi[ss+v]()(0)(0), p_00); + vstream(chi[ss+v]()(0)(1), p_01); + vstream(chi[ss+v]()(0)(2), p_02); + vstream(chi[ss+v]()(1)(0), p_10); + vstream(chi[ss+v]()(1)(1), p_11); + vstream(chi[ss+v]()(1)(2), p_12); + vstream(chi[ss+v]()(2)(0), p_20); + vstream(chi[ss+v]()(2)(1), p_21); + vstream(chi[ss+v]()(2)(2), p_22); + vstream(chi[ss+v]()(3)(0), p_30); + vstream(chi[ss+v]()(3)(1), p_31); + vstream(chi[ss+v]()(3)(2), p_32); + } + + #endif + } + + this->M5Dtime += usecond(); + } + + template + void MobiusEOFAFermion::M5D_shift(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper, + std::vector& shift_coeffs) + { + #if 0 + + this->M5D(psi, phi, chi, lower, diag, upper); + + // FIXME: possible gain from vectorizing shift operation as well? + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; spm == 1){ axpby_ssp_pplus(chi, one, chi, shift_coeffs[s], psi, s, Ls-1); } + else{ axpby_ssp_pminus(chi, one, chi, shift_coeffs[s], psi, s, 0); } + } + + #else + + GridBase* grid = psi._grid; + int Ls = this->Ls; + int LLs = grid->_rdimensions[0]; + const int nsimd = Simd::Nsimd(); + + Vector> u(LLs); + Vector> l(LLs); + Vector> d(LLs); + Vector> s(LLs); + + assert(Ls/LLs == nsimd); + assert(phi.checkerboard == psi.checkerboard); + + chi.checkerboard = psi.checkerboard; + + // just directly address via type pun + typedef typename Simd::scalar_type scalar_type; + scalar_type* u_p = (scalar_type*) &u[0]; + scalar_type* l_p = (scalar_type*) &l[0]; + scalar_type* d_p = (scalar_type*) &d[0]; + scalar_type* s_p = (scalar_type*) &s[0]; + + for(int o=0; oM5Dcalls++; + this->M5Dtime -= usecond(); + + assert(Nc == 3); + + parallel_for(int ss=0; ssoSites(); ss+=LLs){ // adds LLs + + int vs = (this->pm == 1) ? LLs-1 : 0; + Simd hs_00 = (this->pm == 1) ? psi[ss+vs]()(2)(0) : psi[ss+vs]()(0)(0); + Simd hs_01 = (this->pm == 1) ? psi[ss+vs]()(2)(1) : psi[ss+vs]()(0)(1); + Simd hs_02 = (this->pm == 1) ? psi[ss+vs]()(2)(2) : psi[ss+vs]()(0)(2); + Simd hs_10 = (this->pm == 1) ? psi[ss+vs]()(3)(0) : psi[ss+vs]()(1)(0); + Simd hs_11 = (this->pm == 1) ? psi[ss+vs]()(3)(1) : psi[ss+vs]()(1)(1); + Simd hs_12 = (this->pm == 1) ? psi[ss+vs]()(3)(2) : psi[ss+vs]()(1)(2); + + for(int v=0; v(hp_00.v); + hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v); + hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v); + hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v); + hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v); + hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v); + } + + if(this->pm == 1 && vs <= v){ + hs_00.v = Optimization::Rotate::tRotate<2>(hs_00.v); + hs_01.v = Optimization::Rotate::tRotate<2>(hs_01.v); + hs_02.v = Optimization::Rotate::tRotate<2>(hs_02.v); + hs_10.v = Optimization::Rotate::tRotate<2>(hs_10.v); + hs_11.v = Optimization::Rotate::tRotate<2>(hs_11.v); + hs_12.v = Optimization::Rotate::tRotate<2>(hs_12.v); + } + + if(vm >= v){ + hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v); + hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v); + hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v); + hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v); + hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v); + hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v); + } + + if(this->pm == -1 && vs >= v){ + hs_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_00.v); + hs_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_01.v); + hs_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_02.v); + hs_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_10.v); + hs_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_11.v); + hs_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_12.v); + } + + // Can force these to real arithmetic and save 2x. + Simd p_00 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(l[v]()()(), hm_00) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(l[v]()()(), hm_00) + + switcheroo::mult(s[v]()()(), hs_00); + Simd p_01 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(l[v]()()(), hm_01) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(l[v]()()(), hm_01) + + switcheroo::mult(s[v]()()(), hs_01); + Simd p_02 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(l[v]()()(), hm_02) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(l[v]()()(), hm_02) + + switcheroo::mult(s[v]()()(), hs_02); + Simd p_10 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(l[v]()()(), hm_10) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(l[v]()()(), hm_10) + + switcheroo::mult(s[v]()()(), hs_10); + Simd p_11 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(l[v]()()(), hm_11) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(l[v]()()(), hm_11) + + switcheroo::mult(s[v]()()(), hs_11); + Simd p_12 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(l[v]()()(), hm_12) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(l[v]()()(), hm_12) + + switcheroo::mult(s[v]()()(), hs_12); + Simd p_20 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(u[v]()()(), hp_00) + + switcheroo::mult(s[v]()()(), hs_00) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(u[v]()()(), hp_00); + Simd p_21 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(u[v]()()(), hp_01) + + switcheroo::mult(s[v]()()(), hs_01) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(u[v]()()(), hp_01); + Simd p_22 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(u[v]()()(), hp_02) + + switcheroo::mult(s[v]()()(), hs_02) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(u[v]()()(), hp_02); + Simd p_30 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(u[v]()()(), hp_10) + + switcheroo::mult(s[v]()()(), hs_10) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(u[v]()()(), hp_10); + Simd p_31 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(u[v]()()(), hp_11) + + switcheroo::mult(s[v]()()(), hs_11) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(u[v]()()(), hp_11); + Simd p_32 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(u[v]()()(), hp_12) + + switcheroo::mult(s[v]()()(), hs_12) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(u[v]()()(), hp_12); + + vstream(chi[ss+v]()(0)(0), p_00); + vstream(chi[ss+v]()(0)(1), p_01); + vstream(chi[ss+v]()(0)(2), p_02); + vstream(chi[ss+v]()(1)(0), p_10); + vstream(chi[ss+v]()(1)(1), p_11); + vstream(chi[ss+v]()(1)(2), p_12); + vstream(chi[ss+v]()(2)(0), p_20); + vstream(chi[ss+v]()(2)(1), p_21); + vstream(chi[ss+v]()(2)(2), p_22); + vstream(chi[ss+v]()(3)(0), p_30); + vstream(chi[ss+v]()(3)(1), p_31); + vstream(chi[ss+v]()(3)(2), p_32); + } + } + + this->M5Dtime += usecond(); + + #endif + } + + template + void MobiusEOFAFermion::M5Ddag(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper) + { + GridBase* grid = psi._grid; + int Ls = this->Ls; + int LLs = grid->_rdimensions[0]; + int nsimd = Simd::Nsimd(); + + Vector> u(LLs); + Vector> l(LLs); + Vector> d(LLs); + + assert(Ls/LLs == nsimd); + assert(phi.checkerboard == psi.checkerboard); + + chi.checkerboard = psi.checkerboard; + + // just directly address via type pun + typedef typename Simd::scalar_type scalar_type; + scalar_type* u_p = (scalar_type*) &u[0]; + scalar_type* l_p = (scalar_type*) &l[0]; + scalar_type* d_p = (scalar_type*) &d[0]; + + for(int o=0; oM5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=LLs){ // adds LLs + + #if 0 + + alignas(64) SiteHalfSpinor hp; + alignas(64) SiteHalfSpinor hm; + alignas(64) SiteSpinor fp; + alignas(64) SiteSpinor fm; + + for(int v=0; v= v){ rotate(hm, hm, nsimd-1); } + + hp = hp*0.5; + hm = hm*0.5; + spRecon5p(fp, hp); + spRecon5m(fm, hm); + + chi[ss+v] = d[v]*phi[ss+v]+u[v]*fp; + chi[ss+v] = chi[ss+v] +l[v]*fm; + + } + + #else + + for(int v=0; v(hp_00.v); + hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v); + hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v); + hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v); + hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v); + hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v); + } + + if(vm >= v){ + hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v); + hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v); + hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v); + hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v); + hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v); + hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v); + } + + Simd p_00 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(u[v]()()(), hp_00); + Simd p_01 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(u[v]()()(), hp_01); + Simd p_02 = switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(u[v]()()(), hp_02); + Simd p_10 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(u[v]()()(), hp_10); + Simd p_11 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(u[v]()()(), hp_11); + Simd p_12 = switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(u[v]()()(), hp_12); + Simd p_20 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(l[v]()()(), hm_00); + Simd p_21 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(l[v]()()(), hm_01); + Simd p_22 = switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(l[v]()()(), hm_02); + Simd p_30 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(l[v]()()(), hm_10); + Simd p_31 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(l[v]()()(), hm_11); + Simd p_32 = switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(l[v]()()(), hm_12); + + vstream(chi[ss+v]()(0)(0), p_00); + vstream(chi[ss+v]()(0)(1), p_01); + vstream(chi[ss+v]()(0)(2), p_02); + vstream(chi[ss+v]()(1)(0), p_10); + vstream(chi[ss+v]()(1)(1), p_11); + vstream(chi[ss+v]()(1)(2), p_12); + vstream(chi[ss+v]()(2)(0), p_20); + vstream(chi[ss+v]()(2)(1), p_21); + vstream(chi[ss+v]()(2)(2), p_22); + vstream(chi[ss+v]()(3)(0), p_30); + vstream(chi[ss+v]()(3)(1), p_31); + vstream(chi[ss+v]()(3)(2), p_32); + + } + + #endif + + } + + this->M5Dtime += usecond(); + } + + template + void MobiusEOFAFermion::M5Ddag_shift(const FermionField& psi, const FermionField& phi, + FermionField& chi, std::vector& lower, std::vector& diag, std::vector& upper, + std::vector& shift_coeffs) + { + #if 0 + + this->M5Ddag(psi, phi, chi, lower, diag, upper); + + // FIXME: possible gain from vectorizing shift operation as well? + Coeff_t one(1.0); + int Ls = this->Ls; + for(int s=0; spm == 1){ axpby_ssp_pplus(chi, one, chi, shift_coeffs[s], psi, Ls-1, s); } + else{ axpby_ssp_pminus(chi, one, chi, shift_coeffs[s], psi, 0, s); } + } + + #else + + GridBase* grid = psi._grid; + int Ls = this->Ls; + int LLs = grid->_rdimensions[0]; + int nsimd = Simd::Nsimd(); + + Vector> u(LLs); + Vector> l(LLs); + Vector> d(LLs); + Vector> s(LLs); + + assert(Ls/LLs == nsimd); + assert(phi.checkerboard == psi.checkerboard); + + chi.checkerboard = psi.checkerboard; + + // just directly address via type pun + typedef typename Simd::scalar_type scalar_type; + scalar_type* u_p = (scalar_type*) &u[0]; + scalar_type* l_p = (scalar_type*) &l[0]; + scalar_type* d_p = (scalar_type*) &d[0]; + scalar_type* s_p = (scalar_type*) &s[0]; + + for(int o=0; oM5Dcalls++; + this->M5Dtime -= usecond(); + + parallel_for(int ss=0; ssoSites(); ss+=LLs){ // adds LLs + + int vs = (this->pm == 1) ? LLs-1 : 0; + Simd hs_00 = (this->pm == 1) ? psi[ss+vs]()(0)(0) : psi[ss+vs]()(2)(0); + Simd hs_01 = (this->pm == 1) ? psi[ss+vs]()(0)(1) : psi[ss+vs]()(2)(1); + Simd hs_02 = (this->pm == 1) ? psi[ss+vs]()(0)(2) : psi[ss+vs]()(2)(2); + Simd hs_10 = (this->pm == 1) ? psi[ss+vs]()(1)(0) : psi[ss+vs]()(3)(0); + Simd hs_11 = (this->pm == 1) ? psi[ss+vs]()(1)(1) : psi[ss+vs]()(3)(1); + Simd hs_12 = (this->pm == 1) ? psi[ss+vs]()(1)(2) : psi[ss+vs]()(3)(2); + + for(int v=0; v(hp_00.v); + hp_01.v = Optimization::Rotate::tRotate<2>(hp_01.v); + hp_02.v = Optimization::Rotate::tRotate<2>(hp_02.v); + hp_10.v = Optimization::Rotate::tRotate<2>(hp_10.v); + hp_11.v = Optimization::Rotate::tRotate<2>(hp_11.v); + hp_12.v = Optimization::Rotate::tRotate<2>(hp_12.v); + } + + if(this->pm == 1 && vs <= v){ + hs_00.v = Optimization::Rotate::tRotate<2>(hs_00.v); + hs_01.v = Optimization::Rotate::tRotate<2>(hs_01.v); + hs_02.v = Optimization::Rotate::tRotate<2>(hs_02.v); + hs_10.v = Optimization::Rotate::tRotate<2>(hs_10.v); + hs_11.v = Optimization::Rotate::tRotate<2>(hs_11.v); + hs_12.v = Optimization::Rotate::tRotate<2>(hs_12.v); + } + + if(vm >= v){ + hm_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_00.v); + hm_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_01.v); + hm_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_02.v); + hm_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_10.v); + hm_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_11.v); + hm_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hm_12.v); + } + + if(this->pm == -1 && vs >= v){ + hs_00.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_00.v); + hs_01.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_01.v); + hs_02.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_02.v); + hs_10.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_10.v); + hs_11.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_11.v); + hs_12.v = Optimization::Rotate::tRotate<2*Simd::Nsimd()-2>(hs_12.v); + } + + Simd p_00 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(u[v]()()(), hp_00) + + switcheroo::mult(s[v]()()(), hs_00) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(0)) + switcheroo::mult(u[v]()()(), hp_00); + Simd p_01 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(u[v]()()(), hp_01) + + switcheroo::mult(s[v]()()(), hs_01) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(1)) + switcheroo::mult(u[v]()()(), hp_01); + Simd p_02 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(u[v]()()(), hp_02) + + switcheroo::mult(s[v]()()(), hs_02) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(0)(2)) + switcheroo::mult(u[v]()()(), hp_02); + Simd p_10 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(u[v]()()(), hp_10) + + switcheroo::mult(s[v]()()(), hs_10) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(0)) + switcheroo::mult(u[v]()()(), hp_10); + Simd p_11 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(u[v]()()(), hp_11) + + switcheroo::mult(s[v]()()(), hs_11) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(1)) + switcheroo::mult(u[v]()()(), hp_11); + Simd p_12 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(u[v]()()(), hp_12) + + switcheroo::mult(s[v]()()(), hs_12) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(1)(2)) + switcheroo::mult(u[v]()()(), hp_12); + Simd p_20 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(l[v]()()(), hm_00) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(0)) + switcheroo::mult(l[v]()()(), hm_00) + + switcheroo::mult(s[v]()()(), hs_00); + Simd p_21 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(l[v]()()(), hm_01) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(1)) + switcheroo::mult(l[v]()()(), hm_01) + + switcheroo::mult(s[v]()()(), hs_01); + Simd p_22 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(l[v]()()(), hm_02) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(2)(2)) + switcheroo::mult(l[v]()()(), hm_02) + + switcheroo::mult(s[v]()()(), hs_02); + Simd p_30 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(l[v]()()(), hm_10) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(0)) + switcheroo::mult(l[v]()()(), hm_10) + + switcheroo::mult(s[v]()()(), hs_10); + Simd p_31 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(l[v]()()(), hm_11) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(1)) + switcheroo::mult(l[v]()()(), hm_11) + + switcheroo::mult(s[v]()()(), hs_11); + Simd p_32 = (this->pm == 1) ? switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(l[v]()()(), hm_12) + : switcheroo::mult(d[v]()()(), phi[ss+v]()(3)(2)) + switcheroo::mult(l[v]()()(), hm_12) + + switcheroo::mult(s[v]()()(), hs_12); + + vstream(chi[ss+v]()(0)(0), p_00); + vstream(chi[ss+v]()(0)(1), p_01); + vstream(chi[ss+v]()(0)(2), p_02); + vstream(chi[ss+v]()(1)(0), p_10); + vstream(chi[ss+v]()(1)(1), p_11); + vstream(chi[ss+v]()(1)(2), p_12); + vstream(chi[ss+v]()(2)(0), p_20); + vstream(chi[ss+v]()(2)(1), p_21); + vstream(chi[ss+v]()(2)(2), p_22); + vstream(chi[ss+v]()(3)(0), p_30); + vstream(chi[ss+v]()(3)(1), p_31); + vstream(chi[ss+v]()(3)(2), p_32); + + } + + } + + this->M5Dtime += usecond(); + + #endif + } + + #ifdef AVX512 + #include + #include + #include + #endif + + template + void MobiusEOFAFermion::MooeeInternalAsm(const FermionField& psi, FermionField& chi, + int LLs, int site, Vector >& Matp, Vector >& Matm) + { + #ifndef AVX512 + { + SiteHalfSpinor BcastP; + SiteHalfSpinor BcastM; + SiteHalfSpinor SiteChiP; + SiteHalfSpinor SiteChiM; + + // Ls*Ls * 2 * 12 * vol flops + for(int s1=0; s1); + + for(int s1=0; s1 + void MobiusEOFAFermion::MooeeInternalZAsm(const FermionField& psi, FermionField& chi, + int LLs, int site, Vector >& Matp, Vector >& Matm) + { + std::cout << "Error: zMobius not implemented for EOFA" << std::endl; + exit(-1); + }; + + template + void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv) + { + int Ls = this->Ls; + int LLs = psi._grid->_rdimensions[0]; + int vol = psi._grid->oSites()/LLs; + + chi.checkerboard = psi.checkerboard; + + Vector> Matp; + Vector> Matm; + Vector>* _Matp; + Vector>* _Matm; + + // MooeeInternalCompute(dag,inv,Matp,Matm); + if(inv && dag){ + _Matp = &this->MatpInvDag; + _Matm = &this->MatmInvDag; + } + + if(inv && (!dag)){ + _Matp = &this->MatpInv; + _Matm = &this->MatmInv; + } + + if(!inv){ + MooeeInternalCompute(dag, inv, Matp, Matm); + _Matp = &Matp; + _Matm = &Matm; + } + + assert(_Matp->size() == Ls*LLs); + + this->MooeeInvCalls++; + this->MooeeInvTime -= usecond(); + + if(switcheroo::iscomplex()){ + parallel_for(auto site=0; siteMooeeInvTime += usecond(); + } + + #ifdef MOBIUS_EOFA_DPERP_VEC + + INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplD); + INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplD); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplF); + + INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(DomainWallVec5dImplFH); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplDF); + INSTANTIATE_DPERP_MOBIUS_EOFA(ZDomainWallVec5dImplFH); + + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + template void MobiusEOFAFermion::MooeeInternal(const FermionField& psi, FermionField& chi, int dag, int inv); + + #endif + +}} diff --git a/lib/qcd/action/fermion/WilsonKernelsHand.cc b/lib/qcd/action/fermion/WilsonKernelsHand.cc index 96b8ab0a..80b81714 100644 --- a/lib/qcd/action/fermion/WilsonKernelsHand.cc +++ b/lib/qcd/action/fermion/WilsonKernelsHand.cc @@ -30,60 +30,181 @@ Author: paboyle #define REGISTER -#define LOAD_CHIMU \ - {const SiteSpinor & ref (in._odata[offset]); \ - Chimu_00=ref()(0)(0);\ - Chimu_01=ref()(0)(1);\ - Chimu_02=ref()(0)(2);\ - Chimu_10=ref()(1)(0);\ - Chimu_11=ref()(1)(1);\ - Chimu_12=ref()(1)(2);\ - Chimu_20=ref()(2)(0);\ - Chimu_21=ref()(2)(1);\ - Chimu_22=ref()(2)(2);\ - Chimu_30=ref()(3)(0);\ - Chimu_31=ref()(3)(1);\ - Chimu_32=ref()(3)(2);} +#define LOAD_CHIMU_BODY(F) \ + Chimu_00=ref(F)(0)(0); \ + Chimu_01=ref(F)(0)(1); \ + Chimu_02=ref(F)(0)(2); \ + Chimu_10=ref(F)(1)(0); \ + Chimu_11=ref(F)(1)(1); \ + Chimu_12=ref(F)(1)(2); \ + Chimu_20=ref(F)(2)(0); \ + Chimu_21=ref(F)(2)(1); \ + Chimu_22=ref(F)(2)(2); \ + Chimu_30=ref(F)(3)(0); \ + Chimu_31=ref(F)(3)(1); \ + Chimu_32=ref(F)(3)(2) -#define LOAD_CHI\ - {const SiteHalfSpinor &ref(buf[offset]); \ - Chi_00 = ref()(0)(0);\ - Chi_01 = ref()(0)(1);\ - Chi_02 = ref()(0)(2);\ - Chi_10 = ref()(1)(0);\ - Chi_11 = ref()(1)(1);\ - Chi_12 = ref()(1)(2);} +#define LOAD_CHIMU(DIR,F,PERM) \ + { const SiteSpinor & ref (in._odata[offset]); LOAD_CHIMU_BODY(F); } + +#define LOAD_CHI_BODY(F) \ + Chi_00 = ref(F)(0)(0);\ + Chi_01 = ref(F)(0)(1);\ + Chi_02 = ref(F)(0)(2);\ + Chi_10 = ref(F)(1)(0);\ + Chi_11 = ref(F)(1)(1);\ + Chi_12 = ref(F)(1)(2) + +#define LOAD_CHI(DIR,F,PERM) \ + {const SiteHalfSpinor &ref(buf[offset]); LOAD_CHI_BODY(F); } + + +//G-parity implementations using in-place intrinsic ops + +//1l 1h -> 1h 1l +//0l 0h , 1h 1l -> 0l 1h 0h,1l +//0h,1l -> 1l,0h +//if( (distance == 1 && !perm_will_occur) || (distance == -1 && perm_will_occur) ) +//Pulled fermion through forwards face, GPBC on upper component +//Need 0= 0l 1h 1= 1l 0h +//else if( (distance == -1 && !perm) || (distance == 1 && perm) ) +//Pulled fermion through backwards face, GPBC on lower component +//Need 0= 1l 0h 1= 0l 1h + +//1l 1h -> 1h 1l +//0l 0h , 1h 1l -> 0l 1h 0h,1l +#define DO_TWIST_0L_1H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \ + permute##PERM(tmp1, ref(1)(S)(C)); \ + exchange##PERM(tmp2,tmp3, ref(0)(S)(C), tmp1); \ + INTO = tmp2; + +//0l 0h -> 0h 0l +//1l 1h, 0h 0l -> 1l 0h, 1h 0l +#define DO_TWIST_1L_0H(INTO,S,C,F, PERM, tmp1, tmp2, tmp3) \ + permute##PERM(tmp1, ref(0)(S)(C)); \ + exchange##PERM(tmp2,tmp3, ref(1)(S)(C), tmp1); \ + INTO = tmp2; + + + + +#define LOAD_CHI_SETUP(DIR,F) \ + g = F; \ + direction = st._directions[DIR]; \ + distance = st._distances[DIR]; \ + sl = st._grid->_simd_layout[direction]; \ + inplace_twist = 0; \ + if(SE->_around_the_world && this->Params.twists[DIR % 4]){ \ + if(sl == 1){ \ + g = (F+1) % 2; \ + }else{ \ + inplace_twist = 1; \ + } \ + } + +#define LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM) \ + { const SiteSpinor &ref(in._odata[offset]); \ + LOAD_CHI_SETUP(DIR,F); \ + if(!inplace_twist){ \ + LOAD_CHIMU_BODY(g); \ + }else{ \ + if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \ + ( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \ + DO_TWIST_0L_1H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_0L_1H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_0L_1H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_0L_1H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_0L_1H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_0L_1H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \ + }else{ \ + DO_TWIST_1L_0H(Chimu_00,0,0,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chimu_01,0,1,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_1L_0H(Chimu_02,0,2,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chimu_10,1,0,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_1L_0H(Chimu_11,1,1,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chimu_12,1,2,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_1L_0H(Chimu_20,2,0,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chimu_21,2,1,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_1L_0H(Chimu_22,2,2,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chimu_30,3,0,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_1L_0H(Chimu_31,3,1,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chimu_32,3,2,F,PERM, U_11,U_20,U_21); \ + } \ + } \ + } + + +#define LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM) \ + { const SiteHalfSpinor &ref(buf[offset]); \ + LOAD_CHI_SETUP(DIR,F); \ + if(!inplace_twist){ \ + LOAD_CHI_BODY(g); \ + }else{ \ + if( ( F==0 && ((distance == 1 && !perm) || (distance == -1 && perm)) ) || \ + ( F==1 && ((distance == -1 && !perm) || (distance == 1 && perm)) ) ){ \ + DO_TWIST_0L_1H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_0L_1H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \ + DO_TWIST_0L_1H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \ + DO_TWIST_0L_1H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_0L_1H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \ + }else{ \ + DO_TWIST_1L_0H(Chi_00,0,0,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chi_01,0,1,F,PERM, U_11,U_20,U_21); \ + DO_TWIST_1L_0H(Chi_02,0,2,F,PERM, UChi_00,UChi_01,UChi_02); \ + DO_TWIST_1L_0H(Chi_10,1,0,F,PERM, UChi_10,UChi_11,UChi_12); \ + DO_TWIST_1L_0H(Chi_11,1,1,F,PERM, U_00,U_01,U_10); \ + DO_TWIST_1L_0H(Chi_12,1,2,F,PERM, U_11,U_20,U_21); \ + } \ + } \ + } + + +#define LOAD_CHI_GPARITY(DIR,F,PERM) LOAD_CHI_GPARITY_INPLACE_TWIST(DIR,F,PERM) +#define LOAD_CHIMU_GPARITY(DIR,F,PERM) LOAD_CHIMU_GPARITY_INPLACE_TWIST(DIR,F,PERM) // To splat or not to splat depends on the implementation -#define MULT_2SPIN(A)\ - {auto & ref(U._odata[sU](A)); \ - Impl::loadLinkElement(U_00,ref()(0,0)); \ - Impl::loadLinkElement(U_10,ref()(1,0)); \ - Impl::loadLinkElement(U_20,ref()(2,0)); \ - Impl::loadLinkElement(U_01,ref()(0,1)); \ - Impl::loadLinkElement(U_11,ref()(1,1)); \ - Impl::loadLinkElement(U_21,ref()(2,1)); \ - UChi_00 = U_00*Chi_00;\ - UChi_10 = U_00*Chi_10;\ - UChi_01 = U_10*Chi_00;\ - UChi_11 = U_10*Chi_10;\ - UChi_02 = U_20*Chi_00;\ - UChi_12 = U_20*Chi_10;\ - UChi_00+= U_01*Chi_01;\ - UChi_10+= U_01*Chi_11;\ - UChi_01+= U_11*Chi_01;\ - UChi_11+= U_11*Chi_11;\ - UChi_02+= U_21*Chi_01;\ - UChi_12+= U_21*Chi_11;\ - Impl::loadLinkElement(U_00,ref()(0,2)); \ - Impl::loadLinkElement(U_10,ref()(1,2)); \ - Impl::loadLinkElement(U_20,ref()(2,2)); \ - UChi_00+= U_00*Chi_02;\ - UChi_10+= U_00*Chi_12;\ - UChi_01+= U_10*Chi_02;\ - UChi_11+= U_10*Chi_12;\ - UChi_02+= U_20*Chi_02;\ - UChi_12+= U_20*Chi_12;} +#define MULT_2SPIN_BODY \ + Impl::loadLinkElement(U_00,ref()(0,0)); \ + Impl::loadLinkElement(U_10,ref()(1,0)); \ + Impl::loadLinkElement(U_20,ref()(2,0)); \ + Impl::loadLinkElement(U_01,ref()(0,1)); \ + Impl::loadLinkElement(U_11,ref()(1,1)); \ + Impl::loadLinkElement(U_21,ref()(2,1)); \ + UChi_00 = U_00*Chi_00; \ + UChi_10 = U_00*Chi_10; \ + UChi_01 = U_10*Chi_00; \ + UChi_11 = U_10*Chi_10; \ + UChi_02 = U_20*Chi_00; \ + UChi_12 = U_20*Chi_10; \ + UChi_00+= U_01*Chi_01; \ + UChi_10+= U_01*Chi_11; \ + UChi_01+= U_11*Chi_01; \ + UChi_11+= U_11*Chi_11; \ + UChi_02+= U_21*Chi_01; \ + UChi_12+= U_21*Chi_11; \ + Impl::loadLinkElement(U_00,ref()(0,2)); \ + Impl::loadLinkElement(U_10,ref()(1,2)); \ + Impl::loadLinkElement(U_20,ref()(2,2)); \ + UChi_00+= U_00*Chi_02; \ + UChi_10+= U_00*Chi_12; \ + UChi_01+= U_10*Chi_02; \ + UChi_11+= U_10*Chi_12; \ + UChi_02+= U_20*Chi_02; \ + UChi_12+= U_20*Chi_12 + + +#define MULT_2SPIN(A,F) \ + {auto & ref(U._odata[sU](A)); MULT_2SPIN_BODY; } + +#define MULT_2SPIN_GPARITY(A,F) \ + {auto & ref(U._odata[sU](F)(A)); MULT_2SPIN_BODY; } #define PERMUTE_DIR(dir) \ @@ -307,84 +428,87 @@ Author: paboyle result_31-= UChi_11; \ result_32-= UChi_12; -#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON) \ +#define HAND_STENCIL_LEG(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ SE=st.GetEntry(ptype,DIR,ss); \ offset = SE->_offset; \ local = SE->_is_local; \ perm = SE->_permute; \ if ( local ) { \ - LOAD_CHIMU; \ + LOAD_CHIMU_IMPL(DIR,F,PERM); \ PROJ; \ if ( perm) { \ PERMUTE_DIR(PERM); \ } \ } else { \ - LOAD_CHI; \ + LOAD_CHI_IMPL(DIR,F,PERM); \ } \ - MULT_2SPIN(DIR); \ + MULT_2SPIN_IMPL(DIR,F); \ RECON; -#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON) \ + +#define HAND_STENCIL_LEG_INT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ SE=st.GetEntry(ptype,DIR,ss); \ offset = SE->_offset; \ local = SE->_is_local; \ perm = SE->_permute; \ if ( local ) { \ - LOAD_CHIMU; \ + LOAD_CHIMU_IMPL(DIR,F,PERM); \ PROJ; \ if ( perm) { \ PERMUTE_DIR(PERM); \ } \ } else if ( st.same_node[DIR] ) { \ - LOAD_CHI; \ + LOAD_CHI_IMPL(DIR,F,PERM); \ } \ if (local || st.same_node[DIR] ) { \ - MULT_2SPIN(DIR); \ + MULT_2SPIN_IMPL(DIR,F); \ RECON; \ } -#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON) \ +#define HAND_STENCIL_LEG_EXT(PROJ,PERM,DIR,RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ SE=st.GetEntry(ptype,DIR,ss); \ offset = SE->_offset; \ + local = SE->_is_local; \ + perm = SE->_permute; \ if((!SE->_is_local)&&(!st.same_node[DIR]) ) { \ - LOAD_CHI; \ - MULT_2SPIN(DIR); \ + LOAD_CHI_IMPL(DIR,F,PERM); \ + MULT_2SPIN_IMPL(DIR,F); \ RECON; \ nmu++; \ } -#define HAND_RESULT(ss) \ +#define HAND_RESULT(ss,F) \ { \ SiteSpinor & ref (out._odata[ss]); \ - vstream(ref()(0)(0),result_00); \ - vstream(ref()(0)(1),result_01); \ - vstream(ref()(0)(2),result_02); \ - vstream(ref()(1)(0),result_10); \ - vstream(ref()(1)(1),result_11); \ - vstream(ref()(1)(2),result_12); \ - vstream(ref()(2)(0),result_20); \ - vstream(ref()(2)(1),result_21); \ - vstream(ref()(2)(2),result_22); \ - vstream(ref()(3)(0),result_30); \ - vstream(ref()(3)(1),result_31); \ - vstream(ref()(3)(2),result_32); \ + vstream(ref(F)(0)(0),result_00); \ + vstream(ref(F)(0)(1),result_01); \ + vstream(ref(F)(0)(2),result_02); \ + vstream(ref(F)(1)(0),result_10); \ + vstream(ref(F)(1)(1),result_11); \ + vstream(ref(F)(1)(2),result_12); \ + vstream(ref(F)(2)(0),result_20); \ + vstream(ref(F)(2)(1),result_21); \ + vstream(ref(F)(2)(2),result_22); \ + vstream(ref(F)(3)(0),result_30); \ + vstream(ref(F)(3)(1),result_31); \ + vstream(ref(F)(3)(2),result_32); \ } -#define HAND_RESULT_EXT(ss) \ +#define HAND_RESULT_EXT(ss,F) \ if (nmu){ \ SiteSpinor & ref (out._odata[ss]); \ - ref()(0)(0)+=result_00; \ - ref()(0)(1)+=result_01; \ - ref()(0)(2)+=result_02; \ - ref()(1)(0)+=result_10; \ - ref()(1)(1)+=result_11; \ - ref()(1)(2)+=result_12; \ - ref()(2)(0)+=result_20; \ - ref()(2)(1)+=result_21; \ - ref()(2)(2)+=result_22; \ - ref()(3)(0)+=result_30; \ - ref()(3)(1)+=result_31; \ - ref()(3)(2)+=result_32; \ + ref(F)(0)(0)+=result_00; \ + ref(F)(0)(1)+=result_01; \ + ref(F)(0)(2)+=result_02; \ + ref(F)(1)(0)+=result_10; \ + ref(F)(1)(1)+=result_11; \ + ref(F)(1)(2)+=result_12; \ + ref(F)(2)(0)+=result_20; \ + ref(F)(2)(1)+=result_21; \ + ref(F)(2)(2)+=result_22; \ + ref(F)(3)(0)+=result_30; \ + ref(F)(3)(1)+=result_31; \ + ref(F)(3)(2)+=result_32; \ } @@ -463,15 +587,18 @@ WilsonKernels::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGauge int offset,local,perm, ptype; StencilEntry *SE; - HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON); - HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM); - HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM); - HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM); - HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM); - HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM); - HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM); - HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM); - HAND_RESULT(ss); +#define HAND_DOP_SITE(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ + HAND_STENCIL_LEG(XM_PROJ,3,Xp,XM_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_RESULT(ss,F) + + HAND_DOP_SITE(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN); } template @@ -485,16 +612,19 @@ void WilsonKernels::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,Doub StencilEntry *SE; int offset,local,perm, ptype; - - HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON); - HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM); - HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM); - HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM); - HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM); - HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM); - HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM); - HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM); - HAND_RESULT(ss); + +#define HAND_DOP_SITE_DAG(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ + HAND_STENCIL_LEG(XP_PROJ,3,Xp,XP_RECON,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_RESULT(ss,F) + + HAND_DOP_SITE_DAG(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN); } template void @@ -509,16 +639,20 @@ WilsonKernels::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGa int offset,local,perm, ptype; StencilEntry *SE; - ZERO_RESULT; - HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM); - HAND_RESULT(ss); + +#define HAND_DOP_SITE_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ + ZERO_RESULT; \ + HAND_STENCIL_LEG_INT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_RESULT(ss,F) + + HAND_DOP_SITE_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN); } template @@ -532,16 +666,20 @@ void WilsonKernels::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,D StencilEntry *SE; int offset,local,perm, ptype; - ZERO_RESULT; - HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM); - HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM); - HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM); - HAND_RESULT(ss); + +#define HAND_DOP_SITE_DAG_INT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ + ZERO_RESULT; \ + HAND_STENCIL_LEG_INT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_INT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_RESULT(ss,F) + + HAND_DOP_SITE_DAG_INT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN); } template void @@ -557,16 +695,20 @@ WilsonKernels::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGa int offset,local,perm, ptype; StencilEntry *SE; int nmu=0; - ZERO_RESULT; - HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM); - HAND_RESULT_EXT(ss); + +#define HAND_DOP_SITE_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ + ZERO_RESULT; \ + HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xp,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(YM_PROJ,2,Yp,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zp,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tp,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xm,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(YP_PROJ,2,Ym,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zm,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tm,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_RESULT_EXT(ss,F) + + HAND_DOP_SITE_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN); } template @@ -581,16 +723,20 @@ void WilsonKernels::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,D StencilEntry *SE; int offset,local,perm, ptype; int nmu=0; - ZERO_RESULT; - HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM); - HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM); - HAND_RESULT_EXT(ss); + +#define HAND_DOP_SITE_DAG_EXT(F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL) \ + ZERO_RESULT; \ + HAND_STENCIL_LEG_EXT(XP_PROJ,3,Xp,XP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(YP_PROJ,2,Yp,YP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(ZP_PROJ,1,Zp,ZP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(TP_PROJ,0,Tp,TP_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(XM_PROJ,3,Xm,XM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(YM_PROJ,2,Ym,YM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(ZM_PROJ,1,Zm,ZM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_STENCIL_LEG_EXT(TM_PROJ,0,Tm,TM_RECON_ACCUM,F,LOAD_CHI_IMPL,LOAD_CHIMU_IMPL,MULT_2SPIN_IMPL); \ + HAND_RESULT_EXT(ss,F) + + HAND_DOP_SITE_DAG_EXT(, LOAD_CHI,LOAD_CHIMU,MULT_2SPIN); } //////////////////////////////////////////////// @@ -646,11 +792,124 @@ void WilsonKernels::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,D const FermionField &in, \ FermionField &out){ assert(0); } \ - HAND_SPECIALISE_EMPTY(GparityWilsonImplF); - HAND_SPECIALISE_EMPTY(GparityWilsonImplD); - HAND_SPECIALISE_EMPTY(GparityWilsonImplFH); - HAND_SPECIALISE_EMPTY(GparityWilsonImplDF); + +#define HAND_SPECIALISE_GPARITY(IMPL) \ + template<> void \ + WilsonKernels::HandDhopSite(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \ + int ss,int sU,const FermionField &in, FermionField &out) \ + { \ + typedef IMPL Impl; \ + typedef typename Simd::scalar_type S; \ + typedef typename Simd::vector_type V; \ + \ + HAND_DECLARATIONS(ignore); \ + \ + int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \ + StencilEntry *SE; \ + HAND_DOP_SITE(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + HAND_DOP_SITE(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + } \ + \ + template<> \ + void WilsonKernels::HandDhopSiteDag(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \ + int ss,int sU,const FermionField &in, FermionField &out) \ + { \ + typedef IMPL Impl; \ + typedef typename Simd::scalar_type S; \ + typedef typename Simd::vector_type V; \ + \ + HAND_DECLARATIONS(ignore); \ + \ + StencilEntry *SE; \ + int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \ + HAND_DOP_SITE_DAG(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + HAND_DOP_SITE_DAG(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + } \ + \ + template<> void \ + WilsonKernels::HandDhopSiteInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \ + int ss,int sU,const FermionField &in, FermionField &out) \ + { \ + typedef IMPL Impl; \ + typedef typename Simd::scalar_type S; \ + typedef typename Simd::vector_type V; \ + \ + HAND_DECLARATIONS(ignore); \ + \ + int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \ + StencilEntry *SE; \ + HAND_DOP_SITE_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + HAND_DOP_SITE_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + } \ + \ + template<> \ + void WilsonKernels::HandDhopSiteDagInt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \ + int ss,int sU,const FermionField &in, FermionField &out) \ + { \ + typedef IMPL Impl; \ + typedef typename Simd::scalar_type S; \ + typedef typename Simd::vector_type V; \ + \ + HAND_DECLARATIONS(ignore); \ + \ + StencilEntry *SE; \ + int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \ + HAND_DOP_SITE_DAG_INT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + HAND_DOP_SITE_DAG_INT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + } \ + \ + template<> void \ + WilsonKernels::HandDhopSiteExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \ + int ss,int sU,const FermionField &in, FermionField &out) \ + { \ + typedef IMPL Impl; \ + typedef typename Simd::scalar_type S; \ + typedef typename Simd::vector_type V; \ + \ + HAND_DECLARATIONS(ignore); \ + \ + int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \ + StencilEntry *SE; \ + int nmu=0; \ + HAND_DOP_SITE_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + nmu = 0; \ + HAND_DOP_SITE_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + } \ + template<> \ + void WilsonKernels::HandDhopSiteDagExt(StencilImpl &st,LebesgueOrder &lo,DoubledGaugeField &U,SiteHalfSpinor *buf, \ + int ss,int sU,const FermionField &in, FermionField &out) \ + { \ + typedef IMPL Impl; \ + typedef typename Simd::scalar_type S; \ + typedef typename Simd::vector_type V; \ + \ + HAND_DECLARATIONS(ignore); \ + \ + StencilEntry *SE; \ + int offset,local,perm, ptype, g, direction, distance, sl, inplace_twist; \ + int nmu=0; \ + HAND_DOP_SITE_DAG_EXT(0, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + nmu = 0; \ + HAND_DOP_SITE_DAG_EXT(1, LOAD_CHI_GPARITY,LOAD_CHIMU_GPARITY,MULT_2SPIN_GPARITY); \ + } + + +HAND_SPECIALISE_GPARITY(GparityWilsonImplF); +HAND_SPECIALISE_GPARITY(GparityWilsonImplD); +HAND_SPECIALISE_GPARITY(GparityWilsonImplFH); +HAND_SPECIALISE_GPARITY(GparityWilsonImplDF); + + + + + + + + + + + ////////////// Wilson ; uses this implementation ///////////////////// #define INSTANTIATE_THEM(A) \ diff --git a/lib/qcd/action/pseudofermion/ExactOneFlavourRatio.h b/lib/qcd/action/pseudofermion/ExactOneFlavourRatio.h new file mode 100644 index 00000000..9c1e2921 --- /dev/null +++ b/lib/qcd/action/pseudofermion/ExactOneFlavourRatio.h @@ -0,0 +1,264 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./lib/qcd/action/pseudofermion/ExactOneFlavourRatio.h + +Copyright (C) 2017 + +Author: Peter Boyle +Author: David Murphy + +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 */ + +///////////////////////////////////////////////////////////////// +// Implementation of exact one flavour algorithm (EOFA) // +// using fermion classes defined in: // +// Grid/qcd/action/fermion/DomainWallEOFAFermion.h (Shamir) // +// Grid/qcd/action/fermion/MobiusEOFAFermion.h (Mobius) // +// arXiv: 1403.1683, 1706.05843 // +///////////////////////////////////////////////////////////////// + +#ifndef QCD_PSEUDOFERMION_EXACT_ONE_FLAVOUR_RATIO_H +#define QCD_PSEUDOFERMION_EXACT_ONE_FLAVOUR_RATIO_H + +namespace Grid{ +namespace QCD{ + + /////////////////////////////////////////////////////////////// + // Exact one flavour implementation of DWF determinant ratio // + /////////////////////////////////////////////////////////////// + + template + class ExactOneFlavourRatioPseudoFermionAction : public Action + { + public: + INHERIT_IMPL_TYPES(Impl); + typedef OneFlavourRationalParams Params; + Params param; + MultiShiftFunction PowerNegHalf; + + private: + bool use_heatbath_forecasting; + AbstractEOFAFermion& Lop; // the basic LH operator + AbstractEOFAFermion& Rop; // the basic RH operator + SchurRedBlackDiagMooeeSolve Solver; + FermionField Phi; // the pseudofermion field for this trajectory + + public: + ExactOneFlavourRatioPseudoFermionAction(AbstractEOFAFermion& _Lop, AbstractEOFAFermion& _Rop, + OperatorFunction& S, Params& p, bool use_fc=false) : Lop(_Lop), Rop(_Rop), Solver(S), + Phi(_Lop.FermionGrid()), param(p), use_heatbath_forecasting(use_fc) + { + AlgRemez remez(param.lo, param.hi, param.precision); + + // MdagM^(+- 1/2) + std::cout << GridLogMessage << "Generating degree " << param.degree << " for x^(-1/2)" << std::endl; + remez.generateApprox(param.degree, 1, 2); + PowerNegHalf.Init(remez, param.tolerance, true); + }; + + virtual std::string action_name() { return "ExactOneFlavourRatioPseudoFermionAction"; } + + virtual std::string LogParameters() { + std::stringstream sstream; + sstream << GridLogMessage << "[" << action_name() << "] Low :" << param.lo << std::endl; + sstream << GridLogMessage << "[" << action_name() << "] High :" << param.hi << std::endl; + sstream << GridLogMessage << "[" << action_name() << "] Max iterations :" << param.MaxIter << std::endl; + sstream << GridLogMessage << "[" << action_name() << "] Tolerance :" << param.tolerance << std::endl; + sstream << GridLogMessage << "[" << action_name() << "] Degree :" << param.degree << std::endl; + sstream << GridLogMessage << "[" << action_name() << "] Precision :" << param.precision << std::endl; + return sstream.str(); + } + + // Spin projection + void spProj(const FermionField& in, FermionField& out, int sign, int Ls) + { + if(sign == 1){ for(int s=0; s tmp(2, Lop.FermionGrid()); + + // Use chronological inverter to forecast solutions across poles + std::vector prev_solns; + if(use_heatbath_forecasting){ prev_solns.reserve(param.degree); } + ChronoForecast, FermionField> Forecast; + + // Seed with Gaussian noise vector (var = 0.5) + RealD scale = std::sqrt(0.5); + gaussian(pRNG,eta); + eta = eta * scale; + printf("Heatbath source vector: <\\eta|\\eta> = %1.15e\n", norm2(eta)); + + // \Phi = ( \alpha_{0} + \sum_{k=1}^{N_{p}} \alpha_{l} * \gamma_{l} ) * \eta + RealD N(PowerNegHalf.norm); + for(int k=0; k tmp(2, Lop.FermionGrid()); + + // S = <\Phi|\Phi> + RealD action(norm2(Phi)); + + // LH term: S = S - k <\Phi| P_{-} \Omega_{-}^{\dagger} H(mf)^{-1} \Omega_{-} P_{-} |\Phi> + spProj(Phi, spProj_Phi, -1, Lop.Ls); + Lop.Omega(spProj_Phi, tmp[0], -1, 0); + G5R5(tmp[1], tmp[0]); + tmp[0] = zero; + Solver(Lop, tmp[1], tmp[0]); + Lop.Dtilde(tmp[0], tmp[1]); // We actually solved Cayley preconditioned system: transform back + Lop.Omega(tmp[1], tmp[0], -1, 1); + action -= Lop.k * innerProduct(spProj_Phi, tmp[0]).real(); + + // RH term: S = S + k <\Phi| P_{+} \Omega_{+}^{\dagger} ( H(mb) + // - \Delta_{+}(mf,mb) P_{+} )^{-1} \Omega_{-} P_{-} |\Phi> + spProj(Phi, spProj_Phi, 1, Rop.Ls); + Rop.Omega(spProj_Phi, tmp[0], 1, 0); + G5R5(tmp[1], tmp[0]); + tmp[0] = zero; + Solver(Rop, tmp[1], tmp[0]); + Rop.Dtilde(tmp[0], tmp[1]); + Rop.Omega(tmp[1], tmp[0], 1, 1); + action += Rop.k * innerProduct(spProj_Phi, tmp[0]).real(); + + return action; + }; + + // EOFA pseudofermion force: see Eqns. (34)-(36) of arXiv:1706.05843 + virtual void deriv(const GaugeField& U, GaugeField& dSdU) + { + Lop.ImportGauge(U); + Rop.ImportGauge(U); + + FermionField spProj_Phi (Lop.FermionGrid()); + FermionField Omega_spProj_Phi(Lop.FermionGrid()); + FermionField CG_src (Lop.FermionGrid()); + FermionField Chi (Lop.FermionGrid()); + FermionField g5_R5_Chi (Lop.FermionGrid()); + + GaugeField force(Lop.GaugeGrid()); + + // LH: dSdU = k \chi_{L}^{\dagger} \gamma_{5} R_{5} ( \partial_{x,\mu} D_{w} ) \chi_{L} + // \chi_{L} = H(mf)^{-1} \Omega_{-} P_{-} \Phi + spProj(Phi, spProj_Phi, -1, Lop.Ls); + Lop.Omega(spProj_Phi, Omega_spProj_Phi, -1, 0); + G5R5(CG_src, Omega_spProj_Phi); + spProj_Phi = zero; + Solver(Lop, CG_src, spProj_Phi); + Lop.Dtilde(spProj_Phi, Chi); + G5R5(g5_R5_Chi, Chi); + Lop.MDeriv(force, g5_R5_Chi, Chi, DaggerNo); + dSdU = Lop.k * force; + + // RH: dSdU = dSdU - k \chi_{R}^{\dagger} \gamma_{5} R_{5} ( \partial_{x,\mu} D_{w} ) \chi_{} + // \chi_{R} = ( H(mb) - \Delta_{+}(mf,mb) P_{+} )^{-1} \Omega_{+} P_{+} \Phi + spProj(Phi, spProj_Phi, 1, Rop.Ls); + Rop.Omega(spProj_Phi, Omega_spProj_Phi, 1, 0); + G5R5(CG_src, Omega_spProj_Phi); + spProj_Phi = zero; + Solver(Rop, CG_src, spProj_Phi); + Rop.Dtilde(spProj_Phi, Chi); + G5R5(g5_R5_Chi, Chi); + Lop.MDeriv(force, g5_R5_Chi, Chi, DaggerNo); + dSdU = dSdU - Rop.k * force; + }; + }; +}} + +#endif diff --git a/lib/qcd/action/pseudofermion/PseudoFermion.h b/lib/qcd/action/pseudofermion/PseudoFermion.h index bccca3d4..133ebb7d 100644 --- a/lib/qcd/action/pseudofermion/PseudoFermion.h +++ b/lib/qcd/action/pseudofermion/PseudoFermion.h @@ -38,5 +38,6 @@ directory #include #include #include +#include #endif diff --git a/lib/qcd/modules/FermionOperatorModules.h b/lib/qcd/modules/FermionOperatorModules.h index c66842c6..fc9d96a7 100644 --- a/lib/qcd/modules/FermionOperatorModules.h +++ b/lib/qcd/modules/FermionOperatorModules.h @@ -72,7 +72,7 @@ protected: } virtual unsigned int Ls(){ - return 0; + return 0; } virtual void print_parameters(){ @@ -97,7 +97,7 @@ class HMC_FermionOperatorModuleFactory : public Factory < FermionOperatorModuleBase > , Reader > { public: // use SINGLETON FUNCTOR MACRO HERE - typedef Reader TheReader; + typedef Reader TheReader; HMC_FermionOperatorModuleFactory(const HMC_FermionOperatorModuleFactory& e) = delete; void operator=(const HMC_FermionOperatorModuleFactory& e) = delete; @@ -122,7 +122,7 @@ namespace QCD{ // Modules class WilsonFermionParameters : Serializable { public: - GRID_SERIALIZABLE_CLASS_MEMBERS(WilsonFermionParameters, + GRID_SERIALIZABLE_CLASS_MEMBERS(WilsonFermionParameters, RealD, mass); }; @@ -144,7 +144,7 @@ class WilsonFermionModule: public FermionOperatorModuleGridRefs[0]; auto GridMod5d = this->GridRefs[1]; typename FermionImpl::GaugeField U(GridMod->get_full()); - this->FOPtr.reset(new MobiusFermion( U, *(GridMod->get_full()), *(GridMod->get_rb()), + this->FOPtr.reset(new MobiusFermion( U, *(GridMod->get_full()), *(GridMod->get_rb()), *(GridMod5d->get_full()), *(GridMod5d->get_rb()), this->Par_.mass, this->Par_.M5, this->Par_.b, this->Par_.c)); } @@ -175,7 +175,7 @@ class MobiusFermionModule: public FermionOperatorModuleGridRefs[0]; auto GridMod5d = this->GridRefs[1]; typename FermionImpl::GaugeField U(GridMod->get_full()); - this->FOPtr.reset(new DomainWallFermion( U, *(GridMod->get_full()), *(GridMod->get_rb()), + this->FOPtr.reset(new DomainWallFermion( U, *(GridMod->get_full()), *(GridMod->get_rb()), *(GridMod5d->get_full()), *(GridMod5d->get_rb()), this->Par_.mass, this->Par_.M5)); } }; +class DomainWallEOFAFermionParameters : Serializable { + public: + GRID_SERIALIZABLE_CLASS_MEMBERS(DomainWallEOFAFermionParameters, + RealD, mq1, + RealD, mq2, + RealD, mq3, + RealD, shift, + int, pm, + RealD, M5, + unsigned