* Draft of a eigenvalues solver

(does not support complex and does not re-use the QR decomposition)

* Rewrite the cache friendly product to have only one instance per scalar type !
  This significantly speeds up compilation time and reduces executable size.
  The current drawback is that some trivial expressions might be
  evaluated like conjugate or negate.

* Renamed "cache optimal" to "cache friendly"

* Added the ability to directly access matrix data of some expressions via:
  - the stride()/_stride() methods
  - DirectAccessBit flag (replace ReferencableBit)
This commit is contained in:
Gael Guennebaud 2008-05-12 10:23:09 +00:00
parent dca416cace
commit 45cda6704a
15 changed files with 1286 additions and 364 deletions

View File

@ -18,9 +18,9 @@
#ifdef __SSE3__
#include <pmmintrin.h>
#endif
//#ifdef __SSSE3__
//#include <tmmintrin.h>
//#endif
#ifdef __SSSE3__
#include <tmmintrin.h>
#endif
#elif (defined __ALTIVEC__)
#define EIGEN_VECTORIZE
#define EIGEN_VECTORIZE_ALTIVEC
@ -69,7 +69,7 @@ namespace Eigen {
#include "src/Core/CwiseBinaryOp.h"
#include "src/Core/CwiseUnaryOp.h"
#include "src/Core/CwiseNullaryOp.h"
#if (defined EIGEN_WIP_PRODUCT) && (defined EIGEN_VECTORIZE)
#if (defined EIGEN_WIP_PRODUCT)
#include "src/Core/ProductWIP.h"
#else
#include "src/Core/Product.h"

View File

@ -6,6 +6,7 @@
namespace Eigen {
#include "Eigen/src/QR/QR.h"
#include "Eigen/src/QR/EigenSolver.h"
} // namespace Eigen

View File

@ -71,7 +71,7 @@ struct ei_traits<Block<MatrixType, BlockRows, BlockCols> >
|| (ColsAtCompileTime != Dynamic && MatrixType::ColsAtCompileTime == Dynamic))
? ~LargeBit
: ~(unsigned int)0,
Flags = MatrixType::Flags & (DefaultLostFlagMask | VectorizableBit | ReferencableBit) & FlagsMaskLargeBit,
Flags = MatrixType::Flags & (DefaultLostFlagMask | VectorizableBit | DirectAccessBit) & FlagsMaskLargeBit,
CoeffReadCost = MatrixType::CoeffReadCost
};
};
@ -132,6 +132,8 @@ template<typename MatrixType, int BlockRows, int BlockCols> class Block
int _rows() const { return m_blockRows.value(); }
int _cols() const { return m_blockCols.value(); }
int _stride(void) const { return m_matrix.stride(); }
Scalar& _coeffRef(int row, int col)
{
return m_matrix.const_cast_derived()

