generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #3935 from vpisarev:extending_hal_part1

Browse Source
This commit is contained in:
Vadim Pisarevsky
2015-04-21 14:02:02 +00:00
parent dce0405c4d 926754a66e
commit 063e4004ba
30 changed files with 5964 additions and 2173 deletions
Download Patch File Download Diff File Expand all files Collapse all files

176
modules/core/include/opencv2/core/base.hpp
View File

@@ -53,6 +53,7 @@
#include "opencv2/core/cvdef.h"
#include "opencv2/core/cvstd.hpp"
#include "opencv2/hal.hpp"
namespace cv
{
@@ -400,136 +401,6 @@ configurations while CV_DbgAssert is only retained in the Debug configuration.
# define CV_DbgAssert(expr)
#endif
/////////////// saturate_cast (used in image & signal processing) ///////////////////
/**
Template function for accurate conversion from one primitive type to another.
The functions saturate_cast resemble the standard C++ cast operations, such as static_cast\<T\>()
and others. They perform an efficient and accurate conversion from one primitive type to another
(see the introduction chapter). saturate in the name means that when the input value v is out of the
range of the target type, the result is not formed just by taking low bits of the input, but instead
the value is clipped. For example:
@code
uchar a = saturate_cast<uchar>(-100); // a = 0 (UCHAR_MIN)
short b = saturate_cast<short>(33333.33333); // b = 32767 (SHRT_MAX)
@endcode
Such clipping is done when the target type is unsigned char , signed char , unsigned short or
signed short . For 32-bit integers, no clipping is done.
When the parameter is a floating-point value and the target type is an integer (8-, 16- or 32-bit),
the floating-point value is first rounded to the nearest integer and then clipped if needed (when
the target type is 8- or 16-bit).
This operation is used in the simplest or most complex image processing functions in OpenCV.
@param v Function parameter.
@sa add, subtract, multiply, divide, Mat::convertTo
*/
template<typename _Tp> static inline _Tp saturate_cast(uchar v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(schar v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(ushort v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(short v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(unsigned v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(int v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(float v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(double v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(int64 v) { return _Tp(v); }
/** @overload */
template<typename _Tp> static inline _Tp saturate_cast(uint64 v) { return _Tp(v); }
//! @cond IGNORED
template<> inline uchar saturate_cast<uchar>(schar v) { return (uchar)std::max((int)v, 0); }
template<> inline uchar saturate_cast<uchar>(ushort v) { return (uchar)std::min((unsigned)v, (unsigned)UCHAR_MAX); }
template<> inline uchar saturate_cast<uchar>(int v) { return (uchar)((unsigned)v <= UCHAR_MAX ? v : v > 0 ? UCHAR_MAX : 0); }
template<> inline uchar saturate_cast<uchar>(short v) { return saturate_cast<uchar>((int)v); }
template<> inline uchar saturate_cast<uchar>(unsigned v) { return (uchar)std::min(v, (unsigned)UCHAR_MAX); }
template<> inline uchar saturate_cast<uchar>(float v) { int iv = cvRound(v); return saturate_cast<uchar>(iv); }
template<> inline uchar saturate_cast<uchar>(double v) { int iv = cvRound(v); return saturate_cast<uchar>(iv); }
template<> inline uchar saturate_cast<uchar>(int64 v) { return (uchar)((uint64)v <= (uint64)UCHAR_MAX ? v : v > 0 ? UCHAR_MAX : 0); }
template<> inline uchar saturate_cast<uchar>(uint64 v) { return (uchar)std::min(v, (uint64)UCHAR_MAX); }
template<> inline schar saturate_cast<schar>(uchar v) { return (schar)std::min((int)v, SCHAR_MAX); }
template<> inline schar saturate_cast<schar>(ushort v) { return (schar)std::min((unsigned)v, (unsigned)SCHAR_MAX); }
