portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Merge pull request #281 from felixerben/feature/baryonSpeedup

Browse Source 
Feature/baryon speedup
This commit is contained in:
Antonin Portelli 2020-05-07 13:41:29 +01:00 committed by GitHub
commit a887206413
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 156 additions and 91 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

247
Grid/qcd/utils/BaryonUtils.h
View File

@ -46,20 +46,44 @@ public:
typedef typename SpinMatrixField::vector_object sobj;
static const int epsilon[6][3] ;
static const Complex epsilon_sgn[6];
static const double epsilon_sgn[6];
private:
template <class mobj, class robj>
static void baryon_site(const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
const Gamma GammaB_left,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
const int * wick_contractions,
robj &result);
static inline void baryon_site(unsigned int mask, const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
const Gamma GammaB_left,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
robj &result);
template <unsigned int mask, class mobj, class robj>
static inline void baryon_site(const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
const Gamma GammaB_left,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
robj &result);
template <unsigned int maxMask>
struct BaryonSiteHelper
{
template <class mobj, class robj>
static inline void function(const unsigned int mask, const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
const Gamma GammaB_left,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
robj &result);
};
public:
static void ContractBaryons(const PropagatorField &q1_left,
const PropagatorField &q2_left,
@ -152,16 +176,11 @@ public:
template <class FImpl>
const int BaryonUtils<FImpl>::epsilon[6][3] = {{0,1,2},{1,2,0},{2,0,1},{0,2,1},{2,1,0},{1,0,2}};
template <class FImpl>
const Complex BaryonUtils<FImpl>::epsilon_sgn[6] = {Complex(1),
Complex(1),
Complex(1),
Complex(-1),
Complex(-1),
Complex(-1)};
const double BaryonUtils<FImpl>::epsilon_sgn[6] = {1.0,1.0,1.0,-1.0,-1.0,-1.0};
template <class FImpl>
template <class mobj, class robj>
void BaryonUtils<FImpl>::baryon_site(const mobj &D1,
template <unsigned int mask, class mobj, class robj>
inline void BaryonUtils<FImpl>::baryon_site(const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
@ -169,9 +188,14 @@ void BaryonUtils<FImpl>::baryon_site(const mobj &D1,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
const int * wick_contraction,
robj &result)
{
constexpr bool wick_contraction_0 = ((mask & (1 << 5)) >> 5);
constexpr bool wick_contraction_1 = ((mask & (1 << 4)) >> 4);
constexpr bool wick_contraction_2 = ((mask & (1 << 3)) >> 3);
constexpr bool wick_contraction_3 = ((mask & (1 << 2)) >> 2);
constexpr bool wick_contraction_4 = ((mask & (1 << 1)) >> 1);
constexpr bool wick_contraction_5 = ((mask & (1 << 0)) >> 0);
Gamma g4(Gamma::Algebra::GammaT); //needed for parity P_\pm = 0.5*(1 \pm \gamma_4)
@ -180,6 +204,10 @@ void BaryonUtils<FImpl>::baryon_site(const mobj &D1,
auto pD1 = 0.5* (gD1a + (double)parity * gD1b);
auto gD3 = GammaB_right * D3;
auto D2g = D2 * GammaB_left;
auto pD1g = pD1 * GammaB_left;
auto gD3g = gD3 * GammaB_left;
for (int ie_left=0; ie_left < 6 ; ie_left++){
int a_left = epsilon[ie_left][0]; //a
int b_left = epsilon[ie_left][1]; //b
@ -188,64 +216,84 @@ void BaryonUtils<FImpl>::baryon_site(const mobj &D1,
int a_right = epsilon[ie_right][0]; //a'
int b_right = epsilon[ie_right][1]; //b'
int c_right = epsilon[ie_right][2]; //c'
//This is the \delta_{456}^{123} part
if (wick_contraction[0]){
auto D2g = D2 * GammaB_left;
for (int alpha_right=0; alpha_right<Ns; alpha_right++){
for (int beta_left=0; beta_left<Ns; beta_left++){
double ee = epsilon_sgn[ie_left] * epsilon_sgn[ie_right];
//All parts together
for (int gamma_left=0; gamma_left<Ns; gamma_left++){
result()()() += epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,gamma_left)(c_right,c_left)*D2g()(alpha_right,beta_left)(a_right,a_left)*gD3()(alpha_right,beta_left)(b_right,b_left);
}}}
}
//This is the \delta_{456}^{231} part
if (wick_contraction[1]){
auto pD1g = pD1 * GammaB_left;
auto eepD1 = ee * pD1()(gamma_left,gamma_left)(c_right,c_left);
for (int alpha_right=0; alpha_right<Ns; alpha_right++){