int, Ls); +}; + +template +class DomainWallEOFAFermionModule: public FermionOperatorModule { + typedef FermionOperatorModule FermBase; + using FermBase::FermBase; // for constructors + + virtual unsigned int Ls(){ + return this->Par_.Ls; + } + + // acquire resource + virtual void initialize(){ + auto GridMod = this->GridRefs[0]; + auto GridMod5d = this->GridRefs[1]; + typename FermionImpl::GaugeField U(GridMod->get_full()); + this->FOPtr.reset(new DomainWallEOFAFermion( U, *(GridMod->get_full()), *(GridMod->get_rb()), + *(GridMod5d->get_full()), *(GridMod5d->get_rb()), + this->Par_.mq1, this->Par_.mq2, this->Par_.mq3, + this->Par_.shift, this->Par_.pm, this->Par_.M5)); + } +}; + } // QCD } // Grid -#endif //FERMIONOPERATOR_MODULES_H \ No newline at end of file +#endif //FERMIONOPERATOR_MODULES_H diff --git a/lib/qcd/utils/SpaceTimeGrid.cc b/lib/qcd/utils/SpaceTimeGrid.cc index 3ada4a3b..b2b5d9c8 100644 --- a/lib/qcd/utils/SpaceTimeGrid.cc +++ b/lib/qcd/utils/SpaceTimeGrid.cc @@ -60,7 +60,7 @@ GridCartesian *SpaceTimeGrid::makeFiveDimGrid(int Ls,const GridCartesian simd5.push_back(FourDimGrid->_simd_layout[d]); mpi5.push_back(FourDimGrid->_processors[d]); } - return new GridCartesian(latt5,simd5,mpi5); + return new GridCartesian(latt5,simd5,mpi5,*FourDimGrid); } @@ -68,18 +68,14 @@ GridRedBlackCartesian *SpaceTimeGrid::makeFiveDimRedBlackGrid(int Ls,const GridC { int N4=FourDimGrid->_ndimension; int cbd=1; - std::vector latt5(1,Ls); - std::vector simd5(1,1); - std::vector mpi5(1,1); std::vector cb5(1,0); - for(int d=0;d_fdimensions[d]); - simd5.push_back(FourDimGrid->_simd_layout[d]); - mpi5.push_back(FourDimGrid->_processors[d]); cb5.push_back( 1); - } - return new GridRedBlackCartesian(latt5,simd5,mpi5,cb5,cbd); + } + GridCartesian *tmp = makeFiveDimGrid(Ls,FourDimGrid); + GridRedBlackCartesian *ret = new GridRedBlackCartesian(tmp,cb5,cbd); + delete tmp; + return ret; } @@ -97,26 +93,24 @@ GridCartesian *SpaceTimeGrid::makeFiveDimDWFGrid(int Ls,const GridCartes simd5.push_back(1); mpi5.push_back(FourDimGrid->_processors[d]); } - return new GridCartesian(latt5,simd5,mpi5); + return new GridCartesian(latt5,simd5,mpi5,*FourDimGrid); } - +/////////////////////////////////////////////////// +// Interface is inefficient and forces the deletion +// Pass in the non-redblack grid +/////////////////////////////////////////////////// GridRedBlackCartesian *SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int Ls,const GridCartesian *FourDimGrid) { int N4=FourDimGrid->_ndimension; - int nsimd = FourDimGrid->Nsimd(); int cbd=1; - std::vector latt5(1,Ls); - std::vector simd5(1,nsimd); - std::vector mpi5(1,1); std::vector cb5(1,0); - for(int d=0;d_fdimensions[d]); - simd5.push_back(1); - mpi5.push_back(FourDimGrid->_processors[d]); cb5.push_back(1); - } - return new GridRedBlackCartesian(latt5,simd5,mpi5,cb5,cbd); + } + GridCartesian *tmp = makeFiveDimDWFGrid(Ls,FourDimGrid); + GridRedBlackCartesian *ret = new GridRedBlackCartesian(tmp,cb5,cbd); + delete tmp; + return ret; } diff --git a/lib/serialisation/BaseIO.h b/lib/serialisation/BaseIO.h index ec3eefa0..24e1cec7 100644 --- a/lib/serialisation/BaseIO.h +++ b/lib/serialisation/BaseIO.h @@ -86,7 +86,7 @@ namespace Grid { or element::is_number; }; - // Vector flatening utility class //////////////////////////////////////////// + // Vector flattening utility class //////////////////////////////////////////// // Class to flatten a multidimensional std::vector template class Flatten diff --git a/lib/serialisation/JSON_IO.cc b/lib/serialisation/JSON_IO.cc index 0a3968e9..99a9cdd6 100644 --- a/lib/serialisation/JSON_IO.cc +++ b/lib/serialisation/JSON_IO.cc @@ -42,6 +42,7 @@ JSONWriter::~JSONWriter(void) // write prettified JSON to file std::ofstream os(fileName_); + //std::cout << "JSONWriter::~JSONWriter" << std::endl; os << std::setw(2) << json::parse(ss_.str()) << std::endl; } @@ -56,6 +57,7 @@ void JSONWriter::push(const string &s) void JSONWriter::pop(void) { + //std::cout << "JSONWriter::pop" << std::endl; delete_comma(); ss_ << "},"; } @@ -67,20 +69,22 @@ void JSONWriter::delete_comma() ss_.str(dlast); } + // here we are hitting a g++ bug (Bug 56480) // compiles fine with clang // have to wrap in the Grid namespace // annoying, but necessary for TravisCI namespace Grid { - template<> - void JSONWriter::writeDefault(const std::string &s, - const std::string &x) + void JSONWriter::writeDefault(const std::string &s, const std::string &x) { + //std::cout << "JSONWriter::writeDefault(string) : " << s << std::endl; + std::ostringstream os; + os << std::boolalpha << x; if (s.size()) - ss_ << "\""<< s << "\" : \"" << x << "\" ," ; + ss_ << "\""<< s << "\" : \"" << os.str() << "\" ," ; else - ss_ << "\"" << x << "\" ," ; + ss_ << os.str() << " ," ; } }// namespace Grid @@ -138,6 +142,7 @@ void JSONReader::pop(void) bool JSONReader::nextElement(const std::string &s) { + // Work in progress // JSON dictionaries do not support multiple names // Same name objects must be packed in vectors ++it_; diff --git a/lib/serialisation/JSON_IO.h b/lib/serialisation/JSON_IO.h index fc5e9631..00287447 100644 --- a/lib/serialisation/JSON_IO.h +++ b/lib/serialisation/JSON_IO.h @@ -58,10 +58,15 @@ namespace Grid void writeDefault(const std::string &s, const std::complex &x); template void writeDefault(const std::string &s, const std::vector &x); + template + void writeDefault(const std::string &s, const std::pair &x); template void writeDefault(const std::string &s, const char(&x)[N]); + void writeDefault(const std::string &s, const std::string &x); + + private: void delete_comma(); std::string fileName_; @@ -82,6 +87,8 @@ namespace Grid void readDefault(const std::string &s, std::complex &output); template void readDefault(const std::string &s, std::vector &output); + template + void readDefault(const std::string &s, std::pair &output); private: json jobject_; // main object json jcur_; // current json object @@ -106,7 +113,7 @@ namespace Grid template void JSONWriter::writeDefault(const std::string &s, const U &x) { - //std::cout << "JSONReader::writeDefault(U) : " << s << std::endl; + //std::cout << "JSONWriter::writeDefault(U) : " << s << " " << x < void JSONWriter::writeDefault(const std::string &s, const std::complex &x) { - //std::cout << "JSONReader::writeDefault(complex) : " << s << std::endl; + //std::cout << "JSONWriter::writeDefault(complex) : " << s << " " << x << std::endl; std::ostringstream os; os << "["<< std::boolalpha << x.real() << ", " << x.imag() << "]"; if (s.size()) @@ -127,10 +134,22 @@ namespace Grid ss_ << os.str() << " ," ; } + template + void JSONWriter::writeDefault(const std::string &s, const std::pair &x) + { + //std::cout << "JSONWriter::writeDefault(pair) : " << s << " " << x << std::endl; + std::ostringstream os; + os << "["<< std::boolalpha << "\""<< x.first << "\" , \"" << x.second << "\" ]"; + if (s.size()) + ss_ << "\""<< s << "\" : " << os.str() << " ," ; + else + ss_ << os.str() << " ," ; + } + template void JSONWriter::writeDefault(const std::string &s, const std::vector &x) { - //std::cout << "JSONReader::writeDefault(vec U) : " << s << std::endl; + //std::cout << "JSONWriter::writeDefault(vec U) : " << s << std::endl; if (s.size()) ss_ << " \""< void JSONWriter::writeDefault(const std::string &s, const char(&x)[N]){ - //std::cout << "JSONReader::writeDefault(char U) : " << s << std::endl; + //std::cout << "JSONWriter::writeDefault(char U) : " << s << " " << x << std::endl; if (s.size()) - ss_ << "\""<< s << "\" : \"" << x << "\" ," ; + ss_ << "\""<< s << "\" : \"" << x << "\" ," ; else - ss_ << "\"" << x << "\" ," ; + ss_ << "\"" << x << "\" ," ; } // Reader template implementation //////////////////////////////////////////// @@ -173,11 +192,35 @@ namespace Grid } + // Reader template implementation //////////////////////////////////////////// + template + void JSONReader::readDefault(const std::string &s, std::pair &output) + { + U first; + P second; + json j; + if (s.size()){ + //std::cout << "JSONReader::readDefault(pair) : " << s << " | "<< jcur_[s] << std::endl; + j = jcur_[s]; + } else { + j = jcur_; + } + json::iterator it = j.begin(); + jcur_ = *it; + read("", first); + it++; + jcur_ = *it; + read("", second); + output = std::pair(first,second); + } + + + template void JSONReader::readDefault(const std::string &s, std::complex &output) { U tmp1, tmp2; - //std::cout << "JSONReader::readDefault( complex U) : " << s << " : "<< jcur_ << std::endl; + //std::cout << "JSONReader::readDefault(complex U) : " << s << " : "<< jcur_ << std::endl; json j = jcur_; json::iterator it = j.begin(); jcur_ = *it; diff --git a/lib/simd/Grid_neon.h b/lib/simd/Grid_neon.h index d6eb9c5a..9f757fbf 100644 --- a/lib/simd/Grid_neon.h +++ b/lib/simd/Grid_neon.h @@ -82,11 +82,11 @@ namespace Optimization { double tmp[2]={a,b}; return vld1q_f64(tmp); } - //Real double // N:tbc + //Real double inline float64x2_t operator()(double a){ return vdupq_n_f64(a); } - //Integer // N:tbc + //Integer inline uint32x4_t operator()(Integer a){ return vdupq_n_u32(a); } @@ -124,33 +124,32 @@ namespace Optimization { // Nils: Vset untested; not used currently in Grid at all; // git commit 4a8c4ccfba1d05159348d21a9698028ea847e77b struct Vset{ - // Complex float // N:ok + // Complex float inline float32x4_t operator()(Grid::ComplexF *a){ float tmp[4]={a[1].imag(),a[1].real(),a[0].imag(),a[0].real()}; return vld1q_f32(tmp); } - // Complex double // N:ok + // Complex double inline float64x2_t operator()(Grid::ComplexD *a){ double tmp[2]={a[0].imag(),a[0].real()}; return vld1q_f64(tmp); } - // Real float // N:ok + // Real float inline float32x4_t operator()(float *a){ float tmp[4]={a[3],a[2],a[1],a[0]}; return vld1q_f32(tmp); } - // Real double // N:ok + // Real double inline float64x2_t operator()(double *a){ double tmp[2]={a[1],a[0]}; return vld1q_f64(tmp); } - // Integer // N:ok + // Integer inline uint32x4_t operator()(Integer *a){ return vld1q_dup_u32(a); } }; - // N:leaving as is template struct Reduce{ //Need templated class to overload output type @@ -249,9 +248,9 @@ namespace Optimization { return vfmaq_f32(r4, r0, a); // ar*br-ai*bi ai*br+ar*bi ... // no fma, use mul and add - //float32x4_t r5; - //r5 = vmulq_f32(r0, a); - //return vaddq_f32(r4, r5); + // float32x4_t r5; + // r5 = vmulq_f32(r0, a); + // return vaddq_f32(r4, r5); } // Complex double inline float64x2_t operator()(float64x2_t a, float64x2_t b){ @@ -272,9 +271,9 @@ namespace Optimization { return vfmaq_f64(r4, r0, a); // ar*br-ai*bi ai*br+ar*bi // no fma, use mul and add - //float64x2_t r5; - //r5 = vmulq_f64(r0, a); - //return vaddq_f64(r4, r5); + // float64x2_t r5; + // r5 = vmulq_f64(r0, a); + // return vaddq_f64(r4, r5); } }; @@ -421,11 +420,6 @@ namespace Optimization { } } -// working, but no restriction on n -// template static inline float32x4_t tRotate(float32x4_t in){ return vextq_f32(in,in,n); }; -// template static inline float64x2_t tRotate(float64x2_t in){ return vextq_f64(in,in,n); }; - -// restriction on n template static inline float32x4_t tRotate(float32x4_t in){ return vextq_f32(in,in,n%4); }; template static inline float64x2_t tRotate(float64x2_t in){ return vextq_f64(in,in,n%2); }; @@ -441,7 +435,7 @@ namespace Optimization { sb = vcvt_high_f32_f16(h); // there is no direct conversion from lower float32x4_t to float64x2_t // vextq_f16 not supported by clang 3.8 / 4.0 / arm clang - //float16x8_t h1 = vextq_f16(h, h, 4); // correct, but not supported by clang + // float16x8_t h1 = vextq_f16(h, h, 4); // correct, but not supported by clang // workaround for clang uint32x4_t h1u = reinterpret_cast(h); float16x8_t h1 = reinterpret_cast(vextq_u32(h1u, h1u, 2)); @@ -547,7 +541,7 @@ namespace Optimization { //Complex double Reduce - template<> // N:by Boyle + template<> inline Grid::ComplexD Reduce::operator()(float64x2_t in){ u128d conv; conv.v = in; return Grid::ComplexD(conv.f[0],conv.f[1]); @@ -562,9 +556,7 @@ namespace Optimization { //Integer Reduce template<> inline Integer Reduce::operator()(uint32x4_t in){ - // FIXME unimplemented - printf("Reduce : Missing integer implementation -> FIX\n"); - assert(0); + return vaddvq_u32(in); } } @@ -603,4 +595,5 @@ namespace Optimization { typedef Optimization::TimesMinusI TimesMinusISIMD; typedef Optimization::TimesI TimesISIMD; -} \ No newline at end of file +} + diff --git a/lib/simd/Grid_vector_types.h b/lib/simd/Grid_vector_types.h index 27585547..c67e74cb 100644 --- a/lib/simd/Grid_vector_types.h +++ b/lib/simd/Grid_vector_types.h @@ -376,7 +376,18 @@ class Grid_simd { Optimization::Exchange::Exchange0(out1.v,out2.v,in1.v,in2.v); } } - + friend inline void exchange0(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){ + Optimization::Exchange::Exchange0(out1.v,out2.v,in1.v,in2.v); + } + friend inline void exchange1(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){ + Optimization::Exchange::Exchange1(out1.v,out2.v,in1.v,in2.v); + } + friend inline void exchange2(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){ + Optimization::Exchange::Exchange2(out1.v,out2.v,in1.v,in2.v); + } + friend inline void exchange3(Grid_simd &out1,Grid_simd &out2,Grid_simd in1,Grid_simd in2){ + Optimization::Exchange::Exchange3(out1.v,out2.v,in1.v,in2.v); + } //////////////////////////////////////////////////////////////////// // General permute; assumes vector length is same across // all subtypes; may not be a good assumption, but could diff --git a/lib/stencil/Stencil.h b/lib/stencil/Stencil.h index cd0792d5..887d8a7c 100644 --- a/lib/stencil/Stencil.h +++ b/lib/stencil/Stencil.h @@ -400,11 +400,13 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal if ( sshift[0] == sshift[1] ) { if (splice_dim) { splicetime-=usecond(); - same_node = same_node && GatherSimd(source,dimension,shift,0x3,compress,face_idx); + auto tmp = GatherSimd(source,dimension,shift,0x3,compress,face_idx); + same_node = same_node && tmp; splicetime+=usecond(); } else { nosplicetime-=usecond(); - same_node = same_node && Gather(source,dimension,shift,0x3,compress,face_idx); + auto tmp = Gather(source,dimension,shift,0x3,compress,face_idx); + same_node = same_node && tmp; nosplicetime+=usecond(); } } else { @@ -412,13 +414,15 @@ class CartesianStencil { // Stencil runs along coordinate axes only; NO diagonal splicetime-=usecond(); // if checkerboard is unfavourable take two passes // both with block stride loop iteration - same_node = same_node && GatherSimd(source,dimension,shift,0x1,compress,face_idx); - same_node = same_node && GatherSimd(source,dimension,shift,0x2,compress,face_idx); + auto tmp1 = GatherSimd(source,dimension,shift,0x1,compress,face_idx); + auto tmp2 = GatherSimd(source,dimension,shift,0x2,compress,face_idx); + same_node = same_node && tmp1 && tmp2; splicetime+=usecond(); } else { nosplicetime-=usecond(); - same_node = same_node && Gather(source,dimension,shift,0x1,compress,face_idx); - same_node = same_node && Gather(source,dimension,shift,0x2,compress,face_idx); + auto tmp1 = Gather(source,dimension,shift,0x1,compress,face_idx); + auto tmp2 = Gather(source,dimension,shift,0x2,compress,face_idx); + same_node = same_node && tmp1 && tmp2; nosplicetime+=usecond(); } } diff --git a/lib/tensors/Tensor_index.h b/lib/tensors/Tensor_index.h index f114baf8..500f6c8d 100644 --- a/lib/tensors/Tensor_index.h +++ b/lib/tensors/Tensor_index.h @@ -175,7 +175,7 @@ class TensorIndexRecursion { } } template inline static - void pokeIndex(iVector &ret, const iVector::peekIndex(ret._internal[0],0)),N> &arg, int i,int j) + void pokeIndex(iVector &ret, const iVector::peekIndex(ret._internal[0],0,0)),N> &arg, int i,int j) { for(int ii=0;ii::pokeIndex(ret._internal[ii],arg._internal[ii],i,j); @@ -191,7 +191,7 @@ class TensorIndexRecursion { }} } template inline static - void pokeIndex(iMatrix &ret, const iMatrix::peekIndex(ret._internal[0][0],0)),N> &arg, int i,int j) + void pokeIndex(iMatrix &ret, const iMatrix::peekIndex(ret._internal[0][0],0,0)),N> &arg, int i,int j) { for(int ii=0;ii& coor,int index,std::vector &dims){ + static inline void CoorFromIndex (std::vector& coor,int index,const std::vector &dims){ int nd= dims.size(); coor.resize(nd); for(int d=0;d& coor,int &index,std::vector &dims){ + static inline void IndexFromCoor (const std::vector& coor,int &index,const std::vector &dims){ int nd=dims.size(); int stride=1; index=0; diff --git a/tests/IO/Test_serialisation.cc b/tests/IO/Test_serialisation.cc index d5b52044..82638ad9 100644 --- a/tests/IO/Test_serialisation.cc +++ b/tests/IO/Test_serialisation.cc @@ -1,6 +1,6 @@ /************************************************************************************* - Grid physics library, www.github.com/paboyle/Grid + Grid physics library, www.github.com/paboyle/Grid Source file: ./tests/Test_serialisation.cc @@ -29,12 +29,11 @@ Author: Peter Boyle /* END LEGAL */ #include - using namespace Grid; using namespace Grid::QCD; GRID_SERIALIZABLE_ENUM(myenum, undef, red, 1, blue, 2, green, 3); - + class myclass: Serializable { public: GRID_SERIALIZABLE_CLASS_MEMBERS(myclass, @@ -79,14 +78,14 @@ void ioTest(const std::string &filename, const O &object, const std::string &nam // writer needs to be destroyed so that writing physically happens { W writer(filename); - + write(writer, "testobject", object); } - + R reader(filename); O buf; bool good; - + read(reader, "testobject", buf); good = (object == buf); std::cout << name << " IO test: " << (good ? "success" : "failure"); @@ -98,7 +97,7 @@ int main(int argc,char **argv) { std::cout << "==== basic IO" << std::endl; XmlWriter WR("bother.xml"); - + // test basic type writing std::cout << "-- basic writing to 'bother.xml'..." << std::endl; push(WR,"BasicTypes"); @@ -112,12 +111,12 @@ int main(int argc,char **argv) write(WR,"d",d); write(WR,"b",b); pop(WR); - + // test serializable class writing myclass obj(1234); // non-trivial constructor std::vector vec; std::pair pair; - + std::cout << "-- serialisable class writing to 'bother.xml'..." << std::endl; write(WR,"obj",obj); WR.write("obj2", obj); @@ -132,11 +131,11 @@ int main(int argc,char **argv) std::cout << "-- serialisable class comparison:" << std::endl; std::cout << "vec[0] == obj: " << ((vec[0] == obj) ? "true" : "false") << std::endl; std::cout << "vec[1] == obj: " << ((vec[1] == obj) ? "true" : "false") << std::endl; - + write(WR, "objpair", pair); std::cout << "-- pair writing to std::cout:" << std::endl; std::cout << pair << std::endl; - + // read tests std::cout << "\n==== IO self-consistency tests" << std::endl; //// XML @@ -151,6 +150,11 @@ int main(int argc,char **argv) ioTest("iotest.dat", obj, "text (object) "); ioTest("iotest.dat", vec, "text (vector of objects)"); ioTest("iotest.dat", pair, "text (pair of objects)"); + //// text + ioTest("iotest.json", obj, "JSON (object) "); + ioTest("iotest.json", vec, "JSON (vector of objects)"); + ioTest("iotest.json", pair, "JSON (pair of objects)"); + //// HDF5 #undef HAVE_HDF5 #ifdef HAVE_HDF5 @@ -158,13 +162,13 @@ int main(int argc,char **argv) ioTest("iotest.h5", vec, "HDF5 (vector of objects)"); ioTest("iotest.h5", pair, "HDF5 (pair of objects)"); #endif - + std::cout << "\n==== vector flattening/reconstruction" << std::endl; typedef std::vector>> vec3d; - + vec3d dv, buf; double d = 0.; - + dv.resize(4); for (auto &v1: dv) { @@ -180,14 +184,14 @@ int main(int argc,char **argv) } std::cout << "original 3D vector:" << std::endl; std::cout << dv << std::endl; - + Flatten flatdv(dv); - + std::cout << "\ndimensions:" << std::endl; std::cout << flatdv.getDim() << std::endl; std::cout << "\nflattened vector:" << std::endl; std::cout << flatdv.getFlatVector() << std::endl; - + Reconstruct rec(flatdv.getFlatVector(), flatdv.getDim()); std::cout << "\nreconstructed vector:" << std::endl; std::cout << flatdv.getVector() << std::endl; @@ -199,10 +203,12 @@ int main(int argc,char **argv) { JSONWriter JW("bother.json"); - + // test basic type writing + myenum a = myenum::red; push(JW,"BasicTypes"); write(JW,std::string("i16"),i16); + write(JW,"myenum",a); write(JW,"u16",u16); write(JW,"i32",i32); write(JW,"u32",u32); @@ -212,23 +218,25 @@ int main(int argc,char **argv) write(JW,"d",d); write(JW,"b",b); pop(JW); - + + // test serializable class writing myclass obj(1234); // non-trivial constructor + std::cout << obj << std::endl; std::cout << "-- serialisable class writing to 'bother.json'..." << std::endl; write(JW,"obj",obj); JW.write("obj2", obj); - - std::cout << obj << std::endl; - + + std::vector vec; vec.push_back(myclass(1234)); vec.push_back(myclass(5678)); vec.push_back(myclass(3838)); write(JW, "objvec", vec); - + } + { JSONReader RD("bother.json"); myclass jcopy1; @@ -238,8 +246,9 @@ int main(int argc,char **argv) std::cout << "Loaded (JSON) -----------------" << std::endl; std::cout << jcopy1 << std::endl << jveccopy1 << std::endl; } - -/* + + +/* // This is still work in progress { // Testing the next element function diff --git a/tests/Makefile.am b/tests/Makefile.am index a8935268..7928a7fe 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -1,4 +1,4 @@ -SUBDIRS = . core forces hmc solver debug smearing IO +SUBDIRS = . core forces hmc solver debug smearing IO lanczos if BUILD_CHROMA_REGRESSION SUBDIRS+= qdpxx diff --git a/tests/Test_dwf_mixedcg_prec_halfcomms.cc b/tests/Test_dwf_mixedcg_prec_halfcomms.cc index 9cc935d9..d6aaa21e 100644 --- a/tests/Test_dwf_mixedcg_prec_halfcomms.cc +++ b/tests/Test_dwf_mixedcg_prec_halfcomms.cc @@ -80,31 +80,47 @@ int main (int argc, char ** argv) LatticeFermionD src_o(FrbGrid); - LatticeFermionD result_o(FrbGrid); - LatticeFermionD result_o_2(FrbGrid); + LatticeFermionD result_cg(FrbGrid); pickCheckerboard(Odd,src_o,src); - result_o.checkerboard = Odd; - result_o = zero; - result_o_2.checkerboard = Odd; - result_o_2 = zero; + result_cg.checkerboard = Odd; + result_cg = zero; + LatticeFermionD result_mcg(result_cg); + LatticeFermionD result_rlcg(result_cg); SchurDiagMooeeOperator HermOpEO(Ddwf); SchurDiagMooeeOperator HermOpEO_f(Ddwf_f); + //#define DO_MIXED_CG +#define DO_RLUP_CG + +#ifdef DO_MIXED_CG std::cout << "Starting mixed CG" << std::endl; MixedPrecisionConjugateGradient mCG(1.0e-8, 10000, 50, FrbGrid_f, HermOpEO_f, HermOpEO); mCG.InnerTolerance = 3.0e-5; - mCG(src_o,result_o); + mCG(src_o,result_mcg); +#endif +#ifdef DO_RLUP_CG + std::cout << "Starting reliable update CG" << std::endl; + ConjugateGradientReliableUpdate rlCG(1.e-8, 10000, 0.1, FrbGrid_f, HermOpEO_f, HermOpEO); + rlCG(src_o,result_rlcg); +#endif + std::cout << "Starting regular CG" << std::endl; ConjugateGradient CG(1.0e-8,10000); - CG(HermOpEO,src_o,result_o_2); + CG(HermOpEO,src_o,result_cg); - 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; +#ifdef DO_MIXED_CG + LatticeFermionD diff_mcg(FrbGrid); + RealD vdiff_mcg = axpy_norm(diff_mcg, -1.0, result_cg, result_mcg); + std::cout << "Diff between mixed and regular CG: " << vdiff_mcg << std::endl; +#endif +#ifdef DO_RLUP_CG + LatticeFermionD diff_rlcg(FrbGrid); + RealD vdiff_rlcg = axpy_norm(diff_rlcg, -1.0, result_cg, result_rlcg); + std::cout << "Diff between reliable update and regular CG: " << vdiff_rlcg << std::endl; +#endif Grid_finalize(); } diff --git a/tests/Test_stencil.cc b/tests/Test_stencil.cc index fa4b0b57..e9517446 100644 --- a/tests/Test_stencil.cc +++ b/tests/Test_stencil.cc @@ -48,7 +48,7 @@ int main(int argc, char ** argv) { double volume = latt_size[0]*latt_size[1]*latt_size[2]*latt_size[3]; GridCartesian Fine(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian rbFine(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian rbFine(&Fine); GridParallelRNG fRNG(&Fine); // fRNG.SeedFixedIntegers(std::vector({45,12,81,9}); diff --git a/tests/core/Test_cshift_red_black.cc b/tests/core/Test_cshift_red_black.cc index f9269709..c7b0c2f1 100644 --- a/tests/core/Test_cshift_red_black.cc +++ b/tests/core/Test_cshift_red_black.cc @@ -47,7 +47,7 @@ int main (int argc, char ** argv) mask[0]=0; GridCartesian Fine (latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBFine(latt_size,simd_layout,mpi_layout,mask,1); + GridRedBlackCartesian RBFine(&Fine,mask,1); GridParallelRNG FineRNG(&Fine); FineRNG.SeedFixedIntegers(std::vector({45,12,81,9})); diff --git a/tests/core/Test_cshift_red_black_rotate.cc b/tests/core/Test_cshift_red_black_rotate.cc index 3ef1cd21..aa9e6104 100644 --- a/tests/core/Test_cshift_red_black_rotate.cc +++ b/tests/core/Test_cshift_red_black_rotate.cc @@ -47,7 +47,7 @@ int main (int argc, char ** argv) mask[0]=0; GridCartesian Fine (latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBFine(latt_size,simd_layout,mpi_layout,mask,1); + GridRedBlackCartesian RBFine(&Fine,mask,1); GridParallelRNG FineRNG(&Fine); FineRNG.SeedFixedIntegers(std::vector({45,12,81,9})); diff --git a/tests/core/Test_dwf_eofa_even_odd.cc b/tests/core/Test_dwf_eofa_even_odd.cc new file mode 100644 index 00000000..5fe0f653 --- /dev/null +++ b/tests/core/Test_dwf_eofa_even_odd.cc @@ -0,0 +1,239 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/core/Test_dwf_eofa_even_odd.