template<> inline schar saturate_cast<schar>(int v) { return (schar)((unsigned)(v-SCHAR_MIN) <= (unsigned)UCHAR_MAX ? v : v > 0 ? SCHAR_MAX : SCHAR_MIN); }
template<> inline schar saturate_cast<schar>(short v) { return saturate_cast<schar>((int)v); }
template<> inline schar saturate_cast<schar>(unsigned v) { return (schar)std::min(v, (unsigned)SCHAR_MAX); }
template<> inline schar saturate_cast<schar>(float v) { int iv = cvRound(v); return saturate_cast<schar>(iv); }
template<> inline schar saturate_cast<schar>(double v) { int iv = cvRound(v); return saturate_cast<schar>(iv); }
template<> inline schar saturate_cast<schar>(int64 v) { return (schar)((uint64)((int64)v-SCHAR_MIN) <= (uint64)UCHAR_MAX ? v : v > 0 ? SCHAR_MAX : SCHAR_MIN); }
template<> inline schar saturate_cast<schar>(uint64 v) { return (schar)std::min(v, (uint64)SCHAR_MAX); }
template<> inline ushort saturate_cast<ushort>(schar v) { return (ushort)std::max((int)v, 0); }
template<> inline ushort saturate_cast<ushort>(short v) { return (ushort)std::max((int)v, 0); }
template<> inline ushort saturate_cast<ushort>(int v) { return (ushort)((unsigned)v <= (unsigned)USHRT_MAX ? v : v > 0 ? USHRT_MAX : 0); }
template<> inline ushort saturate_cast<ushort>(unsigned v) { return (ushort)std::min(v, (unsigned)USHRT_MAX); }
template<> inline ushort saturate_cast<ushort>(float v) { int iv = cvRound(v); return saturate_cast<ushort>(iv); }
template<> inline ushort saturate_cast<ushort>(double v) { int iv = cvRound(v); return saturate_cast<ushort>(iv); }
template<> inline ushort saturate_cast<ushort>(int64 v) { return (ushort)((uint64)v <= (uint64)USHRT_MAX ? v : v > 0 ? USHRT_MAX : 0); }
template<> inline ushort saturate_cast<ushort>(uint64 v) { return (ushort)std::min(v, (uint64)USHRT_MAX); }
template<> inline short saturate_cast<short>(ushort v) { return (short)std::min((int)v, SHRT_MAX); }
template<> inline short saturate_cast<short>(int v) { return (short)((unsigned)(v - SHRT_MIN) <= (unsigned)USHRT_MAX ? v : v > 0 ? SHRT_MAX : SHRT_MIN); }
template<> inline short saturate_cast<short>(unsigned v) { return (short)std::min(v, (unsigned)SHRT_MAX); }
template<> inline short saturate_cast<short>(float v) { int iv = cvRound(v); return saturate_cast<short>(iv); }
template<> inline short saturate_cast<short>(double v) { int iv = cvRound(v); return saturate_cast<short>(iv); }
template<> inline short saturate_cast<short>(int64 v) { return (short)((uint64)((int64)v - SHRT_MIN) <= (uint64)USHRT_MAX ? v : v > 0 ? SHRT_MAX : SHRT_MIN); }
template<> inline short saturate_cast<short>(uint64 v) { return (short)std::min(v, (uint64)SHRT_MAX); }
template<> inline int saturate_cast<int>(float v) { return cvRound(v); }
template<> inline int saturate_cast<int>(double v) { return cvRound(v); }
// we intentionally do not clip negative numbers, to make -1 become 0xffffffff etc.
template<> inline unsigned saturate_cast<unsigned>(float v) { return cvRound(v); }
template<> inline unsigned saturate_cast<unsigned>(double v) { return cvRound(v); }
//! @endcond
//////////////////////////////// low-level functions ////////////////////////////////
CV_EXPORTS int LU(float* A, size_t astep, int m, float* b, size_t bstep, int n);
CV_EXPORTS int LU(double* A, size_t astep, int m, double* b, size_t bstep, int n);
CV_EXPORTS bool Cholesky(float* A, size_t astep, int m, float* b, size_t bstep, int n);
CV_EXPORTS bool Cholesky(double* A, size_t astep, int m, double* b, size_t bstep, int n);
CV_EXPORTS int normL1_(const uchar* a, const uchar* b, int n);
CV_EXPORTS float normL1_(const float* a, const float* b, int n);
CV_EXPORTS float normL2Sqr_(const float* a, const float* b, int n);
CV_EXPORTS void exp(const float* src, float* dst, int n);
CV_EXPORTS void log(const float* src, float* dst, int n);
CV_EXPORTS void fastAtan2(const float* y, const float* x, float* dst, int n, bool angleInDegrees);
