Skip to content

Instantly share code, notes, and snippets.

Embed
What would you like to do?
Benchmark_ITT gprof output.
Flat profile:
Each sample counts as 0.01 seconds.
% cumulative self self total
time seconds seconds calls s/call s/call name
25.03 722.12 722.12 473465689 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&)
11.38 1050.49 328.37 32487566 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
10.33 1348.49 298.00 28067609 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
10.01 1637.34 288.85 33607682 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
9.71 1917.63 280.29 29677671 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
8.77 2170.77 253.14 30706382 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
8.34 2411.47 240.70 21952384 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
6.89 2610.42 198.95 20202520 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
6.72 2804.30 193.88 19253970 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
1.27 2840.88 36.58 Grid::QCD::DomainWallFermion<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MomentumSpacePropagator(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, double)
0.17 2845.88 5.00 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded()
0.15 2850.32 4.44 36 0.12 0.31 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&)
0.14 2854.33 4.01 27861504 0.00 0.00 Grid::GridRedBlackCartesian::CheckerBoardFromOindex(int)
0.13 2858.17 3.84 Grid::QCD::FermionOperator<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::RedBlackGrid()
0.13 2861.92 3.75 88378312 0.00 0.00 Grid::permute(Grid::Grid_simd<std::complex<float>, __Float32x4_t>&, Grid::Grid_simd<std::complex<float>, __Float32x4_t>, int)
0.12 2865.52 3.60 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int)
0.09 2868.12 2.61 32499663 0.00 0.00 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long)
0.07 2870.01 1.89 778 0.00 0.00 frame_dummy
0.06 2871.85 1.84 __multf3
0.06 2873.59 1.74 Grid::QCD::FermionOperator<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooDeriv(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int)
0.05 2875.08 1.49 std::enable_if<Grid::matchGridTensorIndex<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4>, 2>::value, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >::type Grid::QCD::operator*<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >(Grid::QCD::Gamma const&, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> const&)
0.05 2876.42 1.34 25 0.05 0.05 Grid::GridRNGbase::~GridRNGbase()
0.03 2877.40 0.98 32 0.03 0.03 void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int)
0.03 2878.21 0.81 25 0.03 0.03 std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::discrete_distribution<int>*, std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > > >, unsigned long, std::discrete_distribution<int> const&)
0.02 2878.91 0.70 Benchmark::DWF(int, int, double&)
0.02 2879.56 0.65 844288 0.00 0.00 void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&)
0.02 2880.21 0.65 36 0.02 0.02 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::BuildSurfaceList(int, int)
0.02 2880.86 0.65 __trunctfdf2
0.02 2881.40 0.54 1560 0.00 0.00 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&)
0.02 2881.88 0.48 __extenddftf2
0.01 2882.27 0.39 180 0.00 0.00 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::allocate(unsigned long, void const*)
0.01 2882.58 0.31 4 0.08 0.11 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*)
0.01 2882.86 0.28 8 0.04 0.05 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*)
0.01 2883.13 0.27 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::TwoIndexRep<3, (Grid::QCD::TwoIndexSymmetry)1>, Grid::QCD::CoeffReal> >::ContractConservedCurrentSiteBwd(Grid::iScalar<Grid::iMatrix<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6>, 4> > const&, Grid::iScalar<Grid::iMatrix<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6>, 4> > const&, Grid::iScalar<Grid::iMatrix<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6>, 4> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6> >, 8> >&, unsigned int, unsigned int, bool)
0.01 2883.31 0.18 79152 0.00 0.00 Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&)
0.00 2883.45 0.14 1497897 0.00 0.00 Grid::GridBase::iIndex(std::vector<int, std::allocator<int> >&)
0.00 2883.59 0.14 1535850 0.00 0.00 Grid::GridBase::oIndex(std::vector<int, std::allocator<int> >&)
0.00 2883.72 0.13 422144 0.00 0.00 void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&)
0.00 2883.85 0.13 424 0.00 0.00 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*)
0.00 2883.98 0.13 25 0.01 0.01 std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::normal_distribution<double>*, std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > > >, unsigned long, std::normal_distribution<double> const&)
0.00 2884.10 0.12 36 0.00 0.00 Grid::alignedAllocator<Grid::StencilEntry>::allocate(unsigned long, void const*)
0.00 2884.21 0.11 3377152 0.00 0.00 Grid::CartesianCommunicator::ProcessorCoorFromRank(int, std::vector<int, std::allocator<int> >&)
0.00 2884.31 0.10 410303 0.00 0.00 Grid::GridRedBlackCartesian::oIndex(std::vector<int, std::allocator<int> >&)
0.00 2884.40 0.09 625601 0.00 0.00 Grid::GridRedBlackCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&)
0.00 2884.49 0.09 296762 0.00 0.00 sitmo::prng_engine::encrypt_counter()
0.00 2884.58 0.09 Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const Grid::predicatedWhere<Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const>(Grid::Grid_simd<unsigned int, __Uint32x4_t> const& const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&)
0.00 2884.67 0.09 36 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*)
0.00 2884.74 0.07 4 0.02 0.29 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
0.00 2884.80 0.06 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteDag(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
0.00 2884.86 0.06 __divdc3
0.00 2884.91 0.05 25 0.00 0.00 std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_int_distribution<unsigned int>*, std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > > >, unsigned long, std::uniform_int_distribution<unsigned int> const&)
0.00 2884.96 0.05 24 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::allocate(unsigned long, void const*)
0.00 2885.01 0.05 Grid::Gather_plane_table_compute(Grid::GridBase*, int, int, int, int, std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >&)
0.00 2885.06 0.05 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplex> >::HandDhopSiteDagExt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&)
0.00 2885.11 0.05 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::TwoIndexRep<3, (Grid::QCD::TwoIndexSymmetry)-1>, Grid::QCD::CoeffReal> >::DhopDir(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int, int)
0.00 2885.15 0.04 33 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::allocate(unsigned long, void const*)
0.00 2885.19 0.04 25 0.00 0.00 std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_real_distribution<double>*, std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > > >, unsigned long, std::uniform_real_distribution<double> const&)
0.00 2885.23 0.04 16 0.00 0.01 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*)
0.00 2885.27 0.04 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteDag(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&)
0.00 2885.30 0.03 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopDir(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int, int)
0.00 2885.32 0.02 167755 0.00 0.00 Grid::Grid_simd<unsigned int, __Uint32x4_t> Grid::Comparison<Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t>, 0>(Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::Grid_simd<unsigned int, __Uint32x4_t>::scalar_type const&)
0.00 2885.34 0.02 112128 0.00 0.00 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMerge<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge> >&, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress> >&)
0.00 2885.36 0.02 12 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::allocate(unsigned long, void const*)
0.00 2885.37 0.01 3585944 0.00 0.00 Grid::CartesianCommunicator::ThisRank()
0.00 2885.38 0.01 190280 0.00 0.00 Grid::CartesianCommunicator::RankFromProcessorCoor(std::vector<int, std::allocator<int> >&)
0.00 2885.39 0.01 19232 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.40 0.01 54 0.00 0.00 std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >::_M_default_append(unsigned long)
0.00 2885.41 0.01 16 0.00 0.00 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int, int)
0.00 2885.42 0.01 Grid::CartesianCommunicator::ShiftedRanks(int, int, int&, int&)
0.00 2885.43 0.01 Grid::CartesianCommunicator::BossRank()
0.00 2885.44 0.01 Benchmark::Memory()
0.00 2885.45 0.01 std::vector<std::vector<ompi_request_t*, std::allocator<ompi_request_t*> >, std::allocator<std::vector<ompi_request_t*, std::allocator<ompi_request_t*> > > >::~vector()
0.00 2885.45 0.01 24 0.00 0.00 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.01 Grid::GridRedBlackCartesian::~GridRedBlackCartesian()
0.00 2885.46 0.01 std::ctype<char>::do_widen(char) const
0.00 2885.46 0.00 1266432 0.00 0.00 Grid::GridCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 112128 0.00 0.00 Grid::CartesianCommunicator::StencilBarrier()
0.00 2885.46 0.00 56064 0.00 0.00 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void>&, int, int&)
0.00 2885.46 0.00 56064 0.00 0.00 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Prepare()
0.00 2885.46 0.00 38464 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 38432 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 28032 0.00 0.00 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&)
0.00 2885.46 0.00 28032 0.00 0.00 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&)
0.00 2885.46 0.00 19232 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 17600 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 17600 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 8800 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 8800 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 3206 0.00 0.00 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int)
0.00 2885.46 0.00 3024 0.00 0.00 Grid::GridRedBlackCartesian::CheckerBoardShiftForCB(int, int, int, int)
0.00 2885.46 0.00 2790 0.00 0.00 Grid::GridCartesian::CheckerBoarded(int)
0.00 2885.46 0.00 2640 0.00 0.00 Grid::GridRedBlackCartesian::CheckerBoarded(int)
0.00 2885.46 0.00 1325 0.00 0.00 Grid::PointerCache::Insert(void*, unsigned long)
0.00 2885.46 0.00 1325 0.00 0.00 Grid::PointerCache::Lookup(unsigned long)
0.00 2885.46 0.00 1040 0.00 0.00 Grid::LebesgueOrder::IterateO(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&)
0.00 2885.46 0.00 824 0.00 0.00 void Grid::Copy_plane<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int)
0.00 2885.46 0.00 672 0.00 0.00 Grid::CartesianCommunicator::GlobalSum(double&)
0.00 2885.46 0.00 620 0.00 0.00 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&)
0.00 2885.46 0.00 618 0.00 0.00 void Grid::Copy_plane<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int)
0.00 2885.46 0.00 432 0.00 0.00 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >*, unsigned long)
0.00 2885.46 0.00 432 0.00 0.00 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::allocate(unsigned long, void const*)
0.00 2885.46 0.00 424 0.00 0.00 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long)
0.00 2885.46 0.00 224 0.00 0.00 Grid::GridCartesian::CheckerBoardDestination(int, int, int)
0.00 2885.46 0.00 216 0.00 0.00 Grid::SharedMemory::ShmBufferMalloc(unsigned long)
0.00 2885.46 0.00 191 0.00 0.00 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 180 0.00 0.00 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long)
0.00 2885.46 0.00 144 0.00 0.00 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Report()
0.00 2885.46 0.00 144 0.00 0.00 Grid::CartesianCommunicator::Barrier()
0.00 2885.46 0.00 144 0.00 0.00 Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Reporti(int)
0.00 2885.46 0.00 127 0.00 0.00 Grid::SharedMemory::ShmBufferFreeAll()
0.00 2885.46 0.00 120 0.00 0.00 std::vector<double, std::allocator<double> >::_M_default_append(unsigned long)
0.00 2885.46 0.00 120 0.00 0.00 std::vector<double, std::allocator<double> >::resize(unsigned long)
0.00 2885.46 0.00 116 0.00 0.00 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*)
0.00 2885.46 0.00 104 0.00 0.00 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::~Lattice()
0.00 2885.46 0.00 92 0.00 0.00 std::vector<void*, std::allocator<void*> >::_M_default_append(unsigned long)
0.00 2885.46 0.00 91 0.00 0.00 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*)
0.00 2885.46 0.00 76 0.00 0.00 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&)
0.00 2885.46 0.00 76 0.00 0.00 Grid::alignedAllocator<std::complex<double> >::deallocate(std::complex<double>*, unsigned long)
0.00 2885.46 0.00 76 0.00 0.00 Grid::alignedAllocator<std::complex<double> >::allocate(unsigned long, void const*)
0.00 2885.46 0.00 76 0.00 0.00 Grid::CartesianCommunicator::GlobalSumVector(double*, int)
0.00 2885.46 0.00 76 0.00 0.00 Grid::CartesianCommunicator::GlobalSum(std::complex<double>&)
0.00 2885.46 0.00 72 0.00 0.00 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::deallocate(Grid::Grid_simd<unsigned int, __Uint32x4_t>*, unsigned long)
0.00 2885.46 0.00 72 0.00 0.00 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::allocate(unsigned long, void const*)
0.00 2885.46 0.00 72 0.00 0.00 std::vector<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, std::allocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*> >::_M_default_append(unsigned long)
0.00 2885.46 0.00 64 0.00 0.00 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int)
0.00 2885.46 0.00 64 0.00 0.00 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int)
0.00 2885.46 0.00 55 0.00 0.00 Grid::GridCmdVectorIntToString[abi:cxx11](std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 52 0.00 0.00 Grid::CartesianCommunicator::InitFromMPICommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*)
0.00 2885.46 0.00 52 0.00 0.00 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&)
0.00 2885.46 0.00 48 0.00 0.00 time_statistics::statistics(std::vector<double, std::allocator<double> >)
0.00 2885.46 0.00 48 0.00 0.00 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int)
0.00 2885.46 0.00 48 0.00 0.00 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double)
0.00 2885.46 0.00 48 0.00 0.00 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int)
0.00 2885.46 0.00 48 0.00 0.00 Grid::GridBase::PermuteType(int)
0.00 2885.46 0.00 48 0.00 0.00 std::vector<int, std::allocator<int> >::vector(std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 42 0.00 0.00 Grid::GridDefaultMpi()
0.00 2885.46 0.00 42 0.00 0.00 Grid::Grid_sa_signal_handler(int, siginfo_t*, void*)
0.00 2885.46 0.00 39 0.00 0.00 Grid::GlobalSharedMemory::OptimalCommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*&)
0.00 2885.46 0.00 39 0.00 0.00 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 36 0.00 0.00 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::~CartesianStencil()
0.00 2885.46 0.00 36 0.00 0.00 Grid::alignedAllocator<Grid::StencilEntry>::deallocate(Grid::StencilEntry*, unsigned long)
0.00 2885.46 0.00 36 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long)
0.00 2885.46 0.00 36 0.00 0.00 Grid::CartesianCommunicator::BroadcastWorld(int, void*, int)
0.00 2885.46 0.00 36 0.00 0.00 std::vector<Grid::StencilEntry, Grid::alignedAllocator<Grid::StencilEntry> >::_M_default_append(unsigned long)
0.00 2885.46 0.00 36 0.00 0.00 void std::vector<double, std::allocator<double> >::_M_realloc_insert<double const&>(__gnu_cxx::__normal_iterator<double*, std::vector<double, std::allocator<double> > >, double const&)
0.00 2885.46 0.00 33 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::deallocate(Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8>*, unsigned long)
0.00 2885.46 0.00 32 0.00 0.00 Grid::CartesianCommunicator::Broadcast(int, void*, int)
0.00 2885.46 0.00 32 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters()
0.00 2885.46 0.00 32 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopOE(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
0.00 2885.46 0.00 32 0.00 0.01 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report()
0.00 2885.46 0.00 31 0.00 0.00 Grid::GridDefaultSimd(int, int)
0.00 2885.46 0.00 25 0.00 0.00 std::seed_seq::seed_seq<__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > > >(__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >)
0.00 2885.46 0.00 24 0.00 0.01 Grid::LebesgueOrder::CartesianBlocking()
0.00 2885.46 0.00 24 0.00 0.01 Grid::LebesgueOrder::LebesgueOrder(Grid::GridBase*)
0.00 2885.46 0.00 24 0.00 0.04 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*)
0.00 2885.46 0.00 24 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8>*, unsigned long)
0.00 2885.46 0.00 24 0.00 0.00 Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 24 0.00 0.00 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::Lattice(Grid::GridBase*)
0.00 2885.46 0.00 23 0.00 0.00 Grid::SharedMemory::~SharedMemory()
0.00 2885.46 0.00 23 0.00 0.00 Grid::CartesianCommunicator::~CartesianCommunicator()
0.00 2885.46 0.00 22 0.00 0.00 char** std::__find_if<char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const> >(char**, char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const>, std::random_access_iterator_tag)
0.00 2885.46 0.00 20 0.00 0.00 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::Lattice(Grid::GridBase*)
0.00 2885.46 0.00 16 0.00 0.00 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*)
