portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

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

Browse Source
This commit is contained in:
Peter Boyle 2023-03-29 14:37:40 -04:00
commit 58e020b62a
26 changed files with 19 additions and 71 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Grid/algorithms/iterative/ConjugateGradient.h
View File

@ -191,7 +191,7 @@ public:
std::cout << GridLogMessage << "\tAxpyNorm " << AxpyNormTimer.Elapsed() <<std::endl;
std::cout << GridLogMessage << "\tLinearComb " << LinearCombTimer.Elapsed() <<std::endl;
std::cout << GridLogMessage << "\tMobius flop rate " << DwfFlops/ usecs<< " Gflops " <<std::endl;
std::cout << GridLogDebug << "\tMobius flop rate " << DwfFlops/ usecs<< " Gflops " <<std::endl;
if (ErrorOnNoConverge) assert(true_residual / Tolerance < 10000.0);

11
Grid/lattice/Lattice_reduction_gpu.h
View File

@ -211,13 +211,10 @@ inline typename vobj::scalar_objectD sumD_gpu_small(const vobj *lat, Integer osi
assert(ok);
Integer smemSize = numThreads * sizeof(sobj);
// UVM seems to be buggy under later CUDA drivers
// This fails on A100 and driver 5.30.02 / CUDA 12.1
// Fails with multiple NVCC versions back to 11.4,
// which worked with earlier drivers.
// Not sure which driver had first fail and this bears checking
// Is awkward as must install multiple driver versions
#undef UVM_BLOCK_BUFFER
// Move out of UVM
// Turns out I had messed up the synchronise after move to compute stream
// as running this on the default stream fools the synchronise
#undef UVM_BLOCK_BUFFER
#ifndef UVM_BLOCK_BUFFER
commVector<sobj> buffer(numBlocks);
sobj *buffer_v = &buffer[0];

11
Grid/lattice/Lattice_rng.h
View File

@ -440,17 +440,8 @@ public:
_grid->GlobalCoorToGlobalIndex(gcoor,gidx);
_grid->GlobalCoorToRankIndex(rank,o_idx,i_idx,gcoor);
#if 1
assert(rank == _grid->ThisRank() );
#else
//
if (rank != _grid->ThisRank() ){
std::cout <<"rank "<<rank<<" _grid->ThisRank() "<<_grid->ThisRank()<< std::endl;
// exit(-42);
// assert(0);
}
#endif
assert(rank == _grid->ThisRank() );
int l_idx=generator_idx(o_idx,i_idx);
_generators[l_idx] = master_engine;

11
Grid/qcd/action/fermion/implementation/WilsonKernelsImplementation.h
View File

@ -463,11 +463,7 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
if( interior && exterior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSite); return;}
#ifdef SYCL_HACK
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteSycl); return; }
#else
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSite); return;}
#endif
#ifndef GRID_CUDA
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSite); return;}
#endif
@ -478,6 +474,7 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteInt); return;}
#endif
} else if( exterior ) {
acceleratorFenceComputeStream();
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteExt); return;}
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteExt); return;}
#ifndef GRID_CUDA
@ -502,10 +499,9 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
#ifndef GRID_CUDA
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDag); return;}
#endif
acceleratorFenceComputeStream();
} else if( interior ) {
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALL(GenericDhopSiteDagInt); return;}
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALL(HandDhopSiteDagInt); return;}
if (Opt == WilsonKernelsStatic::OptGeneric ) { KERNEL_CALLNB(GenericDhopSiteDagInt); return;}
if (Opt == WilsonKernelsStatic::OptHandUnroll ) { KERNEL_CALLNB(HandDhopSiteDagInt); return;}
#ifndef GRID_CUDA
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagInt); return;}
#endif
@ -516,7 +512,6 @@ void WilsonKernels<Impl>::DhopKernel(int Opt,StencilImpl &st, DoubledGaugeField
#ifndef GRID_CUDA
if (Opt == WilsonKernelsStatic::OptInlineAsm ) { ASM_CALL(AsmDhopSiteDagExt); return;}
#endif
acceleratorFenceComputeStream();
}
assert(0 && " Kernel optimisation case not covered ");
}

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/CayleyFermion5DInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../CayleyFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/ContinuedFractionFermion5DInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../ContinuedFractionFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/DomainWallEOFAFermionInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../DomainWallEOFAFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/MobiusEOFAFermionInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../MobiusEOFAFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/PartialFractionFermion5DInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../PartialFractionFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/WilsonCloverFermionInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonCloverFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/WilsonFermion5DInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/WilsonFermionInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/WilsonKernelsInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonKernelsInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/WilsonTMFermionInstantiationWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonTMFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/WilsonImplD2/impl.h
View File

@ -1 +0,0 @@
#define IMPLEMENTATION WilsonImplD2

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/CayleyFermion5DInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../CayleyFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/ContinuedFractionFermion5DInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../ContinuedFractionFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/DomainWallEOFAFermionInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../DomainWallEOFAFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/MobiusEOFAFermionInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../MobiusEOFAFermionInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/PartialFractionFermion5DInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../PartialFractionFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/WilsonFermion5DInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonFermion5DInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/WilsonKernelsInstantiationZWilsonImplD2.cc
View File