CV_EXPORTS void magnitude(const float* x, const float* y, float* dst, int n);
/** @brief Computes the cube root of an argument.
The function cubeRoot computes \f$\sqrt[3]{\texttt{val}}\f$. Negative arguments are handled correctly.
NaN and Inf are not handled. The accuracy approaches the maximum possible accuracy for
single-precision data.
@param val A function argument.
*/
CV_EXPORTS_W float cubeRoot(float val);
/** @brief Calculates the angle of a 2D vector in degrees.
The function fastAtan2 calculates the full-range angle of an input 2D vector. The angle is measured
in degrees and varies from 0 to 360 degrees. The accuracy is about 0.3 degrees.
@param x x-coordinate of the vector.
@param y y-coordinate of the vector.
*/
CV_EXPORTS_W float fastAtan2(float y, float x);
/*
* Hamming distance functor - counts the bit differences between two strings - useful for the Brief descriptor
* bit count of A exclusive XOR'ed with B
@@ -549,6 +420,11 @@ typedef Hamming HammingLUT;
/////////////////////////////////// inline norms ////////////////////////////////////
template<typename _Tp> inline _Tp cv_abs(_Tp x) { return std::abs(x); }
inline int cv_abs(uchar x) { return x; }
inline int cv_abs(schar x) { return std::abs(x); }
inline int cv_abs(ushort x) { return x; }
inline int cv_abs(short x) { return std::abs(x); }
template<typename _Tp, typename _AccTp> static inline
_AccTp normL2Sqr(const _Tp* a, int n)
@@ -578,12 +454,12 @@ _AccTp normL1(const _Tp* a, int n)
#if CV_ENABLE_UNROLLED
for(; i <= n - 4; i += 4 )
{
s += (_AccTp)std::abs(a[i]) + (_AccTp)std::abs(a[i+1]) +
(_AccTp)std::abs(a[i+2]) + (_AccTp)std::abs(a[i+3]);
s += (_AccTp)cv_abs(a[i]) + (_AccTp)cv_abs(a[i+1]) +
(_AccTp)cv_abs(a[i+2]) + (_AccTp)cv_abs(a[i+3]);
}
#endif
for( ; i < n; i++ )
s += std::abs(a[i]);
s += cv_abs(a[i]);
return s;
}
@@ -592,7 +468,7 @@ _AccTp normInf(const _Tp* a, int n)
{
_AccTp s = 0;
for( int i = 0; i < n; i++ )
s = std::max(s, (_AccTp)std::abs(a[i]));
s = std::max(s, (_AccTp)cv_abs(a[i]));
return s;
}
@@ -616,11 +492,10 @@ _AccTp normL2Sqr(const _Tp* a, const _Tp* b, int n)
return s;
}
template<> inline
float normL2Sqr(const float* a, const float* b, int n)
inline float normL2Sqr(const float* a, const float* b, int n)
{
if( n >= 8 )
return normL2Sqr_(a, b, n);
return hal::normL2Sqr_(a, b, n);
float s = 0;
for( int i = 0; i < n; i++ )
{
@@ -650,11 +525,10 @@ _AccTp normL1(const _Tp* a, const _Tp* b, int n)
return s;
}
template<> inline
float normL1(const float* a, const float* b, int n)
inline float normL1(const float* a, const float* b, int n)
{
if( n >= 8 )
return normL1_(a, b, n);
return hal::normL1_(a, b, n);
float s = 0;
for( int i = 0; i < n; i++ )
{
@@ -664,10 +538,9 @@ float normL1(const float* a, const float* b, int n)
return s;
}
template<> inline
int normL1(const uchar* a, const uchar* b, int n)
inline int normL1(const uchar* a, const uchar* b, int n)
{
return normL1_(a, b, n);
return hal::normL1_(a, b, n);
}
template<typename _Tp, typename _AccTp> static inline
@@ -682,6 +555,23 @@ _AccTp normInf(const _Tp* a, const _Tp* b, int n)
return s;
}
/** @brief Computes the cube root of an argument.
The function cubeRoot computes \f$\sqrt[3]{\texttt{val}}\f$. Negative arguments are handled correctly.
NaN and Inf are not handled. The accuracy approaches the maximum possible accuracy for
single-precision data.
@param val A function argument.
*/
CV_EXPORTS_W float cubeRoot(float val);
/** @brief Calculates the angle of a 2D vector in degrees.
The function fastAtan2 calculates the full-range angle of an input 2D vector. The angle is measured
in degrees and varies from 0 to 360 degrees. The accuracy is about 0.3 degrees.
@param x x-coordinate of the vector.
@param y y-coordinate of the vector.
*/
CV_EXPORTS_W float fastAtan2(float y, float x);
////////////////// forward declarations for important OpenCV types //////////////////