0.00 2885.46 0.00 16 0.00 0.00 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooeeInternalCompute(int, int, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&)
0.00 2885.46 0.00 16 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters()
0.00 2885.46 0.00 16 0.00 0.01 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report()
0.00 2885.46 0.00 12 0.00 0.00 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>*, unsigned long)
0.00 2885.46 0.00 12 0.00 0.00 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int, int)
0.00 2885.46 0.00 12 0.00 0.00 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::Lattice(Grid::GridBase*)
0.00 2885.46 0.00 12 0.00 0.00 std::vector<std::complex<float>, std::allocator<std::complex<float> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::complex<float>*, std::vector<std::complex<float>, std::allocator<std::complex<float> > > >, unsigned long, std::complex<float> const&)
0.00 2885.46 0.00 8 0.00 0.14 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
0.00 2885.46 0.00 8 0.00 0.00 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*)
0.00 2885.46 0.00 8 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonKernels(Grid::QCD::WilsonImplParams const&)
0.00 2885.46 0.00 8 0.00 0.00 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsTanh(Grid::Approx::zolotarev_data*, double, double)
0.00 2885.46 0.00 8 0.00 0.00 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsInternal(double, std::vector<double, std::allocator<double> >&, double, double)
0.00 2885.46 0.00 8 0.00 1.14 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::CayleyFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, double, Grid::QCD::WilsonImplParams const&)
0.00 2885.46 0.00 8 0.00 0.00 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~CayleyFermion5D()
0.00 2885.46 0.00 8 0.00 0.14 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
0.00 2885.46 0.00 8 0.00 1.14 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&)
0.00 2885.46 0.00 8 0.00 0.00 Grid::Approx::zolotarev_free(Grid::Approx::zolotarev_data*)
0.00 2885.46 0.00 8 0.00 0.00 Grid::Approx::higham(double, int)
0.00 2885.46 0.00 8 0.00 0.00 Grid::Approx::construct_contfrac(Grid::Approx::izd*)
0.00 2885.46 0.00 8 0.00 0.00 Grid::Approx::construct_partfrac(Grid::Approx::izd*)
0.00 2885.46 0.00 4 0.00 0.00 _GLOBAL__sub_I__ZN4Grid3QCD13SpaceTimeGrid15makeFourDimGridERKSt6vectorIiSaIiEES6_S6_
0.00 2885.46 0.00 4 0.00 0.00 Grid::QCD::SpaceTimeGrid::makeFourDimDWFGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&)
0.00 2885.46 0.00 4 0.00 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonKernels(Grid::QCD::WilsonImplParams const&)
0.00 2885.46 0.00 4 0.00 0.29 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
0.00 2885.46 0.00 4 0.00 1.28 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&)
0.00 2885.46 0.00 4 0.00 0.00 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D()
0.00 2885.46 0.00 4 0.00 0.00 Grid::Approx::contfrac_A(double*, double*, double*, double*, int, int)
0.00 2885.46 0.00 3 0.00 0.00 void std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >::_M_realloc_insert<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&>(__gnu_cxx::__normal_iterator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)
0.00 2885.46 0.00 3 0.00 0.00 std::_Rb_tree<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::_Select1st<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >, std::less<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::allocator<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >::_M_erase(std::_Rb_tree_node<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >*)
0.00 2885.46 0.00 2 0.00 0.00 _GLOBAL__sub_I__ZN4Grid3QCD22StaggeredKernelsStatic3OptE
0.00 2885.46 0.00 2 0.00 0.00 _GLOBAL__sub_I__ZN4Grid3QCD5Gamma3gmuE
0.00 2885.46 0.00 2 0.00 0.00 Grid::Grid_quiesce_nodes()
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I_L_list
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid10GridThread8_threadsE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid13LebesgueOrder16UseLebesgueOrderE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid14MemoryProfiler5statsE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid18GlobalSharedMemory17MAX_MPI_SHM_BYTESE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid18GlobalSharedMemory4InitEP19ompi_communicator_t
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid21CartesianCommunicator18CommunicatorPolicyE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid21CartesianCommunicator18communicator_worldE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid26Gather_plane_table_computeEPNS_8GridBaseEiiiiRSt6vectorISt4pairIiiESaIS4_EE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid3QCD13WilsonKernelsINS0_17GparityWilsonImplINS_9Grid_simdISt7complexIfE13__Float32x4_tEENS0_14FundamentalRepILi3EEENS0_9CoeffRealEEEE12HandDhopSiteERNS0_13WilsonStencilINS_7iVectorINSE_INSE_IS7_Li3EEELi4EEELi2EEENSE_INSE_ISF_Li2EEELi2EEEEERNS_13LebesgueOrderERNS_7LatticeINSE_INSE_INS_7iScalarINS_7iMatrixIS7_Li3EEEEELi8EEELi2EEEEEPSJ_iiRKNSO_ISH_EERSY_
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid3QCD19WilsonKernelsStatic3OptE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid3QCD21WilsonFermion5DStatic10directionsE
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid8demangleB5cxx11EPKc
0.00 2885.46 0.00 1 0.00 0.00 _GLOBAL__sub_I__ZN4Grid9printHashEv
0.00 2885.46 0.00 1 0.00 0.00 __static_initialization_and_destruction_0(int, int) [clone .constprop.58]
0.00 2885.46 0.00 1 0.00 0.00 Grid::GridParseLayout(char**, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&)
0.00 2885.46 0.00 1 0.00 0.00 Grid::GridCmdOptionCSL(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&)
0.00 2885.46 0.00 1 0.00 0.00 Grid::GridLogConfigure(std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&)
0.00 2885.46 0.00 1 0.00 0.00 Grid::GridLogTimestamp(int)
0.00 2885.46 0.00 1 0.00 0.00 Grid::GlobalSharedMemory::SharedMemoryAllocate(unsigned long, int)
0.00 2885.46 0.00 1 0.00 0.00 Grid::GlobalSharedMemory::Init(ompi_communicator_t*)
0.00 2885.46 0.00 1 0.00 0.00 Grid::CartesianCommunicator::Init(int*, char***)
0.00 2885.46 0.00 1 0.00 0.00 Grid::CartesianCommunicator::RankWorld()
0.00 2885.46 0.00 1 0.00 0.00 Grid::printHash()
0.00 2885.46 0.00 1 0.00 0.00 Grid::operator<<(std::ostream&, Grid::Logger&) [clone .constprop.140]
0.00 2885.46 0.00 1 0.00 0.00 std::_Rb_tree<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::_Select1st<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >, std::less<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::allocator<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >::_M_get_insert_unique_pos(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)
0.00 2885.46 0.00 1 0.00 0.00 void std::__introsort_loop<__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, long, __gnu_cxx::__ops::_Iter_comp_iter<std::less<int> > >(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, long, __gnu_cxx::__ops::_Iter_comp_iter<std::less<int> >)
% the percentage of the total running time of the
time program used by this function.
cumulative a running sum of the number of seconds accounted
seconds for by this function and those listed above it.
self the number of seconds accounted for by this
seconds function alone. This is the major sort for this
listing.
calls the number of times this function was invoked, if
this function is profiled, else blank.
self the average number of milliseconds spent in this
ms/call function per call, if this function is profiled,
else blank.
total the average number of milliseconds spent in this
ms/call function and its descendents per call, if this
function is profiled, else blank.
name the name of the function. This is the minor sort
for this listing. The index shows the location of
the function in the gprof listing. If the index is
in parenthesis it shows where it would appear in
the gprof listing if it were to be printed.
Copyright (C) 2012-2015 Free Software Foundation, Inc.
Copying and distribution of this file, with or without modification,
are permitted in any medium without royalty provided the copyright
notice and this notice are preserved.
Call graph (explanation follows)
granularity: each sample hit covers 2 byte(s) for 0.00% of 2885.46 seconds
index % time self children called name
1193 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
[1] 53.3 5.00 1532.19 0+1193 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
240.70 261.21 21952384/21952384 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [4]
193.88 227.74 19253970/19253970 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [6]
328.37 0.00 32487566/32487566 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [7]
280.29 0.00 29677671/29677671 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [10]
1193 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
-----------------------------------------------
<spontaneous>
[2] 44.3 3.60 1275.86 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int) [2]
198.95 233.70 20202520/20202520 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [5]
298.00 3.22 28067609/28067609 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [8]
288.85 0.00 33607682/33607682 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [9]
253.14 0.00 30706382/30706382 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [11]
-----------------------------------------------
227.74 0.00 149316963/473465689 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [6]
233.70 0.00 153229376/473465689 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [5]
260.68 0.00 170919350/473465689 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [4]
[3] 25.0 722.12 0.00 473465689 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [3]
-----------------------------------------------
240.70 261.21 21952384/21952384 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
[4] 17.4 240.70 261.21 21952384 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [4]
260.68 0.00 170919350/473465689 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [3]
0.53 0.00 12524980/88378312 Grid::permute(Grid::Grid_simd<std::complex<float>, __Float32x4_t>&, Grid::Grid_simd<std::complex<float>, __Float32x4_t>, int) [21]
-----------------------------------------------
198.95 233.70 20202520/20202520 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int) [2]
[5] 15.0 198.95 233.70 20202520 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [5]
233.70 0.00 153229376/473465689 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [3]
-----------------------------------------------
193.88 227.74 19253970/19253970 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
[6] 14.6 193.88 227.74 19253970 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [6]
227.74 0.00 149316963/473465689 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [3]
-----------------------------------------------
328.37 0.00 32487566/32487566 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
[7] 11.4 328.37 0.00 32487566 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [7]
-----------------------------------------------
298.00 3.22 28067609/28067609 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int) [2]
[8] 10.4 298.00 3.22 28067609 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [8]
3.22 0.00 75853332/88378312 Grid::permute(Grid::Grid_simd<std::complex<float>, __Float32x4_t>&, Grid::Grid_simd<std::complex<float>, __Float32x4_t>, int) [21]
-----------------------------------------------
288.85 0.00 33607682/33607682 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int) [2]
[9] 10.0 288.85 0.00 33607682 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [9]
-----------------------------------------------
280.29 0.00 29677671/29677671 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommunicateThreaded() [1]
[10] 9.7 280.29 0.00 29677671 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [10]
-----------------------------------------------
253.14 0.00 30706382/30706382 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int) [2]
[11] 8.8 253.14 0.00 30706382 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [11]
-----------------------------------------------
<spontaneous>
[12] 1.3 36.58 0.00 Grid::QCD::DomainWallFermion<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MomentumSpacePropagator(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, double) [12]
-----------------------------------------------
<spontaneous>
[13] 0.5 0.70 13.24 Benchmark::DWF(int, int, double&) [13]
0.00 9.13 8/8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::CayleyFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, double, Grid::QCD::WilsonImplParams const&) [15]
1.20 0.05 496/778 frame_dummy [23]
0.86 0.00 16/25 Grid::GridRNGbase::~GridRNGbase() [27]
0.00 0.70 16/24 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
0.28 0.16 8/8 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
0.00 0.38 32/32 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
0.00 0.16 72/116 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*) [45]
0.00 0.15 64/64 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int) [54]
0.02 0.02 8/16 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*) [67]
0.04 0.00 16/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.03 0.00 112/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.03 32/48 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
0.00 0.02 38464/38464 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [83]
0.00 0.00 16/24 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [104]
0.00 0.00 16/24 Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [112]
0.00 0.00 8/16 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [113]
0.00 0.00 38432/38432 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [141]
0.00 0.00 136/180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long) [158]
0.00 0.00 112/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
0.00 0.00 96/144 Grid::CartesianCommunicator::Barrier() [159]
0.00 0.00 72/76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
0.00 0.00 72/104 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::~Lattice() [163]
0.00 0.00 32/55 Grid::GridCmdVectorIntToString[abi:cxx11](std::vector<int, std::allocator<int> > const&) [173]
0.00 0.00 32/32 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters() [187]
0.00 0.00 32/32 Grid::CartesianCommunicator::Broadcast(int, void*, int) [186]
0.00 0.00 32/48 time_statistics::statistics(std::vector<double, std::allocator<double> >) [174]
0.00 0.00 32/32 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopOE(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [188]
0.00 0.00 24/42 Grid::GridDefaultMpi() [177]
0.00 0.00 24/36 void std::vector<double, std::allocator<double> >::_M_realloc_insert<double const&>(__gnu_cxx::__normal_iterator<double*, std::vector<double, std::allocator<double> > >, double const&) [184]
0.00 0.00 16/31 Grid::GridDefaultSimd(int, int) [189]
0.00 0.00 16/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
0.00 0.00 8/12 std::vector<std::complex<float>, std::allocator<std::complex<float> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::complex<float>*, std::vector<std::complex<float>, std::allocator<std::complex<float> > > >, unsigned long, std::complex<float> const&) [198]
0.00 0.00 8/8 Grid::Approx::higham(double, int) [204]
0.00 0.00 8/8 Grid::Approx::zolotarev_free(Grid::Approx::zolotarev_data*) [203]
0.00 0.00 8/8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsTanh(Grid::Approx::zolotarev_data*, double, double) [200]
0.00 0.00 8/8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~CayleyFermion5D() [202]
-----------------------------------------------
1.48 2.20 12/36 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
2.96 4.39 24/36 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
[14] 0.4 4.44 6.59 36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
4.01 2.46 27861504/27861504 Grid::GridRedBlackCartesian::CheckerBoardFromOindex(int) [18]
0.00 0.12 36/36 std::vector<Grid::StencilEntry, Grid::alignedAllocator<Grid::StencilEntry> >::_M_default_append(unsigned long) [60]
0.00 0.00 3024/3024 Grid::GridRedBlackCartesian::CheckerBoardShiftForCB(int, int, int, int) [145]
0.00 0.00 2640/2640 Grid::GridRedBlackCartesian::CheckerBoarded(int) [147]
0.00 0.00 1512/3206 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [144]
0.00 0.00 1320/2790 Grid::GridCartesian::CheckerBoarded(int) [146]
0.00 0.00 216/216 Grid::SharedMemory::ShmBufferMalloc(unsigned long) [156]
0.00 0.00 72/191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
0.00 0.00 72/72 std::vector<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, std::allocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*> >::_M_default_append(unsigned long) [172]
0.00 0.00 36/127 Grid::SharedMemory::ShmBufferFreeAll() [160]
-----------------------------------------------
0.00 9.13 8/8 Benchmark::DWF(int, int, double&) [13]
[15] 0.3 0.00 9.13 8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::CayleyFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, double, Grid::QCD::WilsonImplParams const&) [15]
0.00 9.13 8/8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
-----------------------------------------------
0.00 9.13 8/8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::CayleyFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, double, Grid::QCD::WilsonImplParams const&) [15]
[16] 0.3 0.00 9.13 8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
2.96 4.39 24/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
0.00 1.16 8/8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [28]
0.43 0.00 24/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::BuildSurfaceList(int, int) [36]
