GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-04-19 08:09:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

* compilation fixes for gcc 3.3

Browse Source 
* test Part::swap
This commit is contained in:
Gael Guennebaud 2009-05-06 08:43:38 +00:00
parent 23f073625d
commit 1e286464ab
4 changed files with 36 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Eigen/src/Core/Block.h
View File

@ -221,7 +221,7 @@ class Block<MatrixType,BlockRows,BlockCols,PacketAccess,HasDirectAccess>
class InnerIterator; class InnerIterator;
typedef typename ei_traits<Block>::AlignedDerivedType AlignedDerivedType; typedef typename ei_traits<Block>::AlignedDerivedType AlignedDerivedType;
friend class Block<MatrixType,BlockRows,BlockCols,AsRequested,HasDirectAccess>; friend class Block<MatrixType,BlockRows,BlockCols,PacketAccess==AsRequested?ForceAligned:AsRequested,HasDirectAccess>;
EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)

10
Eigen/src/Core/Part.h
View File

@ -26,7 +26,7 @@
#ifndef EIGEN_PART_H #ifndef EIGEN_PART_H
#define EIGEN_PART_H #define EIGEN_PART_H
/** \nonstableyet /** \nonstableyet
* \class Part * \class Part
* *
* \brief Expression of a triangular matrix extracted from a given matrix * \brief Expression of a triangular matrix extracted from a given matrix
@ -117,10 +117,10 @@ template<typename MatrixType, unsigned int Mode> class Part
const Block<Part, RowsAtCompileTime, 1> col(int i) { return Base::col(i); } const Block<Part, RowsAtCompileTime, 1> col(int i) { return Base::col(i); }
const Block<Part, RowsAtCompileTime, 1> col(int i) const { return Base::col(i); } const Block<Part, RowsAtCompileTime, 1> col(int i) const { return Base::col(i); }
template<typename OtherDerived/*, int OtherMode*/> template<typename OtherDerived>
void swap(const MatrixBase<OtherDerived>& other) void swap(const MatrixBase<OtherDerived>& other)
{ {
Part<SwapWrapper<MatrixType>,Mode>(SwapWrapper<MatrixType>(const_cast<MatrixType&>(m_matrix))).lazyAssign(other.derived()); Part<SwapWrapper<MatrixType>,Mode>(const_cast<MatrixType&>(m_matrix)).lazyAssign(other.derived());
} }
protected: protected:
@ -128,7 +128,7 @@ template<typename MatrixType, unsigned int Mode> class Part
const typename MatrixType::Nested m_matrix; const typename MatrixType::Nested m_matrix;
}; };
/** \nonstableyet /** \nonstableyet
* \returns an expression of a triangular matrix extracted from the current matrix * \returns an expression of a triangular matrix extracted from the current matrix
* *
* The parameter \a Mode can have the following values: \c UpperTriangular, \c StrictlyUpperTriangular, \c UnitUpperTriangular, * The parameter \a Mode can have the following values: \c UpperTriangular, \c StrictlyUpperTriangular, \c UnitUpperTriangular,
@ -280,7 +280,7 @@ void Part<MatrixType, Mode>::lazyAssign(const Other& other)
>::run(m_matrix.const_cast_derived(), other.derived()); >::run(m_matrix.const_cast_derived(), other.derived());
} }
/** \nonstableyet /** \nonstableyet
* \returns a lvalue pseudo-expression allowing to perform special operations on \c *this. * \returns a lvalue pseudo-expression allowing to perform special operations on \c *this.
* *
* The \a Mode parameter can have the following values: \c UpperTriangular, \c StrictlyUpperTriangular, \c LowerTriangular, * The \a Mode parameter can have the following values: \c UpperTriangular, \c StrictlyUpperTriangular, \c LowerTriangular,

