portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-08-04 05:37:06 +01:00
Code Releases Activity

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

Browse Source
This commit is contained in:
u61464
2021-05-04 08:40:38 -07:00
parent 834f536b5f c332d9f08b
commit 15ae317858
25 changed files with 358 additions and 159 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
Grid/cshift/Cshift_mpi.h
View File

@@ -122,8 +122,8 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
assert(shift<fd);
int buffer_size = rhs.Grid()->_slice_nblock[dimension]*rhs.Grid()->_slice_block[dimension];
cshiftVector<vobj> send_buf(buffer_size);
cshiftVector<vobj> recv_buf(buffer_size);
static cshiftVector<vobj> send_buf; send_buf.resize(buffer_size);
static cshiftVector<vobj> recv_buf; recv_buf.resize(buffer_size);
int cb= (cbmask==0x2)? Odd : Even;
int sshift= rhs.Grid()->CheckerBoardShiftForCB(rhs.Checkerboard(),dimension,shift,cb);
@@ -198,8 +198,8 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
int buffer_size = grid->_slice_nblock[dimension]*grid->_slice_block[dimension];
// int words = sizeof(vobj)/sizeof(vector_type);
std::vector<cshiftVector<scalar_object> > send_buf_extract(Nsimd);
std::vector<cshiftVector<scalar_object> > recv_buf_extract(Nsimd);
static std::vector<cshiftVector<scalar_object> > send_buf_extract; send_buf_extract.resize(Nsimd);
static std::vector<cshiftVector<scalar_object> > recv_buf_extract; recv_buf_extract.resize(Nsimd);
scalar_object * recv_buf_extract_mpi;
scalar_object * send_buf_extract_mpi;
@@ -294,8 +294,8 @@ template<class vobj> void Cshift_comms(Lattice<vobj> &ret,const Lattice<vobj> &r
assert(shift<fd);
int buffer_size = rhs.Grid()->_slice_nblock[dimension]*rhs.Grid()->_slice_block[dimension];
cshiftVector<vobj> send_buf_v(buffer_size);
cshiftVector<vobj> recv_buf_v(buffer_size);
static cshiftVector<vobj> send_buf_v; send_buf_v.resize(buffer_size);
static cshiftVector<vobj> recv_buf_v; recv_buf_v.resize(buffer_size);
vobj *send_buf;
vobj *recv_buf;
{
@@ -381,8 +381,8 @@ template<class vobj> void Cshift_comms_simd(Lattice<vobj> &ret,const Lattice<vo
int buffer_size = grid->_slice_nblock[dimension]*grid->_slice_block[dimension];
// int words = sizeof(vobj)/sizeof(vector_type);
std::vector<cshiftVector<scalar_object> > send_buf_extract(Nsimd);
std::vector<cshiftVector<scalar_object> > recv_buf_extract(Nsimd);
static std::vector<cshiftVector<scalar_object> > send_buf_extract; send_buf_extract.resize(Nsimd);
static std::vector<cshiftVector<scalar_object> > recv_buf_extract; recv_buf_extract.resize(Nsimd);
scalar_object * recv_buf_extract_mpi;
scalar_object * send_buf_extract_mpi;
{

2
Grid/parallelIO/MetaData.h
View File

@@ -128,7 +128,7 @@ inline void MachineCharacteristics(FieldMetaData &header)
std::time_t t = std::time(nullptr);
std::tm tm_ = *std::localtime(&t);
std::ostringstream oss;
// oss << std::put_time(&tm_, "%c %Z");
oss << std::put_time(&tm_, "%c %Z");
header.creation_date = oss.str();
header.archive_date = header.creation_date;