0.00 0.12 16/24 Grid::LebesgueOrder::LebesgueOrder(Grid::GridBase*) [53]
0.00 0.05 24/24 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::Lattice(Grid::GridBase*) [72]
0.00 0.02 8/116 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*) [45]
0.00 0.00 8/8 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonKernels(Grid::QCD::WilsonImplParams const&) [199]
-----------------------------------------------
<spontaneous>
[17] 0.3 0.00 7.29 Benchmark::DWF5(int, int) [17]
0.00 5.14 4/4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
0.55 0.02 228/778 frame_dummy [23]
0.31 0.15 4/4 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.43 0.00 8/25 Grid::GridRNGbase::~GridRNGbase() [27]
0.00 0.35 8/24 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
0.00 0.19 16/16 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.00 0.07 32/116 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*) [45]
0.02 0.02 8/16 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*) [67]
0.00 0.01 16/48 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
0.01 0.00 4/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.00 0.01 17600/17600 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [101]
0.00 0.00 8/24 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [104]
0.00 0.00 4/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.00 8/24 Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [112]
0.00 0.00 4/4 Grid::QCD::SpaceTimeGrid::makeFourDimDWFGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [118]
0.00 0.00 4/8 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*) [114]
0.00 0.00 17600/17600 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [143]
0.00 0.00 48/144 Grid::CartesianCommunicator::Barrier() [159]
0.00 0.00 32/180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long) [158]
0.00 0.00 32/104 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::~Lattice() [163]
0.00 0.00 16/42 Grid::GridDefaultMpi() [177]
0.00 0.00 16/55 Grid::GridCmdVectorIntToString[abi:cxx11](std::vector<int, std::allocator<int> > const&) [173]
0.00 0.00 16/16 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters() [196]
0.00 0.00 16/48 time_statistics::statistics(std::vector<double, std::allocator<double> >) [174]
0.00 0.00 12/36 void std::vector<double, std::allocator<double> >::_M_realloc_insert<double const&>(__gnu_cxx::__normal_iterator<double*, std::vector<double, std::allocator<double> > >, double const&) [184]
0.00 0.00 8/31 Grid::GridDefaultSimd(int, int) [189]
0.00 0.00 4/76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
0.00 0.00 4/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
0.00 0.00 4/12 std::vector<std::complex<float>, std::allocator<std::complex<float> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::complex<float>*, std::vector<std::complex<float>, std::allocator<std::complex<float> > > >, unsigned long, std::complex<float> const&) [198]
0.00 0.00 4/4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D() [209]
0.00 0.00 4/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
-----------------------------------------------
4.01 2.46 27861504/27861504 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[18] 0.2 4.01 2.46 27861504 Grid::GridRedBlackCartesian::CheckerBoardFromOindex(int) [18]
2.23 0.23 27861504/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
-----------------------------------------------
0.00 5.14 4/4 Benchmark::DWF5(int, int) [17]
[19] 0.2 0.00 5.14 4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
1.48 2.20 12/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
0.00 1.15 4/4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [29]
0.22 0.00 12/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::BuildSurfaceList(int, int) [36]
0.00 0.06 8/24 Grid::LebesgueOrder::LebesgueOrder(Grid::GridBase*) [53]
0.00 0.02 12/12 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::Lattice(Grid::GridBase*) [88]
0.00 0.01 4/116 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*) [45]
0.00 0.00 4/4 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonKernels(Grid::QCD::WilsonImplParams const&) [208]
-----------------------------------------------
<spontaneous>
[20] 0.1 3.84 0.00 Grid::QCD::FermionOperator<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::RedBlackGrid() [20]
-----------------------------------------------
0.53 0.00 12524980/88378312 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSiteInt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [4]
3.22 0.00 75853332/88378312 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::GenericDhopSite(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [8]
[21] 0.1 3.75 0.00 88378312 Grid::permute(Grid::Grid_simd<std::complex<float>, __Float32x4_t>&, Grid::Grid_simd<std::complex<float>, __Float32x4_t>, int) [21]
-----------------------------------------------
923818 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 2/32499663 Grid::GlobalSharedMemory::Init(ompi_communicator_t*) [120]
0.00 0.00 32/32499663 void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int) [32]
0.00 0.00 39/32499663 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
0.00 0.00 48/32499663 Grid::QCD::SpaceTimeGrid::makeFourDimDWFGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [118]
0.00 0.00 52/32499663 Grid::CartesianCommunicator::InitFromMPICommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*) [119]
0.00 0.00 91/32499663 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [116]
0.00 0.00 96/32499663 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*) [114]
0.00 0.00 132/32499663 Benchmark::Memory() [49]
0.00 0.00 192/32499663 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [113]
0.00 0.00 288/32499663 Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [112]
0.00 0.00 25800/32499663 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.01 0.00 178511/32499663 frame_dummy [23]
0.03 0.00 422144/32499663 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.04 0.00 527888/32499663 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*) [67]
0.14 0.01 1688628/32499663 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.14 0.01 1794216/32499663 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
2.23 0.23 27861504/32499663 Grid::GridRedBlackCartesian::CheckerBoardFromOindex(int) [18]
[22] 0.1 2.61 0.27 32499663+923818 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.09 0.00 296737/296762 sitmo::prng_engine::encrypt_counter() [64]
0.07 0.00 269928/410303 Grid::GridRedBlackCartesian::oIndex(std::vector<int, std::allocator<int> >&) [62]
0.06 0.00 403941/625601 Grid::GridRedBlackCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [63]
0.02 0.00 269418/1535850 Grid::GridBase::oIndex(std::vector<int, std::allocator<int> >&) [56]
0.02 0.00 231465/1497897 Grid::GridBase::iIndex(std::vector<int, std::allocator<int> >&) [55]
0.01 0.00 190280/190280 Grid::CartesianCommunicator::RankFromProcessorCoor(std::vector<int, std::allocator<int> >&) [93]
0.00 0.00 208792/3585944 Grid::CartesianCommunicator::ThisRank() [92]
923818 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
-----------------------------------------------
5536 frame_dummy [23]
0.03 0.00 13/778 Benchmark::Decomposition() [81]
0.04 0.00 17/778 Benchmark::Memory() [49]
0.06 0.00 24/778 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
0.55 0.02 228/778 Benchmark::DWF5(int, int) [17]
1.20 0.05 496/778 Benchmark::DWF(int, int, double&) [13]
[23] 0.1 1.89 0.07 778+5536 frame_dummy [23]
0.03 0.00 124573/410303 Grid::GridRedBlackCartesian::oIndex(std::vector<int, std::allocator<int> >&) [62]
0.03 0.00 197535/625601 Grid::GridRedBlackCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [63]
0.01 0.00 178511/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
5536 frame_dummy [23]
-----------------------------------------------
<spontaneous>
[24] 0.1 1.84 0.00 __multf3 [24]
-----------------------------------------------
<spontaneous>
[25] 0.1 1.74 0.00 Grid::QCD::FermionOperator<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooDeriv(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int) [25]
-----------------------------------------------
283544 std::enable_if<Grid::matchGridTensorIndex<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4>, 2>::value, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >::type Grid::QCD::operator*<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >(Grid::QCD::Gamma const&, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> const&) [26]
[26] 0.1 1.49 0.00 0+283544 std::enable_if<Grid::matchGridTensorIndex<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4>, 2>::value, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >::type Grid::QCD::operator*<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >(Grid::QCD::Gamma const&, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> const&) [26]
283544 std::enable_if<Grid::matchGridTensorIndex<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4>, 2>::value, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >::type Grid::QCD::operator*<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >(Grid::QCD::Gamma const&, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> const&) [26]
-----------------------------------------------
0.05 0.00 1/25 Benchmark::Memory() [49]
0.43 0.00 8/25 Benchmark::DWF5(int, int) [17]
0.86 0.00 16/25 Benchmark::DWF(int, int, double&) [13]
[27] 0.0 1.34 0.00 25 Grid::GridRNGbase::~GridRNGbase() [27]
-----------------------------------------------
0.00 1.16 8/8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
[28] 0.0 0.00 1.16 8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [28]
0.00 1.14 8/8 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
0.02 0.00 8/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.00 0.00 8/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
-----------------------------------------------
0.00 1.15 4/4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
[29] 0.0 0.00 1.15 4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [29]
0.07 1.07 4/4 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.01 0.00 4/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.00 0.00 4/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
-----------------------------------------------
0.07 1.07 4/4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [29]
[30] 0.0 0.07 1.07 4 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.65 0.15 844288/844288 void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&) [35]
0.13 0.08 422144/422144 void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&) [46]
0.03 0.00 422144/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.01 0.00 4/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.01 0.00 32/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.01 16/48 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
0.00 0.00 4/20 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::Lattice(Grid::GridBase*) [102]
0.00 0.00 52/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
0.00 0.00 4/36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
-----------------------------------------------
0.00 1.14 8/8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [28]
[31] 0.0 0.00 1.14 8 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
0.98 0.12 32/32 void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int) [32]
0.02 0.00 64/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.01 32/48 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
0.00 0.00 16/20 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::Lattice(Grid::GridBase*) [102]
0.00 0.00 112/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
0.00 0.00 72/72 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::deallocate(Grid::Grid_simd<unsigned int, __Uint32x4_t>*, unsigned long) [170]
0.00 0.00 72/72 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::allocate(unsigned long, void const*) [171]
-----------------------------------------------
0.98 0.12 32/32 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
[32] 0.0 0.98 0.12 32 void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int) [32]
0.11 0.00 3377152/3377152 Grid::CartesianCommunicator::ProcessorCoorFromRank(int, std::vector<int, std::allocator<int> >&) [61]
0.01 0.00 3377152/3585944 Grid::CartesianCommunicator::ThisRank() [92]
0.00 0.00 32/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
-----------------------------------------------
0.00 0.35 8/24 Benchmark::DWF5(int, int) [17]
0.00 0.70 16/24 Benchmark::DWF(int, int, double&) [13]
[33] 0.0 0.00 1.05 24 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
0.78 0.00 24/25 std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::discrete_distribution<int>*, std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > > >, unsigned long, std::discrete_distribution<int> const&) [34]
0.12 0.00 24/25 std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::normal_distribution<double>*, std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > > >, unsigned long, std::normal_distribution<double> const&) [58]
0.06 0.00 24/778 frame_dummy [23]
0.05 0.00 24/25 std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_int_distribution<unsigned int>*, std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > > >, unsigned long, std::uniform_int_distribution<unsigned int> const&) [70]
0.04 0.00 24/25 std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_real_distribution<double>*, std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > > >, unsigned long, std::uniform_real_distribution<double> const&) [79]
-----------------------------------------------
0.03 0.00 1/25 Benchmark::Memory() [49]
0.78 0.00 24/25 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
[34] 0.0 0.81 0.00 25 std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::discrete_distribution<int>*, std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > > >, unsigned long, std::discrete_distribution<int> const&) [34]
-----------------------------------------------
0.65 0.15 844288/844288 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
[35] 0.0 0.65 0.15 844288 void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&) [35]
0.08 0.00 844288/1497897 Grid::GridBase::iIndex(std::vector<int, std::allocator<int> >&) [55]
0.07 0.00 844288/1535850 Grid::GridBase::oIndex(std::vector<int, std::allocator<int> >&) [56]
0.00 0.00 844288/1266432 Grid::GridCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [130]
-----------------------------------------------
0.22 0.00 12/36 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
0.43 0.00 24/36 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
[36] 0.0 0.65 0.00 36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::BuildSurfaceList(int, int) [36]
-----------------------------------------------
<spontaneous>
[37] 0.0 0.65 0.00 __trunctfdf2 [37]
-----------------------------------------------
18720 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.04 0.00 120/1560 Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Reporti(int) [77]
0.11 0.01 304/1560 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.18 0.01 528/1560 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Report() [47]
0.21 0.01 608/1560 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
[38] 0.0 0.54 0.03 1560+18720 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.02 0.00 167755/167755 Grid::Grid_simd<unsigned int, __Uint32x4_t> Grid::Comparison<Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t>, 0>(Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::Grid_simd<unsigned int, __Uint32x4_t>::scalar_type const&) [85]
0.00 0.00 15802/410303 Grid::GridRedBlackCartesian::oIndex(std::vector<int, std::allocator<int> >&) [62]
0.00 0.00 24125/625601 Grid::GridRedBlackCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [63]
0.00 0.00 25800/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
18720 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
-----------------------------------------------
<spontaneous>
[39] 0.0 0.48 0.00 __extenddftf2 [39]
-----------------------------------------------
0.31 0.15 4/4 Benchmark::DWF5(int, int) [17]
[40] 0.0 0.31 0.15 4 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.14 0.01 1688628/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 4/8 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*) [114]
0.00 0.00 4/52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
0.00 0.00 12/620 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&) [151]
0.00 0.00 4/191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
0.00 0.00 4/23 Grid::CartesianCommunicator::~CartesianCommunicator() [193]
-----------------------------------------------
0.28 0.16 8/8 Benchmark::DWF(int, int, double&) [13]
[41] 0.0 0.28 0.16 8 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
0.14 0.01 1794216/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 8/16 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [113]
0.00 0.00 8/52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
0.00 0.00 24/620 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&) [151]
0.00 0.00 8/191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
0.00 0.00 8/23 Grid::CartesianCommunicator::~CartesianCommunicator() [193]
-----------------------------------------------
0.14 0.00 64/180 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int) [54]
0.25 0.00 116/180 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*) [45]
[42] 0.0 0.39 0.00 180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::allocate(unsigned long, void const*) [42]
0.00 0.00 180/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.38 32/32 Benchmark::DWF(int, int, double&) [13]
[43] 0.0 0.00 0.38 32 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
