mirror of
https://github.com/paboyle/Grid.git
synced 2025-06-20 08:46:55 +01:00
Imported changes from feature/gparity_HMC branch:
Added storage of final true residual in mixed-prec CG and enhanced log output Fixed const correctness of multi-shift constructor Added a mixed precision variant of the multi-shift algorithm that uses a single precision operator and applies periodic reliable update to the residual Added tests/solver/Test_dwf_multishift_mixedprec to test the above Fixed local coherence lanczos using the (large!) max approx to the chebyshev eval as the scale from which to judge the quality of convergence, resulting a test that always passes Added a method to local coherence lanczos class that returns the fine eval/evec pair Added iterative log output to power method Added optional disabling of the plaquette check in Nerscio to support loading old G-parity configs which have a factor of 2 error in the plaquette G-parity Dirac op no longer allows GPBC in the time direction; instead we toggle between periodic and antiperiodic Replaced thread_for G-parity 5D force insertion implementation with accelerator_for version capable of running on GPUs Generalized tests/lanczos/Test_dwf_lanczos to support regular DWF as well as Gparity, with the action chosen by a command line option Modified tests/forces/Test_dwf_gpforce,Test_gpdwf_force,Test_gpwilson_force to use GPBC a spatial direction rather than the t-direction, and antiperiodic BCs for time direction tests/core/Test_gparity now supports using APBC in time direction using command line toggle
This commit is contained in:
@ -30,6 +30,18 @@ directory
|
||||
|
||||
NAMESPACE_BEGIN(Grid);
|
||||
|
||||
/*
|
||||
Policy implementation for G-parity boundary conditions
|
||||
|
||||
Rather than treating the gauge field as a flavored field, the Grid implementation of G-parity treats the gauge field as a regular
|
||||
field with complex conjugate boundary conditions. In order to ensure the second flavor interacts with the conjugate links and the first
|
||||
with the regular links we overload the functionality of doubleStore, whose purpose is to store the gauge field and the barrel-shifted gauge field
|
||||
to avoid communicating links when applying the Dirac operator, such that the double-stored field contains also a flavor index which maps to
|
||||
either the link or the conjugate link. This flavored field is then used by multLink to apply the correct link to a spinor.
|
||||
|
||||
Here the first Nd-1 directions are treated as "spatial", and a twist value of 1 indicates G-parity BCs in that direction.
|
||||
mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
|
||||
*/
|
||||
template <class S, class Representation = FundamentalRepresentation, class Options=CoeffReal>
|
||||
class GparityWilsonImpl : public ConjugateGaugeImpl<GaugeImplTypes<S, Representation::Dimension> > {
|
||||
public:
|
||||
@ -113,7 +125,7 @@ public:
|
||||
|| ((distance== 1)&&(icoor[direction]==1))
|
||||
|| ((distance==-1)&&(icoor[direction]==0));
|
||||
|
||||
permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu]; //only if we are going around the world
|
||||
permute_lane = permute_lane && SE->_around_the_world && St.parameters.twists[mmu] && mmu < Nd-1; //only if we are going around the world in a spatial direction
|
||||
|
||||
//Apply the links
|
||||
int f_upper = permute_lane ? 1 : 0;
|
||||
@ -139,10 +151,10 @@ public:
|
||||
assert((distance == 1) || (distance == -1)); // nearest neighbour stencil hard code
|
||||
assert((sl == 1) || (sl == 2));
|
||||
|
||||
if ( SE->_around_the_world && St.parameters.twists[mmu] ) {
|
||||
|
||||
//If this site is an global boundary site, perform the G-parity flavor twist
|
||||
if ( mmu < Nd-1 && SE->_around_the_world && St.parameters.twists[mmu] ) {
|
||||
if ( sl == 2 ) {
|
||||
|
||||
//Only do the twist for lanes on the edge of the physical node
|
||||
ExtractBuffer<sobj> vals(Nsimd);
|
||||
|
||||
extract(chi,vals);
|
||||
@ -197,6 +209,19 @@ public:
|
||||
reg = memory;
|
||||
}
|
||||
|
||||
|
||||
//Poke 'poke_f0' onto flavor 0 and 'poke_f1' onto flavor 1 in direction mu of the doubled gauge field Uds
|
||||
inline void pokeGparityDoubledGaugeField(DoubledGaugeField &Uds, const GaugeLinkField &poke_f0, const GaugeLinkField &poke_f1, const int mu){
|
||||
autoView(poke_f0_v, poke_f0, CpuRead);
|
||||
autoView(poke_f1_v, poke_f1, CpuRead);
|
||||
autoView(Uds_v, Uds, CpuWrite);
|
||||
thread_foreach(ss,poke_f0_v,{
|
||||
Uds_v[ss](0)(mu) = poke_f0_v[ss]();
|
||||
Uds_v[ss](1)(mu) = poke_f1_v[ss]();
|
||||
});
|
||||
}
|
||||
|
||||
|
||||
inline void DoubleStore(GridBase *GaugeGrid,DoubledGaugeField &Uds,const GaugeField &Umu)
|
||||
{
|
||||
conformable(Uds.Grid(),GaugeGrid);
|
||||
@ -207,14 +232,19 @@ public:
|
||||
GaugeLinkField Uconj(GaugeGrid);
|
||||
|
||||
Lattice<iScalar<vInteger> > coor(GaugeGrid);
|
||||
|
||||
for(int mu=0;mu<Nd;mu++){
|
||||
|
||||
LatticeCoordinate(coor,mu);
|
||||
|
||||
//Here the first Nd-1 directions are treated as "spatial", and a twist value of 1 indicates G-parity BCs in that direction.
