GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-10 18:59:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

play with inlining to get better performance when the compiler is not asked to optimize

Browse Source
This commit is contained in:
Benoit Jacob 2007-12-11 13:14:14 +00:00
parent 8117c9aa83
commit 936b0de9cc
16 changed files with 71 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
CMakeLists.txt
View File

@ -7,7 +7,6 @@ set(CMAKE_INCLUDE_CURRENT_DIR ON)
if(CMAKE_COMPILER_IS_GNUCXX) if(CMAKE_COMPILER_IS_GNUCXX)
if (CMAKE_SYSTEM_NAME MATCHES Linux) if (CMAKE_SYSTEM_NAME MATCHES Linux)
set ( CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wno-long-long -ansi -Wundef -Wcast-align -Werror-implicit-function-declaration -Wchar-subscripts -Wall -W -Wpointer-arith -Wwrite-strings -Wformat-security -Wmissing-format-attribute -fno-common -fstrict-aliasing")
set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wnon-virtual-dtor -Wno-long-long -ansi -Wundef -Wcast-align -Wchar-subscripts -Wall -W -Wpointer-arith -Wwrite-strings -Wformat-security -fno-exceptions -fno-check-new -fno-common -fstrict-aliasing") set ( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wnon-virtual-dtor -Wno-long-long -ansi -Wundef -Wcast-align -Wchar-subscripts -Wall -W -Wpointer-arith -Wwrite-strings -Wformat-security -fno-exceptions -fno-check-new -fno-common -fstrict-aliasing")
endif (CMAKE_SYSTEM_NAME MATCHES Linux) endif (CMAKE_SYSTEM_NAME MATCHES Linux)
endif (CMAKE_COMPILER_IS_GNUCXX) endif (CMAKE_COMPILER_IS_GNUCXX)

8
src/Core/Column.h
View File

@ -50,16 +50,16 @@ template<typename MatrixType> class Column
private: private:
const Column& _ref() const { return *this; } const Column& _ref() const { return *this; }
int _rows() const { return m_matrix.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_matrix.rows(); }
int _cols() const { return 1; } int _cols() const EIGEN_ALWAYS_INLINE { return 1; }
Scalar& _write(int row, int col) Scalar& _write(int row, int col) EIGEN_ALWAYS_INLINE
{ {
EIGEN_UNUSED(col); EIGEN_UNUSED(col);
return m_matrix.write(row, m_col); return m_matrix.write(row, m_col);
} }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
EIGEN_UNUSED(col); EIGEN_UNUSED(col);
return m_matrix.read(row, m_col); return m_matrix.read(row, m_col);

6
src/Core/Conjugate.h
View File

@ -47,10 +47,10 @@ template<typename MatrixType> class Conjugate
private: private:
const Conjugate& _ref() const { return *this; } const Conjugate& _ref() const { return *this; }
int _rows() const { return m_matrix.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_matrix.rows(); }
int _cols() const { return m_matrix.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_matrix.cols(); }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return conj(m_matrix.read(row, col)); return conj(m_matrix.read(row, col));
} }

6
src/Core/Difference.h
View File

@ -52,10 +52,10 @@ template<typename Lhs, typename Rhs> class Difference
private: private:
const Difference& _ref() const { return *this; } const Difference& _ref() const { return *this; }
int _rows() const { return m_lhs.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_lhs.rows(); }
int _cols() const { return m_lhs.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_lhs.cols(); }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return m_lhs.read(row, col) - m_rhs.read(row, col); return m_lhs.read(row, col) - m_rhs.read(row, col);
} }

8
src/Core/Matrix.h
View File

@ -40,21 +40,21 @@ class Matrix : public MatrixBase<_Scalar, Matrix<_Scalar, _Rows, _Cols> >,
static const int RowsAtCompileTime = _Rows, ColsAtCompileTime = _Cols; static const int RowsAtCompileTime = _Rows, ColsAtCompileTime = _Cols;
const Scalar* array() const const Scalar* array() const EIGEN_ALWAYS_INLINE
{ return Storage::m_array; } { return Storage::m_array; }
Scalar* array() Scalar* array() EIGEN_ALWAYS_INLINE
{ return Storage::m_array; } { return Storage::m_array; }
private: private:
Ref _ref() const { return Ref(*this); } Ref _ref() const { return Ref(*this); }
const Scalar& _read(int row, int col) const const Scalar& _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return array()[row + col * Storage::_rows()]; return array()[row + col * Storage::_rows()];
} }
Scalar& _write(int row, int col) Scalar& _write(int row, int col) EIGEN_ALWAYS_INLINE
{ {
return array()[row + col * Storage::_rows()]; return array()[row + col * Storage::_rows()];
} }