0.21 0.01 608/1560 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.00 0.13 96/144 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Report() [47]
0.00 0.03 96/144 Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Reporti(int) [77]
0.00 0.00 32/48 std::vector<int, std::allocator<int> >::vector(std::vector<int, std::allocator<int> > const&) [176]
0.00 0.00 32/672 Grid::CartesianCommunicator::GlobalSum(double&) [150]
-----------------------------------------------
<spontaneous>
[44] 0.0 0.27 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::TwoIndexRep<3, (Grid::QCD::TwoIndexSymmetry)1>, Grid::QCD::CoeffReal> >::ContractConservedCurrentSiteBwd(Grid::iScalar<Grid::iMatrix<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6>, 4> > const&, Grid::iScalar<Grid::iMatrix<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6>, 4> > const&, Grid::iScalar<Grid::iMatrix<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6>, 4> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 6> >, 8> >&, unsigned int, unsigned int, bool) [44]
-----------------------------------------------
0.00 0.01 4/116 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
0.00 0.02 8/116 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
0.00 0.07 32/116 Benchmark::DWF5(int, int) [17]
0.00 0.16 72/116 Benchmark::DWF(int, int, double&) [13]
[45] 0.0 0.00 0.25 116 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*) [45]
0.25 0.00 116/180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::allocate(unsigned long, void const*) [42]
-----------------------------------------------
0.13 0.08 422144/422144 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
[46] 0.0 0.13 0.08 422144 void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&) [46]
0.04 0.00 422144/1497897 Grid::GridBase::iIndex(std::vector<int, std::allocator<int> >&) [55]
0.04 0.00 422144/1535850 Grid::GridBase::oIndex(std::vector<int, std::allocator<int> >&) [56]
0.00 0.00 422144/1266432 Grid::GridCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [130]
-----------------------------------------------
0.00 0.06 48/144 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.00 0.13 96/144 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
[47] 0.0 0.00 0.19 144 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Report() [47]
0.18 0.01 528/1560 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.00 0.00 624/672 Grid::CartesianCommunicator::GlobalSum(double&) [150]
-----------------------------------------------
0.00 0.19 16/16 Benchmark::DWF5(int, int) [17]
[48] 0.0 0.00 0.19 16 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.11 0.01 304/1560 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.00 0.06 48/144 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Report() [47]
0.00 0.01 48/144 Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Reporti(int) [77]
0.00 0.00 16/48 std::vector<int, std::allocator<int> >::vector(std::vector<int, std::allocator<int> > const&) [176]
0.00 0.00 16/672 Grid::CartesianCommunicator::GlobalSum(double&) [150]
-----------------------------------------------
<spontaneous>
[49] 0.0 0.01 0.18 Benchmark::Memory() [49]
0.05 0.00 1/25 Grid::GridRNGbase::~GridRNGbase() [27]
0.04 0.00 17/778 frame_dummy [23]
0.04 0.00 33/33 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::allocate(unsigned long, void const*) [78]
0.03 0.00 1/25 std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::discrete_distribution<int>*, std::vector<std::discrete_distribution<int>, std::allocator<std::discrete_distribution<int> > > >, unsigned long, std::discrete_distribution<int> const&) [34]
0.01 0.00 1/25 std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::normal_distribution<double>*, std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > > >, unsigned long, std::normal_distribution<double> const&) [58]
0.00 0.00 1/25 std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_int_distribution<unsigned int>*, std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > > >, unsigned long, std::uniform_int_distribution<unsigned int> const&) [70]
0.00 0.00 1/25 std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_real_distribution<double>*, std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > > >, unsigned long, std::uniform_real_distribution<double> const&) [79]
0.00 0.00 132/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 11/39 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
0.00 0.00 1/296762 sitmo::prng_engine::encrypt_counter() [64]
0.00 0.00 33/33 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::deallocate(Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8>*, unsigned long) [185]
0.00 0.00 12/36 Grid::CartesianCommunicator::BroadcastWorld(int, void*, int) [183]
0.00 0.00 11/23 Grid::CartesianCommunicator::~CartesianCommunicator() [193]
0.00 0.00 1/42 Grid::GridDefaultMpi() [177]
0.00 0.00 1/31 Grid::GridDefaultSimd(int, int) [189]
0.00 0.00 1/25 std::seed_seq::seed_seq<__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > > >(__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >) [190]
-----------------------------------------------
1108128 Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [50]
0.18 0.00 79152/79152 Grid::LebesgueOrder::IterateO(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [51]
[50] 0.0 0.18 0.00 79152+1108128 Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [50]
0.00 0.00 368/620 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&) [151]
1108128 Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [50]
-----------------------------------------------
0.00 0.18 1040/1040 Grid::LebesgueOrder::CartesianBlocking() [52]
[51] 0.0 0.00 0.18 1040 Grid::LebesgueOrder::IterateO(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [51]
0.18 0.00 79152/79152 Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [50]
-----------------------------------------------
0.00 0.18 24/24 Grid::LebesgueOrder::LebesgueOrder(Grid::GridBase*) [53]
[52] 0.0 0.00 0.18 24 Grid::LebesgueOrder::CartesianBlocking() [52]
0.00 0.18 1040/1040 Grid::LebesgueOrder::IterateO(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [51]
-----------------------------------------------
0.00 0.06 8/24 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
0.00 0.12 16/24 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
[53] 0.0 0.00 0.18 24 Grid::LebesgueOrder::LebesgueOrder(Grid::GridBase*) [53]
0.00 0.18 24/24 Grid::LebesgueOrder::CartesianBlocking() [52]
-----------------------------------------------
0.00 0.15 64/64 Benchmark::DWF(int, int, double&) [13]
[54] 0.0 0.00 0.15 64 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int) [54]
0.14 0.00 64/180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::allocate(unsigned long, void const*) [42]
0.00 0.01 64/64 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [90]
0.00 0.00 128/3206 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [144]
0.00 0.00 64/224 Grid::GridCartesian::CheckerBoardDestination(int, int, int) [155]
-----------------------------------------------
0.02 0.00 231465/1497897 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.04 0.00 422144/1497897 void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&) [46]
0.08 0.00 844288/1497897 void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&) [35]
[55] 0.0 0.14 0.00 1497897 Grid::GridBase::iIndex(std::vector<int, std::allocator<int> >&) [55]
-----------------------------------------------
0.02 0.00 269418/1535850 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.04 0.00 422144/1535850 void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&) [46]
0.07 0.00 844288/1535850 void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&) [35]
[56] 0.0 0.14 0.00 1535850 Grid::GridBase::oIndex(std::vector<int, std::allocator<int> >&) [56]
-----------------------------------------------
0.00 0.00 4/424 Benchmark::DWF5(int, int) [17]
0.01 0.00 20/424 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::Lattice(Grid::GridBase*) [102]
0.01 0.00 32/424 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.01 0.00 48/424 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
0.02 0.00 64/424 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
0.03 0.00 112/424 Benchmark::DWF(int, int, double&) [13]
0.04 0.00 144/424 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
[57] 0.0 0.13 0.00 424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.00 424/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.01 0.00 1/25 Benchmark::Memory() [49]
0.12 0.00 24/25 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
[58] 0.0 0.13 0.00 25 std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::normal_distribution<double>*, std::vector<std::normal_distribution<double>, std::allocator<std::normal_distribution<double> > > >, unsigned long, std::normal_distribution<double> const&) [58]
-----------------------------------------------
0.12 0.00 36/36 std::vector<Grid::StencilEntry, Grid::alignedAllocator<Grid::StencilEntry> >::_M_default_append(unsigned long) [60]
[59] 0.0 0.12 0.00 36 Grid::alignedAllocator<Grid::StencilEntry>::allocate(unsigned long, void const*) [59]
0.00 0.00 36/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.12 36/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[60] 0.0 0.00 0.12 36 std::vector<Grid::StencilEntry, Grid::alignedAllocator<Grid::StencilEntry> >::_M_default_append(unsigned long) [60]
0.12 0.00 36/36 Grid::alignedAllocator<Grid::StencilEntry>::allocate(unsigned long, void const*) [59]
-----------------------------------------------
0.11 0.00 3377152/3377152 void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int) [32]
[61] 0.0 0.11 0.00 3377152 Grid::CartesianCommunicator::ProcessorCoorFromRank(int, std::vector<int, std::allocator<int> >&) [61]
-----------------------------------------------
0.00 0.00 15802/410303 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.03 0.00 124573/410303 frame_dummy [23]
0.07 0.00 269928/410303 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
[62] 0.0 0.10 0.00 410303 Grid::GridRedBlackCartesian::oIndex(std::vector<int, std::allocator<int> >&) [62]
-----------------------------------------------
0.00 0.00 24125/625601 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
0.03 0.00 197535/625601 frame_dummy [23]
0.06 0.00 403941/625601 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
[63] 0.0 0.09 0.00 625601 Grid::GridRedBlackCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [63]
-----------------------------------------------
0.00 0.00 1/296762 Benchmark::Memory() [49]
0.00 0.00 24/296762 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [104]
0.09 0.00 296737/296762 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
[64] 0.0 0.09 0.00 296762 sitmo::prng_engine::encrypt_counter() [64]
-----------------------------------------------
397821 Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const Grid::predicatedWhere<Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const>(Grid::Grid_simd<unsigned int, __Uint32x4_t> const& const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&) [65]
[65] 0.0 0.09 0.00 0+397821 Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const Grid::predicatedWhere<Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const>(Grid::Grid_simd<unsigned int, __Uint32x4_t> const& const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&) [65]
397821 Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const Grid::predicatedWhere<Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const>(Grid::Grid_simd<unsigned int, __Uint32x4_t> const& const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&, Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > const&) [65]
-----------------------------------------------
0.01 0.00 4/36 Benchmark::DWF5(int, int) [17]
0.01 0.00 4/36 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.01 0.00 4/36 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [29]
0.02 0.00 8/36 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [28]
0.04 0.00 16/36 Benchmark::DWF(int, int, double&) [13]
[66] 0.0 0.09 0.00 36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.00 0.00 36/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.02 0.02 8/16 Benchmark::DWF(int, int, double&) [13]
0.02 0.02 8/16 Benchmark::DWF5(int, int) [17]
[67] 0.0 0.04 0.05 16 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*) [67]
0.04 0.00 527888/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 16/52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
0.00 0.00 16/191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
-----------------------------------------------
<spontaneous>
[68] 0.0 0.06 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteDag(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [68]
-----------------------------------------------
<spontaneous>
[69] 0.0 0.06 0.00 __divdc3 [69]
-----------------------------------------------
0.00 0.00 1/25 Benchmark::Memory() [49]
0.05 0.00 24/25 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
[70] 0.0 0.05 0.00 25 std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_int_distribution<unsigned int>*, std::vector<std::uniform_int_distribution<unsigned int>, std::allocator<std::uniform_int_distribution<unsigned int> > > >, unsigned long, std::uniform_int_distribution<unsigned int> const&) [70]
-----------------------------------------------
0.05 0.00 24/24 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::Lattice(Grid::GridBase*) [72]
[71] 0.0 0.05 0.00 24 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::allocate(unsigned long, void const*) [71]
0.00 0.00 24/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.05 24/24 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
[72] 0.0 0.00 0.05 24 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::Lattice(Grid::GridBase*) [72]
0.05 0.00 24/24 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::allocate(unsigned long, void const*) [71]
-----------------------------------------------
<spontaneous>
[73] 0.0 0.05 0.00 Grid::Gather_plane_table_compute(Grid::GridBase*, int, int, int, int, std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >&) [73]
-----------------------------------------------
<spontaneous>
[74] 0.0 0.05 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplex> >::HandDhopSiteDagExt(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&) [74]
-----------------------------------------------
<spontaneous>
[75] 0.0 0.05 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::TwoIndexRep<3, (Grid::QCD::TwoIndexSymmetry)-1>, Grid::QCD::CoeffReal> >::DhopDir(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int, int) [75]
-----------------------------------------------
0.00 0.01 16/48 Benchmark::DWF5(int, int) [17]
0.00 0.03 32/48 Benchmark::DWF(int, int, double&) [13]
[76] 0.0 0.00 0.04 48 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
0.04 0.00 144/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.00 432/432 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >*, unsigned long) [152]
0.00 0.00 432/432 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::allocate(unsigned long, void const*) [153]
0.00 0.00 144/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
-----------------------------------------------
0.00 0.01 48/144 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.00 0.03 96/144 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
[77] 0.0 0.00 0.04 144 Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Reporti(int) [77]
0.04 0.00 120/1560 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
-----------------------------------------------
0.04 0.00 33/33 Benchmark::Memory() [49]
[78] 0.0 0.04 0.00 33 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::allocate(unsigned long, void const*) [78]
0.00 0.00 33/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.00 1/25 Benchmark::Memory() [49]
0.04 0.00 24/25 Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [33]
[79] 0.0 0.04 0.00 25 std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::uniform_real_distribution<double>*, std::vector<std::uniform_real_distribution<double>, std::allocator<std::uniform_real_distribution<double> > > >, unsigned long, std::uniform_real_distribution<double> const&) [79]
-----------------------------------------------
<spontaneous>
[80] 0.0 0.04 0.00 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::HandDhopSiteDag(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&) [80]
-----------------------------------------------
<spontaneous>
[81] 0.0 0.00 0.03 Benchmark::Decomposition() [81]
0.03 0.00 13/778 frame_dummy [23]
0.00 0.00 7/55 Grid::GridCmdVectorIntToString[abi:cxx11](std::vector<int, std::allocator<int> > const&) [173]
0.00 0.00 6/31 Grid::GridDefaultSimd(int, int) [189]
0.00 0.00 1/42 Grid::GridDefaultMpi() [177]
-----------------------------------------------
<spontaneous>
[82] 0.0 0.03 0.00 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopDir(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, int, int, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int, int) [82]
-----------------------------------------------
0.00 0.02 38464/38464 Benchmark::DWF(int, int, double&) [13]
[83] 0.0 0.00 0.02 38464 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [83]
0.01 0.01 19232/19232 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [89]
0.00 0.01 19232/19232 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [100]
-----------------------------------------------
0.00 0.01 16/48 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.00 0.01 32/48 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
[84] 0.0 0.00 0.02 48 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