17
Grid/parallelIO/NerscIO.h
View File

@@ -205,11 +205,20 @@ public:
std::cout<<GridLogMessage <<"NERSC Configuration "<<file<< " and plaquette, link trace, and checksum agree"<<std::endl;
}
// Preferred interface
template<class GaugeStats=PeriodicGaugeStatistics>
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
std::string file,
std::string ens_label = std::string("DWF"))
{
writeConfiguration(Umu,file,0,1,ens_label);
}
template<class GaugeStats=PeriodicGaugeStatistics>
static inline void writeConfiguration(Lattice<vLorentzColourMatrixD > &Umu,
std::string file,
int two_row,
int bits32)
int bits32,
std::string ens_label = std::string("DWF"))
{
typedef vLorentzColourMatrixD vobj;
typedef typename vobj::scalar_object sobj;
@@ -219,8 +228,8 @@ public:
// Following should become arguments
///////////////////////////////////////////
header.sequence_number = 1;
header.ensemble_id = "UKQCD";
header.ensemble_label = "DWF";
header.ensemble_id = std::string("UKQCD");
header.ensemble_label = ens_label;
typedef LorentzColourMatrixD fobj3D;
typedef LorentzColour2x3D fobj2D;
@@ -232,7 +241,7 @@ public:
GaugeStats Stats; Stats(Umu,header);
MachineCharacteristics(header);
uint64_t offset;
uint64_t offset;
// Sod it -- always write 3x3 double
header.floating_point = std::string("IEEE64BIG");

16
Grid/qcd/action/fermion/StaggeredImpl.h
View File

@@ -72,19 +72,23 @@ public:
StaggeredImpl(const ImplParams &p = ImplParams()) : Params(p){};
static accelerator_inline void multLink(SiteSpinor &phi,
template<class _Spinor>
static accelerator_inline void multLink(_Spinor &phi,
const SiteDoubledGaugeField &U,
const SiteSpinor &chi,
const _Spinor &chi,
int mu)
{
mult(&phi(), &U(mu), &chi());
auto UU = coalescedRead(U(mu));
mult(&phi(), &UU, &chi());
}
static accelerator_inline void multLinkAdd(SiteSpinor &phi,
template<class _Spinor>
static accelerator_inline void multLinkAdd(_Spinor &phi,
const SiteDoubledGaugeField &U,
const SiteSpinor &chi,
const _Spinor &chi,
int mu)
{
mac(&phi(), &U(mu), &chi());
auto UU = coalescedRead(U(mu));
mac(&phi(), &UU, &chi());
}
template <class ref>

35
Grid/qcd/action/fermion/WilsonImpl.h
View File

@@ -184,18 +184,22 @@ public:
mat = TraceIndex<SpinIndex>(P);
}
inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds)
{
for (int mu = 0; mu < Nd; mu++)
mat[mu] = PeekIndex<LorentzIndex>(Uds, mu);
}
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu){
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField &Atilde,int mu)
{
#undef USE_OLD_INSERT_FORCE
int Ls=Btilde.Grid()->_fdimensions[0];
autoView( mat_v , mat, AcceleratorWrite);
#ifdef USE_OLD_INSERT_FORCE
GaugeLinkField tmp(mat.Grid());
tmp = Zero();
{
const int Nsimd = SiteSpinor::Nsimd();
autoView( tmp_v , tmp, AcceleratorWrite);
autoView( Btilde_v , Btilde, AcceleratorRead);
autoView( Atilde_v , Atilde, AcceleratorRead);
@@ -208,6 +212,29 @@ public:
});
}
PokeIndex<LorentzIndex>(mat,tmp,mu);
#else
{
const int Nsimd = SiteSpinor::Nsimd();
autoView( Btilde_v , Btilde, AcceleratorRead);
autoView( Atilde_v , Atilde, AcceleratorRead);
accelerator_for(sss,mat.Grid()->oSites(),Nsimd,{
int sU=sss;
typedef decltype(coalescedRead(mat_v[sU](mu)() )) ColorMatrixType;
ColorMatrixType sum;
zeroit(sum);
for(int s=0;s<Ls;s++){
int sF = s+Ls*sU;
for(int spn=0;spn<Ns;spn++){ //sum over spin
auto bb = coalescedRead(Btilde_v[sF]()(spn) ); //color vector
auto aa = coalescedRead(Atilde_v[sF]()(spn) );
auto op = outerProduct(bb,aa);
sum = sum + op;
}
}
coalescedWrite(mat_v[sU](mu)(), sum);
});
}
#endif
}
};