auto gD3_ag = gD3()(alpha_right,gamma_left)(b_right,c_left);
auto D2_ag = D2()(alpha_right,gamma_left)(a_right,c_left);
for (int beta_left=0; beta_left<Ns; beta_left++){
for (int gamma_left=0; gamma_left<Ns; gamma_left++){
result()()() += epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1g()(gamma_left,beta_left)(c_right,a_left)*D2()(alpha_right,beta_left)(a_right,b_left)*gD3()(alpha_right,gamma_left)(b_right,c_left);
}}}
}
//This is the \delta_{456}^{312} part
if (wick_contraction[2]){
auto gD3g = gD3 * GammaB_left;
for (int alpha_right=0; alpha_right<Ns; alpha_right++){
for (int beta_left=0; beta_left<Ns; beta_left++){
for (int gamma_left=0; gamma_left<Ns; gamma_left++){
result()()() += epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,beta_left)(c_right,b_left)*D2()(alpha_right,gamma_left)(a_right,c_left)*gD3g()(alpha_right,beta_left)(b_right,a_left);
}}}
}
//This is the \delta_{456}^{132} part
if (wick_contraction[3]){
auto gD3g = gD3 * GammaB_left;
for (int alpha_right=0; alpha_right<Ns; alpha_right++){
for (int beta_left=0; beta_left<Ns; beta_left++){
for (int gamma_left=0; gamma_left<Ns; gamma_left++){
result()()() -= epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,gamma_left)(c_right,c_left)*D2()(alpha_right,beta_left)(a_right,b_left)*gD3g()(alpha_right,beta_left)(b_right,a_left);
}}}
}
//This is the \delta_{456}^{321} part
if (wick_contraction[4]){
auto D2g = D2 * GammaB_left;
for (int alpha_right=0; alpha_right<Ns; alpha_right++){
for (int beta_left=0; beta_left<Ns; beta_left++){
for (int gamma_left=0; gamma_left<Ns; gamma_left++){
result()()() -= epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1()(gamma_left,beta_left)(c_right,b_left)*D2g()(alpha_right,beta_left)(a_right,a_left)*gD3()(alpha_right,gamma_left)(b_right,c_left);
}}}
}
//This is the \delta_{456}^{213} part
if (wick_contraction[5]){
auto pD1g = pD1 * GammaB_left;
for (int alpha_right=0; alpha_right<Ns; alpha_right++){
for (int beta_left=0; beta_left<Ns; beta_left++){
for (int gamma_left=0; gamma_left<Ns; gamma_left++){
result()()() -= epsilon_sgn[ie_left] * epsilon_sgn[ie_right] * pD1g()(gamma_left,beta_left)(c_right,a_left)*D2()(alpha_right,gamma_left)(a_right,c_left)*gD3()(alpha_right,beta_left)(b_right,b_left);
}}}
}
auto eepD1g_gb = ee * pD1g()(gamma_left,beta_left)(c_right,a_left);
auto eepD1_gb = ee * pD1()(gamma_left,beta_left)(c_right,b_left);
auto D2g_ab = D2g()(alpha_right,beta_left)(a_right,a_left);
auto D2_ab = D2()(alpha_right,beta_left)(a_right,b_left);
auto gD3_ab = gD3()(alpha_right,beta_left)(b_right,b_left);
auto gD3g_ab = gD3g()(alpha_right,beta_left)(b_right,a_left);
if(wick_contraction_0){
result()()() += eepD1*D2g_ab*gD3_ab;
}
if(wick_contraction_1){
result()()() += eepD1g_gb*D2_ab*gD3_ag;
}
if(wick_contraction_2){
result()()() += eepD1_gb*D2_ag*gD3g_ab;
}
if(wick_contraction_3){
result()()() -= eepD1*D2_ab*gD3g_ab;
}
if(wick_contraction_4){
result()()() -= eepD1_gb*D2g_ab*gD3_ag;
}
if(wick_contraction_5){
result()()() -= eepD1g_gb*D2_ag*gD3_ab;
}
}}}
}
}
}
template <class FImpl>
template <unsigned int maxMask>
template <class mobj, class robj>
inline void BaryonUtils<FImpl>::BaryonSiteHelper<maxMask>::function(const unsigned int mask, const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
const Gamma GammaB_left,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
robj &result)
{
assert(mask <= maxMask);
if (mask == maxMask)
{
baryon_site<maxMask,decltype(D1),decltype(result)>(D1, D2, D3, GammaA_left, GammaB_left, GammaA_right, GammaB_right, parity, result);
}
else
{
BaryonSiteHelper<(maxMask>0) ? maxMask-1 : 0>::function(mask, D1, D2, D3, GammaA_left, GammaB_left, GammaA_right, GammaB_right, parity, result);
}
}
// top-level function
template <class FImpl>
template <class mobj, class robj>
inline void BaryonUtils<FImpl>::baryon_site(const unsigned int mask, const mobj &D1,
const mobj &D2,
const mobj &D3,
const Gamma GammaA_left,
const Gamma GammaB_left,
const Gamma GammaA_right,
const Gamma GammaB_right,
const int parity,
robj &result)
{
BaryonSiteHelper<63>::function(mask, D1, D2, D3, GammaA_left, GammaB_left, GammaA_right, GammaB_right, parity, result);
}
template<class FImpl>
void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