44
Eigen/src/Core/arch/SSE/MathFunctions.h
View File

@ -41,7 +41,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_plog(Packet4f x)
/* the smallest non denormalized float number */ /* the smallest non denormalized float number */
_EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos, 0x00800000); _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos, 0x00800000);
/* natural logarithm computed for 4 simultaneous float /* natural logarithm computed for 4 simultaneous float
return NaN for x <= 0 return NaN for x <= 0
*/ */
_EIGEN_DECLARE_CONST_Packet4f(cephes_SQRTHF, 0.707106781186547524f); _EIGEN_DECLARE_CONST_Packet4f(cephes_SQRTHF, 0.707106781186547524f);
@ -72,7 +72,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_plog(Packet4f x)
emm0 = _mm_sub_epi32(emm0, ei_p4i_0x7f); emm0 = _mm_sub_epi32(emm0, ei_p4i_0x7f);
Packet4f e = ei_padd(_mm_cvtepi32_ps(emm0), ei_p4f_1); Packet4f e = ei_padd(_mm_cvtepi32_ps(emm0), ei_p4f_1);
/* part2: /* part2:
if( x < SQRTHF ) { if( x < SQRTHF ) {
e -= 1; e -= 1;
x = x + x - 1.0; x = x + x - 1.0;
@ -86,7 +86,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_plog(Packet4f x)
Packet4f x2 = ei_pmul(x,x); Packet4f x2 = ei_pmul(x,x);
Packet4f x3 = ei_pmul(x2,x); Packet4f x3 = ei_pmul(x2,x);
Packet4f y, y1, y2; Packet4f y, y1, y2;
y = ei_pmadd(ei_p4f_cephes_log_p0, x, ei_p4f_cephes_log_p1); y = ei_pmadd(ei_p4f_cephes_log_p0, x, ei_p4f_cephes_log_p1);
y1 = ei_pmadd(ei_p4f_cephes_log_p3, x, ei_p4f_cephes_log_p4); y1 = ei_pmadd(ei_p4f_cephes_log_p3, x, ei_p4f_cephes_log_p4);
@ -97,7 +97,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_plog(Packet4f x)
y = ei_pmadd(y, x3, y1); y = ei_pmadd(y, x3, y1);
y = ei_pmadd(y, x3, y2); y = ei_pmadd(y, x3, y2);
y = ei_pmul(y, x3); y = ei_pmul(y, x3);
y1 = ei_pmul(e, ei_p4f_cephes_log_q1); y1 = ei_pmul(e, ei_p4f_cephes_log_q1);
tmp = ei_pmul(x2, ei_p4f_half); tmp = ei_pmul(x2, ei_p4f_half);
y = ei_padd(y, y1); y = ei_padd(y, y1);
@ -142,7 +142,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_pexp(Packet4f x)
emm0 = _mm_cvttps_epi32(fx); emm0 = _mm_cvttps_epi32(fx);
tmp = _mm_cvtepi32_ps(emm0); tmp = _mm_cvtepi32_ps(emm0);
/* if greater, substract 1 */ /* if greater, substract 1 */
Packet4f mask = _mm_cmpgt_ps(tmp, fx); Packet4f mask = _mm_cmpgt_ps(tmp, fx);
mask = _mm_and_ps(mask, ei_p4f_1); mask = _mm_and_ps(mask, ei_p4f_1);
fx = ei_psub(tmp, mask); fx = ei_psub(tmp, mask);
@ -152,7 +152,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_pexp(Packet4f x)
x = ei_psub(x, z); x = ei_psub(x, z);
z = ei_pmul(x,x); z = ei_pmul(x,x);
Packet4f y = ei_p4f_cephes_exp_p0; Packet4f y = ei_p4f_cephes_exp_p0;
y = ei_pmadd(y, x, ei_p4f_cephes_exp_p1); y = ei_pmadd(y, x, ei_p4f_cephes_exp_p1);
y = ei_pmadd(y, x, ei_p4f_cephes_exp_p2); y = ei_pmadd(y, x, ei_p4f_cephes_exp_p2);
@ -210,12 +210,12 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_psin(Packet4f x)
sign_bit = x; sign_bit = x;
/* take the absolute value */ /* take the absolute value */
x = ei_pabs(x); x = ei_pabs(x);