13
src/Core/MatrixBase.h
View File

@ -40,9 +40,12 @@ template<typename Scalar, typename Derived> class MatrixBase
typedef typename ForwardDecl<Derived>::Ref Ref; typedef typename ForwardDecl<Derived>::Ref Ref;
typedef typename NumTraits<Scalar>::Real RealScalar; typedef typename NumTraits<Scalar>::Real RealScalar;
int rows() const { return static_cast<const Derived *>(this)->_rows(); } int rows() const EIGEN_ALWAYS_INLINE
int cols() const { return static_cast<const Derived *>(this)->_cols(); } { return static_cast<const Derived *>(this)->_rows(); }
int size() const { return rows() * cols(); } int cols() const EIGEN_ALWAYS_INLINE
{ return static_cast<const Derived *>(this)->_cols(); }
int size() const EIGEN_ALWAYS_INLINE
{ return rows() * cols(); }
Ref ref() const Ref ref() const
{ return static_cast<const Derived *>(this)->_ref(); } { return static_cast<const Derived *>(this)->_ref(); }
@ -129,6 +132,7 @@ template<typename Scalar, typename Derived> class MatrixBase
Derived& operator/=(const std::complex<double>& other); Derived& operator/=(const std::complex<double>& other);
Scalar read(int row, int col, AssertLevel assertLevel = InternalDebugging) const Scalar read(int row, int col, AssertLevel assertLevel = InternalDebugging) const
EIGEN_ALWAYS_INLINE
{ {
eigen_assert(assertLevel, row >= 0 && row < rows() eigen_assert(assertLevel, row >= 0 && row < rows()
&& col >= 0 && col < cols()); && col >= 0 && col < cols());
@ -137,6 +141,7 @@ template<typename Scalar, typename Derived> class MatrixBase
Scalar operator()(int row, int col) const { return read(row, col, UserDebugging); } Scalar operator()(int row, int col) const { return read(row, col, UserDebugging); }
Scalar& write(int row, int col, AssertLevel assertLevel = InternalDebugging) Scalar& write(int row, int col, AssertLevel assertLevel = InternalDebugging)
EIGEN_ALWAYS_INLINE
{ {
eigen_assert(assertLevel, row >= 0 && row < rows() eigen_assert(assertLevel, row >= 0 && row < rows()
&& col >= 0 && col < cols()); && col >= 0 && col < cols());
@ -145,6 +150,7 @@ template<typename Scalar, typename Derived> class MatrixBase
Scalar& operator()(int row, int col) { return write(row, col, UserDebugging); } Scalar& operator()(int row, int col) { return write(row, col, UserDebugging); }
Scalar read(int index, AssertLevel assertLevel = InternalDebugging) const Scalar read(int index, AssertLevel assertLevel = InternalDebugging) const
EIGEN_ALWAYS_INLINE
{ {
eigen_assert(assertLevel, IsVector); eigen_assert(assertLevel, IsVector);
if(RowsAtCompileTime == 1) if(RowsAtCompileTime == 1)
@ -161,6 +167,7 @@ template<typename Scalar, typename Derived> class MatrixBase
Scalar operator[](int index) const { return read(index, UserDebugging); } Scalar operator[](int index) const { return read(index, UserDebugging); }
Scalar& write(int index, AssertLevel assertLevel = InternalDebugging) Scalar& write(int index, AssertLevel assertLevel = InternalDebugging)
EIGEN_ALWAYS_INLINE
{ {
eigen_assert(assertLevel, IsVector); eigen_assert(assertLevel, IsVector);
if(RowsAtCompileTime == 1) if(RowsAtCompileTime == 1)

