portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 13:40:46 +01:00
Code Releases Activity

Prior to rationalising 2 versions of BaseIO::write (scalar and vector)

Browse Source
This commit is contained in:
Michael Marshall 2019-02-19 13:29:08 +00:00
parent 6ebb32ffbf
commit 63c97db414
2 changed files with 105 additions and 133 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

46
Grid/serialisation/BaseIO.h
View File

@ -36,13 +36,19 @@ Author: Guido Cossu <guido.cossu@ed.ac.uk>
#include <Grid/Eigen/unsupported/CXX11/Tensor> #include <Grid/Eigen/unsupported/CXX11/Tensor>
namespace Grid { namespace Grid {
// TODO Support Grid::complex from GPU port
template<typename T> using Grid_complex = std::complex<T>;
// Returns original type, except for Grid_complex, where it returns the underlying type
template<typename T> struct RealType { using type = T; };
template<typename T> struct RealType<Grid_complex<T>> { using type = T; };
namespace EigenIO { namespace EigenIO {
template<typename T> struct is_complex : public std::false_type {}; template<typename T> struct is_complex : public std::false_type {};
template<typename T> struct is_complex<std::complex<T>> template<typename T> struct is_complex<Grid_complex<T>>
: std::integral_constant<bool, std::is_arithmetic<T>::value> {}; : std::integral_constant<bool, std::is_arithmetic<T>::value> {};
// Eigen tensors can be composed of arithmetic scalar and complex types // Eigen tensors can be composed of arithmetic scalar and complex types
// TODO Support Grid::comples from GPU port
template<typename T> struct is_scalar : std::integral_constant<bool, template<typename T> struct is_scalar : std::integral_constant<bool,
std::is_arithmetic<T>::value || is_complex<T>::value> {}; std::is_arithmetic<T>::value || is_complex<T>::value> {};
@ -88,11 +94,12 @@ namespace Grid {
template <typename T, typename C = void> struct Traits {}; // C needed for specialisation template <typename T, typename C = void> struct Traits {}; // C needed for specialisation
// This defines the bottom level - i.e. it's a description of the underlying scalar // This defines the bottom level - i.e. it's a description of the underlying scalar
template <typename T> struct Traits<T, typename std::enable_if<is_scalar<T>::value, void>::type> { template <typename T> struct Traits<T, typename std::enable_if<is_scalar<T>::value, void>::type> {
using scalar_type = T; // Type of the underlying scalar
using scalar_real = typename RealType<scalar_type>::type; // real type underlying scalar_type
static constexpr unsigned int depth = 0; // How many levels of Grid Tensor there are (TensorLevel) static constexpr unsigned int depth = 0; // How many levels of Grid Tensor there are (TensorLevel)
static constexpr unsigned int rank = 0; // The rank of the grid tensor (i.e. how many indices used) static constexpr unsigned int rank = 0; // The rank of the grid tensor (i.e. how many indices used)
static constexpr unsigned int rank_non_trivial = 0; // As per rank, but excludes those of dimension 1 static constexpr unsigned int rank_non_trivial = 0; // As per rank, but excludes those of dimension 1
static constexpr unsigned int count = 1; // total number of elements (i.e. product of dimensions) static constexpr unsigned int count = 1; // total number of elements (i.e. product of dimensions)