@ -1 +0,0 @@
../WilsonKernelsInstantiation.cc.master

1
Grid/qcd/action/fermion/instantiation/ZWilsonImplD2/impl.h
View File

@ -1 +0,0 @@
#define IMPLEMENTATION ZWilsonImplD2

13
Grid/qcd/action/pseudofermion/TwoFlavourEvenOddRatio.h
View File

@ -112,40 +112,27 @@ NAMESPACE_BEGIN(Grid);
// NumOp == V
// DenOp == M
//
AUDIT();
FermionField etaOdd (NumOp.FermionRedBlackGrid());
FermionField etaEven(NumOp.FermionRedBlackGrid());
FermionField tmp (NumOp.FermionRedBlackGrid());
AUDIT();
pickCheckerboard(Even,etaEven,eta);
AUDIT();
pickCheckerboard(Odd,etaOdd,eta);
AUDIT();
NumOp.ImportGauge(U);
AUDIT();
DenOp.ImportGauge(U);
std::cout << " TwoFlavourRefresh: Imported gauge "<<std::endl;
AUDIT();
SchurDifferentiableOperator<Impl> Mpc(DenOp);
AUDIT();
SchurDifferentiableOperator<Impl> Vpc(NumOp);
AUDIT();
std::cout << " TwoFlavourRefresh: Diff ops "<<std::endl;
AUDIT();
// Odd det factors
Mpc.MpcDag(etaOdd,PhiOdd);
AUDIT();
std::cout << " TwoFlavourRefresh: MpcDag "<<std::endl;
tmp=Zero();
AUDIT();
std::cout << " TwoFlavourRefresh: Zero() guess "<<std::endl;
AUDIT();
HeatbathSolver(Vpc,PhiOdd,tmp);
AUDIT();
std::cout << " TwoFlavourRefresh: Heatbath solver "<<std::endl;
Vpc.Mpc(tmp,PhiOdd);
std::cout << " TwoFlavourRefresh: Mpc "<<std::endl;

21
Grid/qcd/hmc/integrators/Integrator.h
View File

@ -134,14 +134,12 @@ protected:
double start_force = usecond();
std::cout << GridLogMessage << "AuditForce["<<level<<"]["<<a<<"] before"<<std::endl;
AUDIT();
as[level].actions.at(a)->deriv_timer_start();
as[level].actions.at(a)->deriv(Us, force); // deriv should NOT include Ta
as[level].actions.at(a)->deriv_timer_stop();
std::cout << GridLogMessage << "AuditForce["<<level<<"]["<<a<<"] after"<<std::endl;
AUDIT();
std::cout << GridLogIntegrator << "Smearing (on/off): " << as[level].actions.at(a)->is_smeared << std::endl;
auto name = as[level].actions.at(a)->action_name();
@ -382,12 +380,12 @@ public:
Field& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
std::cout << GridLogMessage << "AuditRefresh["<<level<<"]["<<actionID<<"] before"<<std::endl;
AUDIT();
as[level].actions.at(actionID)->refresh_timer_start();
as[level].actions.at(actionID)->refresh(Us, sRNG, pRNG);
as[level].actions.at(actionID)->refresh_timer_stop();
std::cout << GridLogMessage << "AuditRefresh["<<level<<"]["<<actionID<<"] after"<<std::endl;
AUDIT();
}
// Refresh the higher representation actions
@ -424,7 +422,7 @@ public:
// Actions
for (int level = 0; level < as.size(); ++level) {
for (int actionID = 0; actionID < as[level].actions.size(); ++actionID) {
AUDIT();
// get gauge field from the SmearingPolicy and
// based on the boolean is_smeared in actionID
Field& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
@ -434,7 +432,7 @@ public:
as[level].actions.at(actionID)->S_timer_stop();
std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] H = " << Hterm << std::endl;
H += Hterm;
AUDIT();
}
as[level].apply(S_hireps, Representations, level, H);
}
@ -447,9 +445,9 @@ public:
void operator()(std::vector<Action<FieldType>*> repr_set, Repr& Rep, int level, RealD& H) {
for (int a = 0; a < repr_set.size(); ++a) {
AUDIT();
RealD Hterm = repr_set.at(a)->Sinitial(Rep.U);
AUDIT();
std::cout << GridLogMessage << "Sinitial Level " << level << " term " << a << " H Hirep = " << Hterm << std::endl;
H += Hterm;
@ -474,10 +472,10 @@ public:
Field& Us = Smearer.get_U(as[level].actions.at(actionID)->is_smeared);
std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] action eval " << std::endl;
as[level].actions.at(actionID)->S_timer_start();
AUDIT();
Hterm = as[level].actions.at(actionID)->Sinitial(Us);
as[level].actions.at(actionID)->S_timer_stop();
AUDIT();
std::cout << GridLogMessage << "S [" << level << "][" << actionID << "] H = " << Hterm << std::endl;
H += Hterm;
}
@ -490,7 +488,6 @@ public:
void integrate(Field& U)
{
AUDIT();
// reset the clocks
t_U = 0;
for (int level = 0; level < as.size(); ++level) {
@ -508,10 +505,8 @@ public:
assert(fabs(t_U - t_P[level]) < 1.0e-6); // must be the same
std::cout << GridLogIntegrator << " times[" << level << "]= " << t_P[level] << " " << t_U << std::endl;
}
AUDIT();
FieldImplementation::Project(U);
AUDIT();
// and that we indeed got to the end of the trajectory
assert(fabs(t_U - Params.trajL) < 1.0e-6);

2
Grid/stencil/Stencil.h
View File

@ -665,9 +665,11 @@ public:
for(int i=0;i<mm.size();i++){
decompressor::MergeFace(decompress,mm[i]);
}
if ( mm.size() ) acceleratorFenceComputeStream();
for(int i=0;i<dd.size();i++){
decompressor::DecompressFace(decompress,dd[i]);
}
if ( dd.size() ) acceleratorFenceComputeStream();
}
////////////////////////////////////////
// Set up routines