14
Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h
View File

@@ -880,11 +880,23 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
}
std::vector<RealD> G_s(Ls,1.0);
Integer sign = 1; // sign flip for vector/tadpole
if ( curr_type == Current::Axial ) {
for(int s=0;s<Ls/2;s++){
G_s[s] = -1.0;
}
}
else if ( curr_type == Current::Tadpole ) {
auto b=this->_b;
auto c=this->_c;
if ( b == 1 && c == 0 ) {
sign = -1;
}
else {
std::cerr << "Error: Tadpole implementation currently unavailable for non-Shamir actions." << std::endl;
assert(b==1 && c==0);
}
}
for(int s=0;s<Ls;s++){
@@ -907,7 +919,7 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
tmp = Cshift(tmp,mu,1);
Impl::multLinkField(Utmp,this->Umu,tmp,mu);
tmp = G_s[s]*( Utmp*ph - gmu*Utmp*ph ); // Forward hop
tmp = sign*G_s[s]*( Utmp*ph - gmu*Utmp*ph ); // Forward hop
tmp = where((lcoor>=tmin),tmp,zz); // Mask the time
L_Q = where((lcoor<=tmax),tmp,zz); // Position of current complicated

76
Grid/qcd/action/fermion/implementation/StaggeredKernelsImplementation.h
View File

@@ -35,39 +35,32 @@ NAMESPACE_BEGIN(Grid);
#define GENERIC_STENCIL_LEG(U,Dir,skew,multLink) \
SE = st.GetEntry(ptype, Dir+skew, sF); \
if (SE->_is_local ) { \
if (SE->_permute) { \
chi_p = &chi; \
permute(chi, in[SE->_offset], ptype); \
} else { \
chi_p = &in[SE->_offset]; \
} \
int perm= SE->_permute; \
chi = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);\
} else { \
chi_p = &buf[SE->_offset]; \
chi = coalescedRead(buf[SE->_offset],lane); \
} \
multLink(Uchi, U[sU], *chi_p, Dir);
acceleratorSynchronise(); \
multLink(Uchi, U[sU], chi, Dir);
#define GENERIC_STENCIL_LEG_INT(U,Dir,skew,multLink) \
SE = st.GetEntry(ptype, Dir+skew, sF); \
if (SE->_is_local ) { \
if (SE->_permute) { \
chi_p = &chi; \
permute(chi, in[SE->_offset], ptype); \
} else { \
chi_p = &in[SE->_offset]; \
} \
int perm= SE->_permute; \
chi = coalescedReadPermute(in[SE->_offset],ptype,perm,lane);\
} else if ( st.same_node[Dir] ) { \
chi_p = &buf[SE->_offset]; \
chi = coalescedRead(buf[SE->_offset],lane); \
} \
if (SE->_is_local || st.same_node[Dir] ) { \
multLink(Uchi, U[sU], *chi_p, Dir); \
multLink(Uchi, U[sU], chi, Dir); \
}
#define GENERIC_STENCIL_LEG_EXT(U,Dir,skew,multLink) \
SE = st.GetEntry(ptype, Dir+skew, sF); \
if ((!SE->_is_local) && (!st.same_node[Dir]) ) { \
nmu++; \
chi_p = &buf[SE->_offset]; \
multLink(Uchi, U[sU], *chi_p, Dir); \
chi = coalescedRead(buf[SE->_offset],lane); \
multLink(Uchi, U[sU], chi, Dir); \
}
template <class Impl>
@@ -84,12 +77,14 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilView &st,
SiteSpinor *buf, int sF, int sU,
const FermionFieldView &in, FermionFieldView &out, int dag)
{
const SiteSpinor *chi_p;
SiteSpinor chi;
SiteSpinor Uchi;
typedef decltype(coalescedRead(in[0])) calcSpinor;
calcSpinor chi;
calcSpinor Uchi;
StencilEntry *SE;
int ptype;
int skew;
const int Nsimd = SiteHalfSpinor::Nsimd();
const int lane=acceleratorSIMTlane(Nsimd);
