potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 23:33:05 +04:00
Code Packages Releases Wiki Activity

Merge remote-tracking branch 'upstream/3.4' into merge-3.4

Browse Source
This commit is contained in:
Alexander Alekhin
2018-09-04 19:35:38 +03:00
committed by Alexander Alekhin
parent e0bfe44cac 351ee2e3a5
commit d74b98c3d9
117 changed files with 6913 additions and 3713 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/hal/intrin_avx.hpp
+2
View File
@@ -664,6 +664,8 @@ inline void v_mul_expand(const v_uint32x8& a, const v_uint32x8& b,
v_zip(v_uint64x4(v0), v_uint64x4(v1), c, d);
}
inline v_int16x16 v_mul_hi(const v_int16x16& a, const v_int16x16& b) { return v_int16x16(_mm256_mulhi_epi16(a.val, b.val)); }
inline v_uint16x16 v_mul_hi(const v_uint16x16& a, const v_uint16x16& b) { return v_uint16x16(_mm256_mulhi_epu16(a.val, b.val)); }
/** Non-saturating arithmetics **/
#define OPENCV_HAL_IMPL_AVX_BIN_FUNC(func, _Tpvec, intrin) \
modules/core/include/opencv2/core/hal/intrin_cpp.hpp
+14
View File
@@ -891,6 +891,20 @@ template<typename _Tp, int n> inline void v_mul_expand(const v_reg<_Tp, n>& a, c
}
}
/** @brief Multiply and extract high part
Multiply values two registers and store high part of the results.
Implemented only for 16-bit source types (v_int16x8, v_uint16x8). Returns \f$ a*b >> 16 \f$
*/
template<typename _Tp, int n> inline v_reg<_Tp, n> v_mul_hi(const v_reg<_Tp, n>& a, const v_reg<_Tp, n>& b)
{
typedef typename V_TypeTraits<_Tp>::w_type w_type;
v_reg<_Tp, n> c;
for (int i = 0; i < n; i++)
c.s[i] = (_Tp)(((w_type)a.s[i] * b.s[i]) >> sizeof(_Tp)*8);
return c;
}
//! @cond IGNORED
template<typename _Tp, int n> inline void v_hsum(const v_reg<_Tp, n>& a,
v_reg<typename V_TypeTraits<_Tp>::w_type, n/2>& c)
modules/core/include/opencv2/core/hal/intrin_neon.hpp
+15
View File
@@ -553,6 +553,21 @@ inline void v_mul_expand(const v_uint32x4& a, const v_uint32x4& b,
d.val = vmull_u32(vget_high_u32(a.val), vget_high_u32(b.val));
}
inline v_int16x8 v_mul_hi(const v_int16x8& a, const v_int16x8& b)
{
return v_int16x8(vcombine_s16(
vshrn_n_s32(vmull_s16( vget_low_s16(a.val), vget_low_s16(b.val)), 16),
vshrn_n_s32(vmull_s16(vget_high_s16(a.val), vget_high_s16(b.val)), 16)
));
}
inline v_uint16x8 v_mul_hi(const v_uint16x8& a, const v_uint16x8& b)
{
return v_uint16x8(vcombine_u16(
vshrn_n_u32(vmull_u16( vget_low_u16(a.val), vget_low_u16(b.val)), 16),
vshrn_n_u32(vmull_u16(vget_high_u16(a.val), vget_high_u16(b.val)), 16)
));
}
inline v_int32x4 v_dotprod(const v_int16x8& a, const v_int16x8& b)
{
int32x4_t c = vmull_s16(vget_low_s16(a.val), vget_low_s16(b.val));