|
||||
//mu=Nd-1 is assumed to be the time direction and a twist value of 1 indicates antiperiodic BCs
|
||||
for(int mu=0;mu<Nd-1;mu++){
|
||||
|
||||
if( Params.twists[mu] ){
|
||||
LatticeCoordinate(coor,mu);
|
||||
}
|
||||
|
||||
U = PeekIndex<LorentzIndex>(Umu,mu);
|
||||
Uconj = conjugate(U);
|
||||
|
||||
// Implement the isospin rotation sign on the boundary between f=1 and f=0
|
||||
// This phase could come from a simple bc 1,1,-1,1 ..
|
||||
int neglink = GaugeGrid->GlobalDimensions()[mu]-1;
|
||||
if ( Params.twists[mu] ) {
|
||||
@ -229,7 +259,7 @@ public:
|
||||
thread_foreach(ss,U_v,{
|
||||
Uds_v[ss](0)(mu) = U_v[ss]();
|
||||
Uds_v[ss](1)(mu) = Uconj_v[ss]();
|
||||
});
|
||||
});
|
||||
}
|
||||
|
||||
U = adj(Cshift(U ,mu,-1)); // correct except for spanning the boundary
|
||||
@ -260,6 +290,38 @@ public:
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
{ //periodic / antiperiodic temporal BCs
|
||||
int mu = Nd-1;
|
||||
int L = GaugeGrid->GlobalDimensions()[mu];
|
||||
int Lmu = L - 1;
|
||||
|
||||
LatticeCoordinate(coor, mu);
|
||||
|
||||
U = PeekIndex<LorentzIndex>(Umu, mu); //Get t-directed links
|
||||
|
||||
GaugeLinkField *Upoke = &U;
|
||||
|
||||
if(Params.twists[mu]){ //antiperiodic
|
||||
Utmp = where(coor == Lmu, -U, U);
|
||||
Upoke = &Utmp;
|
||||
}
|
||||
|
||||
Uconj = conjugate(*Upoke); //second flavor interacts with conjugate links
|
||||
pokeGparityDoubledGaugeField(Uds, *Upoke, Uconj, mu);
|
||||
|
||||
//Get the barrel-shifted field
|
||||
Utmp = adj(Cshift(U, mu, -1)); //is a forward shift!
|
||||
Upoke = &Utmp;
|
||||
|
||||
if(Params.twists[mu]){
|
||||
U = where(coor == 0, -Utmp, Utmp); //boundary phase
|
||||
Upoke = &U;
|
||||
}
|
||||
|
||||
Uconj = conjugate(*Upoke);
|
||||
pokeGparityDoubledGaugeField(Uds, *Upoke, Uconj, mu + 4);
|
||||
}
|
||||
}
|
||||
|
||||
inline void InsertForce4D(GaugeField &mat, FermionField &Btilde, FermionField &A, int mu) {
|
||||
@ -298,28 +360,48 @@ public:
|
||||
inline void extractLinkField(std::vector<GaugeLinkField> &mat, DoubledGaugeField &Uds){
|
||||
assert(0);
|
||||
}
|
||||
|
||||
|
||||
inline void InsertForce5D(GaugeField &mat, FermionField &Btilde, FermionField Ã, int mu) {
|
||||
|
||||
int Ls = Btilde.Grid()->_fdimensions[0];
|
||||
|
||||
GaugeLinkField tmp(mat.Grid());
|
||||
tmp = Zero();
|
||||
int Ls=Btilde.Grid()->_fdimensions[0];
|
||||
|
||||
{
|
||||
autoView( tmp_v , tmp, CpuWrite);
|
||||
autoView( Atilde_v , Atilde, CpuRead);
|
||||
autoView( Btilde_v , Btilde, CpuRead);
|
||||
thread_for(ss,tmp.Grid()->oSites(),{
|
||||
for (int s = 0; s < Ls; s++) {
|
||||
int sF = s + Ls * ss;
|
||||
auto ttmp = traceIndex<SpinIndex>(outerProduct(Btilde_v[sF], Atilde_v[sF]));
|
||||
tmp_v[ss]() = tmp_v[ss]() + ttmp(0, 0) + conjugate(ttmp(1, 1));
|
||||
}
|
||||
});
|
||||
GridBase *GaugeGrid = mat.Grid();
|
||||
Lattice<iScalar<vInteger> > coor(GaugeGrid);
|
||||
|
||||
if( Params.twists[mu] ){
|
||||
LatticeCoordinate(coor,mu);
|
||||
}
|
||||
|
||||
autoView( mat_v , mat, AcceleratorWrite);
|
||||
autoView( Btilde_v , Btilde, AcceleratorRead);
|
||||
autoView( Atilde_v , Atilde, AcceleratorRead);
|
||||
accelerator_for(sss,mat.Grid()->oSites(), FermionField::vector_type::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
|
||||
//Flavor 0
|
||||
auto bb = coalescedRead(Btilde_v[sF](0)(spn) ); //color vector
|
||||
auto aa = coalescedRead(Atilde_v[sF](0)(spn) );
|
||||
sum = sum + outerProduct(bb,aa);
|
||||
|
||||
//Flavor 1
|
||||
bb = coalescedRead(Btilde_v[sF](1)(spn) );
|
||||
aa = coalescedRead(Atilde_v[sF](1)(spn) );
|
||||
sum = sum + conjugate(outerProduct(bb,aa));
|
||||
}
|
||||
}
|
||||
coalescedWrite(mat_v[sU](mu)(), sum);
|
||||
});
|
||||
}
|
||||
PokeIndex<LorentzIndex>(mat, tmp, mu);
|
||||
return;
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
||||
|
||||
};
|
||||
|
||||
|
Reference in New Issue
Block a user