const PropagatorField &q2_left,
@ -259,26 +307,30 @@ void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
const int parity,
ComplexField &baryon_corr)
{
std::cout << "Contraction <" << quarks_right[0] << quarks_right[1] << quarks_right[2] << "|" << quarks_left[0] << quarks_left[1] << quarks_left[2] << ">" << std::endl;
assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
std::cout << "Contraction <" << quarks_right[0] << quarks_right[1] << quarks_right[2] << "|" << quarks_left[0] << quarks_left[1] << quarks_left[2] << ">" << std::endl;
std::cout << "GammaA (left) " << (GammaA_left.g) << std::endl;
std::cout << "GammaB (left) " << (GammaB_left.g) << std::endl;
std::cout << "GammaA (right) " << (GammaA_right.g) << std::endl;
std::cout << "GammaB (right) " << (GammaB_right.g) << std::endl;
assert(parity==1 || parity == -1 && "Parity must be +1 or -1");
assert(parity==1 || parity == -1 && "Parity must be +1 or -1");
GridBase *grid = q1_left.Grid();
GridBase *grid = q1_left.Grid();
int wick_contraction[6];
for (int ie=0; ie < 6 ; ie++)
wick_contraction[ie] = (quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0;
int wick_id;
for (int ie=0; ie < 6 ; ie++)
wick_id = ((quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0) << (5-ie);
auto vbaryon_corr= baryon_corr.View();
auto v1 = q1_left.View();
auto v2 = q2_left.View();
auto v3 = q3_left.View();
auto vbaryon_corr= baryon_corr.View();
auto v1 = q1_left.View();
auto v2 = q2_left.View();
auto v3 = q3_left.View();
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
// accelerator_for(ss, grid->oSites(), grid->Nsimd(), {
thread_for(ss,grid->oSites(),{
//for(int ss=0; ss < grid->oSites(); ss++){
@ -287,9 +339,10 @@ void BaryonUtils<FImpl>::ContractBaryons(const PropagatorField &q1_left,
auto D3 = v3[ss];
vobj result=Zero();
baryon_site(D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,wick_contraction,result);
baryon_site(wick_id,D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,result);
vbaryon_corr[ss] = result;
} );//end loop over lattice sites
}
template <class FImpl>
template <class mobj, class robj>
@ -305,20 +358,24 @@ void BaryonUtils<FImpl>::ContractBaryons_Sliced(const mobj &D1,
const int parity,
robj &result)
{
std::cout << "Contraction <" << quarks_right[0] << quarks_right[1] << quarks_right[2] << "|" << quarks_left[0] << quarks_left[1] << quarks_left[2] << ">" << std::endl;
assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
std::cout << "Contraction <" << quarks_right[0] << quarks_right[1] << quarks_right[2] << "|" << quarks_left[0] << quarks_left[1] << quarks_left[2] << ">" << std::endl;
std::cout << "GammaA (left) " << (GammaA_left.g) << std::endl;
std::cout << "GammaB (left) " << (GammaB_left.g) << std::endl;
std::cout << "GammaA (right) " << (GammaA_right.g) << std::endl;
std::cout << "GammaB (right) " << (GammaB_right.g) << std::endl;
assert(parity==1 || parity == -1 && "Parity must be +1 or -1");
assert(parity==1 || parity == -1 && "Parity must be +1 or -1");
int wick_contraction[6];
for (int ie=0; ie < 6 ; ie++)
wick_contraction[ie] = (quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0;
result=Zero();
baryon_site(D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,wick_contraction,result);
int wick_id;
for (int ie=0; ie < 6 ; ie++)
wick_id = ((quarks_left[0] == quarks_right[epsilon[ie][0]] && quarks_left[1] == quarks_right[epsilon[ie][1]] && quarks_left[2] == quarks_right[epsilon[ie][2]]) ? 1 : 0) << (5-ie);
result=Zero();
baryon_site(wick_id,D1,D2,D3,GammaA_left,GammaB_left,GammaA_right,GammaB_right,parity,result);
}
/***********************************************************************
@ -558,6 +615,10 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_Eye(const PropagatorField &qq_loop,
const std::string op,
SpinMatrixField &stn_corr)
{
assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
GridBase *grid = qs_ti.Grid();
auto vcorr= stn_corr.View();
@ -595,6 +656,10 @@ void BaryonUtils<FImpl>::Sigma_to_Nucleon_NonEye(const PropagatorField &qq_ti,
const std::string op,
SpinMatrixField &stn_corr)
{
assert(Ns==4 && "Baryon code only implemented for N_spin = 4");
assert(Nc==3 && "Baryon code only implemented for N_colour = 3");
GridBase *grid = qs_ti.Grid();
auto vcorr= stn_corr.View();