modules/core/include/opencv2/core/hal/intrin_sse.hpp
+3
View File
@@ -737,6 +737,9 @@ inline void v_mul_expand(const v_uint32x4& a, const v_uint32x4& b,
d.val = _mm_unpackhi_epi64(c0, c1);
}
inline v_int16x8 v_mul_hi(const v_int16x8& a, const v_int16x8& b) { return v_int16x8(_mm_mulhi_epi16(a.val, b.val)); }
inline v_uint16x8 v_mul_hi(const v_uint16x8& a, const v_uint16x8& b) { return v_uint16x8(_mm_mulhi_epu16(a.val, b.val)); }
inline v_int32x4 v_dotprod(const v_int16x8& a, const v_int16x8& b)
{
return v_int32x4(_mm_madd_epi16(a.val, b.val));
modules/core/include/opencv2/core/hal/intrin_vsx.hpp
+15
View File
@@ -457,6 +457,21 @@ inline void v_mul_expand(const v_uint32x4& a, const v_uint32x4& b, v_uint64x2& c
d.val = vec_mul(vec_unpacklu(a.val), vec_unpacklu(b.val));
}
inline v_int16x8 v_mul_hi(const v_int16x8& a, const v_int16x8& b)
{
return v_int16x8(vec_packs(
vec_sra(vec_mul(vec_unpackh(a.val), vec_unpackh(b.val)), vec_uint4_sp(16)),
vec_sra(vec_mul(vec_unpackl(a.val), vec_unpackl(b.val)), vec_uint4_sp(16))
));
}
inline v_uint16x8 v_mul_hi(const v_uint16x8& a, const v_uint16x8& b)
{
return v_uint16x8(vec_packs(
vec_sr(vec_mul(vec_unpackhu(a.val), vec_unpackhu(b.val)), vec_uint4_sp(16)),
vec_sr(vec_mul(vec_unpacklu(a.val), vec_unpacklu(b.val)), vec_uint4_sp(16))
));
}
/** Non-saturating arithmetics **/
#define OPENCV_HAL_IMPL_VSX_BIN_FUNC(func, intrin) \
template<typename _Tpvec> \
modules/core/include/opencv2/core/ippasync.hpp
+1 -1
View File
@@ -45,7 +45,7 @@
#ifndef OPENCV_CORE_IPPASYNC_HPP
#define OPENCV_CORE_IPPASYNC_HPP
#ifdef HAVE_IPP_A
#ifdef HAVE_IPP_A // this file will be removed in OpenCV 4.0
#include "opencv2/core.hpp"
#include <ipp_async_op.h>
modules/core/include/opencv2/core/mat.hpp
+22 -12
View File
@@ -146,6 +146,12 @@ synonym is needed to generate Python/Java etc. wrappers properly. At the functio
level their use is similar, but _InputArray::getMat(idx) should be used to get header for the
idx-th component of the outer vector and _InputArray::size().area() should be used to find the
number of components (vectors/matrices) of the outer vector.
In general, type support is limited to cv::Mat types. Other types are forbidden.
But in some cases we need to support passing of custom non-general Mat types, like arrays of cv::KeyPoint, cv::DMatch, etc.
This data is not intented to be interpreted as an image data, or processed somehow like regular cv::Mat.
To pass such custom type use rawIn() / rawOut() / rawInOut() wrappers.
Custom type is wrapped as Mat-compatible `CV_8UC<N>` values (N = sizeof(T), N <= CV_CN_MAX).