181
modules/core/include/opencv2/core/cvdef.h
View File

@@ -70,16 +70,6 @@
# define CV_EXPORTS
#endif
#ifndef CV_INLINE
# if defined __cplusplus
# define CV_INLINE static inline
# elif defined _MSC_VER
# define CV_INLINE __inline
# else
# define CV_INLINE static
# endif
#endif
#ifndef CV_EXTERN_C
# ifdef __cplusplus
# define CV_EXTERN_C extern "C"
@@ -186,19 +176,6 @@
#define CV_ELEM_SIZE(type) \
(CV_MAT_CN(type) << ((((sizeof(size_t)/4+1)*16384|0x3a50) >> CV_MAT_DEPTH(type)*2) & 3))
/****************************************************************************************\
* fast math *
\****************************************************************************************/
#if defined __BORLANDC__
# include <fastmath.h>
#elif defined __cplusplus
# include <cmath>
#else
# include <math.h>
#endif
#ifndef MIN
# define MIN(a,b) ((a) > (b) ? (b) : (a))
#endif
@@ -207,164 +184,6 @@
# define MAX(a,b) ((a) < (b) ? (b) : (a))
#endif
#ifdef HAVE_TEGRA_OPTIMIZATION
# include "tegra_round.hpp"
#endif
//! @addtogroup core_utils
//! @{
#if CV_VFP
// 1. general scheme
#define ARM_ROUND(_value, _asm_string) \
int res; \
float temp; \
asm(_asm_string : [res] "=r" (res), [temp] "=w" (temp) : [value] "w" (_value)); \
return res;
// 2. version for double
#ifdef __clang__
#define ARM_ROUND_DBL(value) ARM_ROUND(value, "vcvtr.s32.f64 %[temp], %[value] \n vmov %[res], %[temp]")
#else
#define ARM_ROUND_DBL(value) ARM_ROUND(value, "vcvtr.s32.f64 %[temp], %P[value] \n vmov %[res], %[temp]")
#endif
// 3. version for float
#define ARM_ROUND_FLT(value) ARM_ROUND(value, "vcvtr.s32.f32 %[temp], %[value]\n vmov %[res], %[temp]")
#endif // CV_VFP
/** @brief Rounds floating-point number to the nearest integer
@param value floating-point number. If the value is outside of INT_MIN ... INT_MAX range, the
result is not defined.
*/
CV_INLINE int cvRound( double value )
{
#if ((defined _MSC_VER && defined _M_X64) || (defined __GNUC__ && defined __x86_64__ && defined __SSE2__ && !defined __APPLE__)) && !defined(__CUDACC__)
__m128d t = _mm_set_sd( value );
return _mm_cvtsd_si32(t);
#elif defined _MSC_VER && defined _M_IX86
int t;
__asm
{
fld value;
fistp t;
}
return t;
#elif ((defined _MSC_VER && defined _M_ARM) || defined CV_ICC || defined __GNUC__) && defined HAVE_TEGRA_OPTIMIZATION
TEGRA_ROUND_DBL(value);
#elif defined CV_ICC || defined __GNUC__
# if CV_VFP
ARM_ROUND_DBL(value)
# else
return (int)lrint(value);
# endif
#else
double intpart, fractpart;
fractpart = modf(value, &intpart);
if ((fabs(fractpart) != 0.5) || ((((int)intpart) % 2) != 0))
return (int)(value + (value >= 0 ? 0.5 : -0.5));
else
return (int)intpart;
#endif
}
#ifdef __cplusplus
/** @overload */
CV_INLINE int cvRound(float value)
{
#if defined ANDROID && (defined CV_ICC || defined __GNUC__) && defined HAVE_TEGRA_OPTIMIZATION
TEGRA_ROUND_FLT(value);
#elif CV_VFP && !defined HAVE_TEGRA_OPTIMIZATION
ARM_ROUND_FLT(value)
#else
return cvRound((double)value);
#endif
}
/** @overload */
CV_INLINE int cvRound(int value)
{
return value;
}
#endif // __cplusplus
/** @brief Rounds floating-point number to the nearest integer not larger than the original.
The function computes an integer i such that:
\f[i \le \texttt{value} < i+1\f]
@param value floating-point number. If the value is outside of INT_MIN ... INT_MAX range, the
result is not defined.
*/
CV_INLINE int cvFloor( double value )
{
#if (defined _MSC_VER && defined _M_X64 || (defined __GNUC__ && defined __SSE2__ && !defined __APPLE__)) && !defined(__CUDACC__)
__m128d t = _mm_set_sd( value );
int i = _mm_cvtsd_si32(t);
return i - _mm_movemask_pd(_mm_cmplt_sd(t, _mm_cvtsi32_sd(t,i)));
#elif defined __GNUC__
int i = (int)value;
return i - (i > value);
#else
int i = cvRound(value);
float diff = (float)(value - i);
return i - (diff < 0);
#endif
}
/** @brief Rounds floating-point number to the nearest integer not larger than the original.
The function computes an integer i such that:
\f[i \le \texttt{value} < i+1\f]
@param value floating-point number. If the value is outside of INT_MIN ... INT_MAX range, the
result is not defined.
*/
CV_INLINE int cvCeil( double value )
{
#if (defined _MSC_VER && defined _M_X64 || (defined __GNUC__ && defined __SSE2__&& !defined __APPLE__)) && !defined(__CUDACC__)
__m128d t = _mm_set_sd( value );
int i = _mm_cvtsd_si32(t);
return i + _mm_movemask_pd(_mm_cmplt_sd(_mm_cvtsi32_sd(t,i), t));
#elif defined __GNUC__
int i = (int)value;
return i + (i < value);
#else
int i = cvRound(value);
float diff = (float)(i - value);
return i + (diff < 0);
#endif
}
/** @brief Determines if the argument is Not A Number.
@param value The input floating-point value
The function returns 1 if the argument is Not A Number (as defined by IEEE754 standard), 0
otherwise. */
CV_INLINE int cvIsNaN( double value )
{
union { uint64 u; double f; } ieee754;
ieee754.f = value;
return ((unsigned)(ieee754.u >> 32) & 0x7fffffff) +
((unsigned)ieee754.u != 0) > 0x7ff00000;
}
/** @brief Determines if the argument is Infinity.
@param value The input floating-point value
The function returns 1 if the argument is a plus or minus infinity (as defined by IEEE754 standard)
and 0 otherwise. */
CV_INLINE int cvIsInf( double value )
{
union { uint64 u; double f; } ieee754;
ieee754.f = value;
return ((unsigned)(ieee754.u >> 32) & 0x7fffffff) == 0x7ff00000 &&
(unsigned)ieee754.u == 0;
}
//! @} core_utils
/****************************************************************************************\
* exchange-add operation for atomic operations on reference counters *
\****************************************************************************************/