using scalar_type = T; // Type of the underlying scalar
static constexpr std::size_t scalar_size = sizeof(T); // Size of the underlying scalar in bytes static constexpr std::size_t scalar_size = sizeof(T); // Size of the underlying scalar in bytes
static constexpr std::size_t size = scalar_size * count; // total size of elements in bytes static constexpr std::size_t size = scalar_size * count; // total size of elements in bytes
static constexpr std::size_t Dimension(unsigned int dim) { return 0; } // Dimension size static constexpr std::size_t Dimension(unsigned int dim) { return 0; } // Dimension size
@ -114,11 +121,12 @@ namespace Grid {
// count = 48 // count = 48
}; };
template <typename T> struct Traits<iScalar<T>> { template <typename T> struct Traits<iScalar<T>> {
using scalar_type = typename Traits<T>::scalar_type;
using scalar_real = typename RealType<scalar_type>::type;
static constexpr unsigned int depth = 1 + Traits<T>::depth; static constexpr unsigned int depth = 1 + Traits<T>::depth;
static constexpr unsigned int rank = 0 + Traits<T>::rank; static constexpr unsigned int rank = 0 + Traits<T>::rank;
static constexpr unsigned int rank_non_trivial = 0 + Traits<T>::rank_non_trivial; static constexpr unsigned int rank_non_trivial = 0 + Traits<T>::rank_non_trivial;
static constexpr unsigned int count = 1 * Traits<T>::count; static constexpr unsigned int count = 1 * Traits<T>::count;
using scalar_type = typename Traits<T>::scalar_type;
static constexpr std::size_t scalar_size = Traits<T>::scalar_size; static constexpr std::size_t scalar_size = Traits<T>::scalar_size;
static constexpr std::size_t size = scalar_size * count; static constexpr std::size_t size = scalar_size * count;
static constexpr std::size_t Dimension(unsigned int dim) { static constexpr std::size_t Dimension(unsigned int dim) {
@ -127,11 +135,12 @@ namespace Grid {
return Traits<T>::DimensionNT(dim); } return Traits<T>::DimensionNT(dim); }
}; };
template <typename T, int N> struct Traits<iVector<T, N>> { template <typename T, int N> struct Traits<iVector<T, N>> {
using scalar_type = typename Traits<T>::scalar_type;
using scalar_real = typename RealType<scalar_type>::type;
static constexpr unsigned int depth = 1 + Traits<T>::depth; static constexpr unsigned int depth = 1 + Traits<T>::depth;
static constexpr unsigned int rank = 1 + Traits<T>::rank; static constexpr unsigned int rank = 1 + Traits<T>::rank;
static constexpr unsigned int rank_non_trivial = (N>1 ? 1 : 0) + Traits<T>::rank_non_trivial; static constexpr unsigned int rank_non_trivial = (N>1 ? 1 : 0) + Traits<T>::rank_non_trivial;
static constexpr unsigned int count = N * Traits<T>::count; static constexpr unsigned int count = N * Traits<T>::count;
using scalar_type = typename Traits<T>::scalar_type;
static constexpr std::size_t scalar_size = Traits<T>::scalar_size; static constexpr std::size_t scalar_size = Traits<T>::scalar_size;
static constexpr std::size_t size = scalar_size * count; static constexpr std::size_t size = scalar_size * count;
static constexpr std::size_t Dimension(unsigned int dim) { static constexpr std::size_t Dimension(unsigned int dim) {
@ -141,11 +150,12 @@ namespace Grid {
