mirror of
https://gitlab.com/libeigen/eigen.git
synced 2026-04-10 11:34:33 +08:00
Merged latest changes from the parent
This commit is contained in:
@@ -49,7 +49,7 @@ class ArrayWrapper : public ArrayBase<ArrayWrapper<ExpressionType> >
|
||||
typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
|
||||
EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
|
||||
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline Index rows() const { return m_expression.rows(); }
|
||||
|
||||
@@ -574,13 +574,13 @@ public:
|
||||
template<int StoreMode, int LoadMode>
|
||||
void assignPacket(Index row, Index col)
|
||||
{
|
||||
m_functor.assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode>(row,col));
|
||||
m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode>(row,col));
|
||||
}
|
||||
|
||||
template<int StoreMode, int LoadMode>
|
||||
void assignPacket(Index index)
|
||||
{
|
||||
m_functor.assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode>(index));
|
||||
m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode>(index));
|
||||
}
|
||||
|
||||
template<int StoreMode, int LoadMode>
|
||||
|
||||
@@ -45,6 +45,18 @@ struct CommaInitializer
|
||||
m_xpr.block(0, 0, other.rows(), other.cols()) = other;
|
||||
}
|
||||
|
||||
/* Copy/Move constructor which transfers ownership. This is crucial in
|
||||
* absence of return value optimization to avoid assertions during destruction. */
|
||||
// FIXME in C++11 mode this could be replaced by a proper RValue constructor
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline CommaInitializer(const CommaInitializer& o)
|
||||
: m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) {
|
||||
// Mark original object as finished. In absence of R-value references we need to const_cast:
|
||||
const_cast<CommaInitializer&>(o).m_row = m_xpr.rows();
|
||||
const_cast<CommaInitializer&>(o).m_col = m_xpr.cols();
|
||||
const_cast<CommaInitializer&>(o).m_currentBlockRows = 0;
|
||||
}
|
||||
|
||||
/* inserts a scalar value in the target matrix */
|
||||
EIGEN_DEVICE_FUNC
|
||||
CommaInitializer& operator,(const Scalar& s)
|
||||
@@ -110,7 +122,7 @@ struct CommaInitializer
|
||||
EIGEN_DEVICE_FUNC
|
||||
inline XprType& finished() { return m_xpr; }
|
||||
|
||||
XprType& m_xpr; // target expression
|
||||
XprType& m_xpr; // target expression
|
||||
Index m_row; // current row id
|
||||
Index m_col; // current col id
|
||||
Index m_currentBlockRows; // current block height
|
||||
|
||||
@@ -356,8 +356,6 @@ template<typename Derived> class MatrixBase
|
||||
|
||||
Scalar trace() const;
|
||||
|
||||
/////////// Array module ///////////
|
||||
|
||||
template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
|
||||
|
||||
EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
|
||||
@@ -365,8 +363,10 @@ template<typename Derived> class MatrixBase
|
||||
|
||||
/** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
|
||||
* \sa ArrayBase::matrix() */
|
||||
EIGEN_DEVICE_FUNC ArrayWrapper<Derived> array() { return derived(); }
|
||||
EIGEN_DEVICE_FUNC const ArrayWrapper<const Derived> array() const { return derived(); }
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper<Derived> array() { return derived(); }
|
||||
/** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix
|
||||
* \sa ArrayBase::matrix() */
|
||||
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper<const Derived> array() const { return derived(); }
|
||||
|
||||
/////////// LU module ///////////
|
||||
|
||||
|
||||
@@ -101,7 +101,7 @@ struct traits<Ref<_PlainObjectType, _Options, _StrideType> >
|
||||
template<typename Derived> struct match {
|
||||
enum {
|
||||
HasDirectAccess = internal::has_direct_access<Derived>::ret,
|
||||
StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
|
||||
StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
|
||||
InnerStrideMatch = int(StrideType::InnerStrideAtCompileTime)==int(Dynamic)
|
||||
|| int(StrideType::InnerStrideAtCompileTime)==int(Derived::InnerStrideAtCompileTime)
|
||||
|| (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
|
||||
@@ -172,8 +172,12 @@ protected:
|
||||
}
|
||||
else
|
||||
::new (static_cast<Base*>(this)) Base(expr.data(), expr.rows(), expr.cols());
|
||||
::new (&m_stride) StrideBase(StrideType::OuterStrideAtCompileTime==0?0:expr.outerStride(),