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +template +struct scal { + d internal; +}; + +Gamma::Algebra Gmu [] = { + Gamma::Algebra::GammaX, + Gamma::Algebra::GammaY, + Gamma::Algebra::GammaZ, + Gamma::Algebra::GammaT +}; + +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; + + const int Ls = 8; + // GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridCartesian* FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian* UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridRedBlackCartesian* FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + std::vector seeds4({1,2,3,4}); + std::vector seeds5({5,6,7,8}); + + GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4); + GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5); + + LatticeFermion src (FGrid); random(RNG5, src); + LatticeFermion phi (FGrid); random(RNG5, phi); + LatticeFermion chi (FGrid); random(RNG5, chi); + LatticeFermion result(FGrid); result = zero; + LatticeFermion ref (FGrid); ref = zero; + LatticeFermion tmp (FGrid); tmp = zero; + LatticeFermion err (FGrid); err = zero; + LatticeGaugeField Umu (UGrid); SU3::HotConfiguration(RNG4, Umu); + std::vector U(4,UGrid); + + // Only one non-zero (y) + Umu = zero; + for(int nn=0; nn0){ U[nn] = zero; } + PokeIndex(Umu, U[nn], nn); + } + + RealD mq1 = 0.1; + RealD mq2 = 0.5; + RealD mq3 = 1.0; + RealD shift = 0.1234; + RealD M5 = 1.8; + int pm = 1; + DomainWallEOFAFermionR Ddwf(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mq1, mq2, mq3, shift, pm, M5); + + LatticeFermion src_e (FrbGrid); + LatticeFermion src_o (FrbGrid); + LatticeFermion r_e (FrbGrid); + LatticeFermion r_o (FrbGrid); + LatticeFermion r_eo (FGrid); + LatticeFermion r_eeoo(FGrid); + + std::cout << GridLogMessage << "==========================================================" << std::endl; + std::cout << GridLogMessage << "= Testing that Meo + Moe + Moo + Mee = Munprec " << std::endl; + std::cout << GridLogMessage << "==========================================================" << std::endl; + + pickCheckerboard(Even, src_e, src); + pickCheckerboard(Odd, src_o, src); + + Ddwf.Meooe(src_e, r_o); std::cout << GridLogMessage << "Applied Meo" << std::endl; + Ddwf.Meooe(src_o, r_e); std::cout << GridLogMessage << "Applied Moe" << std::endl; + setCheckerboard(r_eo, r_o); + setCheckerboard(r_eo, r_e); + + Ddwf.Mooee(src_e, r_e); std::cout << GridLogMessage << "Applied Mee" << std::endl; + Ddwf.Mooee(src_o, r_o); std::cout << GridLogMessage << "Applied Moo" << std::endl; + setCheckerboard(r_eeoo, r_e); + setCheckerboard(r_eeoo, r_o); + + r_eo = r_eo + r_eeoo; + Ddwf.M(src, ref); + + // std::cout << GridLogMessage << r_eo << std::endl; + // std::cout << GridLogMessage << ref << std::endl; + + err = ref - r_eo; + std::cout << GridLogMessage << "EO norm diff " << norm2(err) << " " << norm2(ref) << " " << norm2(r_eo) << std::endl; + + LatticeComplex cerr(FGrid); + cerr = localInnerProduct(err,err); + // std::cout << GridLogMessage << cerr << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test Ddagger is the dagger of D by requiring " << std::endl; + std::cout << GridLogMessage << "= < phi | Deo | chi > * = < chi | Deo^dag| phi> " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + LatticeFermion chi_e (FrbGrid); + LatticeFermion chi_o (FrbGrid); + + LatticeFermion dchi_e(FrbGrid); + LatticeFermion dchi_o(FrbGrid); + + LatticeFermion phi_e (FrbGrid); + LatticeFermion phi_o (FrbGrid); + + LatticeFermion dphi_e(FrbGrid); + LatticeFermion dphi_o(FrbGrid); + + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + pickCheckerboard(Even, phi_e, phi); + pickCheckerboard(Odd , phi_o, phi); + + Ddwf.Meooe (chi_e, dchi_o); + Ddwf.Meooe (chi_o, dchi_e); + Ddwf.MeooeDag(phi_e, dphi_o); + Ddwf.MeooeDag(phi_o, dphi_e); + + ComplexD pDce = innerProduct(phi_e, dchi_e); + ComplexD pDco = innerProduct(phi_o, dchi_o); + ComplexD cDpe = innerProduct(chi_e, dphi_e); + ComplexD cDpo = innerProduct(chi_o, dphi_o); + + std::cout << GridLogMessage << "e " << pDce << " " << cDpe << std::endl; + std::cout << GridLogMessage << "o " << pDco << " " << cDpo << std::endl; + + std::cout << GridLogMessage << "pDce - conj(cDpo) " << pDce-conj(cDpo) << std::endl; + std::cout << GridLogMessage << "pDco - conj(cDpe) " << pDco-conj(cDpe) << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test MeeInv Mee = 1 " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + + Ddwf.Mooee (chi_e, src_e); + Ddwf.MooeeInv(src_e, phi_e); + + Ddwf.Mooee (chi_o, src_o); + Ddwf.MooeeInv(src_o, phi_o); + + setCheckerboard(phi, phi_e); + setCheckerboard(phi, phi_o); + + err = phi - chi; + std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test MeeInvDag MeeDag = 1 " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + + Ddwf.MooeeDag (chi_e, src_e); + Ddwf.MooeeInvDag(src_e, phi_e); + + Ddwf.MooeeDag (chi_o, src_o); + Ddwf.MooeeInvDag(src_o, phi_o); + + setCheckerboard(phi, phi_e); + setCheckerboard(phi, phi_o); + + err = phi - chi; + std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test MpcDagMpc is Hermitian " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + random(RNG5, phi); + random(RNG5, chi); + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + pickCheckerboard(Even, phi_e, phi); + pickCheckerboard(Odd , phi_o, phi); + RealD t1,t2; + + SchurDiagMooeeOperator HermOpEO(Ddwf); + HermOpEO.MpcDagMpc(chi_e, dchi_e, t1, t2); + HermOpEO.MpcDagMpc(chi_o, dchi_o, t1, t2); + + HermOpEO.MpcDagMpc(phi_e, dphi_e, t1, t2); + HermOpEO.MpcDagMpc(phi_o, dphi_o, t1, t2); + + pDce = innerProduct(phi_e, dchi_e); + pDco = innerProduct(phi_o, dchi_o); + cDpe = innerProduct(chi_e, dphi_e); + cDpo = innerProduct(chi_o, dphi_o); + + std::cout << GridLogMessage << "e " << pDce << " " << cDpe << std::endl; + std::cout << GridLogMessage << "o " << pDco << " " << cDpo << std::endl; + + std::cout << GridLogMessage << "pDce - conj(cDpo) " << pDco-conj(cDpo) << std::endl; + std::cout << GridLogMessage << "pDco - conj(cDpe) " << pDce-conj(cDpe) << std::endl; + + Grid_finalize(); +} diff --git a/tests/core/Test_fft.cc b/tests/core/Test_fft.cc index 877683f0..b2336cfa 100644 --- a/tests/core/Test_fft.cc +++ b/tests/core/Test_fft.cc @@ -47,7 +47,7 @@ int main (int argc, char ** argv) vol = vol * latt_size[d]; } GridCartesian GRID(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGRID(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGRID(&GRID); LatticeComplexD one(&GRID); LatticeComplexD zz(&GRID); diff --git a/tests/core/Test_gparity.cc b/tests/core/Test_gparity.cc index cfb5d2c3..81091e9e 100644 --- a/tests/core/Test_gparity.cc +++ b/tests/core/Test_gparity.cc @@ -33,22 +33,68 @@ using namespace std; using namespace Grid; using namespace Grid::QCD; -typedef typename GparityDomainWallFermionR::FermionField FermionField; +//typedef GparityDomainWallFermionD GparityDiracOp; +//typedef DomainWallFermionD StandardDiracOp; +//#define DOP_PARAMS +typedef GparityMobiusFermionD GparityDiracOp; +typedef MobiusFermionD StandardDiracOp; +#define DOP_PARAMS ,1.5, 0.5 + + +typedef typename GparityDiracOp::FermionField GparityFermionField; +typedef typename GparityDiracOp::GaugeField GparityGaugeField; +typedef typename GparityFermionField::vector_type vComplexType; + +typedef typename StandardDiracOp::FermionField StandardFermionField; +typedef typename StandardDiracOp::GaugeField StandardGaugeField; + +enum{ same_vComplex = std::is_same::value }; +static_assert(same_vComplex == 1, "Dirac Operators must have same underlying SIMD complex type"); int main (int argc, char ** argv) { - const int nu = 3; + int nu = 0; Grid_init(&argc,&argv); + for(int i=1;i> nu; + std::cout << GridLogMessage << "Set Gparity direction to " << nu << std::endl; + } + } + + std::cout << GridLogMessage<< "*****************************************************************" < latt_2f(Nd,L); - std::vector latt_1f(Nd,L); latt_1f[nu] = 2*L; + //const int L =4; + //std::vector latt_2f(Nd,L); - std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector latt_2f = GridDefaultLatt(); + std::vector latt_1f(latt_2f); latt_1f[nu] = 2*latt_2f[nu]; + int L = latt_2f[nu]; + + + std::vector simd_layout = GridDefaultSimd(Nd,vComplexType::Nsimd()); + + std::cout << GridLogMessage << "SIMD layout: "; + for(int i=0;i mpi_layout = GridDefaultMpi(); //node layout GridCartesian * UGrid_1f = SpaceTimeGrid::makeFourDimGrid(latt_1f, simd_layout, mpi_layout); @@ -67,13 +113,13 @@ int main (int argc, char ** argv) GridParallelRNG RNG5_2f(FGrid_2f); RNG5_2f.SeedFixedIntegers(seeds5); GridParallelRNG RNG4_2f(UGrid_2f); RNG4_2f.SeedFixedIntegers(seeds4); - LatticeGaugeField Umu_2f(UGrid_2f); + GparityGaugeField Umu_2f(UGrid_2f); SU3::HotConfiguration(RNG4_2f,Umu_2f); - LatticeFermion src (FGrid_2f); - LatticeFermion tmpsrc(FGrid_2f); - FermionField src_2f(FGrid_2f); - LatticeFermion src_1f(FGrid_1f); + StandardFermionField src (FGrid_2f); + StandardFermionField tmpsrc(FGrid_2f); + GparityFermionField src_2f(FGrid_2f); + StandardFermionField src_1f(FGrid_1f); // Replicate fermion source random(RNG5_2f,src); @@ -81,8 +127,8 @@ int main (int argc, char ** argv) tmpsrc=src*2.0; PokeIndex<0>(src_2f,tmpsrc,1); - LatticeFermion result_1f(FGrid_1f); result_1f=zero; - LatticeGaugeField Umu_1f(UGrid_1f); + StandardFermionField result_1f(FGrid_1f); result_1f=zero; + StandardGaugeField Umu_1f(UGrid_1f); Replicate(Umu_2f,Umu_1f); //Coordinate grid for reference @@ -92,7 +138,7 @@ int main (int argc, char ** argv) //Copy-conjugate the gauge field //First C-shift the lattice by Lx/2 { - LatticeGaugeField Umu_shift = conjugate( Cshift(Umu_1f,nu,L) ); + StandardGaugeField Umu_shift = conjugate( Cshift(Umu_1f,nu,L) ); Umu_1f = where( xcoor_1f >= Integer(L), Umu_shift, Umu_1f ); // hack test to check the same @@ -101,7 +147,7 @@ int main (int argc, char ** argv) cout << GridLogMessage << "Umu diff " << norm2(Umu_shift)<(Umu_1f,nu)) Unu(UGrid_1f); Unu = PeekIndex(Umu_1f,nu); Unu = where(xcoor_1f == Integer(2*L-1), -Unu, Unu); PokeIndex(Umu_1f,Unu,nu); @@ -115,33 +161,33 @@ int main (int argc, char ** argv) RealD mass=0.0; RealD M5=1.8; - DomainWallFermionR Ddwf(Umu_1f,*FGrid_1f,*FrbGrid_1f,*UGrid_1f,*UrbGrid_1f,mass,M5); + StandardDiracOp Ddwf(Umu_1f,*FGrid_1f,*FrbGrid_1f,*UGrid_1f,*UrbGrid_1f,mass,M5 DOP_PARAMS); - LatticeFermion src_o_1f(FrbGrid_1f); - LatticeFermion result_o_1f(FrbGrid_1f); + StandardFermionField src_o_1f(FrbGrid_1f); + StandardFermionField result_o_1f(FrbGrid_1f); pickCheckerboard(Odd,src_o_1f,src_1f); result_o_1f=zero; - SchurDiagMooeeOperator HermOpEO(Ddwf); - ConjugateGradient CG(1.0e-8,10000); + SchurDiagMooeeOperator HermOpEO(Ddwf); + ConjugateGradient CG(1.0e-8,10000); CG(HermOpEO,src_o_1f,result_o_1f); // const int nu = 3; std::vector twists(Nd,0); twists[nu] = 1; - GparityDomainWallFermionR::ImplParams params; + GparityDiracOp::ImplParams params; params.twists = twists; - GparityDomainWallFermionR GPDdwf(Umu_2f,*FGrid_2f,*FrbGrid_2f,*UGrid_2f,*UrbGrid_2f,mass,M5,params); + GparityDiracOp GPDdwf(Umu_2f,*FGrid_2f,*FrbGrid_2f,*UGrid_2f,*UrbGrid_2f,mass,M5 DOP_PARAMS,params); for(int disp=-1;disp<=1;disp+=2) for(int mu=0;mu<5;mu++) { - FermionField Dsrc_2f(FGrid_2f); + GparityFermionField Dsrc_2f(FGrid_2f); - LatticeFermion Dsrc_1f(FGrid_1f); - LatticeFermion Dsrc_2freplica(FGrid_1f); - LatticeFermion Dsrc_2freplica0(FGrid_1f); - LatticeFermion Dsrc_2freplica1(FGrid_1f); + StandardFermionField Dsrc_1f(FGrid_1f); + StandardFermionField Dsrc_2freplica(FGrid_1f); + StandardFermionField Dsrc_2freplica0(FGrid_1f); + StandardFermionField Dsrc_2freplica1(FGrid_1f); if ( mu ==0 ) { std::cout << GridLogMessage<< " Cross checking entire hopping term"<(Dsrc_2f,0); - LatticeFermion Dsrc_2f1(FGrid_2f); Dsrc_2f1 = PeekIndex<0>(Dsrc_2f,1); + StandardFermionField Dsrc_2f0(FGrid_2f); Dsrc_2f0 = PeekIndex<0>(Dsrc_2f,0); + StandardFermionField Dsrc_2f1(FGrid_2f); Dsrc_2f1 = PeekIndex<0>(Dsrc_2f,1); // Dsrc_2f1 = Dsrc_2f1 - Dsrc_2f0; // std::cout << GridLogMessage << " Cross check two halves " < CG2f(1.0e-8,10000); - SchurDiagMooeeOperator HermOpEO2f(GPDdwf); + ConjugateGradient CG2f(1.0e-8,10000); + SchurDiagMooeeOperator HermOpEO2f(GPDdwf); CG2f(HermOpEO2f,src_o_2f,result_o_2f); std::cout << "2f cb "< simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< +Author: paboyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +template +struct scal { + d internal; +}; + +Gamma::Algebra Gmu [] = { + Gamma::Algebra::GammaX, + Gamma::Algebra::GammaY, + Gamma::Algebra::GammaZ, + Gamma::Algebra::GammaT +}; + +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; + + const int Ls = 8; + // GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridCartesian* FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian* UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridRedBlackCartesian* FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + std::vector seeds4({1,2,3,4}); + std::vector seeds5({5,6,7,8}); + + GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4); + GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5); + + LatticeFermion src (FGrid); random(RNG5, src); + LatticeFermion phi (FGrid); random(RNG5, phi); + LatticeFermion chi (FGrid); random(RNG5, chi); + LatticeFermion result(FGrid); result = zero; + LatticeFermion ref (FGrid); ref = zero; + LatticeFermion tmp (FGrid); tmp = zero; + LatticeFermion err (FGrid); err = zero; + LatticeGaugeField Umu (UGrid); SU3::HotConfiguration(RNG4, Umu); + std::vector U(4,UGrid); + + // Only one non-zero (y) + Umu = zero; + for(int nn=0; nn0){ U[nn] = zero; } + PokeIndex(Umu, U[nn], nn); + } + + RealD b = 2.5; + RealD c = 1.5; + RealD mq1 = 0.1; + RealD mq2 = 0.5; + RealD mq3 = 1.0; + RealD shift = 0.1234; + RealD M5 = 1.8; + int pm = 1; + MobiusEOFAFermionR Ddwf(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mq1, mq2, mq3, shift, pm, M5, b, c); + + LatticeFermion src_e (FrbGrid); + LatticeFermion src_o (FrbGrid); + LatticeFermion r_e (FrbGrid); + LatticeFermion r_o (FrbGrid); + LatticeFermion r_eo (FGrid); + LatticeFermion r_eeoo(FGrid); + + std::cout << GridLogMessage << "==========================================================" << std::endl; + std::cout << GridLogMessage << "= Testing that Meo + Moe + Moo + Mee = Munprec " << std::endl; + std::cout << GridLogMessage << "==========================================================" << std::endl; + + pickCheckerboard(Even, src_e, src); + pickCheckerboard(Odd, src_o, src); + + Ddwf.Meooe(src_e, r_o); std::cout << GridLogMessage << "Applied Meo" << std::endl; + Ddwf.Meooe(src_o, r_e); std::cout << GridLogMessage << "Applied Moe" << std::endl; + setCheckerboard(r_eo, r_o); + setCheckerboard(r_eo, r_e); + + Ddwf.Mooee(src_e, r_e); std::cout << GridLogMessage << "Applied Mee" << std::endl; + Ddwf.Mooee(src_o, r_o); std::cout << GridLogMessage << "Applied Moo" << std::endl; + setCheckerboard(r_eeoo, r_e); + setCheckerboard(r_eeoo, r_o); + + r_eo = r_eo + r_eeoo; + Ddwf.M(src, ref); + + // std::cout << GridLogMessage << r_eo << std::endl; + // std::cout << GridLogMessage << ref << std::endl; + + err = ref - r_eo; + std::cout << GridLogMessage << "EO norm diff " << norm2(err) << " " << norm2(ref) << " " << norm2(r_eo) << std::endl; + + LatticeComplex cerr(FGrid); + cerr = localInnerProduct(err,err); + // std::cout << GridLogMessage << cerr << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test Ddagger is the dagger of D by requiring " << std::endl; + std::cout << GridLogMessage << "= < phi | Deo | chi > * = < chi | Deo^dag| phi> " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + LatticeFermion chi_e (FrbGrid); + LatticeFermion chi_o (FrbGrid); + + LatticeFermion dchi_e(FrbGrid); + LatticeFermion dchi_o(FrbGrid); + + LatticeFermion phi_e (FrbGrid); + LatticeFermion phi_o (FrbGrid); + + LatticeFermion dphi_e(FrbGrid); + LatticeFermion dphi_o(FrbGrid); + + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + pickCheckerboard(Even, phi_e, phi); + pickCheckerboard(Odd , phi_o, phi); + + Ddwf.Meooe (chi_e, dchi_o); + Ddwf.Meooe (chi_o, dchi_e); + Ddwf.MeooeDag(phi_e, dphi_o); + Ddwf.MeooeDag(phi_o, dphi_e); + + ComplexD pDce = innerProduct(phi_e, dchi_e); + ComplexD pDco = innerProduct(phi_o, dchi_o); + ComplexD cDpe = innerProduct(chi_e, dphi_e); + ComplexD cDpo = innerProduct(chi_o, dphi_o); + + std::cout << GridLogMessage << "e " << pDce << " " << cDpe << std::endl; + std::cout << GridLogMessage << "o " << pDco << " " << cDpo << std::endl; + + std::cout << GridLogMessage << "pDce - conj(cDpo) " << pDce-conj(cDpo) << std::endl; + std::cout << GridLogMessage << "pDco - conj(cDpe) " << pDco-conj(cDpe) << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test MeeInv Mee = 1 " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + + Ddwf.Mooee (chi_e, src_e); + Ddwf.MooeeInv(src_e, phi_e); + + Ddwf.Mooee (chi_o, src_o); + Ddwf.MooeeInv(src_o, phi_o); + + setCheckerboard(phi, phi_e); + setCheckerboard(phi, phi_o); + + err = phi - chi; + std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test MeeInvDag MeeDag = 1 " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + + Ddwf.MooeeDag (chi_e, src_e); + Ddwf.MooeeInvDag(src_e, phi_e); + + Ddwf.MooeeDag (chi_o, src_o); + Ddwf.MooeeInvDag(src_o, phi_o); + + setCheckerboard(phi, phi_e); + setCheckerboard(phi, phi_o); + + err = phi - chi; + std::cout << GridLogMessage << "norm diff " << norm2(err) << std::endl; + + std::cout << GridLogMessage << "==============================================================" << std::endl; + std::cout << GridLogMessage << "= Test MpcDagMpc is Hermitian " << std::endl; + std::cout << GridLogMessage << "==============================================================" << std::endl; + + random(RNG5, phi); + random(RNG5, chi); + pickCheckerboard(Even, chi_e, chi); + pickCheckerboard(Odd , chi_o, chi); + pickCheckerboard(Even, phi_e, phi); + pickCheckerboard(Odd , phi_o, phi); + RealD t1,t2; + + SchurDiagMooeeOperator HermOpEO(Ddwf); + HermOpEO.MpcDagMpc(chi_e, dchi_e, t1, t2); + HermOpEO.MpcDagMpc(chi_o, dchi_o, t1, t2); + + HermOpEO.MpcDagMpc(phi_e, dphi_e, t1, t2); + HermOpEO.MpcDagMpc(phi_o, dphi_o, t1, t2); + + pDce = innerProduct(phi_e, dchi_e); + pDco = innerProduct(phi_o, dchi_o); + cDpe = innerProduct(chi_e, dphi_e); + cDpo = innerProduct(chi_o, dphi_o); + + std::cout << GridLogMessage << "e " << pDce << " " << cDpe << std::endl; + std::cout << GridLogMessage << "o " << pDco << " " << cDpo << std::endl; + + std::cout << GridLogMessage << "pDce - conj(cDpo) " << pDco-conj(cDpo) << std::endl; + std::cout << GridLogMessage << "pDco - conj(cDpe) " << pDce-conj(cDpe) << std::endl; + + Grid_finalize(); +} diff --git a/tests/core/Test_staggered.cc b/tests/core/Test_staggered.cc index 75531c83..f8f7035a 100644 --- a/tests/core/Test_staggered.cc +++ b/tests/core/Test_staggered.cc @@ -40,7 +40,7 @@ int main (int argc, char ** argv) std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< +Author: paboyle +Author: David Murphy + + 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 */ + +////////////////////////////////////////////////////////////////////////////////////////// +// This program sets up the initial pseudofermion field |Phi> = Meofa^{-1/2}*|eta>, and +// then uses this Phi to compute the action . +// If all is working, one should find that = . +////////////////////////////////////////////////////////////////////////////////////////// + +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +// Parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Npoles = 12; +const RealD mf = 0.01; +const RealD mpv = 1.0; +const RealD M5 = 1.8; + +int main(int argc, char** argv) +{ + Grid_init(&argc, &argv); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is set up to use " << threads << " threads" << std::endl; + + // Initialize spacetime grid + std::cout << GridLogMessage << "Lattice dimensions: " << grid_dim << " Ls: " << Ls << std::endl; + GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(grid_dim, + GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian* UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian* FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian* FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Set up RNGs + std::vector seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + DomainWallEOFAFermionR Lop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mpv, 0.0, -1, M5); + DomainWallEOFAFermionR Rop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mpv, mf, mpv, -1.0, 1, M5); + + // Construct the action and test the heatbath (zero initial guess) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, false); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + // Construct the action and test the heatbath (forecasted initial guesses) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, true); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + return 0; +} diff --git a/tests/debug/Test_heatbath_dwf_eofa_gparity.cc b/tests/debug/Test_heatbath_dwf_eofa_gparity.cc new file mode 100644 index 00000000..5c9d4923 --- /dev/null +++ b/tests/debug/Test_heatbath_dwf_eofa_gparity.cc @@ -0,0 +1,108 @@ + /************************************************************************************* + + Grid physics library, www.github.com/paboyle/Grid + + Source file: ./tests/debug/Test_heatbath_dwf_eofa.cc + + Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + + 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 */ + +////////////////////////////////////////////////////////////////////////////////////////// +// This program sets up the initial pseudofermion field |Phi> = Meofa^{-1/2}*|eta>, and +// then uses this Phi to compute the action . +// If all is working, one should find that = . +////////////////////////////////////////////////////////////////////////////////////////// + +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +typedef GparityWilsonImplR FermionImplPolicy; +typedef GparityDomainWallEOFAFermionR FermionAction; +typedef typename FermionAction::FermionField FermionField; + +// Parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Npoles = 12; +const RealD mf = 0.01; +const RealD mpv = 1.0; +const RealD