// for(int s=0;s<LLs;s++){
//
@@ -118,7 +113,7 @@ void StaggeredKernels<Impl>::DhopSiteGeneric(StencilView &st,
if ( dag ) {
Uchi = - Uchi;
}
vstream(out[sF], Uchi);
coalescedWrite(out[sF], Uchi,lane);
}
};
@@ -130,13 +125,16 @@ template <int Naik> accelerator_inline
void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilView &st,
DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
SiteSpinor *buf, int sF, int sU,
const FermionFieldView &in, FermionFieldView &out,int dag) {
const SiteSpinor *chi_p;
SiteSpinor chi;
SiteSpinor Uchi;
const FermionFieldView &in, FermionFieldView &out,int dag)
{
typedef decltype(coalescedRead(in[0])) calcSpinor;
calcSpinor chi;
calcSpinor Uchi;
StencilEntry *SE;
int ptype;
int skew ;
const int Nsimd = SiteHalfSpinor::Nsimd();
const int lane=acceleratorSIMTlane(Nsimd);
// for(int s=0;s<LLs;s++){
// int sF=LLs*sU+s;
@@ -165,7 +163,7 @@ void StaggeredKernels<Impl>::DhopSiteGenericInt(StencilView &st,
if ( dag ) {
Uchi = - Uchi;
}
vstream(out[sF], Uchi);
coalescedWrite(out[sF], Uchi,lane);
}
};
@@ -178,14 +176,17 @@ template <int Naik> accelerator_inline
void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilView &st,
DoubledGaugeFieldView &U, DoubledGaugeFieldView &UUU,
SiteSpinor *buf, int sF, int sU,
const FermionFieldView &in, FermionFieldView &out,int dag) {
const SiteSpinor *chi_p;
// SiteSpinor chi;
SiteSpinor Uchi;
const FermionFieldView &in, FermionFieldView &out,int dag)
{
typedef decltype(coalescedRead(in[0])) calcSpinor;
calcSpinor chi;
calcSpinor Uchi;
StencilEntry *SE;
int ptype;
int nmu=0;
int skew ;
const int Nsimd = SiteHalfSpinor::Nsimd();
const int lane=acceleratorSIMTlane(Nsimd);
// for(int s=0;s<LLs;s++){
// int sF=LLs*sU+s;
@@ -211,11 +212,12 @@ void StaggeredKernels<Impl>::DhopSiteGenericExt(StencilView &st,
GENERIC_STENCIL_LEG_EXT(UUU,Zm,skew,Impl::multLinkAdd);
GENERIC_STENCIL_LEG_EXT(UUU,Tm,skew,Impl::multLinkAdd);
}
if ( nmu ) {
if ( dag ) {
out[sF] = out[sF] - Uchi;
if ( nmu ) {
auto _out = coalescedRead(out[sF],lane);
if ( dag ) {
coalescedWrite(out[sF], _out-Uchi,lane);
} else {
out[sF] = out[sF] + Uchi;
coalescedWrite(out[sF], _out+Uchi,lane);
}
}
}
@@ -261,6 +263,8 @@ void StaggeredKernels<Impl>::DhopImproved(StencilImpl &st, LebesgueOrder &lo,
GridBase *FGrid=in.Grid();
GridBase *UGrid=U.Grid();
typedef StaggeredKernels<Impl> ThisKernel;
const int Nsimd = SiteHalfSpinor::Nsimd();
const int lane=acceleratorSIMTlane(Nsimd);
autoView( UUU_v , UUU, AcceleratorRead);
autoView( U_v , U, AcceleratorRead);
autoView( in_v , in, AcceleratorRead);
@@ -301,6 +305,8 @@ void StaggeredKernels<Impl>::DhopNaive(StencilImpl &st, LebesgueOrder &lo,
GridBase *FGrid=in.Grid();
GridBase *UGrid=U.Grid();
typedef StaggeredKernels<Impl> ThisKernel;
const int Nsimd = SiteHalfSpinor::Nsimd();
const int lane=acceleratorSIMTlane(Nsimd);
autoView( UUU_v , U, AcceleratorRead);
autoView( U_v , U, AcceleratorRead);
autoView( in_v , in, AcceleratorRead);