8
src/Core/MatrixRef.h
View File

@ -40,15 +40,15 @@ template<typename MatrixType> class MatrixRef
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixRef) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixRef)
private: private:
int _rows() const { return m_matrix.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_matrix.rows(); }
int _cols() const { return m_matrix.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_matrix.cols(); }
const Scalar& _read(int row, int col) const const Scalar& _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return m_matrix._read(row, col); return m_matrix._read(row, col);
} }
Scalar& _write(int row, int col) Scalar& _write(int row, int col) EIGEN_ALWAYS_INLINE
{ {
return m_matrix.write(row, col); return m_matrix.write(row, col);
} }

16
src/Core/MatrixStorage.h
View File

@ -41,10 +41,10 @@ class MatrixStorage
assert(rows == RowsAtCompileTime && cols == ColsAtCompileTime); assert(rows == RowsAtCompileTime && cols == ColsAtCompileTime);
} }
int _rows() const int _rows() const EIGEN_ALWAYS_INLINE
{ return RowsAtCompileTime; } { return RowsAtCompileTime; }
int _cols() const int _cols() const EIGEN_ALWAYS_INLINE
{ return ColsAtCompileTime; } { return ColsAtCompileTime; }
public: public:
@ -80,10 +80,10 @@ class MatrixStorage<Scalar, Dynamic, ColsAtCompileTime>
m_rows = rows; m_rows = rows;
} }
int _rows() const int _rows() const EIGEN_ALWAYS_INLINE
{ return m_rows; } { return m_rows; }
int _cols() const int _cols() const EIGEN_ALWAYS_INLINE
{ return ColsAtCompileTime; } { return ColsAtCompileTime; }
public: public:
@ -124,10 +124,10 @@ class MatrixStorage<Scalar, RowsAtCompileTime, Dynamic>
m_cols = cols; m_cols = cols;
} }
int _rows() const int _rows() const EIGEN_ALWAYS_INLINE
{ return RowsAtCompileTime; } { return RowsAtCompileTime; }
int _cols() const int _cols() const EIGEN_ALWAYS_INLINE
{ return m_cols; } { return m_cols; }
public: public:
@ -168,10 +168,10 @@ class MatrixStorage<Scalar, Dynamic, Dynamic>
m_cols = cols; m_cols = cols;
} }
int _rows() const int _rows() const EIGEN_ALWAYS_INLINE
{ return m_rows; } { return m_rows; }
int _cols() const int _cols() const EIGEN_ALWAYS_INLINE
{ return m_cols; } { return m_cols; }
public: public:

22
src/Core/OperatorEquals.h
View File

@ -27,23 +27,23 @@
#ifndef EIGEN_OPERATOREQUALS_H #ifndef EIGEN_OPERATOREQUALS_H
#define EIGEN_OPERATOREQUALS_H #define EIGEN_OPERATOREQUALS_H
template<int UnrollCount, int Rows> struct OperatorEqualsUnroller template<typename Derived1, typename Derived2, int UnrollCount, int Rows>
struct OperatorEqualsUnroller
{ {
static const int col = (UnrollCount-1) / Rows; static const int col = (UnrollCount-1) / Rows;
static const int row = (UnrollCount-1) % Rows; static const int row = (UnrollCount-1) % Rows;
template <typename Derived1, typename Derived2>
static void run(Derived1 &dst, const Derived2 &src) static void run(Derived1 &dst, const Derived2 &src)
{ {
OperatorEqualsUnroller<UnrollCount-1, Rows>::run(dst, src); OperatorEqualsUnroller<Derived1, Derived2, UnrollCount-1, Rows>::run(dst, src);
dst.write(row, col) = src.read(row, col); dst.write(row, col) = src.read(row, col);
} }
}; };
// prevent buggy user code from causing an infinite recursion // prevent buggy user code from causing an infinite recursion
template<int UnrollCount> struct OperatorEqualsUnroller<UnrollCount, 0> template<typename Derived1, typename Derived2, int UnrollCount>
struct OperatorEqualsUnroller<Derived1, Derived2, UnrollCount, 0>
{ {
template <typename Derived1, typename Derived2>
static void run(Derived1 &dst, const Derived2 &src) static void run(Derived1 &dst, const Derived2 &src)
{ {
EIGEN_UNUSED(dst); EIGEN_UNUSED(dst);
@ -51,18 +51,18 @@ template<int UnrollCount> struct OperatorEqualsUnroller<UnrollCount, 0>
} }
}; };
template<int Rows> struct OperatorEqualsUnroller<1, Rows> template<typename Derived1, typename Derived2, int Rows>
struct OperatorEqualsUnroller<Derived1, Derived2, 1, Rows>
{ {
template <typename Derived1, typename Derived2>
static void run(Derived1 &dst, const Derived2 &src) static void run(Derived1 &dst, const Derived2 &src)
{ {
dst.write(0, 0) = src.read(0, 0); dst.write(0, 0) = src.read(0, 0);
} }
}; };
template<int Rows> struct OperatorEqualsUnroller<Dynamic, Rows> template<typename Derived1, typename Derived2, int Rows>
struct OperatorEqualsUnroller<Derived1, Derived2, Dynamic, Rows>
{ {
template <typename Derived1, typename Derived2>
static void run(Derived1 &dst, const Derived2 &src) static void run(Derived1 &dst, const Derived2 &src)
{ {
EIGEN_UNUSED(dst); EIGEN_UNUSED(dst);
@ -77,7 +77,9 @@ Derived& MatrixBase<Scalar, Derived>
{ {
assert(rows() == other.rows() && cols() == other.cols()); assert(rows() == other.rows() && cols() == other.cols());
if(EIGEN_UNROLLED_LOOPS && SizeAtCompileTime != Dynamic && SizeAtCompileTime <= 25) if(EIGEN_UNROLLED_LOOPS && SizeAtCompileTime != Dynamic && SizeAtCompileTime <= 25)
OperatorEqualsUnroller<SizeAtCompileTime, RowsAtCompileTime>::run(*this, other); OperatorEqualsUnroller
<Derived, OtherDerived, SizeAtCompileTime, RowsAtCompileTime>::run
(*static_cast<Derived*>(this), *static_cast<const OtherDerived*>(&other));
else else
for(int j = 0; j < cols(); j++) //traverse in column-dominant order for(int j = 0; j < cols(); j++) //traverse in column-dominant order
for(int i = 0; i < rows(); i++) for(int i = 0; i < rows(); i++)

6
src/Core/Opposite.h
View File

@ -47,10 +47,10 @@ template<typename MatrixType> class Opposite
private: private:
const Opposite& _ref() const { return *this; } const Opposite& _ref() const { return *this; }
int _rows() const { return m_matrix.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_matrix.rows(); }
int _cols() const { return m_matrix.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_matrix.cols(); }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return -(m_matrix.read(row, col)); return -(m_matrix.read(row, col));
} }

8
src/Core/Product.h
View File

@ -30,7 +30,7 @@ template<int Index, int Size, typename Lhs, typename Rhs>
struct ProductUnroller struct ProductUnroller
{ {
static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
typename Lhs::Scalar &res) typename Lhs::Scalar &res)
{ {
ProductUnroller<Index-1, Size, Lhs, Rhs>::run(row, col, lhs, rhs, res); ProductUnroller<Index-1, Size, Lhs, Rhs>::run(row, col, lhs, rhs, res);
res += lhs.read(row, Index) * rhs.read(Index, col); res += lhs.read(row, Index) * rhs.read(Index, col);
@ -41,7 +41,7 @@ template<int Size, typename Lhs, typename Rhs>
struct ProductUnroller<0, Size, Lhs, Rhs> struct ProductUnroller<0, Size, Lhs, Rhs>
{ {
static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
typename Lhs::Scalar &res) typename Lhs::Scalar &res)
{ {
res = lhs.read(row, 0) * rhs.read(0, col); res = lhs.read(row, 0) * rhs.read(0, col);
} }
@ -51,7 +51,7 @@ template<int Index, typename Lhs, typename Rhs>
struct ProductUnroller<Index, Dynamic, Lhs, Rhs> struct ProductUnroller<Index, Dynamic, Lhs, Rhs>
{ {
static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
typename Lhs::Scalar &res) typename Lhs::Scalar &res)
{ {
EIGEN_UNUSED(row); EIGEN_UNUSED(row);
EIGEN_UNUSED(col); EIGEN_UNUSED(col);
@ -66,7 +66,7 @@ template<int Index, typename Lhs, typename Rhs>
struct ProductUnroller<Index, 0, Lhs, Rhs> struct ProductUnroller<Index, 0, Lhs, Rhs>
{ {
static void run(int row, int col, const Lhs& lhs, const Rhs& rhs, static void run(int row, int col, const Lhs& lhs, const Rhs& rhs,
typename Lhs::Scalar &res) typename Lhs::Scalar &res)
{ {
EIGEN_UNUSED(row); EIGEN_UNUSED(row);
EIGEN_UNUSED(col); EIGEN_UNUSED(col);