View File

@ -0,0 +1,353 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra. Eigen itself is part of the KDE project.
//
// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_CACHE_FRIENDLY_PRODUCT_H
#define EIGEN_CACHE_FRIENDLY_PRODUCT_H
template<typename Scalar>
static void ei_cache_friendly_product(
int _rows, int _cols, int depth,
bool _lhsRowMajor, const Scalar* _lhs, int _lhsStride,
bool _rhsRowMajor, const Scalar* _rhs, int _rhsStride,
bool resRowMajor, Scalar* res, int resStride)
{
const Scalar* __restrict__ lhs;
const Scalar* __restrict__ rhs;
int lhsStride, rhsStride, rows, cols;
bool lhsRowMajor;
if (resRowMajor)
{
lhs = _rhs;
rhs = _lhs;
lhsStride = _rhsStride;
rhsStride = _lhsStride;
cols = _rows;
rows = _cols;
lhsRowMajor = _rhsRowMajor;
ei_assert(_lhsRowMajor);
}
else
{
lhs = _lhs;
rhs = _rhs;
lhsStride = _lhsStride;
rhsStride = _rhsStride;
rows = _rows;
cols = _cols;
lhsRowMajor = _lhsRowMajor;
ei_assert(!_rhsRowMajor);
}
typedef typename ei_packet_traits<Scalar>::type PacketType;
enum {
PacketSize = sizeof(PacketType)/sizeof(Scalar),
#if (defined __i386__)
// i386 architecture provides only 8 xmm registers,
// so let's reduce the max number of rows processed at once.
MaxBlockRows = 4,
MaxBlockRows_ClampingMask = 0xFFFFFC,
#else
MaxBlockRows = 8,
MaxBlockRows_ClampingMask = 0xFFFFF8,
#endif
// maximal size of the blocks fitted in L2 cache
MaxL2BlockSize = EIGEN_TUNE_FOR_L2_CACHE_SIZE / sizeof(Scalar)
};
//const bool rhsIsAligned = (PacketSize==1) || (((rhsStride%PacketSize) == 0) && (size_t(rhs)%16==0));
const bool resIsAligned = (PacketSize==1) || (((resStride%PacketSize) == 0) && (size_t(res)%16==0));
const int remainingSize = depth % PacketSize;
const int size = depth - remainingSize; // third dimension of the product clamped to packet boundaries
const int l2BlockRows = MaxL2BlockSize > rows ? rows : MaxL2BlockSize;
const int l2BlockCols = MaxL2BlockSize > cols ? cols : MaxL2BlockSize;
const int l2BlockSize = MaxL2BlockSize > size ? size : MaxL2BlockSize;
Scalar* __restrict__ block = (Scalar*)alloca(sizeof(Scalar)*l2BlockRows*size);
Scalar* __restrict__ rhsCopy = (Scalar*)alloca(sizeof(Scalar)*l2BlockSize);
// loops on each L2 cache friendly blocks of the result
for(int l2i=0; l2i<rows; l2i+=l2BlockRows)
{
const int l2blockRowEnd = std::min(l2i+l2BlockRows, rows);
const int l2blockRowEndBW = l2blockRowEnd & MaxBlockRows_ClampingMask; // end of the rows aligned to bw
const int l2blockRemainingRows = l2blockRowEnd - l2blockRowEndBW; // number of remaining rows
//const int l2blockRowEndBWPlusOne = l2blockRowEndBW + (l2blockRemainingRows?0:MaxBlockRows);
// build a cache friendly blocky matrix
int count = 0;
// copy l2blocksize rows of m_lhs to blocks of ps x bw
asm("#eigen begin buildblocks");
for(int l2k=0; l2k<size; l2k+=l2BlockSize)
{
const int l2blockSizeEnd = std::min(l2k+l2BlockSize, size);
for (int i = l2i; i<l2blockRowEndBW/*PlusOne*/; i+=MaxBlockRows)
{
// TODO merge the if l2blockRemainingRows
// const int blockRows = std::min(i+MaxBlockRows, rows) - i;
for (int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
// TODO write these loops using meta unrolling
// negligible for large matrices but useful for small ones
if (lhsRowMajor)
{
for (int w=0; w<MaxBlockRows; ++w)
for (int s=0; s<PacketSize; ++s)
block[count++] = lhs[(i+w)*lhsStride + (k+s)];
}
else
{
for (int w=0; w<MaxBlockRows; ++w)
for (int s=0; s<PacketSize; ++s)
block[count++] = lhs[(i+w) + (k+s)*lhsStride];
}
}
}
if (l2blockRemainingRows>0)
{
for (int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
if (lhsRowMajor)
{
for (int w=0; w<l2blockRemainingRows; ++w)
for (int s=0; s<PacketSize; ++s)
block[count++] = lhs[(l2blockRowEndBW+w)*lhsStride + (k+s)];
}
else
{
for (int w=0; w<l2blockRemainingRows; ++w)
for (int s=0; s<PacketSize; ++s)
block[count++] = lhs[(l2blockRowEndBW+w) + (k+s)*lhsStride];
}
}
}
}
asm("#eigen end buildblocks");
for(int l2j=0; l2j<cols; l2j+=l2BlockCols)
{
int l2blockColEnd = std::min(l2j+l2BlockCols, cols);
for(int l2k=0; l2k<size; l2k+=l2BlockSize)
{
// acumulate bw rows of lhs time a single column of rhs to a bw x 1 block of res
int l2blockSizeEnd = std::min(l2k+l2BlockSize, size);
// for each bw x 1 result's block
for(int l1i=l2i; l1i<l2blockRowEndBW; l1i+=MaxBlockRows)
{
for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
{
int offsetblock = l2k * (l2blockRowEnd-l2i) + (l1i-l2i)*(l2blockSizeEnd-l2k) - l2k*MaxBlockRows;
const Scalar* __restrict__ localB = &block[offsetblock];
const Scalar* __restrict__ rhsColumn = &(rhs[l1j*rhsStride]);
// copy unaligned rhs data
// YES it seems to be faster to copy some part of rhs multiple times
// to aligned memory rather than using unligned load.
// Moreover this avoids a "if" in the most nested loop :)
if (PacketSize>1 && size_t(rhsColumn)%16)
{
int count = 0;
for (int k = l2k; k<l2blockSizeEnd; ++k)
{
rhsCopy[count++] = rhsColumn[k];
}
rhsColumn = &(rhsCopy[-l2k]);
}
PacketType dst[MaxBlockRows];
dst[0] = ei_pset1(Scalar(0.));
dst[1] = dst[0];
dst[2] = dst[0];
dst[3] = dst[0];
if (MaxBlockRows==8)
{
dst[4] = dst[0];
dst[5] = dst[0];
dst[6] = dst[0];
dst[7] = dst[0];
}
PacketType tmp;
asm("#eigen begincore");
for(int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
tmp = ei_pload(&rhsColumn[k]);
dst[0] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows ])), dst[0]);
dst[1] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+ PacketSize])), dst[1]);
dst[2] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+2*PacketSize])), dst[2]);
dst[3] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+3*PacketSize])), dst[3]);
if (MaxBlockRows==8)
{
dst[4] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+4*PacketSize])), dst[4]);
dst[5] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+5*PacketSize])), dst[5]);
dst[6] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+6*PacketSize])), dst[6]);
dst[7] = ei_pmadd(tmp, ei_pload(&(localB[k*MaxBlockRows+7*PacketSize])), dst[7]);
}
}
Scalar* __restrict__ localRes = &(res[l1i + l1j*resStride]);
if (PacketSize>1 && resIsAligned)
{
ei_pstore(&(localRes[0]), ei_padd(ei_pload(&(localRes[0])), ei_preduxp(dst)));
if (PacketSize==2)
ei_pstore(&(localRes[2]), ei_padd(ei_pload(&(localRes[2])), ei_preduxp(&(dst[2]))));
if (MaxBlockRows==8)
{
ei_pstore(&(localRes[4]), ei_padd(ei_pload(&(localRes[4])), ei_preduxp(&(dst[4]))));
if (PacketSize==2)
ei_pstore(&(localRes[6]), ei_padd(ei_pload(&(localRes[6])), ei_preduxp(&(dst[6]))));
}
}
else
{
localRes[0] += ei_predux(dst[0]);
localRes[1] += ei_predux(dst[1]);
localRes[2] += ei_predux(dst[2]);
localRes[3] += ei_predux(dst[3]);
if (MaxBlockRows==8)
{
localRes[4] += ei_predux(dst[4]);
localRes[5] += ei_predux(dst[5]);
localRes[6] += ei_predux(dst[6]);
localRes[7] += ei_predux(dst[7]);
}
}
asm("#eigen endcore");
}
}
if (l2blockRemainingRows>0)
{
int offsetblock = l2k * (l2blockRowEnd-l2i) + (l2blockRowEndBW-l2i)*(l2blockSizeEnd-l2k) - l2k*l2blockRemainingRows;
const Scalar* localB = &block[offsetblock];
asm("#eigen begin dynkernel");
for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
{
const Scalar* __restrict__ rhsColumn = &(rhs[l1j*rhsStride]);
// copy unaligned rhs data
if (PacketSize>1 && size_t(rhsColumn)%16)
{
int count = 0;
for (int k = l2k; k<l2blockSizeEnd; ++k)
{
rhsCopy[count++] = rhsColumn[k];
}
rhsColumn = &(rhsCopy[-l2k]);
}
PacketType dst[MaxBlockRows];
dst[0] = ei_pset1(Scalar(0.));
dst[1] = dst[0];
dst[2] = dst[0];
dst[3] = dst[0];
if (MaxBlockRows>4)
{
dst[4] = dst[0];
dst[5] = dst[0];
dst[6] = dst[0];
dst[7] = dst[0];
}
// let's declare a few other temporary registers
PacketType tmp;
for(int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
tmp = ei_pload(&rhsColumn[k]);
dst[0] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows ])), dst[0]);
if (l2blockRemainingRows>=2) dst[1] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+ PacketSize])), dst[1]);
if (l2blockRemainingRows>=3) dst[2] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+2*PacketSize])), dst[2]);
if (l2blockRemainingRows>=4) dst[3] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+3*PacketSize])), dst[3]);
if (MaxBlockRows>4)
{
if (l2blockRemainingRows>=5) dst[4] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+4*PacketSize])), dst[4]);
if (l2blockRemainingRows>=6) dst[5] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+5*PacketSize])), dst[5]);
if (l2blockRemainingRows>=7) dst[6] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+6*PacketSize])), dst[6]);
if (l2blockRemainingRows>=8) dst[7] = ei_pmadd(tmp, ei_pload(&(localB[k*l2blockRemainingRows+7*PacketSize])), dst[7]);
}
}
Scalar* __restrict__ localRes = &(res[l2blockRowEndBW + l1j*resStride]);
// process the remaining rows once at a time
localRes[0] += ei_predux(dst[0]);
if (l2blockRemainingRows>=2) localRes[1] += ei_predux(dst[1]);
if (l2blockRemainingRows>=3) localRes[2] += ei_predux(dst[2]);
if (l2blockRemainingRows>=4) localRes[3] += ei_predux(dst[3]);
if (MaxBlockRows>4)
{
if (l2blockRemainingRows>=5) localRes[4] += ei_predux(dst[4]);
if (l2blockRemainingRows>=6) localRes[5] += ei_predux(dst[5]);
if (l2blockRemainingRows>=7) localRes[6] += ei_predux(dst[6]);
if (l2blockRemainingRows>=8) localRes[7] += ei_predux(dst[7]);
}
asm("#eigen end dynkernel");
}
}
}
}
}
if (PacketSize>1 && remainingSize)
{
if (lhsRowMajor)
{
for (int j=0; j<cols; ++j)
for (int i=0; i<rows; ++i)
{
Scalar tmp = lhs[i*lhsStride+size] * rhs[j*rhsStride+size];
for (int k=1; k<remainingSize; ++k)
tmp += lhs[i*lhsStride+size+k] * rhs[j*rhsStride+size+k];
res[i+j*resStride] += tmp;
}
}
else
{
for (int j=0; j<cols; ++j)
for (int i=0; i<rows; ++i)
{
Scalar tmp = lhs[i+size*lhsStride] * rhs[j*rhsStride+size];
for (int k=1; k<remainingSize; ++k)
tmp += lhs[i+(size+k)*lhsStride] * rhs[j*rhsStride+size+k];
res[i+j*resStride] += tmp;
}
}
}
}
#endif // EIGEN_CACHE_FRIENDLY_PRODUCT_H

View File

@ -47,7 +47,7 @@ struct ei_traits<Map<MatrixType> >
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
Flags = MatrixType::Flags & (DefaultLostFlagMask | ReferencableBit),
Flags = MatrixType::Flags & (DefaultLostFlagMask | DirectAccessBit),
CoeffReadCost = NumTraits<Scalar>::ReadCost
};
};

View File

@ -46,7 +46,7 @@ inline int ei_log(int) { ei_assert(false); return 0; }
inline int ei_sin(int) { ei_assert(false); return 0; }
inline int ei_cos(int) { ei_assert(false); return 0; }
#if EIGEN_GNUC_AT_LEAST(4,3)
#if EIGEN_GNUC_AT_LEAST(4,2)
inline int ei_pow(int x, int y) { return int(std::pow(double(x), y)); }
#else
inline int ei_pow(int x, int y) { return std::pow(x, y); }

View File

@ -100,6 +100,14 @@ class Matrix : public MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Flags, _MaxRows,
int _rows() const { return m_storage.rows(); }
int _cols() const { return m_storage.cols(); }
int _stride(void) const
{
if(Flags & RowMajorBit)
return m_storage.cols();
else
return m_storage.rows();
}
const Scalar& _coeff(int row, int col) const
{
if(Flags & RowMajorBit)

View File

@ -185,7 +185,7 @@ template<typename Derived> class MatrixBase
/** Overloaded for optimal product evaluation */
template<typename Derived1, typename Derived2>
Derived& lazyAssign(const Product<Derived1,Derived2,CacheOptimalProduct>& product);
Derived& lazyAssign(const Product<Derived1,Derived2,CacheFriendlyProduct>& product);
CommaInitializer operator<< (const Scalar& s);
@ -419,6 +419,13 @@ template<typename Derived> class MatrixBase
const Lazy<Derived> lazy() const;
const Temporary<Derived> temporary() const;
/** \returns number of elements to skip to pass from one row (resp. column) to another
* for a row-major (resp. column-major) matrix.
* Combined with coeffRef() and the compile times flags, it allows a direct access to the data
* of the underlying matrix.
*/
int stride(void) const { return derived()._stride(); }
//@}
/// \name Coefficient-wise operations