0.01 0.00 48/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.01 48/48 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
0.00 0.00 96/3206 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [144]
0.00 0.00 48/224 Grid::GridCartesian::CheckerBoardDestination(int, int, int) [155]
-----------------------------------------------
0.02 0.00 167755/167755 Grid::QCD::CayleyFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::M5D(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&, std::vector<std::complex<double>, std::allocator<std::complex<double> > >&) [38]
[85] 0.0 0.02 0.00 167755 Grid::Grid_simd<unsigned int, __Uint32x4_t> Grid::Comparison<Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t>, 0>(Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::Grid_simd<unsigned int, __Uint32x4_t>::scalar_type const&) [85]
-----------------------------------------------
0.00 0.00 8800/112128 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [109]
0.00 0.00 19232/112128 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [89]
0.01 0.00 28032/112128 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
0.01 0.00 56064/112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>) [94]
[86] 0.0 0.02 0.00 112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMerge<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge> >&, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress> >&) [86]
-----------------------------------------------
0.02 0.00 12/12 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::Lattice(Grid::GridBase*) [88]
[87] 0.0 0.02 0.00 12 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::allocate(unsigned long, void const*) [87]
0.00 0.00 12/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.02 12/12 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
[88] 0.0 0.00 0.02 12 Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::Lattice(Grid::GridBase*) [88]
0.02 0.00 12/12 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::allocate(unsigned long, void const*) [87]
-----------------------------------------------
0.01 0.01 19232/19232 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [83]
[89] 0.0 0.01 0.01 19232 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [89]
0.00 0.00 19232/56064 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>) [94]
0.00 0.00 19232/112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMerge<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge> >&, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress> >&) [86]
0.00 0.00 19232/56064 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Prepare() [140]
0.00 0.00 19232/28032 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
-----------------------------------------------
0.00 0.01 64/64 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int) [54]
[90] 0.0 0.00 0.01 64 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [90]
0.01 0.00 16/16 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int, int) [91]
0.00 0.00 824/824 void Grid::Copy_plane<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int) [110]
0.00 0.00 840/3206 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [144]
0.00 0.00 64/224 Grid::GridCartesian::CheckerBoardDestination(int, int, int) [155]
-----------------------------------------------
0.01 0.00 16/16 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [90]
[91] 0.0 0.01 0.00 16 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int, int) [91]
0.00 0.00 7/54 std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >::_M_default_append(unsigned long) [96]
0.00 0.00 16/2790 Grid::GridCartesian::CheckerBoarded(int) [146]
-----------------------------------------------
0.00 0.00 208792/3585944 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.01 0.00 3377152/3585944 void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int) [32]
[92] 0.0 0.01 0.00 3585944 Grid::CartesianCommunicator::ThisRank() [92]
-----------------------------------------------
0.01 0.00 190280/190280 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
[93] 0.0 0.01 0.00 190280 Grid::CartesianCommunicator::RankFromProcessorCoor(std::vector<int, std::allocator<int> >&) [93]
-----------------------------------------------
0.00 0.00 8800/56064 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [109]
0.00 0.00 19232/56064 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [89]
0.00 0.01 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[94] 0.0 0.00 0.01 56064 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>) [94]
0.01 0.00 56064/112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMerge<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge> >&, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress> >&) [86]
0.00 0.00 56064/112128 Grid::CartesianCommunicator::StencilBarrier() [131]
-----------------------------------------------
0.00 0.00 8800/28032 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [108]
0.00 0.01 19232/28032 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [100]
[95] 0.0 0.00 0.01 28032 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
0.01 0.00 28032/112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMerge<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge> >&, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress> >&) [86]
0.00 0.01 28032/56064 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>) [94]
0.00 0.00 28032/56064 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Prepare() [140]
0.00 0.00 28032/112128 Grid::CartesianCommunicator::StencilBarrier() [131]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void>&, int, int&) [134]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void>&, int, int&) [136]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void>&, int, int&) [138]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void>&, int, int&) [132]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void>&, int, int&) [135]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void>&, int, int&) [137]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void>&, int, int&) [139]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void>&, int, int&) [133]
-----------------------------------------------
0.00 0.00 7/54 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int, int) [91]
0.00 0.00 10/54 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int, int) [111]
0.00 0.00 15/54 void Grid::Copy_plane<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int) [110]
0.00 0.00 22/54 void Grid::Copy_plane<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int) [107]
[96] 0.0 0.01 0.00 54 std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >::_M_default_append(unsigned long) [96]
-----------------------------------------------
<spontaneous>
[97] 0.0 0.01 0.00 Grid::CartesianCommunicator::ShiftedRanks(int, int, int&, int&) [97]
-----------------------------------------------
<spontaneous>
[98] 0.0 0.01 0.00 Grid::CartesianCommunicator::BossRank() [98]
-----------------------------------------------
<spontaneous>
[99] 0.0 0.01 0.00 std::vector<std::vector<ompi_request_t*, std::allocator<ompi_request_t*> >, std::allocator<std::vector<ompi_request_t*, std::allocator<ompi_request_t*> > > >::~vector() [99]
-----------------------------------------------
0.00 0.01 19232/19232 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [83]
[100] 0.0 0.00 0.01 19232 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [100]
0.00 0.01 19232/28032 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
-----------------------------------------------
0.00 0.01 17600/17600 Benchmark::DWF5(int, int) [17]
[101] 0.0 0.00 0.01 17600 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [101]
0.00 0.00 8800/8800 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [108]
0.00 0.00 8800/8800 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [109]
-----------------------------------------------
0.00 0.00 4/20 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.00 0.00 16/20 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
[102] 0.0 0.00 0.01 20 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::Lattice(Grid::GridBase*) [102]
0.01 0.00 20/424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
-----------------------------------------------
0.00 0.01 48/48 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
[103] 0.0 0.00 0.01 48 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
0.00 0.00 618/618 void Grid::Copy_plane<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int) [107]
0.00 0.00 12/12 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int, int) [111]
0.00 0.00 630/3206 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [144]
0.00 0.00 48/224 Grid::GridCartesian::CheckerBoardDestination(int, int, int) [155]
0.00 0.00 48/48 Grid::GridBase::PermuteType(int) [175]
-----------------------------------------------
0.00 0.00 8/24 Benchmark::DWF5(int, int) [17]
0.00 0.00 16/24 Benchmark::DWF(int, int, double&) [13]
[104] 0.0 0.01 0.00 24 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [104]
0.00 0.00 24/296762 sitmo::prng_engine::encrypt_counter() [64]
0.00 0.00 24/25 std::seed_seq::seed_seq<__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > > >(__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >) [190]
0.00 0.00 24/36 Grid::CartesianCommunicator::BroadcastWorld(int, void*, int) [183]
-----------------------------------------------
<spontaneous>
[105] 0.0 0.01 0.00 Grid::GridRedBlackCartesian::~GridRedBlackCartesian() [105]
-----------------------------------------------
<spontaneous>
[106] 0.0 0.01 0.00 std::ctype<char>::do_widen(char) const [106]
-----------------------------------------------
0.00 0.00 618/618 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
[107] 0.0 0.00 0.00 618 void Grid::Copy_plane<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int) [107]
0.00 0.00 22/54 std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >::_M_default_append(unsigned long) [96]
0.00 0.00 618/2790 Grid::GridCartesian::CheckerBoarded(int) [146]
-----------------------------------------------
0.00 0.00 8800/8800 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [101]
[108] 0.0 0.00 0.00 8800 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [108]
0.00 0.00 8800/28032 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
-----------------------------------------------
0.00 0.00 8800/8800 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [101]
[109] 0.0 0.00 0.00 8800 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [109]
0.00 0.00 8800/56064 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>) [94]
0.00 0.00 8800/112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMerge<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Merge> >&, std::vector<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress, std::allocator<Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Decompress> >&) [86]
0.00 0.00 8800/56064 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Prepare() [140]
0.00 0.00 8800/28032 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
-----------------------------------------------
0.00 0.00 824/824 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [90]
[110] 0.0 0.00 0.00 824 void Grid::Copy_plane<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int) [110]
0.00 0.00 15/54 std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >::_M_default_append(unsigned long) [96]
0.00 0.00 824/2790 Grid::GridCartesian::CheckerBoarded(int) [146]
-----------------------------------------------
0.00 0.00 12/12 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
[111] 0.0 0.00 0.00 12 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int, int) [111]
0.00 0.00 10/54 std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >::_M_default_append(unsigned long) [96]
0.00 0.00 12/2790 Grid::GridCartesian::CheckerBoarded(int) [146]
-----------------------------------------------
0.00 0.00 8/24 Benchmark::DWF5(int, int) [17]
0.00 0.00 16/24 Benchmark::DWF(int, int, double&) [13]
[112] 0.0 0.00 0.00 24 Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [112]
0.00 0.00 288/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 24/39 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
-----------------------------------------------
0.00 0.00 8/16 Benchmark::DWF(int, int, double&) [13]
0.00 0.00 8/16 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
[113] 0.0 0.00 0.00 16 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [113]
0.00 0.00 192/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 16/52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
0.00 0.00 144/620 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&) [151]
-----------------------------------------------
0.00 0.00 4/8 Benchmark::DWF5(int, int) [17]
0.00 0.00 4/8 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
[114] 0.0 0.00 0.00 8 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*) [114]
0.00 0.00 96/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 8/52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
0.00 0.00 72/620 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&) [151]
-----------------------------------------------
0.00 0.00 4/52 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.00 0.00 8/52 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*) [114]
0.00 0.00 8/52 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
0.00 0.00 16/52 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*) [67]
0.00 0.00 16/52 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [113]
[115] 0.0 0.00 0.00 52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
0.00 0.00 52/91 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [116]
0.00 0.00 52/52 Grid::CartesianCommunicator::InitFromMPICommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*) [119]
-----------------------------------------------
0.00 0.00 39/91 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
0.00 0.00 52/91 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
[116] 0.0 0.00 0.00 91 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [116]
0.00 0.00 91/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 91/127 Grid::SharedMemory::ShmBufferFreeAll() [160]
0.00 0.00 91/92 std::vector<void*, std::allocator<void*> >::_M_default_append(unsigned long) [164]
-----------------------------------------------
0.00 0.00 4/39 Grid::QCD::SpaceTimeGrid::makeFourDimDWFGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [118]
0.00 0.00 11/39 Benchmark::Memory() [49]
0.00 0.00 24/39 Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [112]
[117] 0.0 0.00 0.00 39 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
0.00 0.00 39/91 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [116]
0.00 0.00 39/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 39/39 Grid::GlobalSharedMemory::OptimalCommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*&) [179]
0.00 0.00 39/191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
-----------------------------------------------
0.00 0.00 4/4 Benchmark::DWF5(int, int) [17]
[118] 0.0 0.00 0.00 4 Grid::QCD::SpaceTimeGrid::makeFourDimDWFGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [118]
0.00 0.00 48/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 4/39 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
-----------------------------------------------
0.00 0.00 52/52 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [115]
[119] 0.0 0.00 0.00 52 Grid::CartesianCommunicator::InitFromMPICommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*) [119]
0.00 0.00 52/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 52/191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
-----------------------------------------------
0.00 0.00 1/1 Grid::CartesianCommunicator::Init(int*, char***) [121]
[120] 0.0 0.00 0.00 1 Grid::GlobalSharedMemory::Init(ompi_communicator_t*) [120]
0.00 0.00 2/32499663 std::vector<int, std::allocator<int> >::_M_default_append(unsigned long) [22]
0.00 0.00 1/1 void std::__introsort_loop<__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, long, __gnu_cxx::__ops::_Iter_comp_iter<std::less<int> > >(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, long, __gnu_cxx::__ops::_Iter_comp_iter<std::less<int> >) [240]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[121] 0.0 0.00 0.00 1 Grid::CartesianCommunicator::Init(int*, char***) [121]
0.00 0.00 1/1 Grid::GlobalSharedMemory::Init(ompi_communicator_t*) [120]
0.00 0.00 1/2 Grid::Grid_quiesce_nodes() [215]
0.00 0.00 1/1 Grid::GlobalSharedMemory::SharedMemoryAllocate(unsigned long, int) [235]
-----------------------------------------------
<spontaneous>
[122] 0.0 0.00 0.00 Grid::Grid_init(int*, char***) [122]
0.00 0.00 1/1 Grid::CartesianCommunicator::Init(int*, char***) [121]
0.00 0.00 42/42 Grid::Grid_sa_signal_handler(int, siginfo_t*, void*) [178]
0.00 0.00 18/22 char** std::__find_if<char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const> >(char**, char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const>, std::random_access_iterator_tag) [194]
0.00 0.00 1/2 Grid::Grid_quiesce_nodes() [215]
0.00 0.00 1/1 Grid::CartesianCommunicator::RankWorld() [236]
0.00 0.00 1/1 Grid::printHash() [237]
0.00 0.00 1/1 Grid::GridCmdOptionCSL(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&) [232]
0.00 0.00 1/1 Grid::GridLogConfigure(std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&) [233]
0.00 0.00 1/1 Grid::GridParseLayout(char**, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [231]
0.00 0.00 1/1 Grid::operator<<(std::ostream&, Grid::Logger&) [clone .constprop.140] [238]
0.00 0.00 1/1 Grid::GridLogTimestamp(int) [234]
-----------------------------------------------
0.00 0.00 422144/1266432 void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&) [46]