8
src/Core/Row.h
View File

@ -57,16 +57,16 @@ template<typename MatrixType> class Row
private: private:
const Row& _ref() const { return *this; } const Row& _ref() const { return *this; }
int _rows() const { return 1; } int _rows() const EIGEN_ALWAYS_INLINE { return 1; }
int _cols() const { return m_matrix.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_matrix.cols(); }
Scalar& _write(int row, int col) Scalar& _write(int row, int col) EIGEN_ALWAYS_INLINE
{ {
EIGEN_UNUSED(row); EIGEN_UNUSED(row);
return m_matrix.write(m_row, col); return m_matrix.write(m_row, col);
} }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
EIGEN_UNUSED(row); EIGEN_UNUSED(row);
return m_matrix.read(m_row, col); return m_matrix.read(m_row, col);

6
src/Core/ScalarMultiple.h
View File

@ -48,10 +48,10 @@ template<typename MatrixType> class ScalarMultiple
private: private:
const ScalarMultiple& _ref() const { return *this; } const ScalarMultiple& _ref() const { return *this; }
int _rows() const { return m_matrix.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_matrix.rows(); }
int _cols() const { return m_matrix.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_matrix.cols(); }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return m_matrix.read(row, col) * m_scalar; return m_matrix.read(row, col) * m_scalar;
} }

6
src/Core/Sum.h
View File

@ -52,10 +52,10 @@ template<typename Lhs, typename Rhs> class Sum
private: private:
const Sum& _ref() const { return *this; } const Sum& _ref() const { return *this; }
int _rows() const { return m_lhs.rows(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_lhs.rows(); }
int _cols() const { return m_lhs.cols(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_lhs.cols(); }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return m_lhs.read(row, col) + m_rhs.read(row, col); return m_lhs.read(row, col) + m_rhs.read(row, col);
} }

8
src/Core/Transpose.h
View File

@ -46,15 +46,15 @@ template<typename MatrixType> class Transpose
private: private:
const Transpose& _ref() const { return *this; } const Transpose& _ref() const { return *this; }
int _rows() const { return m_matrix.cols(); } int _rows() const EIGEN_ALWAYS_INLINE { return m_matrix.cols(); }
int _cols() const { return m_matrix.rows(); } int _cols() const EIGEN_ALWAYS_INLINE { return m_matrix.rows(); }
Scalar& _write(int row, int col) Scalar& _write(int row, int col) EIGEN_ALWAYS_INLINE
{ {
return m_matrix.write(col, row); return m_matrix.write(col, row);
} }
Scalar _read(int row, int col) const Scalar _read(int row, int col) const EIGEN_ALWAYS_INLINE
{ {
return m_matrix.read(col, row); return m_matrix.read(col, row);
} }

4
test/main.cpp
View File

@ -32,8 +32,8 @@ int main(int argc, char *argv[])
bool has_set_repeat = false; bool has_set_repeat = false;
bool has_set_seed = false; bool has_set_seed = false;
bool need_help = false; bool need_help = false;
unsigned int seed; unsigned int seed = 0;
int repeat; int repeat = 0;
QStringList args = QCoreApplication::instance()->arguments(); QStringList args = QCoreApplication::instance()->arguments();
args.takeFirst(); // throw away the first argument (path to executable) args.takeFirst(); // throw away the first argument (path to executable)