View File

@ -60,6 +60,12 @@ struct ei_product_unroller<Index, 0, Lhs, Rhs>
static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
};
template<typename Lhs, typename Rhs>
struct ei_product_unroller<0, Dynamic, Lhs, Rhs>
{
static void run(int, int, const Lhs&, const Rhs&, typename Lhs::Scalar&) {}
};
template<bool RowMajor, int Index, int Size, typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller;
@ -113,6 +119,12 @@ struct ei_packet_product_unroller<false, Index, Dynamic, Lhs, Rhs, PacketScalar>
static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
};
template<typename Lhs, typename Rhs, typename PacketScalar>
struct ei_packet_product_unroller<false, 0, Dynamic, Lhs, Rhs, PacketScalar>
{
static void run(int, int, const Lhs&, const Rhs&, PacketScalar&) {}
};
template<typename Product, bool RowMajor = true> struct ProductPacketCoeffImpl {
inline static typename Product::PacketScalar execute(const Product& product, int row, int col)
{ return product._packetCoeffRowMajor(row,col); }
@ -142,7 +154,7 @@ template<typename Lhs, typename Rhs> struct ei_product_eval_mode
enum{ value = Lhs::MaxRowsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
&& Rhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
&& (!( (Lhs::Flags&RowMajorBit) && ((Rhs::Flags&RowMajorBit) ^ RowMajorBit)))
? CacheOptimalProduct : NormalProduct };
? CacheFriendlyProduct : NormalProduct };
};
template<typename Lhs, typename Rhs, int EvalMode>
@ -166,7 +178,7 @@ struct ei_traits<Product<Lhs, Rhs, EvalMode> >
_LhsVectorizable = (!(LhsFlags & RowMajorBit)) && (LhsFlags & VectorizableBit) && (RowsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
_Vectorizable = (_LhsVectorizable || _RhsVectorizable) ? 1 : 0,
_RowMajor = (RhsFlags & RowMajorBit)
&& (EvalMode==(int)CacheOptimalProduct ? (int)LhsFlags & RowMajorBit : (!_LhsVectorizable)),
&& (EvalMode==(int)CacheFriendlyProduct ? (int)LhsFlags & RowMajorBit : (!_LhsVectorizable)),
_LostBits = DefaultLostFlagMask & ~(
(_RowMajor ? 0 : RowMajorBit)
| ((RowsAtCompileTime == Dynamic || ColsAtCompileTime == Dynamic) ? 0 : LargeBit)),
@ -312,7 +324,7 @@ MatrixBase<Derived>::operator*=(const MatrixBase<OtherDerived> &other)
template<typename Derived>
template<typename Lhs, typename Rhs>
Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheOptimalProduct>& product)
Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheFriendlyProduct>& product)
{
product.template _cacheOptimalEval<Derived, Aligned>(derived(),
#ifdef EIGEN_VECTORIZE