0.00 0.00 844288/1266432 void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&) [35]
[130] 0.0 0.00 0.00 1266432 Grid::GridCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [130]
-----------------------------------------------
0.00 0.00 28032/112128 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/112128 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
0.00 0.00 56064/112128 void Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CommsMergeSHM<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>) [94]
[131] 0.0 0.00 0.00 112128 Grid::CartesianCommunicator::StencilBarrier() [131]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[132] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void>&, int, int&) [132]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[133] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void>&, int, int&) [133]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[134] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void>&, int, int&) [134]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[135] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void>&, int, int&) [135]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[136] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void>&, int, int&) [136]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[137] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void>&, int, int&) [137]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[138] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void>&, int, int&) [138]
-----------------------------------------------
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[139] 0.0 0.00 0.00 56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void>&, int, int&) [139]
-----------------------------------------------
0.00 0.00 8800/56064 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [109]
0.00 0.00 19232/56064 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [89]
0.00 0.00 28032/56064 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloExchangeOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [95]
[140] 0.0 0.00 0.00 56064 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Prepare() [140]
-----------------------------------------------
0.00 0.00 38432/38432 Benchmark::DWF(int, int, double&) [13]
[141] 0.0 0.00 0.00 38432 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [141]
-----------------------------------------------
0.00 0.00 8800/28032 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [109]
0.00 0.00 19232/28032 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [89]
[142] 0.0 0.00 0.00 28032 void Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherOpt<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonProjector, void>&) [142]
0.00 0.00 28032/112128 Grid::CartesianCommunicator::StencilBarrier() [131]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXmProjector, void>&, int, int&) [134]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYmProjector, void>&, int, int&) [136]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZmProjector, void>&, int, int&) [138]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTmProjector, void>&, int, int&) [132]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonXpProjector, void>&, int, int&) [135]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonYpProjector, void>&, int, int&) [137]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonZpProjector, void>&, int, int&) [139]
0.00 0.00 28032/56064 int Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::HaloGatherDir<Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void> >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::QCD::WilsonCompressorTemplate<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::QCD::WilsonTpProjector, void>&, int, int&) [133]
-----------------------------------------------
0.00 0.00 17600/17600 Benchmark::DWF5(int, int) [17]
[143] 0.0 0.00 0.00 17600 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [143]
-----------------------------------------------
0.00 0.00 96/3206 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
0.00 0.00 128/3206 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int) [54]
0.00 0.00 630/3206 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
0.00 0.00 840/3206 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [90]
0.00 0.00 1512/3206 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[144] 0.0 0.00 0.00 3206 Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [144]
-----------------------------------------------
0.00 0.00 3024/3024 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[145] 0.0 0.00 0.00 3024 Grid::GridRedBlackCartesian::CheckerBoardShiftForCB(int, int, int, int) [145]
-----------------------------------------------
0.00 0.00 12/2790 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int, int) [111]
0.00 0.00 16/2790 void Grid::Copy_plane_permute<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int, int) [91]
0.00 0.00 618/2790 void Grid::Copy_plane<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int) [107]
0.00 0.00 824/2790 void Grid::Copy_plane<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int) [110]
0.00 0.00 1320/2790 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[146] 0.0 0.00 0.00 2790 Grid::GridCartesian::CheckerBoarded(int) [146]
-----------------------------------------------
0.00 0.00 2640/2640 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[147] 0.0 0.00 0.00 2640 Grid::GridRedBlackCartesian::CheckerBoarded(int) [147]
-----------------------------------------------
0.00 0.00 12/1325 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>*, unsigned long) [197]
0.00 0.00 24/1325 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8>*, unsigned long) [191]
0.00 0.00 33/1325 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::deallocate(Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8>*, unsigned long) [185]
0.00 0.00 36/1325 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
0.00 0.00 36/1325 Grid::alignedAllocator<Grid::StencilEntry>::deallocate(Grid::StencilEntry*, unsigned long) [181]
0.00 0.00 72/1325 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::deallocate(Grid::Grid_simd<unsigned int, __Uint32x4_t>*, unsigned long) [170]
0.00 0.00 76/1325 Grid::alignedAllocator<std::complex<double> >::deallocate(std::complex<double>*, unsigned long) [166]
0.00 0.00 180/1325 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long) [158]
0.00 0.00 424/1325 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
0.00 0.00 432/1325 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >*, unsigned long) [152]
[148] 0.0 0.00 0.00 1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 12/1325 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::allocate(unsigned long, void const*) [87]
0.00 0.00 24/1325 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::allocate(unsigned long, void const*) [71]
0.00 0.00 33/1325 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::allocate(unsigned long, void const*) [78]
0.00 0.00 36/1325 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::allocate(unsigned long, void const*) [66]
0.00 0.00 36/1325 Grid::alignedAllocator<Grid::StencilEntry>::allocate(unsigned long, void const*) [59]
0.00 0.00 72/1325 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::allocate(unsigned long, void const*) [171]
0.00 0.00 76/1325 Grid::alignedAllocator<std::complex<double> >::allocate(unsigned long, void const*) [167]
0.00 0.00 180/1325 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::allocate(unsigned long, void const*) [42]
0.00 0.00 424/1325 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::allocate(unsigned long, void const*) [57]
0.00 0.00 432/1325 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::allocate(unsigned long, void const*) [153]
[149] 0.0 0.00 0.00 1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.00 16/672 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.00 0.00 32/672 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
0.00 0.00 624/672 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::Report() [47]
[150] 0.0 0.00 0.00 672 Grid::CartesianCommunicator::GlobalSum(double&) [150]
-----------------------------------------------
0.00 0.00 12/620 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.00 0.00 24/620 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
0.00 0.00 72/620 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFGrid(int, Grid::GridCartesian const*) [114]
0.00 0.00 144/620 Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [113]
0.00 0.00 368/620 Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [50]
[151] 0.0 0.00 0.00 620 void std::vector<int, std::allocator<int> >::_M_realloc_insert<int const&>(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, int const&) [151]
-----------------------------------------------
0.00 0.00 432/432 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
[152] 0.0 0.00 0.00 432 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >*, unsigned long) [152]
0.00 0.00 432/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 432/432 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
[153] 0.0 0.00 0.00 432 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > >::allocate(unsigned long, void const*) [153]
0.00 0.00 432/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.00 4/424 Benchmark::DWF5(int, int) [17]
0.00 0.00 52/424 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.00 0.00 112/424 Benchmark::DWF(int, int, double&) [13]
0.00 0.00 112/424 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
0.00 0.00 144/424 void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double) [76]
[154] 0.0 0.00 0.00 424 Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::deallocate(Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > >*, unsigned long) [154]
0.00 0.00 424/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 48/224 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
0.00 0.00 48/224 Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int) [84]
0.00 0.00 64/224 void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [90]
0.00 0.00 64/224 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int) [54]
[155] 0.0 0.00 0.00 224 Grid::GridCartesian::CheckerBoardDestination(int, int, int) [155]
-----------------------------------------------
0.00 0.00 216/216 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[156] 0.0 0.00 0.00 216 Grid::SharedMemory::ShmBufferMalloc(unsigned long) [156]
-----------------------------------------------
0.00 0.00 4/191 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.00 0.00 8/191 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
0.00 0.00 16/191 Grid::QCD::SpaceTimeGrid::makeFourDimRedBlackGrid(Grid::GridCartesian const*) [67]
0.00 0.00 39/191 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
0.00 0.00 52/191 Grid::CartesianCommunicator::InitFromMPICommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*) [119]
0.00 0.00 72/191 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[157] 0.0 0.00 0.00 191 std::vector<int, std::allocator<int> >::operator=(std::vector<int, std::allocator<int> > const&) [157]
-----------------------------------------------
0.00 0.00 4/180 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D() [209]
0.00 0.00 8/180 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~CayleyFermion5D() [202]
0.00 0.00 32/180 Benchmark::DWF5(int, int) [17]
0.00 0.00 136/180 Benchmark::DWF(int, int, double&) [13]
[158] 0.0 0.00 0.00 180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long) [158]
0.00 0.00 180/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 48/144 Benchmark::DWF5(int, int) [17]
0.00 0.00 96/144 Benchmark::DWF(int, int, double&) [13]
[159] 0.0 0.00 0.00 144 Grid::CartesianCommunicator::Barrier() [159]
-----------------------------------------------
0.00 0.00 36/127 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
0.00 0.00 91/127 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [116]
[160] 0.0 0.00 0.00 127 Grid::SharedMemory::ShmBufferFreeAll() [160]
-----------------------------------------------
0.00 0.00 120/120 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsInternal(double, std::vector<double, std::allocator<double> >&, double, double) [201]
[161] 0.0 0.00 0.00 120 std::vector<double, std::allocator<double> >::_M_default_append(unsigned long) [161]
-----------------------------------------------
0.00 0.00 120/120 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsInternal(double, std::vector<double, std::allocator<double> >&, double, double) [201]
[162] 0.0 0.00 0.00 120 std::vector<double, std::allocator<double> >::resize(unsigned long) [162]
-----------------------------------------------
0.00 0.00 32/104 Benchmark::DWF5(int, int) [17]
0.00 0.00 72/104 Benchmark::DWF(int, int, double&) [13]
[163] 0.0 0.00 0.00 104 Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::~Lattice() [163]
-----------------------------------------------
0.00 0.00 1/92 Grid::GlobalSharedMemory::SharedMemoryAllocate(unsigned long, int) [235]
0.00 0.00 91/92 Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [116]
[164] 0.0 0.00 0.00 92 std::vector<void*, std::allocator<void*> >::_M_default_append(unsigned long) [164]
-----------------------------------------------
0.00 0.00 4/76 Benchmark::DWF5(int, int) [17]
0.00 0.00 72/76 Benchmark::DWF(int, int, double&) [13]
[165] 0.0 0.00 0.00 76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
0.00 0.00 76/76 Grid::CartesianCommunicator::GlobalSumVector(double*, int) [168]
0.00 0.00 76/76 Grid::CartesianCommunicator::GlobalSum(std::complex<double>&) [169]
0.00 0.00 76/76 Grid::alignedAllocator<std::complex<double> >::deallocate(std::complex<double>*, unsigned long) [166]
0.00 0.00 76/76 Grid::alignedAllocator<std::complex<double> >::allocate(unsigned long, void const*) [167]
-----------------------------------------------
0.00 0.00 76/76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
[166] 0.0 0.00 0.00 76 Grid::alignedAllocator<std::complex<double> >::deallocate(std::complex<double>*, unsigned long) [166]
0.00 0.00 76/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 76/76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
[167] 0.0 0.00 0.00 76 Grid::alignedAllocator<std::complex<double> >::allocate(unsigned long, void const*) [167]
0.00 0.00 76/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.00 76/76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
[168] 0.0 0.00 0.00 76 Grid::CartesianCommunicator::GlobalSumVector(double*, int) [168]
-----------------------------------------------
0.00 0.00 76/76 std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [165]
[169] 0.0 0.00 0.00 76 Grid::CartesianCommunicator::GlobalSum(std::complex<double>&) [169]
-----------------------------------------------
0.00 0.00 72/72 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
[170] 0.0 0.00 0.00 72 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::deallocate(Grid::Grid_simd<unsigned int, __Uint32x4_t>*, unsigned long) [170]
0.00 0.00 72/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 72/72 Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [31]
[171] 0.0 0.00 0.00 72 Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::allocate(unsigned long, void const*) [171]
0.00 0.00 72/1325 Grid::PointerCache::Lookup(unsigned long) [149]
-----------------------------------------------
0.00 0.00 72/72 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::CartesianStencil(Grid::GridBase*, int, int, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&) [14]
[172] 0.0 0.00 0.00 72 std::vector<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*, std::allocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> >*> >::_M_default_append(unsigned long) [172]
-----------------------------------------------
0.00 0.00 7/55 Benchmark::Decomposition() [81]
0.00 0.00 16/55 Benchmark::DWF5(int, int) [17]
0.00 0.00 32/55 Benchmark::DWF(int, int, double&) [13]
[173] 0.0 0.00 0.00 55 Grid::GridCmdVectorIntToString[abi:cxx11](std::vector<int, std::allocator<int> > const&) [173]
-----------------------------------------------
0.00 0.00 16/48 Benchmark::DWF5(int, int) [17]
0.00 0.00 32/48 Benchmark::DWF(int, int, double&) [13]
[174] 0.0 0.00 0.00 48 time_statistics::statistics(std::vector<double, std::allocator<double> >) [174]
-----------------------------------------------
0.00 0.00 48/48 void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [103]
[175] 0.0 0.00 0.00 48 Grid::GridBase::PermuteType(int) [175]
-----------------------------------------------
0.00 0.00 16/48 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [48]
0.00 0.00 32/48 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report() [43]
[176] 0.0 0.00 0.00 48 std::vector<int, std::allocator<int> >::vector(std::vector<int, std::allocator<int> > const&) [176]
-----------------------------------------------
0.00 0.00 1/42 Benchmark::Decomposition() [81]
0.00 0.00 1/42 Benchmark::Memory() [49]
0.00 0.00 16/42 Benchmark::DWF5(int, int) [17]
0.00 0.00 24/42 Benchmark::DWF(int, int, double&) [13]
[177] 0.0 0.00 0.00 42 Grid::GridDefaultMpi() [177]
-----------------------------------------------
0.00 0.00 42/42 Grid::Grid_init(int*, char***) [122]
[178] 0.0 0.00 0.00 42 Grid::Grid_sa_signal_handler(int, siginfo_t*, void*) [178]
-----------------------------------------------
0.00 0.00 39/39 Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [117]
[179] 0.0 0.00 0.00 39 Grid::GlobalSharedMemory::OptimalCommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*&) [179]
-----------------------------------------------
0.00 0.00 12/36 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D() [209]