21
Grid/qcd/smearing/StoutSmearing.h
View File

@@ -85,21 +85,18 @@ public:
std::cout << GridLogDebug << "Stout smearing started\n";
// Smear the configurations
// C contains the staples multiplied by some rho
u_smr = U ; // set the smeared field to the current gauge field
SmearBase->smear(C, U);
for (int mu = 0; mu < Nd; mu++) {
if( mu == OrthogDim )
tmp = 1.0; // Don't smear in the orthogonal direction
else {
tmp = peekLorentz(C, mu);
Umu = peekLorentz(U, mu);
iq_mu = Ta(
tmp *
adj(Umu)); // iq_mu = Ta(Omega_mu) to match the signs with the paper
exponentiate_iQ(tmp, iq_mu);
}
pokeLorentz(u_smr, tmp * Umu, mu); // u_smr = exp(iQ_mu)*U_mu
if( mu == OrthogDim ) continue ;
// u_smr = exp(iQ_mu)*U_mu apart from Orthogdim
Umu = peekLorentz(U, mu);
tmp = peekLorentz(C, mu);
iq_mu = Ta( tmp * adj(Umu));
exponentiate_iQ(tmp, iq_mu);
pokeLorentz(u_smr, tmp * Umu, mu);
}
std::cout << GridLogDebug << "Stout smearing completed\n";
};

8
Grid/qcd/utils/Metric.h
View File

@@ -93,13 +93,13 @@ public:
GeneralisedMomenta(GridBase* grid, Metric<MomentaField>& M): M(M), Mom(grid), AuxMom(grid), AuxField(grid){}
// Correct
void MomentaDistribution(GridParallelRNG& pRNG){
void MomentaDistribution(GridSerialRNG & sRNG, GridParallelRNG& pRNG){
// Generate a distribution for
// P^dag G P
// where G = M^-1
// Generate gaussian momenta
Implementation::generate_momenta(Mom, pRNG);
Implementation::generate_momenta(Mom, sRNG, pRNG);
// Modify the distribution with the metric
M.MSquareRoot(Mom);
@@ -107,8 +107,8 @@ public:
// Auxiliary momenta
// do nothing if trivial, so hide in the metric
MomentaField AuxMomTemp(Mom.Grid());
Implementation::generate_momenta(AuxMom, pRNG);
Implementation::generate_momenta(AuxField, pRNG);
Implementation::generate_momenta(AuxMom, sRNG, pRNG);
Implementation::generate_momenta(AuxField, sRNG, pRNG);
// Modify the distribution with the metric
// Aux^dag M Aux
M.MInvSquareRoot(AuxMom); // AuxMom = M^{-1/2} AuxMomTemp

21
Grid/tensors/Tensor_outer.h
View File

@@ -34,6 +34,16 @@ NAMESPACE_BEGIN(Grid);
// outerProduct Scalar x Scalar -> Scalar
// Vector x Vector -> Matrix
///////////////////////////////////////////////////////////////////////////////////////
template<class CC,IfComplex<CC> = 0>
accelerator_inline CC outerProduct(const CC &l, const CC& r)
{
return l*conj(r);
}
template<class RR,IfReal<RR> = 0>
accelerator_inline RR outerProduct(const RR &l, const RR& r)
{
return l*r;
}
template<class l,class r,int N> accelerator_inline
auto outerProduct (const iVector<l,N>& lhs,const iVector<r,N>& rhs) -> iMatrix<decltype(outerProduct(lhs._internal[0],rhs._internal[0])),N>
@@ -57,17 +67,6 @@ auto outerProduct (const iScalar<l>& lhs,const iScalar<r>& rhs) -> iScalar<declt
return ret;
}
template<class CC,IfComplex<CC> = 0>
accelerator_inline CC outerProduct(const CC &l, const CC& r)
{
return l*conj(r);
}
template<class RR,IfReal<RR> = 0>
accelerator_inline RR outerProduct(const RR &l, const RR& r)
{
return l*r;
}
NAMESPACE_END(Grid);
#endif

3
Grid/util/Init.cc
View File

@@ -153,6 +153,9 @@ void GridCmdOptionIntVector(const std::string &str,VectorInt & vec)
return;
}
template void GridCmdOptionIntVector(const std::string &str,std::vector<int> & vec);
template void GridCmdOptionIntVector(const std::string &str,Coordinate & vec);
void GridCmdOptionInt(std::string &str,int & val)
{
std::stringstream ss(str);