*/
class CV_EXPORTS _InputArray
{
@@ -199,6 +205,9 @@ public:
template<typename _Tp, std::size_t _Nm> _InputArray(const std::array<_Tp, _Nm>& arr);
template<std::size_t _Nm> _InputArray(const std::array<Mat, _Nm>& arr);
template<typename _Tp> static _InputArray rawIn(const std::vector<_Tp>& vec);
template<typename _Tp, std::size_t _Nm> static _InputArray rawIn(const std::array<_Tp, _Nm>& arr);
Mat getMat(int idx=-1) const;
Mat getMat_(int idx=-1) const;
UMat getUMat(int idx=-1) const;
@@ -328,12 +337,13 @@ public:
_OutputArray(const UMat& m);
_OutputArray(const std::vector<UMat>& vec);
#ifdef CV_CXX_STD_ARRAY
template<typename _Tp, std::size_t _Nm> _OutputArray(std::array<_Tp, _Nm>& arr);
template<typename _Tp, std::size_t _Nm> _OutputArray(const std::array<_Tp, _Nm>& arr);
template<std::size_t _Nm> _OutputArray(std::array<Mat, _Nm>& arr);
template<std::size_t _Nm> _OutputArray(const std::array<Mat, _Nm>& arr);
#endif
template<typename _Tp> static _OutputArray rawOut(std::vector<_Tp>& vec);
template<typename _Tp, std::size_t _Nm> static _OutputArray rawOut(std::array<_Tp, _Nm>& arr);
bool fixedSize() const;
bool fixedType() const;
@@ -397,15 +407,23 @@ public:
_InputOutputArray(const UMat& m);
_InputOutputArray(const std::vector<UMat>& vec);
#ifdef CV_CXX_STD_ARRAY
template<typename _Tp, std::size_t _Nm> _InputOutputArray(std::array<_Tp, _Nm>& arr);
template<typename _Tp, std::size_t _Nm> _InputOutputArray(const std::array<_Tp, _Nm>& arr);
template<std::size_t _Nm> _InputOutputArray(std::array<Mat, _Nm>& arr);
template<std::size_t _Nm> _InputOutputArray(const std::array<Mat, _Nm>& arr);
#endif
template<typename _Tp> static _InputOutputArray rawInOut(std::vector<_Tp>& vec);
template<typename _Tp, std::size_t _Nm> _InputOutputArray rawInOut(std::array<_Tp, _Nm>& arr);
};
/** Helper to wrap custom types. @see InputArray */
template<typename _Tp> static inline _InputArray rawIn(_Tp& v);
/** Helper to wrap custom types. @see InputArray */
template<typename _Tp> static inline _OutputArray rawOut(_Tp& v);
/** Helper to wrap custom types. @see InputArray */
template<typename _Tp> static inline _InputOutputArray rawInOut(_Tp& v);
CV__DEBUG_NS_END
typedef const _InputArray& InputArray;
@@ -991,11 +1009,9 @@ public:
*/
template<typename _Tp> explicit Mat(const std::initializer_list<int> sizes, const std::initializer_list<_Tp> list);
#ifdef CV_CXX_STD_ARRAY
/** @overload
*/
template<typename _Tp, size_t _Nm> explicit Mat(const std::array<_Tp, _Nm>& arr, bool copyData=false);
#endif
/** @overload
*/
@@ -1630,9 +1646,7 @@ public:
template<typename _Tp, int n> operator Vec<_Tp, n>() const;
template<typename _Tp, int m, int n> operator Matx<_Tp, m, n>() const;
#ifdef CV_CXX_STD_ARRAY
template<typename _Tp, std::size_t _Nm> operator std::array<_Tp, _Nm>() const;
#endif
/** @brief Reports whether the matrix is continuous or not.
@@ -2214,9 +2228,7 @@ public:
Mat_(std::initializer_list<_Tp> values);
explicit Mat_(const std::initializer_list<int> sizes, const std::initializer_list<_Tp> values);
#ifdef CV_CXX_STD_ARRAY
template <std::size_t _Nm> explicit Mat_(const std::array<_Tp, _Nm>& arr, bool copyData=false);
#endif
Mat_& operator = (const Mat& m);
Mat_& operator = (const Mat_& m);
@@ -2314,10 +2326,8 @@ public:
//! conversion to vector.
operator std::vector<_Tp>() const;
#ifdef CV_CXX_STD_ARRAY
//! conversion to array.