View File

@ -26,9 +26,7 @@
#ifndef EIGEN_PRODUCT_H
#define EIGEN_PRODUCT_H
#ifndef EIGEN_VECTORIZE
#error you must enable vectorization to try this experimental product implementation
#endif
#include "CacheFriendlyProduct.h"
template<int Index, int Size, typename Lhs, typename Rhs>
struct ei_product_unroller
@ -145,7 +143,7 @@ template<typename Lhs, typename Rhs> struct ei_product_eval_mode
{
enum{ value = Lhs::MaxRowsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
&& Rhs::MaxColsAtCompileTime >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
? CacheOptimalProduct : NormalProduct };
? CacheFriendlyProduct : NormalProduct };
};
template<typename T> class ei_product_eval_to_column_major
@ -173,7 +171,22 @@ template<typename T, int n=1> struct ei_product_nested_rhs
typename ei_meta_if<
(ei_traits<T>::Flags & EvalBeforeNestingBit)
|| (ei_traits<T>::Flags & RowMajorBit)
|| (!(ei_traits<T>::Flags & ReferencableBit))
|| (!(ei_traits<T>::Flags & DirectAccessBit))
|| (n+1) * NumTraits<typename ei_traits<T>::Scalar>::ReadCost < (n-1) * T::CoeffReadCost,
typename ei_product_eval_to_column_major<T>::type,
const T&
>::ret
>::ret type;
};
template<typename T, int n=1> struct ei_product_nested_lhs
{
typedef typename ei_meta_if<
ei_is_temporary<T>::ret && !(ei_traits<T>::Flags & RowMajorBit),
T,
typename ei_meta_if<
(ei_traits<T>::Flags & EvalBeforeNestingBit)
|| (!(ei_traits<T>::Flags & DirectAccessBit))
|| (n+1) * NumTraits<typename ei_traits<T>::Scalar>::ReadCost < (n-1) * T::CoeffReadCost,
typename ei_product_eval_to_column_major<T>::type,
const T&
@ -187,9 +200,12 @@ struct ei_traits<Product<Lhs, Rhs, EvalMode> >
typedef typename Lhs::Scalar Scalar;
// the cache friendly product evals lhs once only
// FIXME what to do if we chose to dynamically call the normal product from the cache friendly one for small matrices ?
typedef typename ei_nested<Lhs, EvalMode==CacheOptimalProduct ? 0 : Rhs::ColsAtCompileTime>::type LhsNested;
typedef typename ei_meta_if<EvalMode==CacheFriendlyProduct,
typename ei_product_nested_lhs<Rhs,0>::type,
typename ei_nested<Lhs,Rhs::ColsAtCompileTime>::type>::ret LhsNested;
// NOTE that rhs must be ColumnMajor, so we might need a special nested type calculation
typedef typename ei_meta_if<EvalMode==CacheOptimalProduct,
typedef typename ei_meta_if<EvalMode==CacheFriendlyProduct,
typename ei_product_nested_rhs<Rhs,Lhs::RowsAtCompileTime>::type,
typename ei_nested<Rhs,Lhs::RowsAtCompileTime>::type>::ret RhsNested;
typedef typename ei_unref<LhsNested>::type _LhsNested;
@ -209,7 +225,7 @@ struct ei_traits<Product<Lhs, Rhs, EvalMode> >
_LhsVectorizable = (!(LhsFlags & RowMajorBit)) && (LhsFlags & VectorizableBit) && (RowsAtCompileTime % ei_packet_traits<Scalar>::size == 0),
_Vectorizable = (_LhsVectorizable || _RhsVectorizable) ? 0 : 0,
_RowMajor = (RhsFlags & RowMajorBit)
&& (EvalMode==(int)CacheOptimalProduct ? (int)LhsFlags & RowMajorBit : (!_LhsVectorizable)),
&& (EvalMode==(int)CacheFriendlyProduct ? (int)LhsFlags & RowMajorBit : (!_LhsVectorizable)),
_LostBits = DefaultLostFlagMask & ~(
(_RowMajor ? 0 : RowMajorBit)
| ((RowsAtCompileTime == Dynamic || ColsAtCompileTime == Dynamic) ? 0 : LargeBit)),
@ -241,19 +257,7 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
typedef typename ei_traits<Product>::_RhsNested _RhsNested;
enum {
PacketSize = ei_packet_traits<Scalar>::size,
#if (defined __i386__)
// i386 architectures provides only 8 xmmm register,
// so let's reduce the max number of rows processed at once.
// NOTE that so far the maximal supported value is 8.
MaxBlockRows = 4,
MaxBlockRows_ClampingMask = 0xFFFFFC,
#else
MaxBlockRows = 8,
MaxBlockRows_ClampingMask = 0xFFFFF8,
#endif
// maximal size of the blocks fitted in L2 cache
MaxL2BlockSize = EIGEN_TUNE_FOR_L2_CACHE_SIZE / sizeof(Scalar)
PacketSize = ei_packet_traits<Scalar>::size
};
Product(const Lhs& lhs, const Rhs& rhs)
@ -327,14 +331,8 @@ template<typename Lhs, typename Rhs, int EvalMode> class Product : ei_no_assignm
}
/** \internal */
template<typename DestDerived, int RhsAlignment, int ResAlignment>
void _cacheFriendlyEvalImpl(DestDerived& res) const __attribute__ ((noinline));
/** \internal */
template<typename DestDerived, int RhsAlignment, int ResAlignment, int BlockRows>
void _cacheFriendlyEvalKernel(DestDerived& res,
int l2i, int l2j, int l2k, int l1i,
int l2blockRowEnd, int l2blockColEnd, int l2blockSizeEnd, const Scalar* block) const EIGEN_DONT_INLINE;
template<typename DestDerived, int RhsAlignment>
void _cacheFriendlyEvalImpl(DestDerived& res) const EIGEN_DONT_INLINE;
protected:
const LhsNested m_lhs;
@ -370,7 +368,7 @@ MatrixBase<Derived>::operator*=(const MatrixBase<OtherDerived> &other)
template<typename Derived>
template<typename Lhs, typename Rhs>
Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheOptimalProduct>& product)
Derived& MatrixBase<Derived>::lazyAssign(const Product<Lhs,Rhs,CacheFriendlyProduct>& product)
{
product._cacheFriendlyEval(derived());
return derived();
@ -380,326 +378,16 @@ template<typename Lhs, typename Rhs, int EvalMode>
template<typename DestDerived>
void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEval(DestDerived& res) const
{
const bool rhsIsAligned = (m_lhs.cols()%PacketSize == 0);
const bool resIsAligned = ((_rows()%PacketSize) == 0);
if (rhsIsAligned && resIsAligned)
_cacheFriendlyEvalImpl<DestDerived, Aligned, Aligned>(res);
else if (rhsIsAligned && (!resIsAligned))
_cacheFriendlyEvalImpl<DestDerived, Aligned, UnAligned>(res);
else if ((!rhsIsAligned) && resIsAligned)
_cacheFriendlyEvalImpl<DestDerived, UnAligned, Aligned>(res);
else
_cacheFriendlyEvalImpl<DestDerived, UnAligned, UnAligned>(res);
}
template<typename Lhs, typename Rhs, int EvalMode>
template<typename DestDerived, int RhsAlignment, int ResAlignment, int BlockRows>
void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEvalKernel(DestDerived& res,
int l2i, int l2j, int l2k, int l1i,
int l2blockRowEnd, int l2blockColEnd, int l2blockSizeEnd, const Scalar* block) const
{
asm("#eigen begin kernel");
ei_internal_assert(BlockRows<=8);
// NOTE: sounds like we cannot rely on meta-unrolling to access dst[I] without enforcing GCC
// to create the dst's elements in memory, hence killing the performance.
for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
{
int offsetblock = l2k * (l2blockRowEnd-l2i) + (l1i-l2i)*(l2blockSizeEnd-l2k) - l2k*BlockRows;
const Scalar* localB = &block[offsetblock];
// int l1jsize = l1j * m_lhs.cols(); //TODO find a better way to optimize address computation ?
Scalar* rhsColumn = &(m_rhs.const_cast_derived().coeffRef(0, l1j));
// don't worry, dst is a set of registers
PacketScalar dst[BlockRows];
dst[0] = ei_pset1(Scalar(0.));
switch(BlockRows)
{
case 8: dst[7] = dst[0];
case 7: dst[6] = dst[0];
case 6: dst[5] = dst[0];
case 5: dst[4] = dst[0];
case 4: dst[3] = dst[0];
case 3: dst[2] = dst[0];
case 2: dst[1] = dst[0];
default: break;
}
// let's declare a few other temporary registers
PacketScalar tmp, tmp1;
// unaligned loads are expensive, therefore let's preload the next element in advance
if (RhsAlignment==UnAligned)
//tmp1 = ei_ploadu(&m_rhs.data()[l1jsize+l2k]);
tmp1 = ei_ploadu(&rhsColumn[l2k]);
for(int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
// FIXME if we don't cache l1j*m_lhs.cols() then the performance are poor,
// let's directly access to the data
//PacketScalar tmp = m_rhs.template packetCoeff<Aligned>(k, l1j);
if (RhsAlignment==Aligned)
{
//tmp = ei_pload(&m_rhs.data()[l1jsize + k]);
tmp = ei_pload(&rhsColumn[k]);
}
else
{
tmp = tmp1;
if (k+PacketSize<l2blockSizeEnd)
//tmp1 = ei_ploadu(&m_rhs.data()[l1jsize + k+PacketSize]);
tmp1 = ei_ploadu(&rhsColumn[k+PacketSize]);
}
dst[0] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows ])), dst[0]);
if (BlockRows>=2) dst[1] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+ PacketSize])), dst[1]);
if (BlockRows>=3) dst[2] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+2*PacketSize])), dst[2]);
if (BlockRows>=4) dst[3] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+3*PacketSize])), dst[3]);
if (BlockRows>=5) dst[4] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+4*PacketSize])), dst[4]);
if (BlockRows>=6) dst[5] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+5*PacketSize])), dst[5]);
if (BlockRows>=7) dst[6] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+6*PacketSize])), dst[6]);
if (BlockRows>=8) dst[7] = ei_pmadd(tmp, ei_pload(&(localB[k*BlockRows+7*PacketSize])), dst[7]);
}
enum {
// Number of rows we can reduce per packet
PacketRows = (ResAlignment==Aligned && PacketSize>1) ? (BlockRows / PacketSize) : 0,
// First row index from which we have to to do redux once at a time
RemainingStart = PacketSize * PacketRows
};
// we have up to 4 packets (for doubles: 8 rows / 2)
if (PacketRows>=1)
res.template writePacketCoeff<Aligned>(l1i, l1j,