|
||||
StrideType::InnerStrideAtCompileTime==0?0:expr.innerStride());
|
||||
|
||||
if(Expression::IsVectorAtCompileTime && (!PlainObjectType::IsVectorAtCompileTime) && ((Expression::Flags&RowMajorBit)!=(PlainObjectType::Flags&RowMajorBit)))
|
||||
::new (&m_stride) StrideBase(expr.innerStride(), StrideType::InnerStrideAtCompileTime==0?0:1);
|
||||
else
|
||||
::new (&m_stride) StrideBase(StrideType::OuterStrideAtCompileTime==0?0:expr.outerStride(),
|
||||
StrideType::InnerStrideAtCompileTime==0?0:expr.innerStride());
|
||||
}
|
||||
|
||||
StrideBase m_stride;
|
||||
|
||||
@@ -1277,6 +1277,7 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, Pan
|
||||
template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
|
||||
struct gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
|
||||
{
|
||||
typedef typename packet_traits<Scalar>::type Packet;
|
||||
enum { PacketSize = packet_traits<Scalar>::size };
|
||||
EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
|
||||
};
|
||||
@@ -1298,12 +1299,18 @@ EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, Pan
|
||||
if(PanelMode) count += nr * offset;
|
||||
for(Index k=0; k<depth; k++)
|
||||
{
|
||||
const Scalar* b0 = &rhs[k*rhsStride + j2];
|
||||
blockB[count+0] = cj(b0[0]);
|
||||
blockB[count+1] = cj(b0[1]);
|
||||
if(nr==4) blockB[count+2] = cj(b0[2]);
|
||||
if(nr==4) blockB[count+3] = cj(b0[3]);
|
||||
count += nr;
|
||||
if (nr == PacketSize) {
|
||||
Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
|
||||
pstoreu(blockB+count, cj.pconj(A));
|
||||
count += PacketSize;
|
||||
} else {
|
||||
const Scalar* b0 = &rhs[k*rhsStride + j2];
|
||||
blockB[count+0] = cj(b0[0]);
|
||||
blockB[count+1] = cj(b0[1]);
|
||||
if(nr==4) blockB[count+2] = cj(b0[2]);
|
||||
if(nr==4) blockB[count+3] = cj(b0[3]);
|
||||
count += nr;
|
||||
}
|
||||
}
|
||||
// skip what we have after
|
||||
if(PanelMode) count += nr * (stride-offset-depth);
|
||||
|
||||
@@ -113,9 +113,9 @@ EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrd
|
||||
|
||||
for (size_t i=starti; i<alignedStart; ++i)
|
||||
{
|
||||
res[i] += t0 * A0[i] + t1 * A1[i];
|
||||
t2 += numext::conj(A0[i]) * rhs[i];
|
||||
t3 += numext::conj(A1[i]) * rhs[i];
|
||||
res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
|
||||
t2 += cj1.pmul(A0[i], rhs[i]);
|
||||
t3 += cj1.pmul(A1[i], rhs[i]);
|
||||
}
|
||||
// Yes this an optimization for gcc 4.3 and 4.4 (=> huge speed up)
|
||||
// gcc 4.2 does this optimization automatically.
|
||||
|
||||
@@ -274,12 +274,12 @@ inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
|
||||
// The defined(_mm_free) is just here to verify that this MSVC version
|
||||
// implements _mm_malloc/_mm_free based on the corresponding _aligned_
|
||||
// functions. This may not always be the case and we just try to be safe.
|
||||
#if defined(_MSC_VER) && defined(_mm_free)
|
||||
#if defined(_MSC_VER) && (!defined(_WIN32_WCE)) && defined(_mm_free)
|
||||
result = _aligned_realloc(ptr,new_size,EIGEN_ALIGN_BYTES);
|
||||
#else
|
||||
result = generic_aligned_realloc(ptr,new_size,old_size);
|
||||
#endif
|
||||
#elif defined(_MSC_VER)
|
||||
#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
|
||||
result = _aligned_realloc(ptr,new_size,EIGEN_ALIGN_BYTES);
|
||||
#else
|
||||
result = handmade_aligned_realloc(ptr,new_size,old_size);
|
||||
@@ -464,7 +464,7 @@ template<typename T, bool Align> inline void conditional_aligned_delete_auto(T *
|
||||
* There is also the variant first_aligned(const MatrixBase&) defined in DenseCoeffsBase.h.
|
||||
*/
|
||||
template<typename Scalar, typename Index>
|
||||
static inline Index first_aligned(const Scalar* array, Index size)
|
||||
inline Index first_aligned(const Scalar* array, Index size)
|
||||
{
|
||||
enum { PacketSize = packet_traits<Scalar>::size,
|
||||
PacketAlignedMask = PacketSize-1
|
||||
@@ -492,7 +492,7 @@ static inline Index first_aligned(const Scalar* array, Index size)
|
||||
/** \internal Returns the smallest integer multiple of \a base and greater or equal to \a size
|
||||
*/
|
||||
template<typename Index>
|
||||
inline static Index first_multiple(Index size, Index base)
|
||||
inline Index first_multiple(Index size, Index base)
|
||||
{
|
||||
return ((size+base-1)/base)*base;
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user