/* take the modulo */ /* take the modulo */
/* extract the sign bit (upper one) */ /* extract the sign bit (upper one) */
sign_bit = _mm_and_ps(sign_bit, ei_p4f_sign_mask); sign_bit = _mm_and_ps(sign_bit, ei_p4f_sign_mask);
/* scale by 4/Pi */ /* scale by 4/Pi */
y = ei_pmul(x, ei_p4f_cephes_FOPI); y = ei_pmul(x, ei_p4f_cephes_FOPI);
@ -228,7 +228,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_psin(Packet4f x)
/* get the swap sign flag */ /* get the swap sign flag */
emm0 = _mm_and_si128(emm2, ei_p4i_4); emm0 = _mm_and_si128(emm2, ei_p4i_4);
emm0 = _mm_slli_epi32(emm0, 29); emm0 = _mm_slli_epi32(emm0, 29);
/* get the polynom selection mask /* get the polynom selection mask
there is one polynom for 0 <= x <= Pi/4 there is one polynom for 0 <= x <= Pi/4
and another one for Pi/4<x<=Pi/2 and another one for Pi/4<x<=Pi/2
@ -236,12 +236,12 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_psin(Packet4f x)
*/ */
emm2 = _mm_and_si128(emm2, ei_p4i_2); emm2 = _mm_and_si128(emm2, ei_p4i_2);
emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128()); emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
Packet4f swap_sign_bit = _mm_castsi128_ps(emm0); Packet4f swap_sign_bit = _mm_castsi128_ps(emm0);
Packet4f poly_mask = _mm_castsi128_ps(emm2); Packet4f poly_mask = _mm_castsi128_ps(emm2);
sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit); sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit);
/* The magic pass: "Extended precision modular arithmetic" /* The magic pass: "Extended precision modular arithmetic"
x = ((x - y * DP1) - y * DP2) - y * DP3; */ x = ((x - y * DP1) - y * DP2) - y * DP3; */
xmm1 = ei_pmul(y, ei_p4f_minus_cephes_DP1); xmm1 = ei_pmul(y, ei_p4f_minus_cephes_DP1);
xmm2 = ei_pmul(y, ei_p4f_minus_cephes_DP2); xmm2 = ei_pmul(y, ei_p4f_minus_cephes_DP2);
@ -261,7 +261,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_psin(Packet4f x)
Packet4f tmp = ei_pmul(z, ei_p4f_half); Packet4f tmp = ei_pmul(z, ei_p4f_half);
y = ei_psub(y, tmp); y = ei_psub(y, tmp);
y = ei_padd(y, ei_p4f_1); y = ei_padd(y, ei_p4f_1);
/* Evaluate the second polynom (Pi/4 <= x <= 0) */ /* Evaluate the second polynom (Pi/4 <= x <= 0) */
Packet4f y2 = ei_p4f_sincof_p0; Packet4f y2 = ei_p4f_sincof_p0;
@ -271,7 +271,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_psin(Packet4f x)
y2 = ei_pmul(y2, x); y2 = ei_pmul(y2, x);
y2 = ei_padd(y2, x); y2 = ei_padd(y2, x);
/* select the correct result from the two polynoms */ /* select the correct result from the two polynoms */
y2 = _mm_and_ps(poly_mask, y2); y2 = _mm_and_ps(poly_mask, y2);
y = _mm_andnot_ps(poly_mask, y); y = _mm_andnot_ps(poly_mask, y);
y = _mm_or_ps(y,y2); y = _mm_or_ps(y,y2);
@ -303,12 +303,12 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_pcos(Packet4f x)
Packet4f xmm1, xmm2 = _mm_setzero_ps(), xmm3, y; Packet4f xmm1, xmm2 = _mm_setzero_ps(), xmm3, y;
Packet4i emm0, emm2; Packet4i emm0, emm2;
x = ei_pabs(x); x = ei_pabs(x);
/* scale by 4/Pi */ /* scale by 4/Pi */