ei_padd(res.template packetCoeff<Aligned>(l1i, l1j), ei_preduxp(&(dst[0]))));
if (PacketRows>=2)
res.template writePacketCoeff<Aligned>(l1i+PacketSize, l1j,
ei_padd(res.template packetCoeff<Aligned>(l1i+PacketSize, l1j), ei_preduxp(&(dst[PacketSize]))));
if (PacketRows>=3)
res.template writePacketCoeff<Aligned>(l1i+2*PacketSize, l1j,
ei_padd(res.template packetCoeff<Aligned>(l1i+2*PacketSize, l1j), ei_preduxp(&(dst[2*PacketSize]))));
if (PacketRows>=4)
res.template writePacketCoeff<Aligned>(l1i+3*PacketSize, l1j,
ei_padd(res.template packetCoeff<Aligned>(l1i+3*PacketSize, l1j), ei_preduxp(&(dst[3*PacketSize]))));
// process the remaining rows one at a time
if (RemainingStart<=0 && BlockRows>=1) res.coeffRef(l1i+0, l1j) += ei_predux(dst[0]);
if (RemainingStart<=1 && BlockRows>=2) res.coeffRef(l1i+1, l1j) += ei_predux(dst[1]);
if (RemainingStart<=2 && BlockRows>=3) res.coeffRef(l1i+2, l1j) += ei_predux(dst[2]);
if (RemainingStart<=3 && BlockRows>=4) res.coeffRef(l1i+3, l1j) += ei_predux(dst[3]);
if (RemainingStart<=4 && BlockRows>=5) res.coeffRef(l1i+4, l1j) += ei_predux(dst[4]);
if (RemainingStart<=5 && BlockRows>=6) res.coeffRef(l1i+5, l1j) += ei_predux(dst[5]);
if (RemainingStart<=6 && BlockRows>=7) res.coeffRef(l1i+6, l1j) += ei_predux(dst[6]);
if (RemainingStart<=7 && BlockRows>=8) res.coeffRef(l1i+7, l1j) += ei_predux(dst[7]);
asm("#eigen end kernel");
}
}
template<typename Lhs, typename Rhs, int EvalMode>
template<typename DestDerived, int RhsAlignment, int ResAlignment>
void Product<Lhs,Rhs,EvalMode>::_cacheFriendlyEvalImpl(DestDerived& res) const
{
// FIXME find a way to optimize: (an_xpr) + (a * b)
// then we don't need to clear res and avoid and additional mat-mat sum
#ifndef EIGEN_WIP_PRODUCT_DIRTY
// std::cout << "wip product\n";
res.setZero();
#endif
const int rows = _rows();
const int cols = _cols();
const int remainingSize = m_lhs.cols()%PacketSize;
const int size = m_lhs.cols() - remainingSize; // third dimension of the product clamped to packet boundaries
const int l2BlockRows = MaxL2BlockSize > _rows() ? _rows() : MaxL2BlockSize;
const int l2BlockCols = MaxL2BlockSize > _cols() ? _cols() : MaxL2BlockSize;
const int l2BlockSize = MaxL2BlockSize > size ? size : MaxL2BlockSize;
//Scalar* __restrict__ block = new Scalar[l2blocksize*size];;
Scalar* __restrict__ block = (Scalar*)alloca(sizeof(Scalar)*l2BlockRows*size);
// loops on each L2 cache friendly blocks of the result
for(int l2i=0; l2i<_rows(); l2i+=l2BlockRows)
{
const int l2blockRowEnd = std::min(l2i+l2BlockRows, rows);
const int l2blockRowEndBW = l2blockRowEnd & MaxBlockRows_ClampingMask; // end of the rows aligned to bw
const int l2blockRemainingRows = l2blockRowEnd - l2blockRowEndBW; // number of remaining rows
// build a cache friendly block
int count = 0;
// copy l2blocksize rows of m_lhs to blocks of ps x bw
for(int l2k=0; l2k<size; l2k+=l2BlockSize)
{
const int l2blockSizeEnd = std::min(l2k+l2BlockSize, size);
for (int i = l2i; i<l2blockRowEndBW; i+=MaxBlockRows)
{
for (int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
// TODO write these two loops using meta unrolling
// negligible for large matrices but useful for small ones
for (int w=0; w<MaxBlockRows; ++w)
for (int s=0; s<PacketSize; ++s)
block[count++] = m_lhs.coeff(i+w,k+s);
}
}
if (l2blockRemainingRows>0)
{
for (int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
for (int w=0; w<l2blockRemainingRows; ++w)
for (int s=0; s<PacketSize; ++s)
block[count++] = m_lhs.coeff(l2blockRowEndBW+w,k+s);
}
}
}
for(int l2j=0; l2j<cols; l2j+=l2BlockCols)
{
int l2blockColEnd = std::min(l2j+l2BlockCols, cols);
for(int l2k=0; l2k<size; l2k+=l2BlockSize)
{
// acumulate a bw rows of lhs time a single column of rhs to a bw x 1 block of res
int l2blockSizeEnd = std::min(l2k+l2BlockSize, size);
// for each bw x 1 result's block
for(int l1i=l2i; l1i<l2blockRowEndBW; l1i+=MaxBlockRows)
{
_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, MaxBlockRows>(
res, l2i, l2j, l2k, l1i, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block);
#if 0
for(int l1j=l2j; l1j<l2blockColEnd; l1j+=1)
{
int offsetblock = l2k * (l2blockRowEnd-l2i) + (l1i-l2i)*(l2blockSizeEnd-l2k) - l2k*MaxBlockRows;
const Scalar* localB = &block[offsetblock];
int l1jsize = l1j * m_lhs.cols(); //TODO find a better way to optimize address computation ?
PacketScalar dst[bw];
dst[0] = ei_pset1(Scalar(0.));
dst[1] = dst[0];
dst[2] = dst[0];
dst[3] = dst[0];
if (MaxBlockRows==8)
{
dst[4] = dst[0];
dst[5] = dst[0];
dst[6] = dst[0];
dst[7] = dst[0];
}
PacketScalar b0, b1, tmp;
// TODO in unaligned mode, preload the next element
// PacketScalar tmp1 = _mm_load_ps(&m_rhs.derived().data()[l1jsize+l2k]);
asm("#eigen begincore");
for(int k=l2k; k<l2blockSizeEnd; k+=PacketSize)
{
// PacketScalar tmp = m_rhs.template packetCoeff<Aligned>(k, l1j);
// TODO make this branching compile time (costly for doubles)
if (rhsIsAligned)
tmp = ei_pload(&m_rhs.derived().data()[l1jsize + k]);
else
tmp = ei_ploadu(&m_rhs.derived().data()[l1jsize + k]);
b0 = ei_pload(&(localB[k*bw]));
b1 = ei_pload(&(localB[k*bw+ps]));
dst[0] = ei_pmadd(tmp, b0, dst[0]);
b0 = ei_pload(&(localB[k*bw+2*ps]));
dst[1] = ei_pmadd(tmp, b1, dst[1]);
b1 = ei_pload(&(localB[k*bw+3*ps]));
dst[2] = ei_pmadd(tmp, b0, dst[2]);
if (MaxBlockRows==8)
b0 = ei_pload(&(localB[k*bw+4*ps]));
dst[3] = ei_pmadd(tmp, b1, dst[3]);
if (MaxBlockRows==8)
{
b1 = ei_pload(&(localB[k*bw+5*ps]));
dst[4] = ei_pmadd(tmp, b0, dst[4]);
b0 = ei_pload(&(localB[k*bw+6*ps]));
dst[5] = ei_pmadd(tmp, b1, dst[5]);
b1 = ei_pload(&(localB[k*bw+7*ps]));
dst[6] = ei_pmadd(tmp, b0, dst[6]);
dst[7] = ei_pmadd(tmp, b1, dst[7]);
}
}
// if (resIsAligned)
{
res.template writePacketCoeff<Aligned>(l1i, l1j, ei_padd(res.template packetCoeff<Aligned>(l1i, l1j), ei_preduxp(dst)));
if (PacketSize==2)
res.template writePacketCoeff<Aligned>(l1i+2,l1j, ei_padd(res.template packetCoeff<Aligned>(l1i+2,l1j), ei_preduxp(&(dst[2]))));
if (MaxBlockRows==8)
{
res.template writePacketCoeff<Aligned>(l1i+4,l1j, ei_padd(res.template packetCoeff<Aligned>(l1i+4,l1j), ei_preduxp(&(dst[4]))));
if (PacketSize==2)
res.template writePacketCoeff<Aligned>(l1i+6,l1j, ei_padd(res.template packetCoeff<Aligned>(l1i+6,l1j), ei_preduxp(&(dst[6]))));
}
}
// else
// {
// // TODO uncommenting this code kill the perf, even though it is never called !!
// // this is because dst cannot be a set of registers only
// // TODO optimize this loop
// // TODO is it better to do one redux at once or packet reduxes + unaligned store ?
// for (int w = 0; w<bw; ++w)
// res.coeffRef(l1i+w, l1j) += ei_predux(dst[w]);
// std::cout << "!\n";
// }
asm("#eigen endcore");
}
#endif
}
if (l2blockRemainingRows>0)
{
// this is an attempt to build an array of kernels, but I did not manage to get it compiles
// typedef void (*Kernel)(DestDerived& , int, int, int, int, int, int, int, const Scalar*);
// Kernel kernels[8];
// kernels[0] = (Kernel)(&Product<Lhs,Rhs,EvalMode>::template _cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 1>);
// kernels[l2blockRemainingRows](res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block);
switch(l2blockRemainingRows)
{
case 1:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 1>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
case 2:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 2>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
case 3:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 3>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
case 4:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 4>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
case 5:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 5>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
case 6:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 6>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
case 7:_cacheFriendlyEvalKernel<DestDerived, RhsAlignment, ResAlignment, 7>(
res, l2i, l2j, l2k, l2blockRowEndBW, l2blockRowEnd, l2blockColEnd, l2blockSizeEnd, block); break;
default:
ei_internal_assert(false && "internal error"); break;
}
}
}
}
}
// handle the part which cannot be processed by the vectorized path
if (remainingSize)
{
res += Product<
Block<typename ei_unconst<_LhsNested>::type,Dynamic,Dynamic>,
Block<typename ei_unconst<_RhsNested>::type,Dynamic,Dynamic>,
NormalProduct>(
m_lhs.block(0,size, _rows(), remainingSize),
m_rhs.block(size,0, remainingSize, _cols())).lazy();
// res += m_lhs.block(0,size, _rows(), remainingSize)._lazyProduct(m_rhs.block(size,0, remainingSize, _cols()));
}
// delete[] block;
ei_cache_friendly_product<Scalar>(
_rows(), _cols(), m_lhs.cols(),
_LhsNested::Flags&RowMajorBit, &(m_lhs.const_cast_derived().coeffRef(0,0)), m_lhs.stride(),
_RhsNested::Flags&RowMajorBit, &(m_rhs.const_cast_derived().coeffRef(0,0)), m_rhs.stride(),
Flags&RowMajorBit, &(res.coeffRef(0,0)), res.stride()
);
}
#endif // EIGEN_PRODUCT_H

