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

HIP allows conserved currents.

Browse Source 
Tracing replaces self timeing
This commit is contained in:
Peter Boyle 2022-08-31 17:11:18 -04:00
parent ee2d7369b3
commit 24182ca8bf
1 changed files with 1 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

57
Grid/qcd/action/fermion/implementation/CayleyFermion5DImplementation.h
View File

@ -152,58 +152,6 @@ void CayleyFermion5D<Impl>::DminusDag(const FermionField &psi, FermionField &chi
} }
} }
template<class Impl> void CayleyFermion5D<Impl>::CayleyReport(void)
{
this->Report();
Coordinate latt = GridDefaultLatt();
RealD volume = this->Ls; for(int mu=0;mu<Nd;mu++) volume=volume*latt[mu];
RealD NP = this->_FourDimGrid->_Nprocessors;
if ( M5Dcalls > 0 ) {
std::cout << GridLogMessage << "#### M5D calls report " << std::endl;
std::cout << GridLogMessage << "CayleyFermion5D Number of M5D Calls : " << M5Dcalls << std::endl;
std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls : " << M5Dtime / M5Dcalls << " us" << std::endl;
// Flops = 10.0*(Nc*Ns) *Ls*vol
RealD mflops = 10.0*(Nc*Ns)*volume*M5Dcalls/M5Dtime/2; // 2 for red black counting
std::cout << GridLogMessage << "Average mflops/s per call : " << mflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per rank : " << mflops/NP << std::endl;
// Bytes = sizeof(Real) * (Nc*Ns*Nreim) * Ls * vol * (read+write) (/2 for red black counting)
// read = 2 ( psi[ss+s+1] and psi[ss+s-1] count as 1 )
// write = 1
RealD Gbytes = sizeof(Real) * (Nc*Ns*2) * volume * 3 /2. * 1.e-9;
std::cout << GridLogMessage << "Average bandwidth (GB/s) : " << Gbytes/M5Dtime*M5Dcalls*1.e6 << std::endl;
}
if ( MooeeInvCalls > 0 ) {
std::cout << GridLogMessage << "#### MooeeInv calls report " << std::endl;
std::cout << GridLogMessage << "CayleyFermion5D Number of MooeeInv Calls : " << MooeeInvCalls << std::endl;
std::cout << GridLogMessage << "CayleyFermion5D ComputeTime/Calls : " << MooeeInvTime / MooeeInvCalls << " us" << std::endl;
#ifdef GRID_CUDA
RealD mflops = ( -16.*Nc*Ns+this->Ls*(1.+18.*Nc*Ns) )*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
std::cout << GridLogMessage << "Average mflops/s per call : " << mflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per rank : " << mflops/NP << std::endl;
#else
// Flops = MADD * Ls *Ls *4dvol * spin/colour/complex
RealD mflops = 2.0*24*this->Ls*volume*MooeeInvCalls/MooeeInvTime/2; // 2 for red black counting
std::cout << GridLogMessage << "Average mflops/s per call : " << mflops << std::endl;
std::cout << GridLogMessage << "Average mflops/s per call per rank : " << mflops/NP << std::endl;
#endif
}
}
template<class Impl> void CayleyFermion5D<Impl>::CayleyZeroCounters(void)
{
this->ZeroCounters();
M5Dflops=0;
M5Dcalls=0;
M5Dtime=0;
MooeeInvFlops=0;
MooeeInvCalls=0;
MooeeInvTime=0;
}
template<class Impl> template<class Impl>
void CayleyFermion5D<Impl>::M5D (const FermionField &psi, FermionField &chi) void CayleyFermion5D<Impl>::M5D (const FermionField &psi, FermionField &chi)
{ {
@ -646,7 +594,6 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
assert(mass_plus == mass_minus); assert(mass_plus == mass_minus);
RealD mass = mass_plus; RealD mass = mass_plus;
#if (!defined(GRID_HIP))
Gamma::Algebra Gmu [] = { Gamma::Algebra Gmu [] = {
Gamma::Algebra::GammaX, Gamma::Algebra::GammaX,
Gamma::Algebra::GammaY, Gamma::Algebra::GammaY,
@ -765,7 +712,7 @@ void CayleyFermion5D<Impl>::ContractConservedCurrent( PropagatorField &q_in_1,
else q_out += C; else q_out += C;
} }
#endif
} }
template <class Impl> template <class Impl>
@ -832,7 +779,6 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
} }
#endif #endif
#if (!defined(GRID_HIP))
int tshift = (mu == Nd-1) ? 1 : 0; int tshift = (mu == Nd-1) ? 1 : 0;
unsigned int LLt = GridDefaultLatt()[Tp]; unsigned int LLt = GridDefaultLatt()[Tp];
//////////////////////////////////////////////// ////////////////////////////////////////////////
@ -952,7 +898,6 @@ void CayleyFermion5D<Impl>::SeqConservedCurrent(PropagatorField &q_in,
InsertSlice(L_Q, q_out, s , 0); InsertSlice(L_Q, q_out, s , 0);
} }
#endif
} }
#undef Pp #undef Pp
#undef Pm #undef Pm