0.00 0.00 24/36 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~CayleyFermion5D() [202]
[180] 0.0 0.00 0.00 36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::~CartesianStencil() [180]
0.00 0.00 36/36 Grid::alignedAllocator<Grid::StencilEntry>::deallocate(Grid::StencilEntry*, unsigned long) [181]
-----------------------------------------------
0.00 0.00 36/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::~CartesianStencil() [180]
[181] 0.0 0.00 0.00 36 Grid::alignedAllocator<Grid::StencilEntry>::deallocate(Grid::StencilEntry*, unsigned long) [181]
0.00 0.00 36/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 4/36 Benchmark::DWF5(int, int) [17]
0.00 0.00 4/36 Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [30]
0.00 0.00 4/36 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [29]
0.00 0.00 8/36 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [28]
0.00 0.00 16/36 Benchmark::DWF(int, int, double&) [13]
[182] 0.0 0.00 0.00 36 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>*, unsigned long) [182]
0.00 0.00 36/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 12/36 Benchmark::Memory() [49]
0.00 0.00 24/36 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [104]
[183] 0.0 0.00 0.00 36 Grid::CartesianCommunicator::BroadcastWorld(int, void*, int) [183]
-----------------------------------------------
0.00 0.00 12/36 Benchmark::DWF5(int, int) [17]
0.00 0.00 24/36 Benchmark::DWF(int, int, double&) [13]
[184] 0.0 0.00 0.00 36 void std::vector<double, std::allocator<double> >::_M_realloc_insert<double const&>(__gnu_cxx::__normal_iterator<double*, std::vector<double, std::allocator<double> > >, double const&) [184]
-----------------------------------------------
0.00 0.00 33/33 Benchmark::Memory() [49]
[185] 0.0 0.00 0.00 33 Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::deallocate(Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8>*, unsigned long) [185]
0.00 0.00 33/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 32/32 Benchmark::DWF(int, int, double&) [13]
[186] 0.0 0.00 0.00 32 Grid::CartesianCommunicator::Broadcast(int, void*, int) [186]
-----------------------------------------------
0.00 0.00 32/32 Benchmark::DWF(int, int, double&) [13]
[187] 0.0 0.00 0.00 32 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters() [187]
-----------------------------------------------
0.00 0.00 32/32 Benchmark::DWF(int, int, double&) [13]
[188] 0.0 0.00 0.00 32 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopOE(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int) [188]
-----------------------------------------------
0.00 0.00 1/31 Benchmark::Memory() [49]
0.00 0.00 6/31 Benchmark::Decomposition() [81]
0.00 0.00 8/31 Benchmark::DWF5(int, int) [17]
0.00 0.00 16/31 Benchmark::DWF(int, int, double&) [13]
[189] 0.0 0.00 0.00 31 Grid::GridDefaultSimd(int, int) [189]
-----------------------------------------------
0.00 0.00 1/25 Benchmark::Memory() [49]
0.00 0.00 24/25 Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [104]
[190] 0.0 0.00 0.00 25 std::seed_seq::seed_seq<__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > > >(__gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int const*, std::vector<int, std::allocator<int> > >) [190]
-----------------------------------------------
0.00 0.00 24/24 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~CayleyFermion5D() [202]
[191] 0.0 0.00 0.00 24 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8>*, unsigned long) [191]
0.00 0.00 24/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 23/23 Grid::CartesianCommunicator::~CartesianCommunicator() [193]
[192] 0.0 0.00 0.00 23 Grid::SharedMemory::~SharedMemory() [192]
-----------------------------------------------
0.00 0.00 4/23 Grid::QCD::SpaceTimeGrid::makeFiveDimDWFRedBlackGrid(int, Grid::GridCartesian const*) [40]
0.00 0.00 8/23 Grid::QCD::SpaceTimeGrid::makeFiveDimRedBlackGrid(int, Grid::GridCartesian const*) [41]
0.00 0.00 11/23 Benchmark::Memory() [49]
[193] 0.0 0.00 0.00 23 Grid::CartesianCommunicator::~CartesianCommunicator() [193]
0.00 0.00 23/23 Grid::SharedMemory::~SharedMemory() [192]
-----------------------------------------------
0.00 0.00 4/22 Grid::GridParseLayout(char**, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [231]
0.00 0.00 18/22 Grid::Grid_init(int*, char***) [122]
[194] 0.0 0.00 0.00 22 char** std::__find_if<char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const> >(char**, char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const>, std::random_access_iterator_tag) [194]
-----------------------------------------------
0.00 0.00 8/16 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsTanh(Grid::Approx::zolotarev_data*, double, double) [200]
0.00 0.00 8/16 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsInternal(double, std::vector<double, std::allocator<double> >&, double, double) [201]
[195] 0.0 0.00 0.00 16 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooeeInternalCompute(int, int, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&) [195]
-----------------------------------------------
0.00 0.00 16/16 Benchmark::DWF5(int, int) [17]
[196] 0.0 0.00 0.00 16 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters() [196]
-----------------------------------------------
0.00 0.00 12/12 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D() [209]
[197] 0.0 0.00 0.00 12 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>*, unsigned long) [197]
0.00 0.00 12/1325 Grid::PointerCache::Insert(void*, unsigned long) [148]
-----------------------------------------------
0.00 0.00 4/12 Benchmark::DWF5(int, int) [17]
0.00 0.00 8/12 Benchmark::DWF(int, int, double&) [13]
[198] 0.0 0.00 0.00 12 std::vector<std::complex<float>, std::allocator<std::complex<float> > >::_M_fill_insert(__gnu_cxx::__normal_iterator<std::complex<float>*, std::vector<std::complex<float>, std::allocator<std::complex<float> > > >, unsigned long, std::complex<float> const&) [198]
-----------------------------------------------
0.00 0.00 8/8 Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [16]
[199] 0.0 0.00 0.00 8 Grid::QCD::WilsonKernels<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonKernels(Grid::QCD::WilsonImplParams const&) [199]
-----------------------------------------------
0.00 0.00 8/8 Benchmark::DWF(int, int, double&) [13]
[200] 0.0 0.00 0.00 8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsTanh(Grid::Approx::zolotarev_data*, double, double) [200]
0.00 0.00 8/16 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooeeInternalCompute(int, int, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&) [195]
0.00 0.00 8/8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsInternal(double, std::vector<double, std::allocator<double> >&, double, double) [201]
-----------------------------------------------
0.00 0.00 8/8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsTanh(Grid::Approx::zolotarev_data*, double, double) [200]
[201] 0.0 0.00 0.00 8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::SetCoefficientsInternal(double, std::vector<double, std::allocator<double> >&, double, double) [201]
0.00 0.00 120/120 std::vector<double, std::allocator<double> >::_M_default_append(unsigned long) [161]
0.00 0.00 120/120 std::vector<double, std::allocator<double> >::resize(unsigned long) [162]
0.00 0.00 8/16 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooeeInternalCompute(int, int, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&, std::vector<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > >, Grid::alignedAllocator<Grid::iScalar<Grid::iScalar<Grid::iScalar<Grid::Grid_simd<std::complex<float>, __Float32x4_t> > > > > >&) [195]
-----------------------------------------------
0.00 0.00 8/8 Benchmark::DWF(int, int, double&) [13]
[202] 0.0 0.00 0.00 8 Grid::QCD::CayleyFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~CayleyFermion5D() [202]
0.00 0.00 24/24 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8>*, unsigned long) [191]
0.00 0.00 24/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::~CartesianStencil() [180]
0.00 0.00 8/180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long) [158]
-----------------------------------------------
0.00 0.00 8/8 Benchmark::DWF(int, int, double&) [13]
[203] 0.0 0.00 0.00 8 Grid::Approx::zolotarev_free(Grid::Approx::zolotarev_data*) [203]
-----------------------------------------------
0.00 0.00 8/8 Benchmark::DWF(int, int, double&) [13]
[204] 0.0 0.00 0.00 8 Grid::Approx::higham(double, int) [204]
0.00 0.00 8/8 Grid::Approx::construct_partfrac(Grid::Approx::izd*) [206]
0.00 0.00 8/8 Grid::Approx::construct_contfrac(Grid::Approx::izd*) [205]
-----------------------------------------------
0.00 0.00 8/8 Grid::Approx::higham(double, int) [204]
[205] 0.0 0.00 0.00 8 Grid::Approx::construct_contfrac(Grid::Approx::izd*) [205]
0.00 0.00 4/4 Grid::Approx::contfrac_A(double*, double*, double*, double*, int, int) [210]
-----------------------------------------------
0.00 0.00 8/8 Grid::Approx::higham(double, int) [204]
[206] 0.0 0.00 0.00 8 Grid::Approx::construct_partfrac(Grid::Approx::izd*) [206]
-----------------------------------------------
0.00 0.00 4/4 __libc_csu_init [3738]
[207] 0.0 0.00 0.00 4 _GLOBAL__sub_I__ZN4Grid3QCD13SpaceTimeGrid15makeFourDimGridERKSt6vectorIiSaIiEES6_S6_ [207]
-----------------------------------------------
0.00 0.00 4/4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&) [19]
[208] 0.0 0.00 0.00 4 Grid::QCD::WilsonKernels<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonKernels(Grid::QCD::WilsonImplParams const&) [208]
-----------------------------------------------
0.00 0.00 4/4 Benchmark::DWF5(int, int) [17]
[209] 0.0 0.00 0.00 4 Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D() [209]
0.00 0.00 12/12 Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>*, unsigned long) [197]
0.00 0.00 12/36 Grid::CartesianStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >::~CartesianStencil() [180]
0.00 0.00 4/180 Grid::alignedAllocator<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::deallocate(Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >*, unsigned long) [158]
-----------------------------------------------
28 Grid::Approx::contfrac_A(double*, double*, double*, double*, int, int) [210]
0.00 0.00 4/4 Grid::Approx::construct_contfrac(Grid::Approx::izd*) [205]
[210] 0.0 0.00 0.00 4+28 Grid::Approx::contfrac_A(double*, double*, double*, double*, int, int) [210]
28 Grid::Approx::contfrac_A(double*, double*, double*, double*, int, int) [210]
-----------------------------------------------
0.00 0.00 3/3 Grid::GridCmdOptionCSL(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&) [232]
[211] 0.0 0.00 0.00 3 void std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >::_M_realloc_insert<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&>(__gnu_cxx::__normal_iterator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) [211]
-----------------------------------------------
0.00 0.00 3/3 Grid::Colours::~Colours() [3543]
[212] 0.0 0.00 0.00 3 std::_Rb_tree<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::_Select1st<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >, std::less<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::allocator<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >::_M_erase(std::_Rb_tree_node<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >*) [212]
-----------------------------------------------
0.00 0.00 2/2 __libc_csu_init [3738]
[213] 0.0 0.00 0.00 2 _GLOBAL__sub_I__ZN4Grid3QCD22StaggeredKernelsStatic3OptE [213]
-----------------------------------------------
0.00 0.00 2/2 __libc_csu_init [3738]
[214] 0.0 0.00 0.00 2 _GLOBAL__sub_I__ZN4Grid3QCD5Gamma3gmuE [214]
-----------------------------------------------
0.00 0.00 1/2 Grid::Grid_init(int*, char***) [122]
0.00 0.00 1/2 Grid::CartesianCommunicator::Init(int*, char***) [121]
[215] 0.0 0.00 0.00 2 Grid::Grid_quiesce_nodes() [215]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[216] 0.0 0.00 0.00 1 _GLOBAL__sub_I_L_list [216]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[217] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid10GridThread8_threadsE [217]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[218] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid13LebesgueOrder16UseLebesgueOrderE [218]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[219] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid14MemoryProfiler5statsE [219]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[220] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid18GlobalSharedMemory17MAX_MPI_SHM_BYTESE [220]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[221] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid18GlobalSharedMemory4InitEP19ompi_communicator_t [221]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[222] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid21CartesianCommunicator18CommunicatorPolicyE [222]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[223] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid21CartesianCommunicator18communicator_worldE [223]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[224] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid26Gather_plane_table_computeEPNS_8GridBaseEiiiiRSt6vectorISt4pairIiiESaIS4_EE [224]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[225] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid3QCD13WilsonKernelsINS0_17GparityWilsonImplINS_9Grid_simdISt7complexIfE13__Float32x4_tEENS0_14FundamentalRepILi3EEENS0_9CoeffRealEEEE12HandDhopSiteERNS0_13WilsonStencilINS_7iVectorINSE_INSE_IS7_Li3EEELi4EEELi2EEENSE_INSE_ISF_Li2EEELi2EEEEERNS_13LebesgueOrderERNS_7LatticeINSE_INSE_INS_7iScalarINS_7iMatrixIS7_Li3EEEEELi8EEELi2EEEEEPSJ_iiRKNSO_ISH_EERSY_ [225]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[226] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid3QCD19WilsonKernelsStatic3OptE [226]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[227] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid3QCD21WilsonFermion5DStatic10directionsE [227]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[228] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid8demangleB5cxx11EPKc [228]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[229] 0.0 0.00 0.00 1 _GLOBAL__sub_I__ZN4Grid9printHashEv [229]
-----------------------------------------------
0.00 0.00 1/1 __libc_csu_init [3738]
[230] 0.0 0.00 0.00 1 __static_initialization_and_destruction_0(int, int) [clone .constprop.58] [230]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[231] 0.0 0.00 0.00 1 Grid::GridParseLayout(char**, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [231]
0.00 0.00 4/22 char** std::__find_if<char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const> >(char**, char**, __gnu_cxx::__ops::_Iter_equals_val<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const>, std::random_access_iterator_tag) [194]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[232] 0.0 0.00 0.00 1 Grid::GridCmdOptionCSL(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&) [232]
0.00 0.00 3/3 void std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >::_M_realloc_insert<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&>(__gnu_cxx::__normal_iterator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >*, std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) [211]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[233] 0.0 0.00 0.00 1 Grid::GridLogConfigure(std::vector<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::allocator<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >&) [233]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[234] 0.0 0.00 0.00 1 Grid::GridLogTimestamp(int) [234]
-----------------------------------------------
0.00 0.00 1/1 Grid::CartesianCommunicator::Init(int*, char***) [121]
[235] 0.0 0.00 0.00 1 Grid::GlobalSharedMemory::SharedMemoryAllocate(unsigned long, int) [235]
0.00 0.00 1/92 std::vector<void*, std::allocator<void*> >::_M_default_append(unsigned long) [164]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[236] 0.0 0.00 0.00 1 Grid::CartesianCommunicator::RankWorld() [236]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[237] 0.0 0.00 0.00 1 Grid::printHash() [237]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_init(int*, char***) [122]
[238] 0.0 0.00 0.00 1 Grid::operator<<(std::ostream&, Grid::Logger&) [clone .constprop.140] [238]
-----------------------------------------------
0.00 0.00 1/1 Grid::Grid_unquiesce_nodes() [631]
[239] 0.0 0.00 0.00 1 std::_Rb_tree<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >, std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::_Select1st<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > >, std::less<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > >, std::allocator<std::pair<std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > > > >::_M_get_insert_unique_pos(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) [239]
-----------------------------------------------
0.00 0.00 1/1 Grid::GlobalSharedMemory::Init(ompi_communicator_t*) [120]
[240] 0.0 0.00 0.00 1 void std::__introsort_loop<__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, long, __gnu_cxx::__ops::_Iter_comp_iter<std::less<int> > >(__gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, __gnu_cxx::__normal_iterator<int*, std::vector<int, std::allocator<int> > >, long, __gnu_cxx::__ops::_Iter_comp_iter<std::less<int> >) [240]
-----------------------------------------------
This table describes the call tree of the program, and was sorted by
the total amount of time spent in each function and its children.
Each entry in this table consists of several lines. The line with the
index number at the left hand margin lists the current function.