View File

@ -69,6 +69,8 @@ template<typename MatrixType> class Transpose
int _rows() const { return m_matrix.cols(); }
int _cols() const { return m_matrix.rows(); }
int _stride(void) const { return m_matrix.stride(); }
Scalar& _coeffRef(int row, int col)
{
return m_matrix.const_cast_derived().coeffRef(col, row);

View File

@ -67,7 +67,7 @@ struct ei_traits<Triangular<Mode, MatrixType> >
ColsAtCompileTime = MatrixType::ColsAtCompileTime,
MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
Flags = (_MatrixTypeNested::Flags & ~(VectorizableBit | Like1DArrayBit)) | Mode,
Flags = (_MatrixTypeNested::Flags & ~(VectorizableBit | Like1DArrayBit | DirectAccessBit)) | Mode,
CoeffReadCost = _MatrixTypeNested::CoeffReadCost
};
};

View File

@ -43,18 +43,18 @@ const unsigned int NullDiagBit = 0x40; ///< means all diagonal coefficients
const unsigned int UnitDiagBit = 0x80; ///< means all diagonal coefficients are equal to 1
const unsigned int NullLowerBit = 0x200; ///< means the strictly triangular lower part is 0
const unsigned int NullUpperBit = 0x400; ///< means the strictly triangular upper part is 0
const unsigned int ReferencableBit = 0x800; ///< means the expression is writable through MatrixBase::coeffRef(int,int)
const unsigned int DirectAccessBit = 0x800; ///< means the underlying matrix data can be direclty accessed
enum { Upper=NullLowerBit, Lower=NullUpperBit };
enum { Aligned=0, UnAligned=1 };
// list of flags that are lost by default
const unsigned int DefaultLostFlagMask = ~(VectorizableBit | Like1DArrayBit | ReferencableBit
const unsigned int DefaultLostFlagMask = ~(VectorizableBit | Like1DArrayBit | DirectAccessBit
| NullDiagBit | UnitDiagBit | NullLowerBit | NullUpperBit);
enum { ConditionalJumpCost = 5 };
enum CornerType { TopLeft, TopRight, BottomLeft, BottomRight };
enum DirectionType { Vertical, Horizontal };
enum ProductEvaluationMode { NormalProduct, CacheOptimalProduct, LazyProduct};
enum ProductEvaluationMode { NormalProduct, CacheFriendlyProduct, LazyProduct};
#endif // EIGEN_CONSTANTS_H