M5 = 1.8; + +int main(int argc, char** argv) +{ + Grid_init(&argc, &argv); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is set up to use " << threads << " threads" << std::endl; + + // Initialize spacetime grid + std::cout << GridLogMessage << "Lattice dimensions: " << grid_dim << " Ls: " << Ls << std::endl; + GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(grid_dim, + GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian* UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian* FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian* FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Set up RNGs + std::vector seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + // GparityDomainWallFermionR::ImplParams params; + FermionAction::ImplParams params; + FermionAction Lop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mpv, 0.0, -1, M5, params); + FermionAction Rop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mpv, mf, mpv, -1.0, 1, M5, params); + + // Construct the action and test the heatbath (zero initial guess) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, false); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + // Construct the action and test the heatbath (forecasted initial guesses) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, true); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + return 0; +} diff --git a/tests/debug/Test_heatbath_mobius_eofa.cc b/tests/debug/Test_heatbath_mobius_eofa.cc new file mode 100644 index 00000000..a952873d --- /dev/null +++ b/tests/debug/Test_heatbath_mobius_eofa.cc @@ -0,0 +1,104 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/debug/Test_heatbath_dwf_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 */ + +////////////////////////////////////////////////////////////////////////////////////////// +// This program sets up the initial pseudofermion field |Phi> = Meofa^{-1/2}*|eta>, and +// then uses this Phi to compute the action . +// If all is working, one should find that = . +////////////////////////////////////////////////////////////////////////////////////////// + +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +// Parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Npoles = 12; +const RealD b = 2.5; +const RealD c = 1.5; +const RealD mf = 0.01; +const RealD mpv = 1.0; +const RealD M5 = 1.8; + +int main(int argc, char** argv) +{ + Grid_init(&argc, &argv); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is set up to use " << threads << " threads" << std::endl; + + // Initialize spacetime grid + std::cout << GridLogMessage << "Lattice dimensions: " << grid_dim << " Ls: " << Ls << std::endl; + GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(grid_dim, + GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian* UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian* FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian* FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Set up RNGs + std::vector seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + MobiusEOFAFermionR Lop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mpv, 0.0, -1, M5, b, c); + MobiusEOFAFermionR Rop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mpv, mf, mpv, -1.0, 1, M5, b, c); + + // Construct the action and test the heatbath (zero initial guess) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, false); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + // Construct the action and test the heatbath (forecasted initial guesses) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, true); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + return 0; +} diff --git a/tests/debug/Test_heatbath_mobius_eofa_gparity.cc b/tests/debug/Test_heatbath_mobius_eofa_gparity.cc new file mode 100644 index 00000000..08c6d566 --- /dev/null +++ b/tests/debug/Test_heatbath_mobius_eofa_gparity.cc @@ -0,0 +1,109 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/debug/Test_heatbath_dwf_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 */ + +////////////////////////////////////////////////////////////////////////////////////////// +// This program sets up the initial pseudofermion field |Phi> = Meofa^{-1/2}*|eta>, and +// then uses this Phi to compute the action . +// If all is working, one should find that = . +////////////////////////////////////////////////////////////////////////////////////////// + +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +typedef GparityWilsonImplR FermionImplPolicy; +typedef GparityMobiusEOFAFermionR FermionAction; +typedef typename FermionAction::FermionField FermionField; + +// Parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Npoles = 12; +const RealD b = 2.5; +const RealD c = 1.5; +const RealD mf = 0.01; +const RealD mpv = 1.0; +const RealD M5 = 1.8; + +int main(int argc, char** argv) +{ + Grid_init(&argc, &argv); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is set up to use " << threads << " threads" << std::endl; + + // Initialize spacetime grid + std::cout << GridLogMessage << "Lattice dimensions: " << grid_dim << " Ls: " << Ls << std::endl; + GridCartesian* UGrid = SpaceTimeGrid::makeFourDimGrid(grid_dim, + GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian* UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian* FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian* FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Set up RNGs + std::vector seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + FermionAction::ImplParams params; + FermionAction Lop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mpv, 0.0, -1, M5, b, c, params); + FermionAction Rop(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mpv, mf, mpv, -1.0, 1, M5, b, c, params); + + // Construct the action and test the heatbath (zero initial guess) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, false); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + // Construct the action and test the heatbath (forecasted initial guesses) + { + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, Npoles); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, true); + + Meofa.refresh(Umu, RNG5); + printf(" = %1.15e\n", Meofa.S(Umu)); + } + + return 0; +} diff --git a/tests/debug/Test_reweight_dwf_eofa.cc b/tests/debug/Test_reweight_dwf_eofa.cc new file mode 100644 index 00000000..98a17e2f --- /dev/null +++ b/tests/debug/Test_reweight_dwf_eofa.cc @@ -0,0 +1,206 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/debug/Test_reweight_dwf_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +// parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Nhits = 25; +const int max_iter = 5000; +const RealD mf = 0.1; +const RealD mb = 0.11; +const RealD M5 = 1.8; +const RealD stop_tol = 1.0e-12; + +RealD mean(const std::vector& data) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& data, int sample) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& jacks, RealD mean) +{ + int N = jacks.size(); + RealD std(0.0); + for(int i=0; i jack_stats(const std::vector& data) +{ + int N = data.size(); + std::vector jack_samples(N); + std::vector jack_stats(2); + + jack_stats[0] = mean(data); + for(int i=0; i seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + // Initialize RHMC fermion operators + DomainWallFermionR Ddwf_f(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, M5); + DomainWallFermionR Ddwf_b(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, M5); + SchurDiagMooeeOperator MdagM(Ddwf_f); + SchurDiagMooeeOperator VdagV(Ddwf_b); + + // Degree 12 rational approximations to x^(1/4) and x^(-1/4) + double lo = 0.0001; + double hi = 95.0; + int precision = 64; + int degree = 12; + AlgRemez remez(lo, hi, precision); + std::cout << GridLogMessage << "Generating degree " << degree << " for x^(1/4)" << std::endl; + remez.generateApprox(degree, 1, 4); + MultiShiftFunction PowerQuarter(remez, stop_tol, false); + MultiShiftFunction PowerNegQuarter(remez, stop_tol, true); + + // Stochastically estimate reweighting factor via RHMC + RealD scale = std::sqrt(0.5); + std::vector rw_rhmc(Nhits); + ConjugateGradientMultiShift msCG_V(max_iter, PowerQuarter); + ConjugateGradientMultiShift msCG_M(max_iter, PowerNegQuarter); + std::cout.precision(12); + + for(int hit=0; hit tmp(2, Ddwf_f.FermionRedBlackGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + pickCheckerboard(Odd, PhiOdd, Phi); + + // evaluate -log(rw) + msCG_V(VdagV, PhiOdd, tmp[0]); + msCG_M(MdagM, tmp[0], tmp[1]); + rw_rhmc[hit] = norm2(tmp[1]) - norm2(PhiOdd); + std::cout << std::endl << "==================================================" << std::endl; + std::cout << " --- RHMC: Hit " << hit << ": rw = " << rw_rhmc[hit]; + std::cout << std::endl << "==================================================" << std::endl << std::endl; + + } + + // Initialize EOFA fermion operators + RealD shift_L = 0.0; + RealD shift_R = -1.0; + int pm = 1; + DomainWallEOFAFermionR Deofa_L(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, shift_L, pm, M5); + DomainWallEOFAFermionR Deofa_R(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, shift_R, pm, M5); + MdagMLinearOperator LdagL(Deofa_L); + MdagMLinearOperator RdagR(Deofa_R); + + // Stochastically estimate reweighting factor via EOFA + RealD k = Deofa_L.k; + std::vector rw_eofa(Nhits); + ConjugateGradient CG(stop_tol, max_iter); + SchurRedBlackDiagMooeeSolve SchurSolver(CG); + + for(int hit=0; hit tmp(2, Deofa_L.FermionGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + // evaluate -log(rw) + // LH term + for(int s=0; s rhmc_result = jack_stats(rw_rhmc); + std::vector eofa_result = jack_stats(rw_eofa); + std::cout << std::endl << "RHMC: rw = " << rhmc_result[0] << " +/- " << rhmc_result[1] << std::endl; + std::cout << std::endl << "EOFA: rw = " << eofa_result[0] << " +/- " << eofa_result[1] << std::endl; + + Grid_finalize(); +} diff --git a/tests/debug/Test_reweight_dwf_eofa_gparity.cc b/tests/debug/Test_reweight_dwf_eofa_gparity.cc new file mode 100644 index 00000000..bb0fd98e --- /dev/null +++ b/tests/debug/Test_reweight_dwf_eofa_gparity.cc @@ -0,0 +1,209 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/debug/Test_reweight_dwf_eofa_gparity.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +typedef typename GparityDomainWallFermionR::FermionField FermionField; + +// parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Nhits = 10; +const int max_iter = 5000; +const RealD mf = 0.1; +const RealD mb = 0.11; +const RealD M5 = 1.8; +const RealD stop_tol = 1.0e-12; + +RealD mean(const std::vector& data) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& data, int sample) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& jacks, RealD mean) +{ + int N = jacks.size(); + RealD std(0.0); + for(int i=0; i jack_stats(const std::vector& data) +{ + int N = data.size(); + std::vector jack_samples(N); + std::vector jack_stats(2); + + jack_stats[0] = mean(data); + for(int i=0; i seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + // Initialize RHMC fermion operators + GparityDomainWallFermionR::ImplParams params; + GparityDomainWallFermionR Ddwf_f(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, M5, params); + GparityDomainWallFermionR Ddwf_b(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, M5, params); + SchurDiagMooeeOperator MdagM(Ddwf_f); + SchurDiagMooeeOperator VdagV(Ddwf_b); + + // Degree 12 rational approximations to x^(1/4) and x^(-1/4) + double lo = 0.0001; + double hi = 95.0; + int precision = 64; + int degree = 12; + AlgRemez remez(lo, hi, precision); + std::cout << GridLogMessage << "Generating degree " << degree << " for x^(1/4)" << std::endl; + remez.generateApprox(degree, 1, 4); + MultiShiftFunction PowerQuarter(remez, stop_tol, false); + MultiShiftFunction PowerNegQuarter(remez, stop_tol, true); + + // Stochastically estimate reweighting factor via RHMC + RealD scale = std::sqrt(0.5); + std::vector rw_rhmc(Nhits); + ConjugateGradientMultiShift msCG_V(max_iter, PowerQuarter); + ConjugateGradientMultiShift msCG_M(max_iter, PowerNegQuarter); + std::cout.precision(12); + + for(int hit=0; hit tmp(2, Ddwf_f.FermionRedBlackGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + pickCheckerboard(Odd, PhiOdd, Phi); + + // evaluate -log(rw) + msCG_V(VdagV, PhiOdd, tmp[0]); + msCG_M(MdagM, tmp[0], tmp[1]); + rw_rhmc[hit] = norm2(tmp[1]) - norm2(PhiOdd); + std::cout << std::endl << "==================================================" << std::endl; + std::cout << " --- RHMC: Hit " << hit << ": rw = " << rw_rhmc[hit]; + std::cout << std::endl << "==================================================" << std::endl << std::endl; + + } + + // Initialize EOFA fermion operators + RealD shift_L = 0.0; + RealD shift_R = -1.0; + int pm = 1; + GparityDomainWallEOFAFermionR Deofa_L(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, shift_L, pm, M5, params); + GparityDomainWallEOFAFermionR Deofa_R(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, shift_R, pm, M5, params); + MdagMLinearOperator LdagL(Deofa_L); + MdagMLinearOperator RdagR(Deofa_R); + + // Stochastically estimate reweighting factor via EOFA + RealD k = Deofa_L.k; + std::vector rw_eofa(Nhits); + ConjugateGradient CG(stop_tol, max_iter); + SchurRedBlackDiagMooeeSolve SchurSolver(CG); + + for(int hit=0; hit tmp(2, Deofa_L.FermionGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + // evaluate -log(rw) + // LH term + for(int s=0; s rhmc_result = jack_stats(rw_rhmc); + std::vector eofa_result = jack_stats(rw_eofa); + std::cout << std::endl << "RHMC: rw = " << rhmc_result[0] << " +/- " << rhmc_result[1] << std::endl; + std::cout << std::endl << "EOFA: rw = " << eofa_result[0] << " +/- " << eofa_result[1] << std::endl; + + Grid_finalize(); +} diff --git a/tests/debug/Test_reweight_mobius_eofa.cc b/tests/debug/Test_reweight_mobius_eofa.cc new file mode 100644 index 00000000..c4fa78d0 --- /dev/null +++ b/tests/debug/Test_reweight_mobius_eofa.cc @@ -0,0 +1,215 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/debug/Test_reweight_dwf_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +// parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Nhits = 10; +const int max_iter = 5000; +const RealD b = 2.5; +const RealD c = 1.5; +const RealD mf = 0.1; +const RealD mb = 0.11; +const RealD M5 = 1.8; +const RealD stop_tol = 1.0e-12; + +RealD mean(const std::vector& data) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& data, int sample) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& jacks, RealD mean) +{ + int N = jacks.size(); + RealD std(0.0); + for(int i=0; i jack_stats(const std::vector& data) +{ + int N = data.size(); + std::vector jack_samples(N); + std::vector jack_stats(2); + + jack_stats[0] = mean(data); + for(int i=0; i seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + // Initialize RHMC fermion operators + MobiusFermionR Ddwf_f(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, M5, b, c); + MobiusFermionR Ddwf_b(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, M5, b, c); + SchurDiagMooeeOperator MdagM(Ddwf_f); + SchurDiagMooeeOperator VdagV(Ddwf_b); + + // Degree 12 rational approximations to x^(1/4) and x^(-1/4) + double lo = 0.0001; + double hi = 95.0; + int precision = 64; + int degree = 12; + AlgRemez remez(lo, hi, precision); + std::cout << GridLogMessage << "Generating degree " << degree << " for x^(1/4)" << std::endl; + remez.generateApprox(degree, 1, 4); + MultiShiftFunction PowerQuarter(remez, stop_tol, false); + MultiShiftFunction PowerNegQuarter(remez, stop_tol, true); + + // Stochastically estimate reweighting factor via RHMC + RealD scale = std::sqrt(0.5); + std::vector rw_rhmc(Nhits); + ConjugateGradientMultiShift msCG_V(max_iter, PowerQuarter); + ConjugateGradientMultiShift msCG_M(max_iter, PowerNegQuarter); + std::cout.precision(12); + + for(int hit=0; hit tmp(2, Ddwf_f.FermionRedBlackGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + pickCheckerboard(Odd, PhiOdd, Phi); + + // evaluate -log(rw) + msCG_V(VdagV, PhiOdd, tmp[0]); + msCG_M(MdagM, tmp[0], tmp[1]); + rw_rhmc[hit] = norm2(tmp[1]) - norm2(PhiOdd); + std::cout << std::endl << "==================================================" << std::endl; + std::cout << " --- RHMC: Hit " << hit << ": rw = " << rw_rhmc[hit]; + std::cout << std::endl << "==================================================" << std::endl << std::endl; + + } + + // Initialize EOFA fermion operators + RealD shift_L = 0.0; + RealD shift_R = -1.0; + int pm = 1; + MobiusEOFAFermionR Deofa_L(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, shift_L, pm, M5, b, c); + MobiusEOFAFermionR Deofa_R(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, shift_R, pm, M5, b, c); + MdagMLinearOperator LdagL(Deofa_L); + MdagMLinearOperator RdagR(Deofa_R); + + // Stochastically estimate reweighting factor via EOFA + RealD k = Deofa_L.k; + std::vector rw_eofa(Nhits); + ConjugateGradient CG(stop_tol, max_iter); + SchurRedBlackDiagMooeeSolve SchurSolver(CG); + + // Compute -log(Z), where: ( RHMC det ratio ) = Z * ( EOFA det ratio ) + RealD Z = std::pow(b+c+1.0,Ls) + mf*std::pow(b+c-1.0,Ls); + Z /= std::pow(b+c+1.0,Ls) + mb*std::pow(b+c-1.0,Ls); + Z = -12.0*grid_dim[0]*grid_dim[1]*grid_dim[2]*grid_dim[3]*std::log(Z); + + for(int hit=0; hit tmp(2, Deofa_L.FermionGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + // evaluate -log(rw) + // LH term + for(int s=0; s rhmc_result = jack_stats(rw_rhmc); + std::vector eofa_result = jack_stats(rw_eofa); + std::cout << std::endl << "RHMC: rw = " << rhmc_result[0] << " +/- " << rhmc_result[1] << std::endl; + std::cout << std::endl << "EOFA: rw = " << eofa_result[0] << " +/- " << eofa_result[1] << std::endl; + + Grid_finalize(); +} diff --git a/tests/debug/Test_reweight_mobius_eofa_gparity.cc b/tests/debug/Test_reweight_mobius_eofa_gparity.cc new file mode 100644 index 00000000..11a242d2 --- /dev/null +++ b/tests/debug/Test_reweight_mobius_eofa_gparity.cc @@ -0,0 +1,218 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/debug/Test_reweight_dwf_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: paboyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +typedef typename GparityDomainWallFermionR::FermionField FermionField; + +// parameters for test +const std::vector grid_dim = { 8, 8, 8, 8 }; +const int Ls = 8; +const int Nhits = 10; +const int max_iter = 5000; +const RealD b = 2.5; +const RealD c = 1.5; +const RealD mf = 0.1; +const RealD mb = 0.11; +const RealD M5 = 1.8; +const RealD stop_tol = 1.0e-12; + +RealD mean(const std::vector& data) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& data, int sample) +{ + int N = data.size(); + RealD mean(0.0); + for(int i=0; i& jacks, RealD mean) +{ + int N = jacks.size(); + RealD std(0.0); + for(int i=0; i jack_stats(const std::vector& data) +{ + int N = data.size(); + std::vector jack_samples(N); + std::vector jack_stats(2); + + jack_stats[0] = mean(data); + for(int i=0; i seeds4({1, 2, 3, 4}); + std::vector seeds5({5, 6, 7, 8}); + GridParallelRNG RNG5(FGrid); + RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); + RNG4.SeedFixedIntegers(seeds4); + + // Random gauge field + LatticeGaugeField Umu(UGrid); + SU3::HotConfiguration(RNG4, Umu); + + // Initialize RHMC fermion operators + GparityDomainWallFermionR::ImplParams params; + GparityMobiusFermionR Ddwf_f(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, M5, b, c, params); + GparityMobiusFermionR Ddwf_b(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, M5, b, c, params); + SchurDiagMooeeOperator MdagM(Ddwf_f); + SchurDiagMooeeOperator VdagV(Ddwf_b); + + // Degree 12 rational approximations to x^(1/4) and x^(-1/4) + double lo = 0.0001; + double hi = 95.0; + int precision = 64; + int degree = 12; + AlgRemez remez(lo, hi, precision); + std::cout << GridLogMessage << "Generating degree " << degree << " for x^(1/4)" << std::endl; + remez.generateApprox(degree, 1, 4); + MultiShiftFunction PowerQuarter(remez, stop_tol, false); + MultiShiftFunction PowerNegQuarter(remez, stop_tol, true); + + // Stochastically estimate reweighting factor via RHMC + RealD scale = std::sqrt(0.5); + std::vector rw_rhmc(Nhits); + ConjugateGradientMultiShift msCG_V(max_iter, PowerQuarter); + ConjugateGradientMultiShift msCG_M(max_iter, PowerNegQuarter); + std::cout.precision(12); + + for(int hit=0; hit tmp(2, Ddwf_f.FermionRedBlackGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + pickCheckerboard(Odd, PhiOdd, Phi); + + // evaluate -log(rw) + msCG_V(VdagV, PhiOdd, tmp[0]); + msCG_M(MdagM, tmp[0], tmp[1]); + rw_rhmc[hit] = norm2(tmp[1]) - norm2(PhiOdd); + std::cout << std::endl << "==================================================" << std::endl; + std::cout << " --- RHMC: Hit " << hit << ": rw = " << rw_rhmc[hit]; + std::cout << std::endl << "==================================================" << std::endl << std::endl; + + } + + // Initialize EOFA fermion operators + RealD shift_L = 0.0; + RealD shift_R = -1.0; + int pm = 1; + GparityMobiusEOFAFermionR Deofa_L(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, shift_L, pm, M5, b, c, params); + GparityMobiusEOFAFermionR Deofa_R(Umu, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, shift_R, pm, M5, b, c, params); + MdagMLinearOperator LdagL(Deofa_L); + MdagMLinearOperator RdagR(Deofa_R); + + // Stochastically estimate reweighting factor via EOFA + RealD k = Deofa_L.k; + std::vector rw_eofa(Nhits); + ConjugateGradient CG(stop_tol, max_iter); + SchurRedBlackDiagMooeeSolve SchurSolver(CG); + + // Compute -log(Z), where: ( RHMC det ratio ) = Z * ( EOFA det ratio ) + RealD Z = std::pow(b+c+1.0,Ls) + mf*std::pow(b+c-1.0,Ls); + Z /= std::pow(b+c+1.0,Ls) + mb*std::pow(b+c-1.0,Ls); + Z = -12.0*grid_dim[0]*grid_dim[1]*grid_dim[2]*grid_dim[3]*std::log(Z); + + for(int hit=0; hit tmp(2, Deofa_L.FermionGrid()); + gaussian(RNG5, Phi); + Phi = Phi*scale; + + // evaluate -log(rw) + // LH term + for(int s=0; s rhmc_result = jack_stats(rw_rhmc); + std::vector eofa_result = jack_stats(rw_eofa); + std::cout << std::endl << "RHMC: rw = " << rhmc_result[0] << " +/- " << rhmc_result[1] << std::endl; + std::cout << std::endl << "EOFA: rw = " << eofa_result[0] << " +/- " << eofa_result[1] << std::endl; + + Grid_finalize(); +} diff --git a/tests/forces/Test_dwf_force_eofa.cc b/tests/forces/Test_dwf_force_eofa.cc new file mode 100644 index 00000000..f17579ae --- /dev/null +++ b/tests/forces/Test_dwf_force_eofa.cc @@ -0,0 +1,164 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/forces/Test_dwf_force_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +int main (int argc, char** argv) +{ + Grid_init(&argc, &argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + + const int Ls = 8; + + GridCartesian *UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian *UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian *FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian *FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Want a different conf at every run + // First create an instance of an engine. + std::random_device rnd_device; + // Specify the engine and distribution. + std::mt19937 mersenne_engine(rnd_device()); + std::uniform_int_distribution dist(1, 100); + + auto gen = std::bind(dist, mersenne_engine); + std::vector seeds4(4); + generate(begin(seeds4), end(seeds4), gen); + + //std::vector seeds4({1,2,3,5}); + std::vector seeds5({5,6,7,8}); + GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl; + + LatticeFermion phi (FGrid); gaussian(RNG5, phi); + LatticeFermion Mphi (FGrid); + LatticeFermion MphiPrime (FGrid); + + LatticeGaugeField