template<std::size_t _Nm> operator std::array<_Tp, _Nm>() const;
#endif
//! conversion to Vec
template<int n> operator Vec<typename DataType<_Tp>::channel_type, n>() const;
modules/core/include/opencv2/core/mat.inl.hpp
+72
View File
@@ -61,6 +61,16 @@ CV__DEBUG_NS_BEGIN
//! @cond IGNORED
////////////////////////// Custom (raw) type wrapper //////////////////////////
template<typename _Tp> static inline
int rawType()
{
CV_StaticAssert(sizeof(_Tp) <= CV_CN_MAX, "sizeof(_Tp) is too large");
const int elemSize = sizeof(_Tp);
return (int)CV_MAKETYPE(CV_8U, elemSize);
}
//////////////////////// Input/Output Arrays ////////////////////////
inline void _InputArray::init(int _flags, const void* _obj)
@@ -134,6 +144,25 @@ inline _InputArray::_InputArray(const ogl::Buffer& buf)
inline _InputArray::_InputArray(const cuda::HostMem& cuda_mem)
{ init(CUDA_HOST_MEM + ACCESS_READ, &cuda_mem); }
template<typename _Tp> inline
_InputArray _InputArray::rawIn(const std::vector<_Tp>& vec)
{
_InputArray v;
v.flags = _InputArray::FIXED_TYPE + _InputArray::STD_VECTOR + rawType<_Tp>() + ACCESS_READ;
v.obj = (void*)&vec;
return v;
}
template<typename _Tp, std::size_t _Nm> inline
_InputArray _InputArray::rawIn(const std::array<_Tp, _Nm>& arr)
{
_InputArray v;
v.flags = FIXED_TYPE + FIXED_SIZE + STD_ARRAY + traits::Type<_Tp>::value + ACCESS_READ;
v.obj = (void*)arr.data();
v.sz = Size(1, _Nm);
return v;
}
inline _InputArray::~_InputArray() {}
inline Mat _InputArray::getMat(int i) const
@@ -261,6 +290,25 @@ inline _OutputArray::_OutputArray(const ogl::Buffer& buf)
inline _OutputArray::_OutputArray(const cuda::HostMem& cuda_mem)
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_HOST_MEM + ACCESS_WRITE, &cuda_mem); }
template<typename _Tp> inline
_OutputArray _OutputArray::rawOut(std::vector<_Tp>& vec)
{
_OutputArray v;
v.flags = _InputArray::FIXED_TYPE + _InputArray::STD_VECTOR + rawType<_Tp>() + ACCESS_WRITE;
v.obj = (void*)&vec;
return v;
}
template<typename _Tp, std::size_t _Nm> inline
_OutputArray _OutputArray::rawOut(std::array<_Tp, _Nm>& arr)
{
_OutputArray v;
v.flags = FIXED_TYPE + FIXED_SIZE + STD_ARRAY + traits::Type<_Tp>::value + ACCESS_WRITE;
v.obj = (void*)arr.data();
v.sz = Size(1, _Nm);
return v;
}
///////////////////////////////////////////////////////////////////////////////////////////
inline _InputOutputArray::_InputOutputArray() { init(ACCESS_RW, 0); }
@@ -370,6 +418,30 @@ inline _InputOutputArray::_InputOutputArray(const ogl::Buffer& buf)
inline _InputOutputArray::_InputOutputArray(const cuda::HostMem& cuda_mem)
{ init(FIXED_TYPE + FIXED_SIZE + CUDA_HOST_MEM + ACCESS_RW, &cuda_mem); }
template<typename _Tp> inline
_InputOutputArray _InputOutputArray::rawInOut(std::vector<_Tp>& vec)
{
_InputOutputArray v;
v.flags = _InputArray::FIXED_TYPE + _InputArray::STD_VECTOR + rawType<_Tp>() + ACCESS_RW;
v.obj = (void*)&vec;
return v;
}
template<typename _Tp, std::size_t _Nm> inline
_InputOutputArray _InputOutputArray::rawInOut(std::array<_Tp, _Nm>& arr)
{
_InputOutputArray v;
v.flags = FIXED_TYPE + FIXED_SIZE + STD_ARRAY + traits::Type<_Tp>::value + ACCESS_RW;
v.obj = (void*)arr.data();
v.sz = Size(1, _Nm);
return v;
}
template<typename _Tp> static inline _InputArray rawIn(_Tp& v) { return _InputArray::rawIn(v); }
template<typename _Tp> static inline _OutputArray rawOut(_Tp& v) { return _OutputArray::rawOut(v); }
template<typename _Tp> static inline _InputOutputArray rawInOut(_Tp& v) { return _InputOutputArray::rawInOut(v); }
CV__DEBUG_NS_END
//////////////////////////////////////////// Mat //////////////////////////////////////////