} }
}; };
template <typename T, int N> struct Traits<iMatrix<T, N>> { template <typename T, int N> struct Traits<iMatrix<T, N>> {
using scalar_type = typename Traits<T>::scalar_type;
using scalar_real = typename RealType<scalar_type>::type;
static constexpr unsigned int depth = 1 + Traits<T>::depth; static constexpr unsigned int depth = 1 + Traits<T>::depth;
static constexpr unsigned int rank = 2 + Traits<T>::rank; static constexpr unsigned int rank = 2 + Traits<T>::rank;
static constexpr unsigned int rank_non_trivial = (N>1 ? 2 : 0) + Traits<T>::rank_non_trivial; static constexpr unsigned int rank_non_trivial = (N>1 ? 2 : 0) + Traits<T>::rank_non_trivial;
static constexpr unsigned int count = N * N * Traits<T>::count; static constexpr unsigned int count = N * N * Traits<T>::count;
using scalar_type = typename Traits<T>::scalar_type;
static constexpr std::size_t scalar_size = Traits<T>::scalar_size; static constexpr std::size_t scalar_size = Traits<T>::scalar_size;
static constexpr std::size_t size = scalar_size * count; static constexpr std::size_t size = scalar_size * count;
static constexpr std::size_t Dimension(unsigned int dim) { static constexpr std::size_t Dimension(unsigned int dim) {
@ -230,16 +240,22 @@ namespace Grid {
} }
} }
// Used for sequential initialisations // Sequential initialisation of tensors
template<typename T> constexpr T Flag = 1; // Would have preferred to define template variables for this, but that's c++ 17
template<typename T> constexpr std::complex<T> Flag<std::complex<T>> {1, -1}; template <typename ETensor>
// Returns the type of the real part of an arithmetic type typename std::enable_if<EigenIO::is_tensor<ETensor>::value && !EigenIO::is_complex<typename EigenIO::Traits<typename ETensor::Scalar>::scalar_type>::value, void>::type
template<typename T> struct RealType { using type = T; }; SequentialInit( ETensor &ET, typename EigenIO::Traits<typename ETensor::Scalar>::scalar_type Inc = 1 )
template<typename T> struct RealType<std::complex<T>> { using type = T; }; {
using Traits = EigenIO::Traits<typename ETensor::Scalar>;
using scalar_type = typename Traits::scalar_type;
for_all( ET, [&](scalar_type &c, typename ETensor::Index n, const std::array<size_t, ETensor::NumIndices + Traits::rank_non_trivial> &Dims ) {
c = Inc * static_cast<scalar_type>(n);
} );
}
template <typename ETensor> template <typename ETensor>
typename std::enable_if<EigenIO::is_tensor<ETensor>::value, void>::type typename std::enable_if<EigenIO::is_tensor<ETensor>::value && EigenIO::is_complex<typename EigenIO::Traits<typename ETensor::Scalar>::scalar_type>::value, void>::type
Sequential_Init( ETensor &ET, typename EigenIO::Traits<typename ETensor::Scalar>::scalar_type Inc = Flag<typename EigenIO::Traits<typename ETensor::Scalar>::scalar_type> ) SequentialInit( ETensor &ET, typename EigenIO::Traits<typename ETensor::Scalar>::scalar_type Inc={1,-1})
{ {
using Traits = EigenIO::Traits<typename ETensor::Scalar>; using Traits = EigenIO::Traits<typename ETensor::Scalar>;
using scalar_type = typename Traits::scalar_type; using scalar_type = typename Traits::scalar_type;
@ -247,7 +263,7 @@ namespace Grid {
c = Inc * static_cast<typename RealType<scalar_type>::type>(n); c = Inc * static_cast<typename RealType<scalar_type>::type>(n);