U(UGrid); + SU3::HotConfiguration(RNG4,U); + + //////////////////////////////////// + // Unmodified matrix element + //////////////////////////////////// + RealD mf = 0.01; + RealD mb = 1.0; + RealD M5 = 1.8; + DomainWallEOFAFermionR Lop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, 0.0, -1, M5); + DomainWallEOFAFermionR Rop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, -1.0, 1, M5); + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, 12); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, true); + + Meofa.refresh(U, RNG5); + RealD S = Meofa.S(U); // pdag M p + + // get the deriv of phidag M phi with respect to "U" + LatticeGaugeField UdSdU(UGrid); + Meofa.deriv(U, UdSdU); + + //////////////////////////////////// + // Modify the gauge field a little + //////////////////////////////////// + RealD dt = 0.0001; + + LatticeColourMatrix mommu(UGrid); + LatticeColourMatrix forcemu(UGrid); + LatticeGaugeField mom(UGrid); + LatticeGaugeField Uprime(UGrid); + + for(int mu=0; mu(mom, mommu, mu); + + // fourth order exponential approx + parallel_for(auto i=mom.begin(); i(UdSdU, mu); + mommu = Ta(mommu)*2.0; + PokeIndex(UdSdU, mommu, mu); + } + + for(int mu=0; mu(UdSdU, mu); + mommu = PeekIndex(mom, mu); + + // Update PF action density + dS = dS + trace(mommu*forcemu)*dt; + } + + ComplexD dSpred = sum(dS); + + /*std::cout << GridLogMessage << " S " << S << std::endl; + std::cout << GridLogMessage << " Sprime " << Sprime << std::endl; + std::cout << GridLogMessage << "dS " << Sprime-S << std::endl; + std::cout << GridLogMessage << "predict dS " << dSpred << std::endl;*/ + printf("\nS = %1.15e\n", S); + printf("Sprime = %1.15e\n", Sprime); + printf("dS = %1.15e\n", Sprime - S); + printf("real(dS_predict) = %1.15e\n", dSpred.real()); + printf("imag(dS_predict) = %1.15e\n\n", dSpred.imag()); + + assert( fabs(real(Sprime-S-dSpred)) < 1.0 ) ; + + std::cout << GridLogMessage << "Done" << std::endl; + Grid_finalize(); +} diff --git a/tests/forces/Test_dwf_gpforce_eofa.cc b/tests/forces/Test_dwf_gpforce_eofa.cc new file mode 100644 index 00000000..3afeaa43 --- /dev/null +++ b/tests/forces/Test_dwf_gpforce_eofa.cc @@ -0,0 +1,169 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/forces/Test_dwf_force_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +typedef GparityWilsonImplR FermionImplPolicy; +typedef GparityDomainWallEOFAFermionR FermionAction; +typedef typename FermionAction::FermionField FermionField; + +int main (int argc, char** argv) +{ + Grid_init(&argc, &argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + + const int Ls = 8; + + GridCartesian *UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian *UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian *FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian *FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Want a different conf at every run + // First create an instance of an engine. + std::random_device rnd_device; + // Specify the engine and distribution. + std::mt19937 mersenne_engine(rnd_device()); + std::uniform_int_distribution dist(1, 100); + + auto gen = std::bind(dist, mersenne_engine); + std::vector seeds4(4); + generate(begin(seeds4), end(seeds4), gen); + + //std::vector seeds4({1,2,3,5}); + std::vector seeds5({5,6,7,8}); + GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl; + + FermionField phi (FGrid); gaussian(RNG5, phi); + FermionField Mphi (FGrid); + FermionField MphiPrime (FGrid); + + LatticeGaugeField U(UGrid); + SU3::HotConfiguration(RNG4,U); + + //////////////////////////////////// + // Unmodified matrix element + //////////////////////////////////// + RealD mf = 0.01; + RealD mb = 1.0; + RealD M5 = 1.8; + FermionAction::ImplParams params; + FermionAction Lop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, 0.0, -1, M5, params); + FermionAction Rop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, -1.0, 1, M5, params); + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, 12); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, true); + + Meofa.refresh(U, RNG5); + RealD S = Meofa.S(U); // pdag M p + + // get the deriv of phidag M phi with respect to "U" + LatticeGaugeField UdSdU(UGrid); + Meofa.deriv(U, UdSdU); + + //////////////////////////////////// + // Modify the gauge field a little + //////////////////////////////////// + RealD dt = 0.0001; + + LatticeColourMatrix mommu(UGrid); + LatticeColourMatrix forcemu(UGrid); + LatticeGaugeField mom(UGrid); + LatticeGaugeField Uprime(UGrid); + + for(int mu=0; mu(mom, mommu, mu); + + // fourth order exponential approx + parallel_for(auto i=mom.begin(); i(UdSdU, mu); + mommu = Ta(mommu)*2.0; + PokeIndex(UdSdU, mommu, mu); + } + + for(int mu=0; mu(UdSdU, mu); + mommu = PeekIndex(mom, mu); + + // Update PF action density + dS = dS + trace(mommu*forcemu)*dt; + } + + ComplexD dSpred = sum(dS); + + /*std::cout << GridLogMessage << " S " << S << std::endl; + std::cout << GridLogMessage << " Sprime " << Sprime << std::endl; + std::cout << GridLogMessage << "dS " << Sprime-S << std::endl; + std::cout << GridLogMessage << "predict dS " << dSpred << std::endl;*/ + printf("\nS = %1.15e\n", S); + printf("Sprime = %1.15e\n", Sprime); + printf("dS = %1.15e\n", Sprime - S); + printf("real(dS_predict) = %1.15e\n", dSpred.real()); + printf("imag(dS_predict) = %1.15e\n\n", dSpred.imag()); + + assert( fabs(real(Sprime-S-dSpred)) < 1.0 ) ; + + std::cout << GridLogMessage << "Done" << std::endl; + Grid_finalize(); +} diff --git a/tests/forces/Test_gp_rect_force.cc b/tests/forces/Test_gp_rect_force.cc index bf308749..bb35c77a 100644 --- a/tests/forces/Test_gp_rect_force.cc +++ b/tests/forces/Test_gp_rect_force.cc @@ -42,7 +42,7 @@ int main (int argc, char ** argv) std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +int main (int argc, char** argv) +{ + Grid_init(&argc, &argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + + const int Ls = 8; + + GridCartesian *UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian *UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian *FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian *FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Want a different conf at every run + // First create an instance of an engine. + std::random_device rnd_device; + // Specify the engine and distribution. + std::mt19937 mersenne_engine(rnd_device()); + std::uniform_int_distribution dist(1, 100); + + auto gen = std::bind(dist, mersenne_engine); + std::vector seeds4(4); + generate(begin(seeds4), end(seeds4), gen); + + //std::vector seeds4({1,2,3,5}); + std::vector seeds5({5,6,7,8}); + GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl; + + LatticeFermion phi (FGrid); gaussian(RNG5, phi); + LatticeFermion Mphi (FGrid); + LatticeFermion MphiPrime (FGrid); + + LatticeGaugeField U(UGrid); + SU3::HotConfiguration(RNG4,U); + + //////////////////////////////////// + // Unmodified matrix element + //////////////////////////////////// + RealD b = 2.5; + RealD c = 1.5; + RealD mf = 0.01; + RealD mb = 1.0; + RealD M5 = 1.8; + MobiusEOFAFermionR Lop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, 0.0, -1, M5, b, c); + MobiusEOFAFermionR Rop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, -1.0, 1, M5, b, c); + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, 12); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, false); + + Meofa.refresh(U, RNG5); + RealD S = Meofa.S(U); // pdag M p + + // get the deriv of phidag M phi with respect to "U" + LatticeGaugeField UdSdU(UGrid); + Meofa.deriv(U, UdSdU); + + //////////////////////////////////// + // Modify the gauge field a little + //////////////////////////////////// + RealD dt = 0.0001; + + LatticeColourMatrix mommu(UGrid); + LatticeColourMatrix forcemu(UGrid); + LatticeGaugeField mom(UGrid); + LatticeGaugeField Uprime(UGrid); + + for(int mu=0; mu(mom, mommu, mu); + + // fourth order exponential approx + parallel_for(auto i=mom.begin(); i(UdSdU, mu); + mommu = Ta(mommu)*2.0; + PokeIndex(UdSdU, mommu, mu); + } + + for(int mu=0; mu(UdSdU, mu); + mommu = PeekIndex(mom, mu); + + // Update PF action density + dS = dS + trace(mommu*forcemu)*dt; + } + + ComplexD dSpred = sum(dS); + + /*std::cout << GridLogMessage << " S " << S << std::endl; + std::cout << GridLogMessage << " Sprime " << Sprime << std::endl; + std::cout << GridLogMessage << "dS " << Sprime-S << std::endl; + std::cout << GridLogMessage << "predict dS " << dSpred << std::endl;*/ + printf("\nS = %1.15e\n", S); + printf("Sprime = %1.15e\n", Sprime); + printf("dS = %1.15e\n", Sprime - S); + printf("real(dS_predict) = %1.15e\n", dSpred.real()); + printf("imag(dS_predict) = %1.15e\n\n", dSpred.imag()); + + assert( fabs(real(Sprime-S-dSpred)) < 1.0 ) ; + + std::cout << GridLogMessage << "Done" << std::endl; + Grid_finalize(); +} diff --git a/tests/forces/Test_mobius_gpforce_eofa.cc b/tests/forces/Test_mobius_gpforce_eofa.cc new file mode 100644 index 00000000..72f1dee2 --- /dev/null +++ b/tests/forces/Test_mobius_gpforce_eofa.cc @@ -0,0 +1,171 @@ +/************************************************************************************* + +Grid physics library, www.github.com/paboyle/Grid + +Source file: ./tests/forces/Test_dwf_force_eofa.cc + +Copyright (C) 2017 + +Author: Peter Boyle +Author: David Murphy + +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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +typedef GparityWilsonImplR FermionImplPolicy; +typedef GparityMobiusEOFAFermionR FermionAction; +typedef typename FermionAction::FermionField FermionField; + +int main (int argc, char** argv) +{ + Grid_init(&argc, &argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + + const int Ls = 8; + + GridCartesian *UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()), GridDefaultMpi()); + GridRedBlackCartesian *UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian *FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls, UGrid); + GridRedBlackCartesian *FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls, UGrid); + + // Want a different conf at every run + // First create an instance of an engine. + std::random_device rnd_device; + // Specify the engine and distribution. + std::mt19937 mersenne_engine(rnd_device()); + std::uniform_int_distribution dist(1, 100); + + auto gen = std::bind(dist, mersenne_engine); + std::vector seeds4(4); + generate(begin(seeds4), end(seeds4), gen); + + //std::vector seeds4({1,2,3,5}); + std::vector seeds5({5,6,7,8}); + GridParallelRNG RNG5(FGrid); RNG5.SeedFixedIntegers(seeds5); + GridParallelRNG RNG4(UGrid); RNG4.SeedFixedIntegers(seeds4); + + int threads = GridThread::GetThreads(); + std::cout << GridLogMessage << "Grid is setup to use " << threads << " threads" << std::endl; + + FermionField phi (FGrid); gaussian(RNG5, phi); + FermionField Mphi (FGrid); + FermionField MphiPrime (FGrid); + + LatticeGaugeField U(UGrid); + SU3::HotConfiguration(RNG4,U); + + //////////////////////////////////// + // Unmodified matrix element + //////////////////////////////////// + RealD b = 2.5; + RealD c = 1.5; + RealD mf = 0.01; + RealD mb = 1.0; + RealD M5 = 1.8; + FermionAction::ImplParams params; + FermionAction Lop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mf, mf, mb, 0.0, -1, M5, b, c, params); + FermionAction Rop(U, *FGrid, *FrbGrid, *UGrid, *UrbGrid, mb, mf, mb, -1.0, 1, M5, b, c, params); + OneFlavourRationalParams Params(0.95, 100.0, 5000, 1.0e-12, 12); + ConjugateGradient CG(1.0e-12, 5000); + ExactOneFlavourRatioPseudoFermionAction Meofa(Lop, Rop, CG, Params, false); + + Meofa.refresh(U, RNG5); + RealD S = Meofa.S(U); // pdag M p + + // get the deriv of phidag M phi with respect to "U" + LatticeGaugeField UdSdU(UGrid); + Meofa.deriv(U, UdSdU); + + //////////////////////////////////// + // Modify the gauge field a little + //////////////////////////////////// + RealD dt = 0.0001; + + LatticeColourMatrix mommu(UGrid); + LatticeColourMatrix forcemu(UGrid); + LatticeGaugeField mom(UGrid); + LatticeGaugeField Uprime(UGrid); + + for(int mu=0; mu(mom, mommu, mu); + + // fourth order exponential approx + parallel_for(auto i=mom.begin(); i(UdSdU, mu); + mommu = Ta(mommu)*2.0; + PokeIndex(UdSdU, mommu, mu); + } + + for(int mu=0; mu(UdSdU, mu); + mommu = PeekIndex(mom, mu); + + // Update PF action density + dS = dS + trace(mommu*forcemu)*dt; + } + + ComplexD dSpred = sum(dS); + + /*std::cout << GridLogMessage << " S " << S << std::endl; + std::cout << GridLogMessage << " Sprime " << Sprime << std::endl; + std::cout << GridLogMessage << "dS " << Sprime-S << std::endl; + std::cout << GridLogMessage << "predict dS " << dSpred << std::endl;*/ + printf("\nS = %1.15e\n", S); + printf("Sprime = %1.15e\n", Sprime); + printf("dS = %1.15e\n", Sprime - S); + printf("real(dS_predict) = %1.15e\n", dSpred.real()); + printf("imag(dS_predict) = %1.15e\n\n", dSpred.imag()); + + assert( fabs(real(Sprime-S-dSpred)) < 1.0 ) ; + + std::cout << GridLogMessage << "Done" << std::endl; + Grid_finalize(); +} diff --git a/tests/forces/Test_rect_force.cc b/tests/forces/Test_rect_force.cc index c312abeb..a846af30 100644 --- a/tests/forces/Test_rect_force.cc +++ b/tests/forces/Test_rect_force.cc @@ -42,7 +42,7 @@ int main (int argc, char ** argv) std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); int threads = GridThread::GetThreads(); std::cout< step) { + crc = crc32(crc,&data[blk],step); + blk += step; + len -= step; + } + + crc = crc32(crc,&data[blk],len); + return crc; + + } + + static int get_bfm_index( int* pos, int co, int* s ) { + + int ls = s[0]; + int NtHalf = s[4] / 2; + int simd_coor = pos[4] / NtHalf; + int regu_coor = (pos[1] + s[1] * (pos[2] + s[2] * ( pos[3] + s[3] * (pos[4] % NtHalf) ) )) / 2; + + return regu_coor * ls * 48 + pos[0] * 48 + co * 4 + simd_coor * 2; + } + + static void get_read_geometry(const GridBase* _grid,const std::vector& cnodes, + std::map >& slots, + std::vector& slot_lvol, + std::vector& lvol, + int64_t& slot_lsites,int& ntotal) { + + int _nd = (int)cnodes.size(); + std::vector nodes = cnodes; + + slots.clear(); + slot_lvol.clear(); + lvol.clear(); + + int i; + ntotal = 1; + int64_t lsites = 1; + slot_lsites = 1; + for (i=0;i<_nd;i++) { + assert(_grid->_fdimensions[i] % nodes[i] == 0); + slot_lvol.push_back(_grid->_fdimensions[i] / nodes[i]); + lvol.push_back(_grid->_fdimensions[i] / _grid->_processors[i]); + lsites *= lvol.back(); + slot_lsites *= slot_lvol.back(); + ntotal *= nodes[i]; + } + + std::vector lcoor, gcoor, scoor; + lcoor.resize(_nd); gcoor.resize(_nd); scoor.resize(_nd); + + // create mapping of indices to slots + for (int lidx = 0; lidx < lsites; lidx++) { + Lexicographic::CoorFromIndex(lcoor,lidx,lvol); + for (int i=0;i<_nd;i++) { + gcoor[i] = lcoor[i] + _grid->_processor_coor[i]*lvol[i]; + scoor[i] = gcoor[i] / slot_lvol[i]; + } + int slot; + Lexicographic::IndexFromCoor(scoor,slot,nodes); + auto sl = slots.find(slot); + if (sl == slots.end()) + slots[slot] = std::vector(); + slots[slot].push_back(lidx); + } + } + + static void canonical_block_to_coarse_coordinates(GridBase* _coarsegrid,int nb,int& ii,int& oi) { + // canonical nb needs to be mapped in a coordinate on my coarsegrid (ii,io) + std::vector _l = _coarsegrid->LocalDimensions(); + std::vector _cl = { _l[1], _l[2], _l[3], _l[4], _l[0] }; + std::vector _cc(_l.size()); + Lexicographic::CoorFromIndex(_cc,nb,_cl); + std::vector _c = { _cc[4], _cc[0], _cc[1], _cc[2], _cc[3] }; + ii = _coarsegrid->iIndex(_c); + oi = _coarsegrid->oIndex(_c); + } + + template + static bool read_argonne(BasisFieldVector& ret,const char* dir, const std::vector& cnodes) { + + GridBase* _grid = ret._v[0]._grid; + + std::map > slots; + std::vector slot_lvol, lvol; + int64_t slot_lsites; + int ntotal; + get_read_geometry(_grid,cnodes, + slots,slot_lvol,lvol,slot_lsites, + ntotal); + int _nd = (int)lvol.size(); + + // this is slow code to read the argonne file format for debugging purposes + int nperdir = ntotal / 32; + if (nperdir < 1) + nperdir=1; + std::cout << GridLogMessage << " Read " << dir << " nodes = " << cnodes << std::endl; + std::cout << GridLogMessage << " lvol = " << lvol << std::endl; + + // for error messages + char hostname[1024]; + gethostname(hostname, 1024); + + // now load one slot at a time and fill the vector + for (auto sl=slots.begin();sl!=slots.end();sl++) { + std::vector& idx = sl->second; + int slot = sl->first; + std::vector rdata; + + char buf[4096]; + + sprintf(buf,"%s/checksums.txt",dir); printf("read_argonne: Reading from %s\n",buf); + FILE* f = fopen(buf,"rt"); + if (!f) { + fprintf(stderr,"Node %s cannot read %s\n",hostname,buf); fflush(stderr); + return false; + } + + for (int l=0;l<3+slot;l++) + fgets(buf,sizeof(buf),f); + uint32_t crc_exp = strtol(buf, NULL, 16); + fclose(f); + + // load one slot vector + sprintf(buf,"%s/%2.2d/%10.10d",dir,slot/nperdir,slot); + f = fopen(buf,"rb"); + if (!f) { + fprintf(stderr,"Node %s cannot read %s\n",hostname,buf); fflush(stderr); + return false; + } + + fseeko(f,0,SEEK_END); + off_t total_size = ftello(f); + fseeko(f,0,SEEK_SET); + + int64_t size = slot_lsites / 2 * 24*4; + rdata.resize(size); + + assert(total_size % size == 0); + + int _Nfull = total_size / size; + ret._v.resize(_Nfull,ret._v[0]); + ret._Nm = _Nfull; + + uint32_t crc = 0x0; + GridStopWatch gsw,gsw2; + for (int nev = 0;nev < _Nfull;nev++) { + + gsw.Start(); + assert(fread(&rdata[0],size,1,f) == 1); + gsw.Stop(); + + gsw2.Start(); + crc = crc32_threaded((unsigned char*)&rdata[0],size,crc); + gsw2.Stop(); + + for (int i=0;i lcoor, gcoor, scoor, slcoor; + lcoor.resize(_nd); gcoor.resize(_nd); + slcoor.resize(_nd); scoor.resize(_nd); + +#pragma omp for + for (int64_t lidx = 0; lidx < idx.size(); lidx++) { + int llidx = idx[lidx]; + Lexicographic::CoorFromIndex(lcoor,llidx,lvol); + for (int i=0;i<_nd;i++) { + gcoor[i] = lcoor[i] + _grid->_processor_coor[i]*lvol[i]; + scoor[i] = gcoor[i] / slot_lvol[i]; + slcoor[i] = gcoor[i] - scoor[i]*slot_lvol[i]; + } + + if ((lcoor[1]+lcoor[2]+lcoor[3]+lcoor[4]) % 2 == 1) { + // poke + iScalar, 4> > sc; + for (int s=0;s<4;s++) + for (int c=0;c<3;c++) + sc()(s)(c) = *(std::complex*)&rdata[get_bfm_index(&slcoor[0],c+s*3, &slot_lvol[0] )]; + + pokeLocalSite(sc,ret._v[nev],lcoor); + } + + } + } + } + + fclose(f); + std::cout << GridLogMessage << "Loading slot " << slot << " with " << idx.size() << " points and " + << _Nfull << " vectors in " + << gsw.Elapsed() << " at " + << ( (double)size * _Nfull / 1024./1024./1024. / gsw.useconds()*1000.*1000. ) + << " GB/s " << " crc32 = " << std::hex << crc << " crc32_expected = " << crc_exp << std::dec + << " computed at " + << ( (double)size * _Nfull / 1024./1024./1024. / gsw2.useconds()*1000.*1000. ) + << " GB/s " + << std::endl; + + assert(crc == crc_exp); + } + + _grid->Barrier(); + std::cout << GridLogMessage << "Loading complete" << std::endl; + + return true; + } + + template + static bool read_argonne(BasisFieldVector& ret,const char* dir) { + + + GridBase* _grid = ret._v[0]._grid; + + char buf[4096]; + sprintf(buf,"%s/nodes.txt",dir); + FILE* f = fopen(buf,"rt"); + if (!f) { + if (_grid->IsBoss()) { + fprintf(stderr,"Attempting to load eigenvectors without secifying node layout failed due to absence of nodes.txt\n"); + fflush(stderr); + } + return false; + } + + + std::vector nodes((int)_grid->_processors.size()); + for (int i =0;i<(int)_grid->_processors.size();i++) + assert(fscanf(f,"%d\n",&nodes[i])==1); + fclose(f); + + return read_argonne(ret,dir,nodes); + } + + static void flush_bytes(FILE* f, std::vector& fbuf) { + if (fbuf.size()) { + + if (fwrite(&fbuf[0],fbuf.size(),1,f) != 1) { + fprintf(stderr,"Write failed of %g GB!\n",(double)fbuf.size() / 1024./1024./1024.); + exit(2); + } + + fbuf.resize(0); + + } + } + + static void write_bytes(void* buf, int64_t s, FILE* f, std::vector& fbuf, uint32_t& crc) { + static double data_counter = 0.0; + static GridStopWatch gsw_crc, gsw_flush1,gsw_flush2,gsw_write,gsw_memcpy; + if (s == 0) + return; + + // checksum + gsw_crc.Start(); + crc = crc32_threaded((unsigned char*)buf,s,crc); + gsw_crc.Stop(); + + if (s > fbuf.capacity()) { + // cannot buffer this, so first flush current buffer contents and then write this directly to file + gsw_flush1.Start(); + flush_bytes(f,fbuf); + gsw_flush1.Stop(); + + gsw_write.Start(); + if (fwrite(buf,s,1,f) != 1) { + fprintf(stderr,"Write failed of %g GB!\n",(double)s / 1024./1024./1024.); + exit(2); + } + gsw_write.Stop(); + + } + + // no room left in buffer, flush to disk + if (fbuf.size() + s > fbuf.capacity()) { + gsw_flush2.Start(); + flush_bytes(f,fbuf); + gsw_flush2.Stop(); + } + + // then fill buffer again + { + gsw_memcpy.Start(); + size_t t = fbuf.size(); + fbuf.resize(t + s); + memcpy(&fbuf[t],buf,s); + gsw_memcpy.Stop(); + } + + data_counter += (double)s; + if (data_counter > 1024.*1024.*20.) { + std::cout << GridLogMessage << "Writing " << ((double)data_counter / 1024./1024./1024.) << " GB at" + " crc = " << gsw_crc.Elapsed() << " flush1 = " << gsw_flush1.Elapsed() << " flush2 = " << gsw_flush2.Elapsed() << + " write = " << gsw_write.Elapsed() << " memcpy = " << gsw_memcpy.Elapsed() << std::endl; + data_counter = 0.0; + gsw_crc.Reset(); + gsw_write.Reset(); + gsw_memcpy.Reset(); + gsw_flush1.Reset(); + gsw_flush2.Reset(); + } + } + + static void write_floats(FILE* f, std::vector& fbuf, uint32_t& crc, float* buf, int64_t n) { + write_bytes(buf,n*sizeof(float),f,fbuf,crc); + } + + static void read_floats(char* & ptr, float* out, int64_t n) { + float* in = (float*)ptr; + ptr += 4*n; + + for (int64_t i=0;i 0, [0,6] -> 1; reconstruct 0 -> -3, 1-> 3 + // + // N=2 + // [-6,-2] -> 0, [-2,2] -> 1, [2,6] -> 2; reconstruct 0 -> -4, 1->0, 2->4 + int ret = (int) ( (float)(N+1) * ( (in - min) / (max - min) ) ); + if (ret == N+1) { + ret = N; + } + return ret; + } + + static float fp_unmap(int val, float min, float max, int N) { + return min + (float)(val + 0.5) * (max - min) / (float)( N + 1 ); + } + +#define SHRT_UMAX 65535 +#define FP16_BASE 1.4142135623730950488 +#define FP16_COEF_EXP_SHARE_FLOATS 10 + static float unmap_fp16_exp(unsigned short e) { + float de = (float)((int)e - SHRT_UMAX / 2); + return ::pow( FP16_BASE, de ); + } + + // can assume that v >=0 and need to guarantee that unmap_fp16_exp(map_fp16_exp(v)) >= v + static unsigned short map_fp16_exp(float v) { + // float has exponents 10^{-44.85} .. 10^{38.53} + int exp = (int)ceil(::log(v) / ::log(FP16_BASE)) + SHRT_UMAX / 2; + if (exp < 0 || exp > SHRT_UMAX) { + fprintf(stderr,"Error in map_fp16_exp(%g,%d)\n",v,exp); + exit(3); + } + + return (unsigned short)exp; + } + + template + static void read_floats_fp16(char* & ptr, OPT* out, int64_t n, int nsc) { + + int64_t nsites = n / nsc; + if (n % nsc) { + fprintf(stderr,"Invalid size in write_floats_fp16\n"); + exit(4); + } + + unsigned short* in = (unsigned short*)ptr; + ptr += 2*(n+nsites); + + // do for each site + for (int64_t site = 0;site + static void write_floats_fp16(FILE* f, std::vector& fbuf, uint32_t& crc, OPT* in, int64_t n, int nsc) { + + int64_t nsites = n / nsc; + if (n % nsc) { + fprintf(stderr,"Invalid size in write_floats_fp16\n"); + exit(4); + } + + unsigned short* buf = (unsigned short*)malloc( sizeof(short) * (n + nsites) ); + if (!buf) { + fprintf(stderr,"Out of mem\n"); + exit(1); + } + + // do for each site +#pragma omp parallel for + for (int64_t site = 0;site max) + max = fabs(ev[i]); + } + + unsigned short exp = map_fp16_exp(max); + max = unmap_fp16_exp(exp); + min = -max; + + *bptr++ = exp; + + for (int i=0;i SHRT_UMAX) { + fprintf(stderr,"Assert failed: val = %d (%d), ev[i] = %.15g, max = %.15g, exp = %d\n",val,SHRT_UMAX,ev[i],max,(int)exp); + exit(48); + } + *bptr++ = (unsigned short)val; + } + + } + + write_bytes(buf,sizeof(short)*(n + nsites),f,fbuf,crc); + + free(buf); + } + + template + static bool read_compressed_vectors(const char* dir,BlockProjector& pr,BasisFieldVector& coef, int ngroups = 1) { + + const BasisFieldVector& basis = pr._evec; + GridBase* _grid = basis._v[0]._grid; + + // for error messages + char hostname[1024]; + gethostname(hostname, 1024); + + std::cout << GridLogMessage << "Ready on host " << hostname << " with " << ngroups << " reader groups" << std::endl; + + // first read metadata + char buf[4096]; + sprintf(buf,"%s/metadata.txt",dir); + + std::vector s,b,nb,nn,crc32; + s.resize(5); b.resize(5); nb.resize(5); nn.resize(5); + uint32_t neig, nkeep, nkeep_single, blocks, _FP16_COEF_EXP_SHARE_FLOATS; + uint32_t nprocessors = 1; + + FILE* f = 0; + uint32_t status = 0; + if (_grid->IsBoss()) { + f = fopen(buf,"rb"); + status=f ? 