848
Eigen/src/QR/EigenSolver.h Normal file
View File

@ -0,0 +1,848 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra. Eigen itself is part of the KDE project.
//
// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_EIGENSOLVER_H
#define EIGEN_EIGENSOLVER_H
/** \class EigenSolver
*
* \brief Eigen values/vectors solver
*
* \param MatrixType the type of the matrix of which we are computing the eigen decomposition
*
* \note this code was adapted from JAMA (public domain)
*
* \sa MatrixBase::eigenvalues()
*/
template<typename _MatrixType> class EigenSolver
{
public:
typedef _MatrixType MatrixType;
typedef typename MatrixType::Scalar Scalar;
typedef Matrix<Scalar, MatrixType::ColsAtCompileTime, 1> VectorType;
EigenSolver(const MatrixType& matrix)
: m_eivec(matrix.rows(), matrix.cols()),
m_eivalr(matrix.cols()), m_eivali(matrix.cols()),
m_H(matrix.rows(), matrix.cols()),
m_ort(matrix.cols())
{
_compute(matrix);
}
MatrixType eigenvectors(void) const { return m_eivec; }
VectorType eigenvalues(void) const { return m_eivalr; }
private:
void _compute(const MatrixType& matrix);
void tridiagonalization(void);
void tql2(void);
void orthes(void);
void hqr2(void);
protected:
MatrixType m_eivec;
VectorType m_eivalr, m_eivali;
MatrixType m_H;
VectorType m_ort;
bool m_isSymmetric;
};
template<typename MatrixType>
void EigenSolver<MatrixType>::_compute(const MatrixType& matrix)
{
assert(matrix.cols() == matrix.rows());
m_isSymmetric = true;
int n = matrix.cols();
for (int j = 0; (j < n) && m_isSymmetric; j++) {
for (int i = 0; (i < j) && m_isSymmetric; i++) {
m_isSymmetric = (matrix(i,j) == matrix(j,i));
}
}
m_eivalr.resize(n,1);
m_eivali.resize(n,1);
if (m_isSymmetric)
{
m_eivec = matrix;
// Tridiagonalize.
tridiagonalization();
// Diagonalize.
tql2();
}
else
{
m_H = matrix;
m_ort.resize(n, 1);
// Reduce to Hessenberg form.
orthes();
// Reduce Hessenberg to real Schur form.
hqr2();
}
std::cout << m_eivali.transpose() << "\n";
}
// Symmetric Householder reduction to tridiagonal form.
template<typename MatrixType>
void EigenSolver<MatrixType>::tridiagonalization(void)
{
// This is derived from the Algol procedures tred2 by
// Bowdler, Martin, Reinsch, and Wilkinson, Handbook for
// Auto. Comp., Vol.ii-Linear Algebra, and the corresponding
// Fortran subroutine in EISPACK.
int n = m_eivec.cols();
m_eivalr = m_eivec.row(m_eivalr.size()-1);
// Householder reduction to tridiagonal form.
for (int i = n-1; i > 0; i--)
{
// Scale to avoid under/overflow.
Scalar scale = 0.0;
Scalar h = 0.0;
scale = m_eivalr.start(i).cwiseAbs().sum();
if (scale == 0.0)
{
m_eivali[i] = m_eivalr[i-1];
m_eivalr.start(i) = m_eivec.row(i-1).start(i);
m_eivec.corner(TopLeft, i, i) = m_eivec.corner(TopLeft, i, i).diagonal().asDiagonal();
}
else
{
// Generate Householder vector.
m_eivalr.start(i) /= scale;
h = m_eivalr.start(i).cwiseAbs2().sum();
Scalar f = m_eivalr[i-1];
Scalar g = ei_sqrt(h);
if (f > 0)
g = -g;
m_eivali[i] = scale * g;
h = h - f * g;
m_eivalr[i-1] = f - g;
m_eivali.start(i).setZero();
// Apply similarity transformation to remaining columns.
for (int j = 0; j < i; j++)
{
f = m_eivalr[j];
m_eivec(j,i) = f;
g = m_eivali[j] + m_eivec(j,j) * f;
int bSize = i-j-1;
if (bSize>0)
{
g += (m_eivec.col(j).block(j+1, bSize).transpose() * m_eivalr.block(j+1, bSize))(0,0);
m_eivali.block(j+1, bSize) += m_eivec.col(j).block(j+1, bSize) * f;
}
m_eivali[j] = g;
}
f = (m_eivali.start(i).transpose() * m_eivalr.start(i))(0,0);
m_eivali.start(i) = (m_eivali.start(i) - (f / (h + h)) * m_eivalr.start(i))/h;
m_eivec.corner(TopLeft, i, i).lower() -=
( (m_eivali.start(i) * m_eivalr.start(i).transpose()).lazy()
+ (m_eivalr.start(i) * m_eivali.start(i).transpose()).lazy());
m_eivalr.start(i) = m_eivec.row(i-1).start(i);
m_eivec.row(i).start(i).setZero();
}
m_eivalr[i] = h;
}
// Accumulate transformations.
for (int i = 0; i < n-1; i++)
{
m_eivec(n-1,i) = m_eivec(i,i);
m_eivec(i,i) = 1.0;
Scalar h = m_eivalr[i+1];
// FIXME this does not looks very stable ;)
if (h != 0.0)
{
m_eivalr.start(i+1) = m_eivec.col(i+1).start(i+1) / h;
m_eivec.corner(TopLeft, i+1, i+1) -= m_eivalr.start(i+1)
* ( m_eivec.col(i+1).start(i+1).transpose() * m_eivec.corner(TopLeft, i+1, i+1) );
}
m_eivec.col(i+1).start(i+1).setZero();
}
m_eivalr = m_eivec.row(m_eivalr.size()-1);
m_eivec.row(m_eivalr.size()-1).setZero();
m_eivec(n-1,n-1) = 1.0;
m_eivali[0] = 0.0;
}
// Symmetric tridiagonal QL algorithm.
template<typename MatrixType>
void EigenSolver<MatrixType>::tql2(void)
{
// This is derived from the Algol procedures tql2, by
// Bowdler, Martin, Reinsch, and Wilkinson, Handbook for
// Auto. Comp., Vol.ii-Linear Algebra, and the corresponding
// Fortran subroutine in EISPACK.
int n = m_eivalr.size();
for (int i = 1; i < n; i++) {
m_eivali[i-1] = m_eivali[i];
}
m_eivali[n-1] = 0.0;
Scalar f = 0.0;
Scalar tst1 = 0.0;
Scalar eps = std::pow(2.0,-52.0);
for (int l = 0; l < n; l++)
{
// Find small subdiagonal element
tst1 = std::max(tst1,ei_abs(m_eivalr[l]) + ei_abs(m_eivali[l]));
int m = l;
while ( (m < n) && (ei_abs(m_eivali[m]) > eps*tst1) )
m++;
// If m == l, m_eivalr[l] is an eigenvalue,
// otherwise, iterate.
if (m > l)
{
int iter = 0;
do
{
iter = iter + 1;
// Compute implicit shift
Scalar g = m_eivalr[l];
Scalar p = (m_eivalr[l+1] - g) / (2.0 * m_eivali[l]);
Scalar r = hypot(p,1.0);
if (p < 0)
r = -r;
m_eivalr[l] = m_eivali[l] / (p + r);
m_eivalr[l+1] = m_eivali[l] * (p + r);
Scalar dl1 = m_eivalr[l+1];
Scalar h = g - m_eivalr[l];
if (l+2<n)
m_eivalr.end(n-l-2) -= VectorType::constant(n-l-2, h);
f = f + h;
// Implicit QL transformation.
p = m_eivalr[m];
Scalar c = 1.0;
Scalar c2 = c;
Scalar c3 = c;
Scalar el1 = m_eivali[l+1];
Scalar s = 0.0;
Scalar s2 = 0.0;
for (int i = m-1; i >= l; i--)
{
c3 = c2;
c2 = c;
s2 = s;
g = c * m_eivali[i];
h = c * p;
r = hypot(p,m_eivali[i]);
m_eivali[i+1] = s * r;
s = m_eivali[i] / r;
c = p / r;
p = c * m_eivalr[i] - s * g;
m_eivalr[i+1] = h + s * (c * g + s * m_eivalr[i]);
// Accumulate transformation.
for (int k = 0; k < n; k++)
{
h = m_eivec(k,i+1);
m_eivec(k,i+1) = s * m_eivec(k,i) + c * h;
m_eivec(k,i) = c * m_eivec(k,i) - s * h;
}
}
p = -s * s2 * c3 * el1 * m_eivali[l] / dl1;
m_eivali[l] = s * p;
m_eivalr[l] = c * p;
// Check for convergence.
} while (ei_abs(m_eivali[l]) > eps*tst1);
}
m_eivalr[l] = m_eivalr[l] + f;
m_eivali[l] = 0.0;
}
// Sort eigenvalues and corresponding vectors.
// TODO use a better sort algorithm !!
for (int i = 0; i < n-1; i++)
{
int k = i;
Scalar minValue = m_eivalr[i];
for (int j = i+1; j < n; j++)
{
if (m_eivalr[j] < minValue)
{
k = j;
minValue = m_eivalr[j];
}
}
if (k != i)
{
std::swap(m_eivalr[i], m_eivalr[k]);
m_eivec.col(i).swap(m_eivec.col(k));
}
}
}
// Nonsymmetric reduction to Hessenberg form.
template<typename MatrixType>
void EigenSolver<MatrixType>::orthes(void)
{
// This is derived from the Algol procedures orthes and ortran,
// by Martin and Wilkinson, Handbook for Auto. Comp.,
// Vol.ii-Linear Algebra, and the corresponding
// Fortran subroutines in EISPACK.
int n = m_eivec.cols();
int low = 0;
int high = n-1;
for (int m = low+1; m <= high-1; m++)
{
// Scale column.
Scalar scale = m_H.block(m, m-1, high-m+1, 1).cwiseAbs().sum();
if (scale != 0.0)
{
// Compute Householder transformation.
Scalar h = 0.0;
// FIXME could be rewritten, but this one looks better wrt cache
for (int i = high; i >= m; i--)
{
m_ort[i] = m_H(i,m-1)/scale;
h += m_ort[i] * m_ort[i];
}
Scalar g = ei_sqrt(h);
if (m_ort[m] > 0)
g = -g;
h = h - m_ort[m] * g;
m_ort[m] = m_ort[m] - g;
// Apply Householder similarity transformation
// H = (I-u*u'/h)*H*(I-u*u')/h)
int bSize = high-m+1;
m_H.block(m, m, bSize, n-m) -= ((m_ort.block(m, bSize)/h)
* (m_ort.block(m, bSize).transpose() * m_H.block(m, m, bSize, n-m)).lazy()).lazy();
m_H.block(0, m, high+1, bSize) -= ((m_H.block(0, m, high+1, bSize) * m_ort.block(m, bSize)).lazy()
* (m_ort.block(m, bSize)/h).transpose()).lazy();
m_ort[m] = scale*m_ort[m];
m_H(m,m-1) = scale*g;
}
}
// Accumulate transformations (Algol's ortran).
m_eivec.setIdentity();
for (int m = high-1; m >= low+1; m--)
{
if (m_H(m,m-1) != 0.0)
{
m_ort.block(m+1, high-m) = m_H.col(m-1).block(m+1, high-m);
int bSize = high-m+1;
m_eivec.block(m, m, bSize, bSize) += ( (m_ort.block(m, bSize) / (m_H(m,m-1) * m_ort[m] ) )
* (m_ort.block(m, bSize).transpose() * m_eivec.block(m, m, bSize, bSize)).lazy());
}
}
}
// Complex scalar division.
template<typename Scalar>
std::complex<Scalar> cdiv(Scalar xr, Scalar xi, Scalar yr, Scalar yi)
{
Scalar r,d;
if (ei_abs(yr) > ei_abs(yi))
{
r = yi/yr;
d = yr + r*yi;
return std::complex<Scalar>((xr + r*xi)/d, (xi - r*xr)/d);
}
else
{
r = yr/yi;
d = yi + r*yr;
return std::complex<Scalar>((r*xr + xi)/d, (r*xi - xr)/d);