} ); } );
} }
// Helper to dump a tensor // Helper to dump a tensor
#ifdef DEBUG #ifdef DEBUG
#define dump_tensor(args...) dump_tensor_func(args) #define dump_tensor(args...) dump_tensor_func(args)

192
tests/IO/Test_serialisation.cc
View File

@ -82,7 +82,7 @@ bool b = false;
template <typename W, typename R, typename O> template <typename W, typename R, typename O>
void ioTest(const std::string &filename, const O &object, const std::string &name) void ioTest(const std::string &filename, const O &object, const std::string &name)
{ {
std::cout << name << " IO test: writing ..."; std::cout << "IO test: " << name << " -> " << filename << " ...";
// writer needs to be destroyed so that writing physically happens // writer needs to be destroyed so that writing physically happens
{ {
W writer(filename); W writer(filename);
@ -92,149 +92,105 @@ void ioTest(const std::string &filename, const O &object, const std::string &nam
std::cout << " done. reading..."; std::cout << " done. reading...";
R reader(filename); R reader(filename);
O buf; std::unique_ptr<O> buf( new O ); // In case object too big for stack
read(reader, "testobject", buf); read(reader, "testobject", *buf);
bool good = Serializable::CompareMember(object, buf); bool good = Serializable::CompareMember(object, *buf);
if (!good) { if (!good) {
std::cout << " failure!" << std::endl; std::cout << " failure!" << std::endl;
if constexpr (EigenIO::is_tensor<O>::value) if constexpr (EigenIO::is_tensor<O>::value)
dump_tensor(buf,"???"); dump_tensor(*buf,"???");
exit(EXIT_FAILURE); exit(EXIT_FAILURE);
} }
std::cout << " done." << std::endl; std::cout << " done." << std::endl;
} }
#ifdef HAVE_HDF5 #ifdef HAVE_HDF5
typedef Eigen::Tensor<int, 5> ShortRank5Tensor;
//typedef int TestScalar;
typedef std::complex<double> TestScalar; typedef std::complex<double> TestScalar;
typedef Eigen::Tensor<TestScalar, 3, Eigen::StorageOptions::RowMajor> TestTensor; using TensorSingle = Eigen::TensorFixedSize<int, Eigen::Sizes<1>>;
typedef Eigen::TensorFixedSize<TestScalar, Eigen::Sizes<9,4,2>, Eigen::StorageOptions::RowMajor> TestTensorFixed; using TensorSimple = Eigen::Tensor<iMatrix<TestScalar,1>, 6>;
typedef std::vector<TestTensorFixed> aTestTensorFixed; typedef Eigen::Tensor<unsigned short, 5> TensorRank5UShort;
typedef Eigen::TensorFixedSize<SpinColourVector, Eigen::Sizes<11,3,2>, Eigen::StorageOptions::RowMajor> LSCTensor; typedef Eigen::Tensor<int, 5, Eigen::StorageOptions::RowMajor> TensorRank5IntAlt;
typedef Eigen::TensorFixedSize<LorentzColourMatrix, Eigen::Sizes<5,7,2>, Eigen::StorageOptions::RowMajor> LCMTensor; typedef Eigen::Tensor<TestScalar, 3, Eigen::StorageOptions::RowMajor> TensorRank3;
// From Test_serialisation.cc typedef Eigen::TensorFixedSize<TestScalar, Eigen::Sizes<9,4,2>, Eigen::StorageOptions::RowMajor> Tensor_9_4_2;
typedef std::vector<Tensor_9_4_2> aTensor_9_4_2;
typedef Eigen::TensorFixedSize<SpinColourVector, Eigen::Sizes<6,5>> LSCTensor;
#ifdef DEBUG
typedef Eigen::TensorFixedSize<iMatrix<iVector<iMatrix<iVector<LorentzColourMatrix,5>,2>,7>,3>, Eigen::Sizes<2,2,11,10,9>, Eigen::StorageOptions::RowMajor> LCMTensor;
#endif
class PerambIOTestClass: Serializable { class PerambIOTestClass: Serializable {
public: public:
using PerambTensor = Eigen::Tensor<SpinColourVector, 6, Eigen::StorageOptions::RowMajor>; using PerambTensor = Eigen::Tensor<SpinColourVector, 6, Eigen::StorageOptions::RowMajor>;
GRID_SERIALIZABLE_CLASS_MEMBERS(PerambIOTestClass GRID_SERIALIZABLE_CLASS_MEMBERS(PerambIOTestClass
, ShortRank5Tensor, shortRank5Tensor , SpinColourVector, spinColourVector
, PerambTensor, Perambulator , SpinColourMatrix, spinColourMatrix
, std::vector<std::string>, DistilParameterNames , std::vector<std::string>, DistilParameterNames
, std::vector<int>, DistilParameterValues , std::vector<int>, DistilParameterValues
, PerambTensor, Perambulator
, PerambTensor, Perambulator2 , PerambTensor, Perambulator2
, SpinColourVector, scv , TensorRank5UShort, tensorRank5UShort
, SpinColourMatrix, scm , TensorRank3, tensorRank3
//, TestTensor, Critter , Tensor_9_4_2, tensor_9_4_2
//, TestTensorFixed, FixedCritter , aTensor_9_4_2, atensor_9_4_2
//, aTestTensorFixed, aFixedCritter , LSCTensor, MyLSCTensor
//, LSCTensor, MyLSCTensor #ifdef DEBUG
//, LCMTensor, MyLCMTensor , LCMTensor, MyLCMTensor
#endif
); );
PerambIOTestClass() : Perambulator(2,3,1,4,5,1), Perambulator2(7,1,6,1,5,1), PerambIOTestClass()
DistilParameterNames {"alpha", "beta", "gamma", "delta", "epsilon", "what's f?"}, : DistilParameterNames {"alpha", "beta", "gamma", "delta", "epsilon", "zeta"}
DistilParameterValues{2,3,1,4,5,1} , DistilParameterValues{2,3,1,4,5,1}
, shortRank5Tensor{5,4,3,2,1} , Perambulator(2,3,1,4,5,1)
//, Critter(7,3,2)//, aFixedCritter(3) , Perambulator2(7,1,6,1,5,1)
, tensorRank5UShort{5,4,3,2,1}
, tensorRank3(7,3,2)
{ {
Sequential_Init(Perambulator); Grid_complex<double> Flag{1,-3.1415927};
Sequential_Init(Perambulator2, {-3.1415927,7}); SequentialInit(Perambulator, Flag);
Sequential_Init(shortRank5Tensor); SequentialInit(Perambulator2, Flag);
SequentialInit(tensorRank5UShort);
SequentialInit(tensorRank3, Flag);
SequentialInit(tensor_9_4_2, Flag);
for( auto &t : atensor_9_4_2 )
SequentialInit(t, Flag);
SequentialInit( MyLSCTensor );
#ifdef DEBUG
SequentialInit( MyLCMTensor );
#endif
} }
}; };
void EigenHdf5IOTest(void) { #define TensorWriteReadInnerNoInit( T ) \
SpinColourVector scv, scv2; ioTest<Hdf5Writer, Hdf5Reader, T>("iotest_"s + std::to_string(++TestNum) + "_" #T ".h5", t, #T);
scv2 = scv; #define TensorWriteReadInner( T ) SequentialInit( t ); TensorWriteReadInnerNoInit( T )
ioTest<Hdf5Writer, Hdf5Reader, SpinColourVector>("iotest_vector.h5", scv, "SpinColourVector"); #define TensorWriteRead( T ) { T t ; TensorWriteReadInner( T ) }
SpinColourMatrix scm; #define TensorWriteReadV(T, ... ) { T t( __VA_ARGS__ ); TensorWriteReadInner( T ) }
ioTest<Hdf5Writer, Hdf5Reader, SpinColourMatrix>("iotest_matrix.h5", scm, "SpinColourMatrix"); #define TensorWriteReadLarge( T ) { std::unique_ptr<T> p{new T}; T &t{*p}; TensorWriteReadInnerNoInit(T) }
constexpr TestScalar Inc{1,-1}; void EigenHdf5IOTest(void)
{
unsigned int TestNum = 0;
using namespace std::string_literals;
TensorWriteRead( TensorSingle )
TensorWriteReadV( TensorSimple, 1, 1, 1, 1, 1, 1 )
TensorWriteReadV( TensorRank3, 6, 3, 2 )
TensorWriteRead ( Tensor_9_4_2 )
TensorWriteRead ( LSCTensor )
TensorWriteReadLarge( PerambIOTestClass )
#ifdef DEBUG
std::cout << "sizeof( LCMTensor ) = " << sizeof( LCMTensor ) / 1024 / 1024 << " MB" << std::endl;