y = ei_pmul(x, ei_p4f_cephes_FOPI); y = ei_pmul(x, ei_p4f_cephes_FOPI);
/* get the integer part of y */ /* get the integer part of y */
emm2 = _mm_cvttps_epi32(y); emm2 = _mm_cvttps_epi32(y);
/* j=(j+1) & (~1) (see the cephes sources) */ /* j=(j+1) & (~1) (see the cephes sources) */
@ -317,18 +317,18 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_pcos(Packet4f x)
y = _mm_cvtepi32_ps(emm2); y = _mm_cvtepi32_ps(emm2);
emm2 = _mm_sub_epi32(emm2, ei_p4i_2); emm2 = _mm_sub_epi32(emm2, ei_p4i_2);
/* get the swap sign flag */ /* get the swap sign flag */
emm0 = _mm_andnot_si128(emm2, ei_p4i_4); emm0 = _mm_andnot_si128(emm2, ei_p4i_4);
emm0 = _mm_slli_epi32(emm0, 29); emm0 = _mm_slli_epi32(emm0, 29);
/* get the polynom selection mask */ /* get the polynom selection mask */
emm2 = _mm_and_si128(emm2, ei_p4i_2); emm2 = _mm_and_si128(emm2, ei_p4i_2);
emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128()); emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
Packet4f sign_bit = _mm_castsi128_ps(emm0); Packet4f sign_bit = _mm_castsi128_ps(emm0);
Packet4f poly_mask = _mm_castsi128_ps(emm2); Packet4f poly_mask = _mm_castsi128_ps(emm2);
/* The magic pass: "Extended precision modular arithmetic" /* The magic pass: "Extended precision modular arithmetic"
x = ((x - y * DP1) - y * DP2) - y * DP3; */ x = ((x - y * DP1) - y * DP2) - y * DP3; */
xmm1 = ei_pmul(y, ei_p4f_minus_cephes_DP1); xmm1 = ei_pmul(y, ei_p4f_minus_cephes_DP1);
xmm2 = ei_pmul(y, ei_p4f_minus_cephes_DP2); xmm2 = ei_pmul(y, ei_p4f_minus_cephes_DP2);
@ -348,7 +348,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_pcos(Packet4f x)
Packet4f tmp = _mm_mul_ps(z, ei_p4f_half); Packet4f tmp = _mm_mul_ps(z, ei_p4f_half);
y = ei_psub(y, tmp); y = ei_psub(y, tmp);
y = ei_padd(y, ei_p4f_1); y = ei_padd(y, ei_p4f_1);
/* Evaluate the second polynom (Pi/4 <= x <= 0) */ /* Evaluate the second polynom (Pi/4 <= x <= 0) */
Packet4f y2 = ei_p4f_sincof_p0; Packet4f y2 = ei_p4f_sincof_p0;
y2 = ei_pmadd(y2, z, ei_p4f_sincof_p1); y2 = ei_pmadd(y2, z, ei_p4f_sincof_p1);
@ -360,7 +360,7 @@ static EIGEN_DONT_INLINE EIGEN_UNUSED Packet4f ei_pcos(Packet4f x)
y2 = _mm_and_ps(poly_mask, y2); y2 = _mm_and_ps(poly_mask, y2);
y = _mm_andnot_ps(poly_mask, y); y = _mm_andnot_ps(poly_mask, y);
y = _mm_or_ps(y,y2); y = _mm_or_ps(y,y2);
/* update the sign */ /* update the sign */
return _mm_xor_ps(y, sign_bit); return _mm_xor_ps(y, sign_bit);
} }

8
test/triangular.cpp
View File

@ -114,6 +114,14 @@ template<typename MatrixType> void triangular(const MatrixType& m)
VERIFY((m1.template part<Eigen::UpperTriangular>() * m2.template part<Eigen::UpperTriangular>()).isUpperTriangular()); VERIFY((m1.template part<Eigen::UpperTriangular>() * m2.template part<Eigen::UpperTriangular>()).isUpperTriangular());
// test swap
m1.setOnes();
m2.setZero();
m2.template part<Eigen::UpperTriangular>().swap(m1);
m3.setZero();
m3.template part<Eigen::UpperTriangular>().setOnes();
VERIFY_IS_APPROX(m2,m3);
} }
void test_triangular() void test_triangular()