}
}
// Nonsymmetric reduction from Hessenberg to real Schur form.
template<typename MatrixType>
void EigenSolver<MatrixType>::hqr2(void)
{
// This is derived from the Algol procedure hqr2,
// by Martin and Wilkinson, Handbook for Auto. Comp.,
// Vol.ii-Linear Algebra, and the corresponding
// Fortran subroutine in EISPACK.
// Initialize
int nn = m_eivec.cols();
int n = nn-1;
int low = 0;
int high = nn-1;
Scalar eps = pow(2.0,-52.0);
Scalar exshift = 0.0;
Scalar p=0,q=0,r=0,s=0,z=0,t,w,x,y;
// Store roots isolated by balanc and compute matrix norm
// FIXME to be efficient the following would requires a triangular reduxion code
// Scalar norm = m_H.upper().cwiseAbs().sum() + m_H.corner(BottomLeft,n,n).diagonal().cwiseAbs().sum();
Scalar norm = 0.0;
for (int j = 0; j < nn; j++)
{
// FIXME what's the purpose of the following since the condition is always false
if ((j < low) || (j > high))
{
m_eivalr[j] = m_H(j,j);
m_eivali[j] = 0.0;
}
norm += m_H.col(j).start(std::min(j+1,nn)).cwiseAbs().sum();
}
// Outer loop over eigenvalue index
int iter = 0;
while (n >= low)
{
// Look for single small sub-diagonal element
int l = n;
while (l > low)
{
s = ei_abs(m_H(l-1,l-1)) + ei_abs(m_H(l,l));
if (s == 0.0)
s = norm;
if (ei_abs(m_H(l,l-1)) < eps * s)
break;
l--;
}
// Check for convergence
// One root found
if (l == n)
{
m_H(n,n) = m_H(n,n) + exshift;
m_eivalr[n] = m_H(n,n);
m_eivali[n] = 0.0;
n--;
iter = 0;
}
else if (l == n-1) // Two roots found
{
w = m_H(n,n-1) * m_H(n-1,n);
p = (m_H(n-1,n-1) - m_H(n,n)) / 2.0;
q = p * p + w;
z = ei_sqrt(ei_abs(q));
m_H(n,n) = m_H(n,n) + exshift;
m_H(n-1,n-1) = m_H(n-1,n-1) + exshift;
x = m_H(n,n);
// Scalar pair
if (q >= 0)
{
if (p >= 0)
z = p + z;
else
z = p - z;
m_eivalr[n-1] = x + z;
m_eivalr[n] = m_eivalr[n-1];
if (z != 0.0)
m_eivalr[n] = x - w / z;
m_eivali[n-1] = 0.0;
m_eivali[n] = 0.0;
x = m_H(n,n-1);
s = ei_abs(x) + ei_abs(z);
p = x / s;
q = z / s;
r = ei_sqrt(p * p+q * q);
p = p / r;
q = q / r;
// Row modification
for (int j = n-1; j < nn; j++)
{
z = m_H(n-1,j);
m_H(n-1,j) = q * z + p * m_H(n,j);
m_H(n,j) = q * m_H(n,j) - p * z;
}
// Column modification
for (int i = 0; i <= n; i++)
{
z = m_H(i,n-1);
m_H(i,n-1) = q * z + p * m_H(i,n);
m_H(i,n) = q * m_H(i,n) - p * z;
}
// Accumulate transformations
for (int i = low; i <= high; i++)
{
z = m_eivec(i,n-1);
m_eivec(i,n-1) = q * z + p * m_eivec(i,n);
m_eivec(i,n) = q * m_eivec(i,n) - p * z;
}
}
else // Complex pair
{
m_eivalr[n-1] = x + p;
m_eivalr[n] = x + p;
m_eivali[n-1] = z;
m_eivali[n] = -z;
}
n = n - 2;
iter = 0;
}
else // No convergence yet
{
// Form shift
x = m_H(n,n);
y = 0.0;
w = 0.0;
if (l < n)
{
y = m_H(n-1,n-1);
w = m_H(n,n-1) * m_H(n-1,n);
}
// Wilkinson's original ad hoc shift
if (iter == 10)
{
exshift += x;
for (int i = low; i <= n; i++)
m_H(i,i) -= x;
s = ei_abs(m_H(n,n-1)) + ei_abs(m_H(n-1,n-2));
x = y = 0.75 * s;
w = -0.4375 * s * s;
}
// MATLAB's new ad hoc shift
if (iter == 30)
{
s = (y - x) / 2.0;
s = s * s + w;
if (s > 0)
{
s = ei_sqrt(s);
if (y < x)
s = -s;
s = x - w / ((y - x) / 2.0 + s);
for (int i = low; i <= n; i++)
m_H(i,i) -= s;
exshift += s;
x = y = w = 0.964;
}
}
iter = iter + 1; // (Could check iteration count here.)
// Look for two consecutive small sub-diagonal elements
int m = n-2;
while (m >= l)
{
z = m_H(m,m);
r = x - z;
s = y - z;
p = (r * s - w) / m_H(m+1,m) + m_H(m,m+1);
q = m_H(m+1,m+1) - z - r - s;
r = m_H(m+2,m+1);
s = ei_abs(p) + ei_abs(q) + ei_abs(r);
p = p / s;
q = q / s;
r = r / s;
if (m == l) {
break;
}
if (ei_abs(m_H(m,m-1)) * (ei_abs(q) + ei_abs(r)) <
eps * (ei_abs(p) * (ei_abs(m_H(m-1,m-1)) + ei_abs(z) +
ei_abs(m_H(m+1,m+1)))))
{
break;
}
m--;
}
for (int i = m+2; i <= n; i++)
{
m_H(i,i-2) = 0.0;
if (i > m+2)
m_H(i,i-3) = 0.0;
}
// Double QR step involving rows l:n and columns m:n
for (int k = m; k <= n-1; k++)
{
int notlast = (k != n-1);
if (k != m) {
p = m_H(k,k-1);
q = m_H(k+1,k-1);
r = (notlast ? m_H(k+2,k-1) : 0.0);
x = ei_abs(p) + ei_abs(q) + ei_abs(r);
if (x != 0.0)
{
p = p / x;
q = q / x;
r = r / x;
}
}
if (x == 0.0)
break;
s = ei_sqrt(p * p + q * q + r * r);
if (p < 0)
s = -s;
if (s != 0)
{
if (k != m)
m_H(k,k-1) = -s * x;
else if (l != m)
m_H(k,k-1) = -m_H(k,k-1);
p = p + s;
x = p / s;
y = q / s;
z = r / s;
q = q / p;
r = r / p;
// Row modification
for (int j = k; j < nn; j++)
{
p = m_H(k,j) + q * m_H(k+1,j);
if (notlast)
{
p = p + r * m_H(k+2,j);
m_H(k+2,j) = m_H(k+2,j) - p * z;
}
m_H(k,j) = m_H(k,j) - p * x;
m_H(k+1,j) = m_H(k+1,j) - p * y;
}
// Column modification
for (int i = 0; i <= std::min(n,k+3); i++)
{
p = x * m_H(i,k) + y * m_H(i,k+1);
if (notlast)
{
p = p + z * m_H(i,k+2);
m_H(i,k+2) = m_H(i,k+2) - p * r;
}
m_H(i,k) = m_H(i,k) - p;
m_H(i,k+1) = m_H(i,k+1) - p * q;
}
// Accumulate transformations
for (int i = low; i <= high; i++)
{
p = x * m_eivec(i,k) + y * m_eivec(i,k+1);
if (notlast)
{
p = p + z * m_eivec(i,k+2);
m_eivec(i,k+2) = m_eivec(i,k+2) - p * r;
}
m_eivec(i,k) = m_eivec(i,k) - p;
m_eivec(i,k+1) = m_eivec(i,k+1) - p * q;
}
} // (s != 0)
} // k loop
} // check convergence
} // while (n >= low)
// Backsubstitute to find vectors of upper triangular form
if (norm == 0.0)
{
return;
}
for (n = nn-1; n >= 0; n--)
{
p = m_eivalr[n];
q = m_eivali[n];
// Scalar vector
if (q == 0)
{
int l = n;
m_H(n,n) = 1.0;
for (int i = n-1; i >= 0; i--)
{
w = m_H(i,i) - p;
r = (m_H.row(i).end(nn-l) * m_H.col(n).end(nn-l))(0,0);
if (m_eivali[i] < 0.0)
{
z = w;
s = r;
}
else
{
l = i;
if (m_eivali[i] == 0.0)
{
if (w != 0.0)
m_H(i,n) = -r / w;
else
m_H(i,n) = -r / (eps * norm);
}
else // Solve real equations
{
x = m_H(i,i+1);
y = m_H(i+1,i);
q = (m_eivalr[i] - p) * (m_eivalr[i] - p) + m_eivali[i] * m_eivali[i];
t = (x * s - z * r) / q;
m_H(i,n) = t;
if (ei_abs(x) > ei_abs(z))
m_H(i+1,n) = (-r - w * t) / x;
else
m_H(i+1,n) = (-s - y * t) / z;
}
// Overflow control
t = ei_abs(m_H(i,n));
if ((eps * t) * t > 1)
m_H.col(n).end(nn-i) /= t;
}
}
}
else if (q < 0) // Complex vector
{
std::complex<Scalar> cc;
int l = n-1;
// Last vector component imaginary so matrix is triangular
if (ei_abs(m_H(n,n-1)) > ei_abs(m_H(n-1,n)))
{
m_H(n-1,n-1) = q / m_H(n,n-1);
m_H(n-1,n) = -(m_H(n,n) - p) / m_H(n,n-1);
}
else
{
cc = cdiv<Scalar>(0.0,-m_H(n-1,n),m_H(n-1,n-1)-p,q);
m_H(n-1,n-1) = ei_real(cc);
m_H(n-1,n) = ei_imag(cc);
}
m_H(n,n-1) = 0.0;
m_H(n,n) = 1.0;
for (int i = n-2; i >= 0; i--)
{
Scalar ra,sa,vr,vi;
ra = (m_H.row(i).end(nn-l) * m_H.col(n-1).end(nn-l)).lazy()(0,0);
sa = (m_H.row(i).end(nn-l) * m_H.col(n).end(nn-l)).lazy()(0,0);
w = m_H(i,i) - p;
if (m_eivali[i] < 0.0)
{
z = w;
r = ra;
s = sa;
}
else
{
l = i;
if (m_eivali[i] == 0)
{
cc = cdiv(-ra,-sa,w,q);
m_H(i,n-1) = ei_real(cc);
m_H(i,n) = ei_imag(cc);
}
else
{
// Solve complex equations
x = m_H(i,i+1);
y = m_H(i+1,i);
vr = (m_eivalr[i] - p) * (m_eivalr[i] - p) + m_eivali[i] * m_eivali[i] - q * q;
vi = (m_eivalr[i] - p) * 2.0 * q;
if ((vr == 0.0) && (vi == 0.0))
vr = eps * norm * (ei_abs(w) + ei_abs(q) + ei_abs(x) + ei_abs(y) + ei_abs(z));
cc= cdiv(x*r-z*ra+q*sa,x*s-z*sa-q*ra,vr,vi);
m_H(i,n-1) = ei_real(cc);
m_H(i,n) = ei_imag(cc);
if (ei_abs(x) > (ei_abs(z) + ei_abs(q)))
{
m_H(i+1,n-1) = (-ra - w * m_H(i,n-1) + q * m_H(i,n)) / x;
m_H(i+1,n) = (-sa - w * m_H(i,n) - q * m_H(i,n-1)) / x;
}
else
{
cc = cdiv(-r-y*m_H(i,n-1),-s-y*m_H(i,n),z,q);
m_H(i+1,n-1) = ei_real(cc);
m_H(i+1,n) = ei_imag(cc);
}
}
// Overflow control
t = std::max(ei_abs(m_H(i,n-1)),ei_abs(m_H(i,n)));
if ((eps * t) * t > 1)
m_H.block(i, n-1, nn-i, 2) /= t;
}
}
}
}
// Vectors of isolated roots
for (int i = 0; i < nn; i++)
{
// FIXME again what's the purpose of this test ?
// in this algo low==0 and high==nn-1 !!
if (i < low || i > high)
{
m_eivec.row(i).end(nn-i) = m_H.row(i).end(nn-i);
}
}
// Back transformation to get eigenvectors of original matrix
int bRows = high-low+1;
for (int j = nn-1; j >= low; j--)
{
int bSize = std::min(j,high)-low+1;
m_eivec.col(j).block(low, bRows) = (m_eivec.block(low, low, bRows, bSize) * m_H.col(j).block(low, bSize));
}
}
#endif // EIGEN_EIGENSOLVER_H

View File

@ -95,10 +95,11 @@ void QR<MatrixType>::_compute(const MatrixType& matrix)
m_qr(k,k) += 1.0;
// apply transformation to remaining columns
for (int j = k+1; j < cols; j++)
int remainingCols = cols - k -1;
if (remainingCols>0)
{
Scalar s = -(m_qr.col(k).end(remainingSize).transpose() * m_qr.col(j).end(remainingSize))(0,0) / m_qr(k,k);
m_qr.col(j).end(remainingSize) += s * m_qr.col(k).end(remainingSize);
m_qr.corner(BottomRight, remainingSize, remainingCols) -= (1./m_qr(k,k)) * m_qr.col(k).end(remainingSize)
* (m_qr.col(k).end(remainingSize).transpose() * m_qr.corner(BottomRight, remainingSize, remainingCols));
}
}
m_norms[k] = -nrm;