1 : 0; + } + _grid->GlobalSum(status); + std::cout << GridLogMessage << "Read params status " << status << std::endl; + + if (!status) { + return false; + } + +#define _IRL_READ_INT(buf,p) if (f) { assert(fscanf(f,buf,p)==1); } else { *(p) = 0; } _grid->GlobalSum(*(p)); + + for (int i=0;i<5;i++) { + sprintf(buf,"s[%d] = %%d\n",i); + _IRL_READ_INT(buf,&s[(i+1)%5]); + } + for (int i=0;i<5;i++) { + sprintf(buf,"b[%d] = %%d\n",i); + _IRL_READ_INT(buf,&b[(i+1)%5]); + } + for (int i=0;i<5;i++) { + sprintf(buf,"nb[%d] = %%d\n",i); + _IRL_READ_INT(buf,&nb[(i+1)%5]); + } + _IRL_READ_INT("neig = %d\n",&neig); + _IRL_READ_INT("nkeep = %d\n",&nkeep); + _IRL_READ_INT("nkeep_single = %d\n",&nkeep_single); + _IRL_READ_INT("blocks = %d\n",&blocks); + _IRL_READ_INT("FP16_COEF_EXP_SHARE_FLOATS = %d\n",&_FP16_COEF_EXP_SHARE_FLOATS); + + for (int i=0;i<5;i++) { + assert(_grid->FullDimensions()[i] % s[i] == 0); + nn[i] = _grid->FullDimensions()[i] / s[i]; + nprocessors *= nn[i]; + } + + std::cout << GridLogMessage << "Reading data that was generated on node-layout " << nn << std::endl; + + crc32.resize(nprocessors); + for (int i =0;i > slots; + std::vector slot_lvol, lvol; + int64_t slot_lsites; + int ntotal; + std::vector _nn(nn.begin(),nn.end()); + get_read_geometry(_grid,_nn, + slots,slot_lvol,lvol,slot_lsites, + ntotal); + int _nd = (int)lvol.size(); + + // types + typedef typename Field::scalar_type Coeff_t; + typedef typename CoarseField::scalar_type CoeffCoarse_t; + + // slot layout + int nperdir = ntotal / 32; + if (nperdir < 1) + nperdir=1; + + // add read groups + for (int ngroup=0;ngroupThisRank() % ngroups == ngroup; + + std::cout << GridLogMessage << "Reading in group " << ngroup << " / " << ngroups << std::endl; + + // load all necessary slots and store them appropriately + for (auto sl=slots.begin();sl!=slots.end();sl++) { + + std::vector& idx = sl->second; + int slot = sl->first; + std::vector rdata; + + char buf[4096]; + + if (action) { + // load one slot vector + sprintf(buf,"%s/%2.2d/%10.10d.compressed",dir,slot/nperdir,slot); + f = fopen(buf,"rb"); + if (!f) { + fprintf(stderr,"Node %s cannot read %s\n",hostname,buf); fflush(stderr); + return false; + } + } + + uint32_t crc = 0x0; + off_t size; + + GridStopWatch gsw; + _grid->Barrier(); + gsw.Start(); + + std::vector raw_in(0); + if (action) { + fseeko(f,0,SEEK_END); + size = ftello(f); + fseeko(f,0,SEEK_SET); + + raw_in.resize(size); + assert(fread(&raw_in[0],size,1,f) == 1); + } + + _grid->Barrier(); + gsw.Stop(); + + RealD totalGB = (RealD)size / 1024./1024./1024 * _grid->_Nprocessors; + RealD seconds = gsw.useconds() / 1e6; + + if (action) { + std::cout << GridLogMessage << "[" << slot << "] Read " << totalGB << " GB of compressed data at " << totalGB/seconds << " GB/s" << std::endl; + + uint32_t crc_comp = crc32_threaded((unsigned char*)&raw_in[0],size,0); + + if (crc_comp != crc32[slot]) { + std::cout << "Node " << hostname << " found crc mismatch for file " << buf << " (" << std::hex << crc_comp << " vs " << crc32[slot] << std::dec << ")" << std::endl; + std::cout << "Byte size: " << size << std::endl; + } + + assert(crc_comp == crc32[slot]); + } + + _grid->Barrier(); + + if (action) { + fclose(f); + } + + char* ptr = &raw_in[0]; + + GridStopWatch gsw2; + gsw2.Start(); + if (action) { + int nsingleCap = nkeep_single; + if (pr._evec.size() < nsingleCap) + nsingleCap = pr._evec.size(); + + int _cf_block_size = slot_lsites * 12 / 2 / blocks; + +#define FP_16_SIZE(a,b) (( (a) + (a/b) )*2) + + // first read single precision basis vectors +#pragma omp parallel + { + std::vector buf(_cf_block_size * 2); +#pragma omp for + for (int nb=0;nb buf(_cf_block_size * 2); +#pragma omp for + for (int nb=0;nb buf1(nkeep_single*2); + std::vector buf2((nkeep - nkeep_single)*2); + +#pragma omp for + for (int j=0;j<(int)coef.size();j++) + for (int nb=0;nb + static void write_compressed_vectors(const char* dir,const BlockProjector& pr, + const BasisFieldVector& coef, + int nsingle,int writer_nodes = 0) { + + GridStopWatch gsw; + + const BasisFieldVector& basis = pr._evec; + GridBase* _grid = basis._v[0]._grid; + std::vector _l = _grid->FullDimensions(); + for (int i=0;i<(int)_l.size();i++) + _l[i] /= _grid->_processors[i]; + + _grid->Barrier(); + gsw.Start(); + + char buf[4096]; + + // Making the directories is somewhat tricky. + // If we run on a joint filesystem we would just + // have the boss create the directories and then + // have a barrier. We also want to be able to run + // on local /scratch, so potentially all nodes need + // to create their own directories. So do the following + // for now. + for (int j=0;j<_grid->_Nprocessors;j++) { + if (j == _grid->ThisRank()) { + conditionalMkDir(dir); + for (int i=0;i<32;i++) { + sprintf(buf,"%s/%2.2d",dir,i); + conditionalMkDir(buf); + } + _grid->Barrier(); // make sure directories are ready + } + } + + + typedef typename Field::scalar_type Coeff_t; + typedef typename CoarseField::scalar_type CoeffCoarse_t; + + int nperdir = _grid->_Nprocessors / 32; + if (nperdir < 1) + nperdir=1; + + int slot; + Lexicographic::IndexFromCoor(_grid->_processor_coor,slot,_grid->_processors); + + int64_t off = 0x0; + uint32_t crc = 0x0; + if (writer_nodes < 1) + writer_nodes = _grid->_Nprocessors; + int groups = _grid->_Nprocessors / writer_nodes; + if (groups<1) + groups = 1; + + std::cout << GridLogMessage << " Write " << dir << " nodes = " << writer_nodes << std::endl; + + for (int group=0;groupBarrier(); + if (_grid->ThisRank() % groups == group) { + + sprintf(buf,"%s/%2.2d/%10.10d.compressed",dir,slot/nperdir,slot); + FILE* f = fopen(buf,"wb"); + assert(f); + + //buffer does not seem to help + //assert(!setvbuf ( f , NULL , _IOFBF , 1024*1024*2 )); + + int nsingleCap = nsingle; + if (pr._evec.size() < nsingleCap) + nsingleCap = pr._evec.size(); + + GridStopWatch gsw1,gsw2,gsw3,gsw4,gsw5; + + gsw1.Start(); + + std::vector fbuf; + fbuf.reserve( 1024 * 1024 * 8 ); + + // first write single precision basis vectors + for (int nb=0;nb buf; + pr._bgrid.peekBlockOfVectorCanonical(nb,pr._evec._v[i],buf); + +#if 0 + { + RealD nrm = 0.0; + for (int j=0;j<(int)buf.size();j++) + nrm += buf[j]*buf[j]; + std::cout << GridLogMessage << "Norm: " << nrm << std::endl; + } +#endif + write_floats(f,fbuf,crc, &buf[0], buf.size() ); + } + } + + gsw1.Stop(); + gsw2.Start(); + + // then write fixed precision basis vectors + for (int nb=0;nb buf; + pr._bgrid.peekBlockOfVectorCanonical(nb,pr._evec._v[i],buf); + write_floats_fp16(f,fbuf,crc, &buf[0], buf.size(), 24); + } + } + + gsw2.Stop(); + assert(coef._v[0]._grid->_isites*coef._v[0]._grid->_osites == pr._bgrid._blocks); + + gsw3.Start(); + for (int j=0;j<(int)coef.size();j++) { + + int64_t size1 = nsingleCap*2; + int64_t size2 = 2*(pr._evec.size()-nsingleCap); + int64_t size = size1; + if (size2>size) + size=size2; + std::vector buf(size); + + //RealD nrmTest = 0.0; + for (int nb=0;nbGlobalSum(nrmTest); + //std::cout << GridLogMessage << "Test norm: " << nrmTest << std::endl; + } + gsw3.Stop(); + + flush_bytes(f,fbuf); + + off = ftello(f); + fclose(f); + + std::cout<Barrier(); + gsw.Stop(); + + RealD totalGB = (RealD)off / 1024./1024./1024 * _grid->_Nprocessors; + RealD seconds = gsw.useconds() / 1e6; + std::cout << GridLogMessage << "Write " << totalGB << " GB of compressed data at " << totalGB/seconds << " GB/s in " << seconds << " s" << std::endl; + + // gather crcs + std::vector crcs(_grid->_Nprocessors); + for (int i=0;i<_grid->_Nprocessors;i++) { + crcs[i] = 0x0; + } + crcs[slot] = crc; + for (int i=0;i<_grid->_Nprocessors;i++) { + _grid->GlobalSum(crcs[i]); + } + + if (_grid->IsBoss()) { + sprintf(buf,"%s/metadata.txt",dir); + FILE* f = fopen(buf,"wb"); + assert(f); + for (int i=0;i<5;i++) + fprintf(f,"s[%d] = %d\n",i,_grid->FullDimensions()[(i+1)%5] / _grid->_processors[(i+1)%5]); + for (int i=0;i<5;i++) + fprintf(f,"b[%d] = %d\n",i,pr._bgrid._bs[(i+1)%5]); + for (int i=0;i<5;i++) + fprintf(f,"nb[%d] = %d\n",i,pr._bgrid._nb[(i+1)%5]); + fprintf(f,"neig = %d\n",(int)coef.size()); + fprintf(f,"nkeep = %d\n",(int)pr._evec.size()); + fprintf(f,"nkeep_single = %d\n",nsingle); + fprintf(f,"blocks = %d\n",pr._bgrid._blocks); + fprintf(f,"FP16_COEF_EXP_SHARE_FLOATS = %d\n",FP16_COEF_EXP_SHARE_FLOATS); + for (int i =0;i<_grid->_Nprocessors;i++) + fprintf(f,"crc32[%d] = %X\n",i,crcs[i]); + fclose(f); + } + + } + + template + static void write_argonne(const BasisFieldVector& ret,const char* dir) { + + GridBase* _grid = ret._v[0]._grid; + std::vector _l = _grid->FullDimensions(); + for (int i=0;i<(int)_l.size();i++) + _l[i] /= _grid->_processors[i]; + + char buf[4096]; + + if (_grid->IsBoss()) { + mkdir(dir,ACCESSPERMS); + + for (int i=0;i<32;i++) { + sprintf(buf,"%s/%2.2d",dir,i); + mkdir(buf,ACCESSPERMS); + } + } + + _grid->Barrier(); // make sure directories are ready + + + int nperdir = _grid->_Nprocessors / 32; + if (nperdir < 1) + nperdir=1; + std::cout << GridLogMessage << " Write " << dir << " nodes = " << _grid->_Nprocessors << std::endl; + + int slot; + Lexicographic::IndexFromCoor(_grid->_processor_coor,slot,_grid->_processors); + //printf("Slot: %d <> %d\n",slot, _grid->ThisRank()); + + sprintf(buf,"%s/%2.2d/%10.10d",dir,slot/nperdir,slot); + FILE* f = fopen(buf,"wb"); + assert(f); + + int N = (int)ret._v.size(); + uint32_t crc = 0x0; + int64_t cf_size = _grid->oSites()*_grid->iSites()*12; + std::vector< float > rdata(cf_size*2); + + GridStopWatch gsw1,gsw2; + + for (int i=0;i coor(_l.size()); + for (coor[1] = 0;coor[1]<_l[1];coor[1]++) { + for (coor[2] = 0;coor[2]<_l[2];coor[2]++) { + for (coor[3] = 0;coor[3]<_l[3];coor[3]++) { + for (coor[4] = 0;coor[4]<_l[4];coor[4]++) { + for (coor[0] = 0;coor[0]<_l[0];coor[0]++) { + + if ((coor[1]+coor[2]+coor[3]+coor[4]) % 2 == 1) { + // peek + iScalar, 4> > sc; + peekLocalSite(sc,ret._v[i],coor); + for (int s=0;s<4;s++) + for (int c=0;c<3;c++) + *(std::complex*)&rdata[get_bfm_index(&coor[0],c+s*3, &_l[0] )] = sc()(s)(c); + } + } + } + } + } + } + + // endian flip + for (int i=0;i crcs(_grid->_Nprocessors); + for (int i=0;i<_grid->_Nprocessors;i++) { + crcs[i] = 0x0; + } + crcs[slot] = crc; + for (int i=0;i<_grid->_Nprocessors;i++) { + _grid->GlobalSum(crcs[i]); + } + + if (_grid->IsBoss()) { + sprintf(buf,"%s/checksums.txt",dir); + FILE* f = fopen(buf,"wt"); + assert(f); + fprintf(f,"00000000\n\n"); + for (int i =0;i<_grid->_Nprocessors;i++) + fprintf(f,"%X\n",crcs[i]); + fclose(f); + + sprintf(buf,"%s/nodes.txt",dir); + f = fopen(buf,"wt"); + assert(f); + for (int i =0;i<(int)_grid->_processors.size();i++) + fprintf(f,"%d\n",_grid->_processors[i]); + fclose(f); + } + + + std::cout << GridLogMessage << "Writing slot " << slot << " with " + << N << " vectors in " + << gsw2.Elapsed() << " at " + << ( (double)cf_size*2*4 * N / 1024./1024./1024. / gsw2.useconds()*1000.*1000. ) + << " GB/s with crc computed at " + << ( (double)cf_size*2*4 * N / 1024./1024./1024. / gsw1.useconds()*1000.*1000. ) + << " GB/s " + << std::endl; + + _grid->Barrier(); + std::cout << GridLogMessage << "Writing complete" << std::endl; + + } + } + +} diff --git a/tests/lanczos/Makefile.am b/tests/lanczos/Makefile.am new file mode 100644 index 00000000..60b82dd7 --- /dev/null +++ b/tests/lanczos/Makefile.am @@ -0,0 +1 @@ +include Make.inc diff --git a/tests/lanczos/Params.h b/tests/lanczos/Params.h new file mode 100644 index 00000000..d9a6d3b3 --- /dev/null +++ b/tests/lanczos/Params.h @@ -0,0 +1,136 @@ +/* + Params IO + + Author: Christoph Lehner + Date: 2017 +*/ + +#define PADD(p,X) p.get(#X,X); + +class Params { + protected: + + std::string trim(const std::string& sc) { + std::string s = sc; + s.erase(s.begin(), std::find_if(s.begin(), s.end(), + std::not1(std::ptr_fun(std::isspace)))); + s.erase(std::find_if(s.rbegin(), s.rend(), + std::not1(std::ptr_fun(std::isspace))).base(), s.end()); + return s; + } + + public: + + std::map< std::string, std::string > lines; + std::string _fn; + + Params(const char* fn) : _fn(fn) { + FILE* f = fopen(fn,"rt"); + assert(f); + while (!feof(f)) { + char buf[4096]; + if (fgets(buf,sizeof(buf),f)) { + if (buf[0] != '#' && buf[0] != '\r' && buf[0] != '\n') { + char* sep = strchr(buf,'='); + assert(sep); + *sep = '\0'; + lines[trim(buf)] = trim(sep+1); + } + } + } + fclose(f); + } + + ~Params() { + } + + std::string loghead() { + return _fn + ": "; + } + + bool has(const char* name) { + auto f = lines.find(name); + return (f != lines.end()); + } + + const std::string& get(const char* name) { + auto f = lines.find(name); + if (f == lines.end()) { + std::cout << Grid::GridLogMessage << loghead() << "Could not find value for " << name << std::endl; + abort(); + } + return f->second; + } + + void parse(std::string& s, const std::string& cval) { + std::stringstream trimmer; + trimmer << cval; + s.clear(); + trimmer >> s; + } + + void parse(int& i, const std::string& cval) { + assert(sscanf(cval.c_str(),"%d",&i)==1); + } + + void parse(long long& i, const std::string& cval) { + assert(sscanf(cval.c_str(),"%lld",&i)==1); + } + + void parse(double& f, const std::string& cval) { + assert(sscanf(cval.c_str(),"%lf",&f)==1); + } + + void parse(float& f, const std::string& cval) { + assert(sscanf(cval.c_str(),"%f",&f)==1); + } + + void parse(bool& b, const std::string& cval) { + std::string lcval = cval; + std::transform(lcval.begin(), lcval.end(), lcval.begin(), ::tolower); + if (lcval == "true" || lcval == "yes") { + b = true; + } else if (lcval == "false" || lcval == "no") { + b = false; + } else { + std::cout << "Invalid value for boolean: " << b << std::endl; + assert(0); + } + } + + void parse(std::complex& f, const std::string& cval) { + double r,i; + assert(sscanf(cval.c_str(),"%lf %lf",&r,&i)==2); + f = std::complex(r,i); + } + + void parse(std::complex& f, const std::string& cval) { + float r,i; + assert(sscanf(cval.c_str(),"%f %f",&r,&i)==2); + f = std::complex(r,i); + } + + template + void get(const char* name, std::vector& v) { + int i = 0; + v.resize(0); + while (true) { + char buf[4096]; + sprintf(buf,"%s[%d]",name,i++); + if (!has(buf)) + break; + T val; + parse(val,get(buf)); + std::cout << Grid::GridLogMessage << loghead() << "Set " << buf << " to " << val << std::endl; + v.push_back(val); + } + } + + template + void get(const char* name, T& f) { + parse(f,get(name)); + std::cout << Grid::GridLogMessage << loghead() << "Set " << name << " to " << f << std::endl; + } + + +}; diff --git a/tests/lanczos/Test_dwf_compressed_lanczos.cc b/tests/lanczos/Test_dwf_compressed_lanczos.cc new file mode 100644 index 00000000..7fe37387 --- /dev/null +++ b/tests/lanczos/Test_dwf_compressed_lanczos.cc @@ -0,0 +1,727 @@ +/* + Authors: Christoph Lehner + Date: 2017 + + Multigrid Lanczos + + + + TODO: + + High priority: + - Explore filtering of starting vector again, should really work: If cheby has 4 for low mode region and 1 for high mode, applying 15 iterations has 1e9 suppression + of high modes, which should create the desired invariant subspace already? Missing something here??? Maybe dynamic range dangerous, i.e., could also kill interesting + eigenrange if not careful. + + Better: Use all Cheby up to order N in order to approximate a step function; try this! Problem: width of step function. Can kill eigenspace > 1e-3 and have < 1e-5 equal + to 1 + + Low priority: + - Given that I seem to need many restarts and high degree poly to create the base and this takes about 1 day, seriously consider a simple method to create a basis + (ortho krylov low poly); and then fix up lowest say 200 eigenvalues by 1 run with high-degree poly (600 could be enough) +*/ +#include +#include +#include "FieldVectorIO.h" +#include "Params.h" + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +bool read_evals(GridBase* _grid, char* fn, std::vector& evals) { + + FILE* f = 0; + uint32_t status = 0; + if (_grid->IsBoss()) { + f = fopen(fn,"rt"); + status = f ? 1 : 0; + } + _grid->GlobalSum(status); + + if (!status) + return false; + + uint32_t N; + if (f) + assert(fscanf(f,"%d\n",&N)==1); + else + N = 0; + _grid->GlobalSum(N); + + std::cout << "Reading " << N << " eigenvalues" << std::endl; + + evals.resize(N); + + for (int i=0;iGlobalSumVector(&evals[0],evals.size()); + + if (f) + fclose(f); + return true; +} + +void write_evals(char* fn, std::vector& evals) { + FILE* f = fopen(fn,"wt"); + assert(f); + + int N = (int)evals.size(); + fprintf(f,"%d\n",N); + + for (int i=0;i& hist) { + FILE* f = fopen(fn,"wt"); + assert(f); + + int N = (int)hist.size(); + for (int i=0;i +class FunctionHermOp : public LinearFunction { +public: + OperatorFunction & _poly; + LinearOperatorBase &_Linop; + + FunctionHermOp(OperatorFunction & poly,LinearOperatorBase& linop) : _poly(poly), _Linop(linop) { + } + + void operator()(const Field& in, Field& out) { + _poly(_Linop,in,out); + } +}; + +template +class CheckpointedLinearFunction : public LinearFunction { +public: + LinearFunction& _op; + std::string _dir; + int _max_apply; + int _apply, _apply_actual; + GridBase* _grid; + FILE* _f; + + CheckpointedLinearFunction(GridBase* grid, LinearFunction& op, const char* dir,int max_apply) : _op(op), _dir(dir), _grid(grid), _f(0), + _max_apply(max_apply), _apply(0), _apply_actual(0) { + + FieldVectorIO::conditionalMkDir(dir); + + char fn[4096]; + sprintf(fn,"%s/ckpt_op.%4.4d",_dir.c_str(),_grid->ThisRank()); + printf("CheckpointLinearFunction:: file %s\n",fn); + _f = fopen(fn,"r+b"); + if (!_f) + _f = fopen(fn,"w+b"); + assert(_f); + fseek(_f,0,SEEK_CUR); + + } + + ~CheckpointedLinearFunction() { + if (_f) { + fclose(_f); + _f = 0; + } + } + + bool load_ckpt(const Field& in, Field& out) { + + off_t cur = ftello(_f); + fseeko(_f,0,SEEK_END); + if (cur == ftello(_f)) + return false; + fseeko(_f,cur,SEEK_SET); + + size_t sz = sizeof(out._odata[0]) * out._odata.size(); + + GridStopWatch gsw; + gsw.Start(); + uint32_t crc_exp; + assert(fread(&crc_exp,4,1,_f)==1); + assert(fread(&out._odata[0],sz,1,_f)==1); + assert(FieldVectorIO::crc32_threaded((unsigned char*)&out._odata[0],sz,0x0)==crc_exp); + gsw.Stop(); + + printf("CheckpointLinearFunction:: reading %lld\n",(long long)sz); + std::cout << GridLogMessage << "Loading " << ((RealD)sz/1024./1024./1024.) << " GB in " << gsw.Elapsed() << std::endl; + return true; + } + + void save_ckpt(const Field& in, Field& out) { + + fseek(_f,0,SEEK_CUR); // switch to write + + size_t sz = sizeof(out._odata[0]) * out._odata.size(); + + GridStopWatch gsw; + gsw.Start(); + uint32_t crc = FieldVectorIO::crc32_threaded((unsigned char*)&out._odata[0],sz,0x0); + assert(fwrite(&crc,4,1,_f)==1); + assert(fwrite(&out._odata[0],sz,1,_f)==1); + fflush(_f); // try this on the GPFS to suppress OPA usage for disk during dslash; this is not needed at Lustre/JLAB + gsw.Stop(); + + printf("CheckpointLinearFunction:: writing %lld\n",(long long)sz); + std::cout << GridLogMessage << "Saving " << ((RealD)sz/1024./1024./1024.) << " GB in " << gsw.Elapsed() << std::endl; + } + + void operator()(const Field& in, Field& out) { + + _apply++; + + if (load_ckpt(in,out)) + return; + + _op(in,out); + + save_ckpt(in,out); + + if (_apply_actual++ >= _max_apply) { + std::cout << GridLogMessage << "Maximum application of operator reached, checkpoint and finish in future job" << std::endl; + if (_f) { fclose(_f); _f=0; } + in._grid->Barrier(); + Grid_finalize(); + exit(3); + } + } +}; + +template +class ProjectedFunctionHermOp : public LinearFunction { +public: + OperatorFunction & _poly; + LinearOperatorBase &_Linop; + BlockProjector& _pr; + + ProjectedFunctionHermOp(BlockProjector& pr,OperatorFunction & poly,LinearOperatorBase& linop) : _poly(poly), _Linop(linop), _pr(pr) { + } + + void operator()(const CoarseField& in, CoarseField& out) { + assert(_pr._bgrid._o_blocks == in._grid->oSites()); + + Field fin(_pr._bgrid._grid); + Field fout(_pr._bgrid._grid); + + GridStopWatch gsw1,gsw2,gsw3; + // fill fin + gsw1.Start(); + _pr.coarseToFine(in,fin); + gsw1.Stop(); + + // apply poly + gsw2.Start(); + _poly(_Linop,fin,fout); + gsw2.Stop(); + + // fill out + gsw3.Start(); + _pr.fineToCoarse(fout,out); + gsw3.Stop(); + + auto eps = innerProduct(in,out); + std::cout << GridLogMessage << "Operator timing details: c2f = " << gsw1.Elapsed() << " poly = " << gsw2.Elapsed() << " f2c = " << gsw3.Elapsed() << + " Complimentary Hermiticity check: " << eps.imag() / std::abs(eps) << std::endl; + + } +}; + +template +class ProjectedHermOp : public LinearFunction { +public: + LinearOperatorBase &_Linop; + BlockProjector& _pr; + + ProjectedHermOp(BlockProjector& pr,LinearOperatorBase& linop) : _Linop(linop), _pr(pr) { + } + + void operator()(const CoarseField& in, CoarseField& out) { + assert(_pr._bgrid._o_blocks == in._grid->oSites()); + Field fin(_pr._bgrid._grid); + Field fout(_pr._bgrid._grid); + _pr.coarseToFine(in,fin); + _Linop.HermOp(fin,fout); + _pr.fineToCoarse(fout,out); + + } +}; + +template +class PlainHermOp : public LinearFunction { +public: + LinearOperatorBase &_Linop; + + PlainHermOp(LinearOperatorBase& linop) : _Linop(linop) { + } + + void operator()(const Field& in, Field& out) { + _Linop.HermOp(in,out); + } +}; + +template using CoarseSiteFieldGeneral = iScalar< iVector >; +template using CoarseSiteFieldD = CoarseSiteFieldGeneral< vComplexD, N >; +template using CoarseSiteFieldF = CoarseSiteFieldGeneral< vComplexF, N >; +template using CoarseSiteField = CoarseSiteFieldGeneral< vComplex, N >; +template using CoarseLatticeFermion = Lattice< CoarseSiteField >; +template using CoarseLatticeFermionD = Lattice< CoarseSiteFieldD >; + +template +void CoarseGridLanczos(BlockProjector& pr,RealD alpha2,RealD beta,int Npoly2, + int Nstop2,int Nk2,int Nm2,RealD resid2,RealD betastp2,int MaxIt,int MinRes2, + LinearOperatorBase& HermOp, std::vector& eval1, bool cg_test_enabled, + int cg_test_maxiter,int nsingle,int SkipTest2, int MaxApply2,bool smoothed_eval_enabled, + int smoothed_eval_inner,int smoothed_eval_outer,int smoothed_eval_begin, + int smoothed_eval_end,RealD smoothed_eval_inner_resid) { + + BlockedGrid& bgrid = pr._bgrid; + BasisFieldVector& basis = pr._evec; + + + std::vector coarseFourDimLatt; + for (int i=0;i<4;i++) + coarseFourDimLatt.push_back(bgrid._nb[1+i] * bgrid._grid->_processors[1+i]); + assert(bgrid._grid->_processors[0] == 1); + + std::cout << GridLogMessage << "CoarseGrid = " << coarseFourDimLatt << " with basis = " << Nstop1 << std::endl; + GridCartesian * UCoarseGrid = SpaceTimeGrid::makeFourDimGrid(coarseFourDimLatt, GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi()); + GridCartesian * FCoarseGrid = SpaceTimeGrid::makeFiveDimGrid(bgrid._nb[0],UCoarseGrid); + + Chebyshev Cheb2(alpha2,beta,Npoly2); + CoarseLatticeFermion src_coarse(FCoarseGrid); + + // Second round of Lanczos