The lines above it list the functions that called this function,
and the lines below it list the functions this one called.
This line lists:
index A unique number given to each element of the table.
Index numbers are sorted numerically.
The index number is printed next to every function name so
it is easier to look up where the function is in the table.
% time This is the percentage of the `total' time that was spent
in this function and its children. Note that due to
different viewpoints, functions excluded by options, etc,
these numbers will NOT add up to 100%.
self This is the total amount of time spent in this function.
children This is the total amount of time propagated into this
function by its children.
called This is the number of times the function was called.
If the function called itself recursively, the number
only includes non-recursive calls, and is followed by
a `+' and the number of recursive calls.
name The name of the current function. The index number is
printed after it. If the function is a member of a
cycle, the cycle number is printed between the
function's name and the index number.
For the function's parents, the fields have the following meanings:
self This is the amount of time that was propagated directly
from the function into this parent.
children This is the amount of time that was propagated from
the function's children into this parent.
called This is the number of times this parent called the
function `/' the total number of times the function
was called. Recursive calls to the function are not
included in the number after the `/'.
name This is the name of the parent. The parent's index
number is printed after it. If the parent is a
member of a cycle, the cycle number is printed between
the name and the index number.
If the parents of the function cannot be determined, the word
`<spontaneous>' is printed in the `name' field, and all the other
fields are blank.
For the function's children, the fields have the following meanings:
self This is the amount of time that was propagated directly
from the child into the function.
children This is the amount of time that was propagated from the
child's children to the function.
called This is the number of times the function called
this child `/' the total number of times the child
was called. Recursive calls by the child are not
listed in the number after the `/'.
name This is the name of the child. The child's index
number is printed after it. If the child is a
member of a cycle, the cycle number is printed
between the name and the index number.
If there are any cycles (circles) in the call graph, there is an
entry for the cycle-as-a-whole. This entry shows who called the
cycle (as parents) and the members of the cycle (as children.)
The `+' recursive calls entry shows the number of function calls that
were internal to the cycle, and the calls entry for each member shows,
for that member, how many times it was called from other members of
the cycle.
Copyright (C) 2012-2015 Free Software Foundation, Inc.
Copying and distribution of this file, with or without modification,
are permitted in any medium without royalty provided the copyright
notice and this notice are preserved.
Index by function name
[216] _GLOBAL__sub_I_L_list [191] Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8>*, unsigned long) [20] Grid::QCD::FermionOperator<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::RedBlackGrid()
[217] _GLOBAL__sub_I__ZN4Grid10GridThread8_threadsE [71] Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::allocate(unsigned long, void const*) [25] Grid::QCD::FermionOperator<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MooDeriv(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int)
[218] _GLOBAL__sub_I__ZN4Grid13LebesgueOrder16UseLebesgueOrderE [197] Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::deallocate(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>*, unsigned long) [28] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
[219] _GLOBAL__sub_I__ZN4Grid14MemoryProfiler5statsE [87] Grid::alignedAllocator<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::allocate(unsigned long, void const*) [83] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[220] _GLOBAL__sub_I__ZN4Grid18GlobalSharedMemory17MAX_MPI_SHM_BYTESE [185] Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::deallocate(Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8>*, unsigned long) [187] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters()
[221] _GLOBAL__sub_I__ZN4Grid18GlobalSharedMemory4InitEP19ompi_communicator_t [78] Grid::alignedAllocator<Grid::iVector<Grid::Grid_simd<float, __Float32x4_t>, 8> >::allocate(unsigned long, void const*) [100] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[222] _GLOBAL__sub_I__ZN4Grid21CartesianCommunicator18CommunicatorPolicyE [170] Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::deallocate(Grid::Grid_simd<unsigned int, __Uint32x4_t>*, unsigned long) [89] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[223] _GLOBAL__sub_I__ZN4Grid21CartesianCommunicator18communicator_worldE [171] Grid::alignedAllocator<Grid::Grid_simd<unsigned int, __Uint32x4_t> >::allocate(unsigned long, void const*) [141] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[224] _GLOBAL__sub_I__ZN4Grid26Gather_plane_table_computeEPNS_8GridBaseEiiiiRSt6vectorISt4pairIiiESaIS4_EE [166] Grid::alignedAllocator<std::complex<double> >::deallocate(std::complex<double>*, unsigned long) [188] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopOE(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[207] _GLOBAL__sub_I__ZN4Grid3QCD13SpaceTimeGrid15makeFourDimGridERKSt6vectorIiSaIiEES6_S6_ [167] Grid::alignedAllocator<std::complex<double> >::allocate(unsigned long, void const*) [43] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report()
[225] _GLOBAL__sub_I__ZN4Grid3QCD13WilsonKernelsINS0_17GparityWilsonImplINS_9Grid_simdISt7complexIfE13__Float32x4_tEENS0_14FundamentalRepILi3EEENS0_9CoeffRealEEEE12HandDhopSiteERNS0_13WilsonStencilINS_7iVectorINSE_INSE_IS7_Li3EEELi4EEELi2EEENSE_INSE_ISF_Li2EEELi2EEEEERNS_13LebesgueOrderERNS_7LatticeINSE_INSE_INS_7iScalarINS_7iMatrixIS7_Li3EEEEELi8EEELi2EEEEEPSJ_iiRKNSO_ISH_EERSY_ [32] void Grid::LatticeCoordinate<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >(Grid::Lattice<Grid::iScalar<Grid::Grid_simd<unsigned int, __Uint32x4_t> > >&, int) [16] Grid::QCD::WilsonFermion5D<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&)
[226] _GLOBAL__sub_I__ZN4Grid3QCD19WilsonKernelsStatic3OptE [111] void Grid::Copy_plane_permute<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int, int) [2] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::Mdir(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3>, 4> > >&, int, int)
[227] _GLOBAL__sub_I__ZN4Grid3QCD21WilsonFermion5DStatic10directionsE [91] void Grid::Copy_plane_permute<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int, int) [3] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffComplexHalfComms> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<double>, __Float64x2_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&)
[213] _GLOBAL__sub_I__ZN4Grid3QCD22StaggeredKernelsStatic3OptE [179] Grid::GlobalSharedMemory::OptimalCommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*&) [29] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ImportGauge(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
[214] _GLOBAL__sub_I__ZN4Grid3QCD5Gamma3gmuE [235] Grid::GlobalSharedMemory::SharedMemoryAllocate(unsigned long, int) [101] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternal(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[228] _GLOBAL__sub_I__ZN4Grid8demangleB5cxx11EPKc [120] Grid::GlobalSharedMemory::Init(ompi_communicator_t*) [196] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::ZeroCounters()
[229] _GLOBAL__sub_I__ZN4Grid9printHashEv [215] Grid::Grid_quiesce_nodes() [108] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalSerialComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[230] __static_initialization_and_destruction_0(int, int) [clone .constprop.58] [97] Grid::CartesianCommunicator::ShiftedRanks(int, int, int&, int&) [109] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopInternalOverlappedComms(Grid::QCD::WilsonStencil<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >, Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 2> > >&, Grid::LebesgueOrder&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[174] time_statistics::statistics(std::vector<double, std::allocator<double> >) [183] Grid::CartesianCommunicator::BroadcastWorld(int, void*, int) [143] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::DhopEO(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, int)
[85] Grid::Grid_simd<unsigned int, __Uint32x4_t> Grid::Comparison<Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t>, 0>(Grid::seq<unsigned int, unsigned int>, Grid::Grid_simd<unsigned int, __Uint32x4_t> const&, Grid::Grid_simd<unsigned int, __Uint32x4_t>::scalar_type const&) [131] Grid::CartesianCommunicator::StencilBarrier() [48] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::Report()
[107] void Grid::Copy_plane<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int, int) [168] Grid::CartesianCommunicator::GlobalSumVector(double*, int) [19] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::WilsonFermion5D(Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> >&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, Grid::GridCartesian&, Grid::GridRedBlackCartesian&, double, Grid::QCD::WilsonImplParams const&)
[110] void Grid::Copy_plane<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int, int) [61] Grid::CartesianCommunicator::ProcessorCoorFromRank(int, std::vector<int, std::allocator<int> >&) [209] Grid::QCD::WilsonFermion5D<Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::~WilsonFermion5D()
[27] Grid::GridRNGbase::~GridRNGbase() [93] Grid::CartesianCommunicator::RankFromProcessorCoor(std::vector<int, std::allocator<int> >&) [12] Grid::QCD::DomainWallFermion<Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal> >::MomentumSpacePropagator(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, double)
[103] void Grid::Cshift_local<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int, int) [119] Grid::CartesianCommunicator::InitFromMPICommunicator(std::vector<int, std::allocator<int> > const&, ompi_communicator_t*) [30] Grid::QCD::DomainWallVec5dImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&)
[90] void Grid::Cshift_local<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int, int) [121] Grid::CartesianCommunicator::Init(int*, char***) [76] void Grid::QCD::SU<3>::LieRandomize<Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > >(Grid::GridParallelRNG&, Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >&, double)
[148] Grid::PointerCache::Insert(void*, unsigned long) [159] Grid::CartesianCommunicator::Barrier() [26] std::enable_if<Grid::matchGridTensorIndex<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4>, 2>::value, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> >::type Grid::QCD::operator*<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >(Grid::QCD::Gamma const&, Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> const&)
[149] Grid::PointerCache::Lookup(unsigned long) [98] Grid::CartesianCommunicator::BossRank() [203] Grid::Approx::zolotarev_free(Grid::Approx::zolotarev_data*)
[116] Grid::SharedMemory::SetCommunicator(ompi_communicator_t*) [92] Grid::CartesianCommunicator::ThisRank() [204] Grid::Approx::higham(double, int)
[156] Grid::SharedMemory::ShmBufferMalloc(unsigned long) [186] Grid::CartesianCommunicator::Broadcast(int, void*, int) [210] Grid::Approx::contfrac_A(double*, double*, double*, double*, int, int)
[160] Grid::SharedMemory::ShmBufferFreeAll() [169] Grid::CartesianCommunicator::GlobalSum(std::complex<double>&) [205] Grid::Approx::construct_contfrac(Grid::Approx::izd*)
[192] Grid::SharedMemory::~SharedMemory() [150] Grid::CartesianCommunicator::GlobalSum(double&) [206] Grid::Approx::construct_partfrac(Grid::Approx::izd*)
[165] std::complex<double> Grid::innerProduct<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&) [236] Grid::CartesianCommunicator::RankWorld() [84] Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > Grid::Cshift<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >(Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > > const&, int, int)
[130] Grid::GridCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [115] Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&, Grid::CartesianCommunicator const&, int&) [54] Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > Grid::Cshift<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >(Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > > const&, int, int)
[146] Grid::GridCartesian::CheckerBoarded(int) [117] Grid::CartesianCommunicator::CartesianCommunicator(std::vector<int, std::allocator<int> > const&) [102] Grid::Lattice<Grid::iScalar<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> > > >::Lattice(Grid::GridBase*)
[144] Grid::GridCartesian::CheckerBoardShiftForCB(int, int, int, int) [193] Grid::CartesianCommunicator::~CartesianCommunicator() [45] Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::Lattice(Grid::GridBase*)
[155] Grid::GridCartesian::CheckerBoardDestination(int, int, int) [63] Grid::GridRedBlackCartesian::CheckerBoard(std::vector<int, std::allocator<int> > const&) [163] Grid::Lattice<Grid::iScalar<Grid::iVector<Grid::iVector<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3>, 4> > >::~Lattice()
[52] Grid::LebesgueOrder::CartesianBlocking() [147] Grid::GridRedBlackCartesian::CheckerBoarded(int) [72] Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >::Lattice(Grid::GridBase*)
[50] Grid::LebesgueOrder::IterateI(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [18] Grid::GridRedBlackCartesian::CheckerBoardFromOindex(int) [88] Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >::Lattice(Grid::GridBase*)
[51] Grid::LebesgueOrder::IterateO(int, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [145] Grid::GridRedBlackCartesian::CheckerBoardShiftForCB(int, int, int, int) [21] Grid::permute(Grid::Grid_simd<std::complex<float>, __Float32x4_t>&, Grid::Grid_simd<std::complex<float>, __Float32x4_t>, int)
[53] Grid::LebesgueOrder::LebesgueOrder(Grid::GridBase*) [62] Grid::GridRedBlackCartesian::oIndex(std::vector<int, std::allocator<int> >&) [175] Grid::GridBase::PermuteType(int)
[35] void Grid::peekLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 4>&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&, std::vector<int, std::allocator<int> >&) [105] Grid::GridRedBlackCartesian::~GridRedBlackCartesian() [55] Grid::GridBase::iIndex(std::vector<int, std::allocator<int> >&)
[46] void Grid::pokeLocalSite<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8>, Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >(Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> const&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<std::complex<float>, 3> >, 8> >&, std::vector<int, std::allocator<int> >&) [178] Grid::Grid_sa_signal_handler(int, siginfo_t*, void*) [56] Grid::GridBase::oIndex(std::vector<int, std::allocator<int> >&)
[177] Grid::GridDefaultMpi() [173] Grid::GridCmdVectorIntToString[abi:cxx11](std::vector<int, std::allocator<int> > const&) [237] Grid::printHash()
[189] Grid::GridDefaultSimd(int, int) [73] Grid::Gather_plane_table_compute(Grid::GridBase*, int, int, int, int, std::vector<std::pair<int, int>, std::allocator<std::pair<int, int> > >&) [238] Grid::operator<<(std::ostream&, Grid::Logger&) [clone .constprop.140]
[104] Grid::GridParallelRNG::SeedFixedIntegers(std::vector<int, std::allocator<int> > const&) [31] Grid::QCD::WilsonImpl<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, Grid::QCD::FundamentalRep<3>, Grid::QCD::CoeffReal>::DoubleStore(Grid::GridBase*, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 8> >&, Grid::Lattice<Grid::iVector<Grid::iScalar<Grid::iMatrix<Grid::Grid_simd<std::complex<float>, __Float32x4_t>, 3> >, 4> > const&) [64] sitmo::prng_engine::encrypt_counter()
[33] Grid::GridParallelRNG::GridParallelRNG(Grid::GridBase*) [113] Grid::QCD::SpaceTimeGrid::makeFiveDimGrid(int, Grid::GridCartesian const*) [13] Benchmark::DWF(int, int, double&)
[231] Grid::GridParseLayout(char**, int, std::vector<int, std::allocator<int> >&, std::vector<int, std::allocator<int> >&) [112] Grid::QCD::SpaceTimeGrid::makeFourDimGrid(std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int> > const&, std::vector<int, std::allocator<int>