TensorWriteReadLarge ( LCMTensor )
// Also write > 4GB of complex numbers (I suspect this will fail inside Hdf5)
{ {
using TestTensorSingle = Eigen::TensorFixedSize<int, Eigen::Sizes<1>>; static constexpr size_t Num = 0x11000000;
TestTensorSingle ts; std::cout << "Stress test: " << Num * sizeof( Grid_complex<double> ) / 1024 / 1024
ts(0) = 7; // lucky << " MB array of complex<double>" << std::endl;
ioTest<Hdf5Writer, Hdf5Reader, TestTensorSingle>("iotest_single.h5", ts, "Tensor_single"); using Stress = std::vector<Grid_complex<double>>;
Stress t (Num);
TensorWriteReadInnerNoInit( Stress );
} }
#endif
{
using TestTensorSimple = Eigen::Tensor<iMatrix<TestScalar,1>, 6>;
TestTensorSimple ts(1,1,1,1,1,1);
ts(0,0,0,0,0,0) = Inc * 3.1415927;
ioTest<Hdf5Writer, Hdf5Reader, TestTensorSimple>("iotest_simple.h5", ts, "Tensor_simple");
}
TestTensor t(6,3,2);
TestScalar Val{Inc};
for( int i = 0 ; i < t.dimension(0) ; i++)
for( int j = 0 ; j < t.dimension(1) ; j++)
for( int k = 0 ; k < t.dimension(2) ; k++) {
t(i,j,k) = Val;
Val += Inc;
}
ioTest<Hdf5Writer, Hdf5Reader, TestTensor>("iotest_tensor.h5", t, "eigen_tensor_instance_name");
//dump_tensor(t, "t");
// Now serialise a fixed size tensor
using FixedTensor = Eigen::TensorFixedSize<TestScalar, Eigen::Sizes<8,4,3>>;
FixedTensor tf;
Val = Inc;
for( int i = 0 ; i < tf.dimension(0) ; i++)
for( int j = 0 ; j < tf.dimension(1) ; j++)
for( int k = 0 ; k < tf.dimension(2) ; k++) {
tf(i,j,k) = Val;
Val += Inc;
}
ioTest<Hdf5Writer, Hdf5Reader, FixedTensor>("iotest_tensor_fixed.h5", tf, "eigen_tensor_fixed_name");
//dump_tensor(tf, "tf");
PerambIOTestClass o;
//dump_tensor(o.Perambulator, "Perambulator" );
dump_tensor(o.shortRank5Tensor, "shortRank5Tensor");
/*for_all( o.FixedCritter, [&](TestScalar &c, float f, const std::array<size_t,TestTensorFixed::NumIndices> &Dims ){
c = TestScalar{f,-f};
//std::cout << " a(" << Dims[0] << "," << Dims[1] << "," << Dims[2] << ")=" << c;
} );
for( auto &z : o.aFixedCritter )
for_all( z, [&](TestScalar &c, float f, const std::array<size_t,TestTensorFixed::NumIndices> &Dims ){
c = TestScalar{f,-f};
//std::cout << " a(" << Dims[0] << "," << Dims[1] << "," << Dims[2] << ")=" << c;
} );*/
ioTest<Hdf5Writer, Hdf5Reader, PerambIOTestClass>("iotest_object.h5", o, "PerambIOTestClass_object_instance_name");
//DumpMemoryOrder(o.Perambulator);
// Tensor of spin colour
LSCTensor l;
Val = 0;
for( int i = 0 ; i < l.dimension(0) ; i++)
for( int j = 0 ; j < l.dimension(1) ; j++)
for( int k = 0 ; k < l.dimension(2) ; k++)
for( int s = 0 ; s < Ns ; s++ )
for( int c = 0 ; c < Nc ; c++ )
{
l(i,j,k)()(s)(c) = Val;
Val += Inc;
}
ioTest<Hdf5Writer, Hdf5Reader, LSCTensor>("iotest_LSCTensor.h5", l, "LSCTensor_object_instance_name");
//dump_tensor(l, "l");
// Tensor of spin colour
LCMTensor l2;
Val = 0;
for( int i = 0 ; i < l2.dimension(0) ; i++)
for( int j = 0 ; j < l2.dimension(1) ; j++)
for( int k = 0 ; k < l2.dimension(2) ; k++)
for( int l = 0 ; l < Ns ; l++ )
for( int c = 0 ; c < Nc ; c++ )
for( int c2 = 0 ; c2 < Nc ; c2++ )
{
l2(i,j,k)(l)()(c,c2) = Val;
Val += Inc;
}
ioTest<Hdf5Writer, Hdf5Reader, LCMTensor>("iotest_LCMTensor.h5", l2, "LCMTensor_object_instance_name");
//dump_tensor(l2, "l2");
} }
#endif #endif