in blocked space + std::vector eval2(Nm2); + std::vector eval3(Nm2); + BasisFieldVector > coef(Nm2,FCoarseGrid); + + ProjectedFunctionHermOp,LatticeFermion> Op2plain(pr,Cheb2,HermOp); + CheckpointedLinearFunction > Op2ckpt(src_coarse._grid,Op2plain,"checkpoint",MaxApply2); + LinearFunction< CoarseLatticeFermion >* Op2; + if (MaxApply2) { + Op2 = &Op2ckpt; + } else { + Op2 = &Op2plain; + } + ProjectedHermOp,LatticeFermion> Op2nopoly(pr,HermOp); + BlockImplicitlyRestartedLanczos > IRL2(*Op2,*Op2,Nstop2,Nk2,Nm2,resid2,betastp2,MaxIt,MinRes2); + + + src_coarse = 1.0; + + // Precision test + { + Field tmp(bgrid._grid); + CoarseLatticeFermion tmp2(FCoarseGrid); + CoarseLatticeFermion tmp3(FCoarseGrid); + tmp2 = 1.0; + tmp3 = 1.0; + + pr.coarseToFine(tmp2,tmp); + pr.fineToCoarse(tmp,tmp2); + + tmp2 -= tmp3; + std::cout << GridLogMessage << "Precision Test c->f->c: " << norm2(tmp2) / norm2(tmp3) << std::endl; + + //bgrid._grid->Barrier(); + //return; + } + + int Nconv; + if (!FieldVectorIO::read_compressed_vectors("lanczos.output",pr,coef) || + !read_evals(UCoarseGrid,(char *)"lanczos.output/eigen-values.txt",eval3) || + !read_evals(UCoarseGrid,(char *)"lanczos.output/eigen-values.txt.linear",eval1) || + !read_evals(UCoarseGrid,(char *)"lanczos.output/eigen-values.txt.poly",eval2) + ) { + + + IRL2.calc(eval2,coef,src_coarse,Nconv,true,SkipTest2); + + coef.resize(Nstop2); + eval2.resize(Nstop2); + eval3.resize(Nstop2); + + std::vector step3_cache; + + // reconstruct eigenvalues of original operator + for (int i=0;iIsBoss()) { + write_evals((char *)"lanczos.output/eigen-values.txt",eval3); + write_evals((char *)"lanczos.output/eigen-values.txt.linear",eval1); + write_evals((char *)"lanczos.output/eigen-values.txt.poly",eval2); + } + + } + + // fix up eigenvalues + if (!read_evals(UCoarseGrid,(char *)"lanczos.output/eigen-values.txt.smoothed",eval3) && smoothed_eval_enabled) { + + ConjugateGradient CG(smoothed_eval_inner_resid, smoothed_eval_inner, false); + + LatticeFermion v_i(basis[0]._grid); + auto tmp = v_i; + auto tmp2 = v_i; + + for (int i=smoothed_eval_begin;iIsBoss()) { + write_evals((char *)"lanczos.output/eigen-values.txt.smoothed",eval3); + write_evals((char *)"lanczos.output/eigen-values.txt",eval3); // also reset this to the best ones we have available + } + } + + // do CG test with and without deflation + if (cg_test_enabled) { + ConjugateGradient CG(1.0e-8, cg_test_maxiter, false); + LatticeFermion src_orig(bgrid._grid); + src_orig.checkerboard = Odd; + src_orig = 1.0; + src_orig = src_orig * (1.0 / ::sqrt(norm2(src_orig)) ); + auto result = src_orig; + + // undeflated solve + result = zero; + CG(HermOp, src_orig, result); + // if (UCoarseGrid->IsBoss()) + // write_history("cg_test.undefl",CG.ResHistory); + // CG.ResHistory.clear(); + + // deflated solve with all eigenvectors + result = zero; + pr.deflate(coef,eval2,Nstop2,src_orig,result); + CG(HermOp, src_orig, result); + // if (UCoarseGrid->IsBoss()) + // write_history("cg_test.defl_all",CG.ResHistory); + // CG.ResHistory.clear(); + + // deflated solve with non-blocked eigenvectors + result = zero; + pr.deflate(coef,eval1,Nstop1,src_orig,result); + CG(HermOp, src_orig, result); + // if (UCoarseGrid->IsBoss()) + // write_history("cg_test.defl_full",CG.ResHistory); + // CG.ResHistory.clear(); + + // deflated solve with all eigenvectors and original eigenvalues from proj + result = zero; + pr.deflate(coef,eval3,Nstop2,src_orig,result); + CG(HermOp, src_orig, result); + // if (UCoarseGrid->IsBoss()) + // write_history("cg_test.defl_all_ev3",CG.ResHistory); + // CG.ResHistory.clear(); + + } + +} + + +template +void quick_krylov_basis(BasisFieldVector& evec,Field& src,LinearFunction& Op,int Nstop) { + Field tmp = src; + Field tmp2 = tmp; + + for (int i=0;i HermOp(Ddwf); + + // Eigenvector storage + const int Nm1 = Np1 + Nk1; + const int Nm2 = Np2 + Nk2; // maximum number of vectors we need to keep + std::cout << GridLogMessage << "Keep " << Nm1 << " full vectors" << std::endl; + std::cout << GridLogMessage << "Keep " << Nm2 << " total vectors" << std::endl; + assert(Nm2 >= Nm1); + BasisFieldVector evec(Nm1,FrbGrid); // start off with keeping full vectors + + // First and second cheby + Chebyshev Cheb1(alpha1,beta,Npoly1); + FunctionHermOp Op1(Cheb1,HermOp); + PlainHermOp Op1test(HermOp); + + // Eigenvalue storage + std::vector eval1(evec.size()); + + // Construct source vector + LatticeFermion src(FrbGrid); + { + src=1.0; + src.checkerboard = Odd; + + // normalize + RealD nn = norm2(src); + nn = Grid::sqrt(nn); + src = src * (1.0/nn); + } + + // Do a benchmark and a quick exit if performance is too little (ugly but needed due to performance fluctuations) + if (max_cheb_time_ms) { + // one round of warmup + auto tmp = src; + GridStopWatch gsw1,gsw2; + gsw1.Start(); + Cheb1(HermOp,src,tmp); + gsw1.Stop(); + Ddwf.ZeroCounters(); + gsw2.Start(); + Cheb1(HermOp,src,tmp); + gsw2.Stop(); + Ddwf.Report(); + std::cout << GridLogMessage << "Performance check; warmup = " << gsw1.Elapsed() << " test = " << gsw2.Elapsed() << std::endl; + int ms = (int)(gsw2.useconds()/1e3); + if (ms > max_cheb_time_ms) { + std::cout << GridLogMessage << "Performance too poor: " << ms << " ms, cutoff = " << max_cheb_time_ms << " ms" << std::endl; + Grid_finalize(); + return 2; + } + + } + + // First round of Lanczos to get low mode basis + BlockImplicitlyRestartedLanczos IRL1(Op1,Op1test,Nstop1,Nk1,Nm1,resid1,betastp1,MaxIt,MinRes1); + int Nconv; + + char tag[1024]; + if (!FieldVectorIO::read_argonne(evec,(char *)"checkpoint") || !read_evals(UGrid,(char *)"checkpoint/eigen-values.txt",eval1)) { + + if (simple_krylov_basis) { + quick_krylov_basis(evec,src,Op1,Nstop1); + } else { + IRL1.calc(eval1,evec,src,Nconv,false,1); + } + evec.resize(Nstop1); // and throw away superfluous + eval1.resize(Nstop1); + if (checkpoint_basis) + FieldVectorIO::write_argonne(evec,(char *)"checkpoint"); + if (UGrid->IsBoss() && checkpoint_basis) + write_evals((char *)"checkpoint/eigen-values.txt",eval1); + + Ddwf.Report(); + + if (exit_after_basis_calculation) { + Grid_finalize(); + return 0; + } + } + + // now test eigenvectors + if (!simple_krylov_basis) { + for (int i=0;i simd_layout = GridDefaultSimd(Nd, vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size, simd_layout, mpi_layout); - GridRedBlackCartesian RBGrid(latt_size, simd_layout, mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1, 2, 3, 4, 5}); GridSerialRNG sRNG; @@ -149,4 +149,4 @@ JSON } -*/ \ No newline at end of file +*/ diff --git a/tests/solver/Test_dwf_mrhs_cg.cc b/tests/solver/Test_dwf_mrhs_cg.cc new file mode 100644 index 00000000..d9215db2 --- /dev/null +++ b/tests/solver/Test_dwf_mrhs_cg.cc @@ -0,0 +1,229 @@ + /************************************************************************************* + + Grid physics library, www.github.com/paboyle/Grid + + Source file: ./tests/Test_dwf_mrhs_cg.cc + + Copyright (C) 2015 + +Author: Peter Boyle + + 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 +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +int main (int argc, char ** argv) +{ + typedef typename DomainWallFermionR::FermionField FermionField; + typedef typename DomainWallFermionR::ComplexField ComplexField; + typename DomainWallFermionR::ImplParams params; + + const int Ls=4; + + Grid_init(&argc,&argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + std::vector mpi_split (mpi_layout.size(),1); + + GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi()); + GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid); + GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid); + + int nrhs = UGrid->RankCount() ; + + ///////////////////////////////////////////// + // Split into 1^4 mpi communicators + ///////////////////////////////////////////// + GridCartesian * SGrid = new GridCartesian(GridDefaultLatt(), + GridDefaultSimd(Nd,vComplex::Nsimd()), + mpi_split, + *UGrid); + + GridCartesian * SFGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,SGrid); + GridRedBlackCartesian * SrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(SGrid); + GridRedBlackCartesian * SFrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,SGrid); + + /////////////////////////////////////////////// + // Set up the problem as a 4d spreadout job + /////////////////////////////////////////////// + std::vector seeds({1,2,3,4}); + + GridParallelRNG pRNG(UGrid ); pRNG.SeedFixedIntegers(seeds); + GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds); + std::vector src(nrhs,FGrid); + std::vector src_chk(nrhs,FGrid); + std::vector result(nrhs,FGrid); + FermionField tmp(FGrid); + + for(int s=0;sThisRank(); + LatticeGaugeField s_Umu(SGrid); + FermionField s_src(SFGrid); + FermionField s_src_split(SFGrid); + FermionField s_tmp(SFGrid); + FermionField s_res(SFGrid); + + { + FGrid->Barrier(); + ScidacWriter _ScidacWriter; + _ScidacWriter.open(file); + std::cout << GridLogMessage << "****************************************************************** "<Barrier(); + std::cout << GridLogMessage << "****************************************************************** "<Barrier(); + std::cout << GridLogMessage << "****************************************************************** "<Barrier(); + + std::cout << GridLogMessage << "****************************************************************** "<Barrier(); + } + + /////////////////////////////////////////////////////////////// + // split the source out using MPI instead of I/O + /////////////////////////////////////////////////////////////// + std::cout << GridLogMessage << " Splitting the grid data "<Barrier(); + if ( n==me ) { + std::cerr << GridLogMessage<<"Split "<< me << " " << norm2(s_src_split) << " " << norm2(s_src)<< " diff " << norm2(s_tmp)<Barrier(); + } + + + /////////////////////////////////////////////////////////////// + // Set up N-solvers as trivially parallel + /////////////////////////////////////////////////////////////// + + RealD mass=0.01; + RealD M5=1.8; + DomainWallFermionR Dchk(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5); + DomainWallFermionR Ddwf(s_Umu,*SFGrid,*SFrbGrid,*SGrid,*SrbGrid,mass,M5); + + std::cout << GridLogMessage << "****************************************************************** "< HermOp(Ddwf); + MdagMLinearOperator HermOpCk(Dchk); + ConjugateGradient CG((1.0e-8/(me+1)),10000); + s_res = zero; + CG(HermOp,s_src,s_res); + + ///////////////////////////////////////////////////////////// + // Report how long they all took + ///////////////////////////////////////////////////////////// + std::vector iterations(nrhs,0); + iterations[me] = CG.IterationsToComplete; + + for(int n=0;nGlobalSum(iterations[n]); + std::cout << GridLogMessage<<" Rank "< + + 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 +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +int main (int argc, char ** argv) +{ + typedef typename DomainWallFermionR::FermionField FermionField; + typedef typename DomainWallFermionR::ComplexField ComplexField; + typename DomainWallFermionR::ImplParams params; + + const int Ls=4; + + Grid_init(&argc,&argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + std::vector mpi_split (mpi_layout.size(),1); + + GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi()); + GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid); + GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid); + + int nrhs = UGrid->RankCount() ; + + ///////////////////////////////////////////// + // Split into 1^4 mpi communicators + ///////////////////////////////////////////// + GridCartesian * SGrid = new GridCartesian(GridDefaultLatt(), + GridDefaultSimd(Nd,vComplex::Nsimd()), + mpi_split, + *UGrid); + + GridCartesian * SFGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,SGrid); + GridRedBlackCartesian * SrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(SGrid); + GridRedBlackCartesian * SFrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,SGrid); + + /////////////////////////////////////////////// + // Set up the problem as a 4d spreadout job + /////////////////////////////////////////////// + std::vector seeds({1,2,3,4}); + + GridParallelRNG pRNG(UGrid ); pRNG.SeedFixedIntegers(seeds); + GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds); + std::vector src(nrhs,FGrid); + std::vector src_chk(nrhs,FGrid); + std::vector result(nrhs,FGrid); + FermionField tmp(FGrid); + + for(int s=0;sThisRank(); + + LatticeGaugeField s_Umu(SGrid); + FermionField s_src(SFGrid); + FermionField s_tmp(SFGrid); + FermionField s_res(SFGrid); + + /////////////////////////////////////////////////////////////// + // split the source out using MPI instead of I/O + /////////////////////////////////////////////////////////////// + Grid_split (Umu,s_Umu); + Grid_split (src,s_src); + + /////////////////////////////////////////////////////////////// + // Set up N-solvers as trivially parallel + /////////////////////////////////////////////////////////////// + RealD mass=0.01; + RealD M5=1.8; + DomainWallFermionR Dchk(Umu,*FGrid,*FrbGrid,*UGrid,*rbGrid,mass,M5); + DomainWallFermionR Ddwf(s_Umu,*SFGrid,*SFrbGrid,*SGrid,*SrbGrid,mass,M5); + + std::cout << GridLogMessage << "****************************************************************** "< HermOp(Ddwf); + MdagMLinearOperator HermOpCk(Dchk); + ConjugateGradient CG((1.0e-8/(me+1)),10000); + s_res = zero; + CG(HermOp,s_src,s_res); + + ///////////////////////////////////////////////////////////// + // Report how long they all took + ///////////////////////////////////////////////////////////// + std::vector iterations(nrhs,0); + iterations[me] = CG.IterationsToComplete; + + for(int n=0;nGlobalSum(iterations[n]); + std::cout << GridLogMessage<<" Rank "< + + 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 +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +int main (int argc, char ** argv) +{ + typedef typename DomainWallFermionR::FermionField FermionField; + typedef typename DomainWallFermionR::ComplexField ComplexField; + typename DomainWallFermionR::ImplParams params; + + const int Ls=4; + + Grid_init(&argc,&argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + std::vector mpi_split (mpi_layout.size(),1); + + GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi()); + GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid); + GridRedBlackCartesian * rbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid); + + int nrhs = UGrid->RankCount() ; + + ///////////////////////////////////////////// + // Split into 1^4 mpi communicators + ///////////////////////////////////////////// + GridCartesian * SGrid = new GridCartesian(GridDefaultLatt(), + GridDefaultSimd(Nd,vComplex::Nsimd()), + mpi_split, + *UGrid); + + GridCartesian * SFGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,SGrid); + GridRedBlackCartesian * SrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(SGrid); + GridRedBlackCartesian * SFrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,SGrid); + + /////////////////////////////////////////////// + // Set up the problem as a 4d spreadout job + /////////////////////////////////////////////// + std::vector seeds({1,2,3,4}); + + GridParallelRNG pRNG(UGrid ); pRNG.SeedFixedIntegers(seeds); + GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds); + std::vector src(nrhs,FGrid); + std::vector src_chk(nrhs,FGrid); + std::vector result(nrhs,FGrid); + FermionField tmp(FGrid); + + std::vector src_e(nrhs,FrbGrid); + std::vector src_o(nrhs,FrbGrid); + + for(int s=0;sThisRank(); + + LatticeGaugeField s_Umu(SGrid); + FermionField s_src(SFGrid); + FermionField s_src_e(SFrbGrid); + FermionField s_src_o(SFrbGrid); + FermionField s_tmp(SFGrid); + FermionField s_res(SFGrid); + + /////////////////////////////////////////////////////////////// + // split the source out using MPI instead of I/O + /////////////////////////////////////////////////////////////// + Grid_split (Umu,s_Umu); + Grid_split (src,s_src); + + /////////////////////////////////////////////////////////////// + // Check even odd cases + /////////////////////////////////////////////////////////////// + for(int s=0;s HermOp(Ddwf); + MdagMLinearOperator HermOpCk(Dchk); + ConjugateGradient CG((1.0e-8/(me+1)),10000); + s_res = zero; + CG(HermOp,s_src,s_res); + + ///////////////////////////////////////////////////////////// + // Report how long they all took + ///////////////////////////////////////////////////////////// + std::vector iterations(nrhs,0); + iterations[me] = CG.IterationsToComplete; + + for(int n=0;nGlobalSum(iterations[n]); + std::cout << GridLogMessage<<" Rank "< simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1,2,3,4,5}); GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds); diff --git a/tests/solver/Test_staggered_block_cg_prec.cc b/tests/solver/Test_staggered_block_cg_prec.cc new file mode 100644 index 00000000..0076e5a0 --- /dev/null +++ b/tests/solver/Test_staggered_block_cg_prec.cc @@ -0,0 +1,130 @@ + /************************************************************************************* + + Grid physics library, www.github.com/paboyle/Grid + + Source file: ./tests/Test_wilson_cg_unprec.cc + + Copyright (C) 2015 + +Author: Azusa Yamaguchi +Author: Peter Boyle + + 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 + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +template +struct scal { + d internal; +}; + + Gamma::Algebra Gmu [] = { + Gamma::Algebra::GammaX, + Gamma::Algebra::GammaY, + Gamma::Algebra::GammaZ, + Gamma::Algebra::GammaT + }; + +int main (int argc, char ** argv) +{ + typedef typename ImprovedStaggeredFermion5DR::FermionField FermionField; + typedef typename ImprovedStaggeredFermion5DR::ComplexField ComplexField; + typename ImprovedStaggeredFermion5DR::ImplParams params; + + const int Ls=8; + + Grid_init(&argc,&argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + + GridCartesian * UGrid = SpaceTimeGrid::makeFourDimGrid(GridDefaultLatt(), GridDefaultSimd(Nd,vComplex::Nsimd()),GridDefaultMpi()); + GridRedBlackCartesian * UrbGrid = SpaceTimeGrid::makeFourDimRedBlackGrid(UGrid); + GridCartesian * FGrid = SpaceTimeGrid::makeFiveDimGrid(Ls,UGrid); + GridRedBlackCartesian * FrbGrid = SpaceTimeGrid::makeFiveDimRedBlackGrid(Ls,UGrid); + + std::vector seeds({1,2,3,4}); + GridParallelRNG pRNG(UGrid ); pRNG.SeedFixedIntegers(seeds); + GridParallelRNG pRNG5(FGrid); pRNG5.SeedFixedIntegers(seeds); + + FermionField src(FGrid); random(pRNG5,src); + FermionField src_o(FrbGrid); pickCheckerboard(Odd,src_o,src); + FermionField result_o(FrbGrid); result_o=zero; + RealD nrm = norm2(src); + + LatticeGaugeField Umu(UGrid); SU3::HotConfiguration(pRNG,Umu); + + RealD mass=0.003; + ImprovedStaggeredFermion5DR Ds(Umu,Umu,*FGrid,*FrbGrid,*UGrid,*UrbGrid,mass); + SchurStaggeredOperator HermOp(Ds); + + ConjugateGradient CG(1.0e-8,10000); + int blockDim = 0; + BlockConjugateGradient BCGrQ(BlockCGrQ,blockDim,1.0e-8,10000); + BlockConjugateGradient BCG (BlockCG,blockDim,1.0e-8,10000); + BlockConjugateGradient mCG (CGmultiRHS,blockDim,1.0e-8,10000); + + std::cout << GridLogMessage << "****************************************************************** "< HermOp4d(Ds4d); + FermionField src4d(UGrid); random(pRNG,src4d); + FermionField src4d_o(UrbGrid); pickCheckerboard(Odd,src4d_o,src4d); + FermionField result4d_o(UrbGrid); + + result4d_o=zero; + CG(HermOp4d,src4d_o,result4d_o); + std::cout << GridLogMessage << "************************************************************************ "< *************************************************************************************/ /* END LEGAL */ #include -#include using namespace std; using namespace Grid; diff --git a/tests/solver/Test_staggered_cg_prec.cc b/tests/solver/Test_staggered_cg_prec.cc new file mode 100644 index 00000000..9a458f1f --- /dev/null +++ b/tests/solver/Test_staggered_cg_prec.cc @@ -0,0 +1,92 @@ + /************************************************************************************* + + Grid physics library, www.github.com/paboyle/Grid + + Source file: ./tests/Test_wilson_cg_unprec.cc + + Copyright (C) 2015 + +Author: Azusa Yamaguchi +Author: Peter Boyle + + 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 +#include + +using namespace std; +using namespace Grid; +using namespace Grid::QCD; + +template +struct scal { + d internal; +}; + + Gamma::Algebra Gmu [] = { + Gamma::Algebra::GammaX, + Gamma::Algebra::GammaY, + Gamma::Algebra::GammaZ, + Gamma::Algebra::GammaT + }; + +int main (int argc, char ** argv) +{ + typedef typename ImprovedStaggeredFermionR::FermionField FermionField; + typedef typename ImprovedStaggeredFermionR::ComplexField ComplexField; + typename ImprovedStaggeredFermionR::ImplParams params; + + Grid_init(&argc,&argv); + + std::vector latt_size = GridDefaultLatt(); + std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); + std::vector mpi_layout = GridDefaultMpi(); + GridCartesian Grid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); + + std::vector seeds({1,2,3,4}); + GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds); + + FermionField src(&Grid); random(pRNG,src); + RealD nrm = norm2(src); + LatticeGaugeField Umu(&Grid); SU3::HotConfiguration(pRNG,Umu); + + double volume=1; + for(int mu=0;mu HermOpEO(Ds); + ConjugateGradient CG(1.0e-8,10000); + CG(HermOpEO,src_o,res_o); + + FermionField tmp(&RBGrid); + + HermOpEO.Mpc(res_o,tmp); + std::cout << "check Mpc resid " << axpy_norm(tmp,-1.0,src_o,tmp)/norm2(src_o) << "\n"; + + Grid_finalize(); +} diff --git a/tests/solver/Test_staggered_cg_unprec.cc b/tests/solver/Test_staggered_cg_unprec.cc index 5e0358d7..eb33c004 100644 --- a/tests/solver/Test_staggered_cg_unprec.cc +++ b/tests/solver/Test_staggered_cg_unprec.cc @@ -57,7 +57,7 @@ int main (int argc, char ** argv) std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1,2,3,4}); GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds); diff --git a/tests/solver/Test_wilson_cg_prec.cc b/tests/solver/Test_wilson_cg_prec.cc index 011bc70b..99ddfceb 100644 --- a/tests/solver/Test_wilson_cg_prec.cc +++ b/tests/solver/Test_wilson_cg_prec.cc @@ -52,7 +52,7 @@ int main (int argc, char ** argv) std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1,2,3,4}); GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds); diff --git a/tests/solver/Test_wilson_cg_schur.cc b/tests/solver/Test_wilson_cg_schur.cc index 7bbf74d3..13ac0090 100644 --- a/tests/solver/Test_wilson_cg_schur.cc +++ b/tests/solver/Test_wilson_cg_schur.cc @@ -52,7 +52,7 @@ int main (int argc, char ** argv) std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1,2,3,4}); GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds); diff --git a/tests/solver/Test_wilson_cg_unprec.cc b/tests/solver/Test_wilson_cg_unprec.cc index 19c5f854..db227ec8 100644 --- a/tests/solver/Test_wilson_cg_unprec.cc +++ b/tests/solver/Test_wilson_cg_unprec.cc @@ -52,7 +52,7 @@ int main (int argc, char ** argv) std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1,2,3,4}); GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds); diff --git a/tests/solver/Test_wilson_cr_unprec.cc b/tests/solver/Test_wilson_cr_unprec.cc index 4182c04e..eccd7e74 100644 --- a/tests/solver/Test_wilson_cr_unprec.cc +++ b/tests/solver/Test_wilson_cr_unprec.cc @@ -52,7 +52,7 @@ int main (int argc, char ** argv) std::vector simd_layout = GridDefaultSimd(Nd,vComplex::Nsimd()); std::vector mpi_layout = GridDefaultMpi(); GridCartesian Grid(latt_size,simd_layout,mpi_layout); - GridRedBlackCartesian RBGrid(latt_size,simd_layout,mpi_layout); + GridRedBlackCartesian RBGrid(&Grid); std::vector seeds({1,2,3,4}); GridParallelRNG pRNG(&Grid); pRNG.SeedFixedIntegers(seeds);