2
modules/core/include/opencv2/core/matx.hpp
View File

@@ -427,7 +427,7 @@ template<typename _Tp, int m> struct Matx_DetOp
double operator ()(const Matx<_Tp, m, m>& a) const
{
Matx<_Tp, m, m> temp = a;
double p = LU(temp.val, m*sizeof(_Tp), m, 0, 0, 0);
double p = hal::LU(temp.val, m*sizeof(_Tp), m, 0, 0, 0);
if( p == 0 )
return p;
for( int i = 0; i < m; i++ )

4
modules/core/include/opencv2/core/operations.hpp
View File

@@ -72,9 +72,9 @@ template<typename _Tp, int m> struct Matx_FastInvOp
b(i, i) = (_Tp)1;
if( method == DECOMP_CHOLESKY )
return Cholesky(temp.val, m*sizeof(_Tp), m, b.val, m*sizeof(_Tp), m);
return hal::Cholesky(temp.val, m*sizeof(_Tp), m, b.val, m*sizeof(_Tp), m);
return LU(temp.val, m*sizeof(_Tp), m, b.val, m*sizeof(_Tp), m) != 0;
return hal::LU(temp.val, m*sizeof(_Tp), m, b.val, m*sizeof(_Tp), m) != 0;
}
};

8
modules/core/include/opencv2/core/private.hpp
View File

@@ -136,14 +136,6 @@ namespace cv
/* the alignment of all the allocated buffers */
#define CV_MALLOC_ALIGN 16
#ifdef __GNUC__
# define CV_DECL_ALIGNED(x) __attribute__ ((aligned (x)))
#elif defined _MSC_VER
# define CV_DECL_ALIGNED(x) __declspec(align(x))
#else
# define CV_DECL_ALIGNED(x)
#endif
/* IEEE754 constants and macros */
#define CV_TOGGLE_FLT(x) ((x)^((int)(x) < 0 ? 0x7fffffff : 0))
#define CV_TOGGLE_DBL(x) ((x)^((int64)(x) < 0 ? CV_BIG_INT(0x7fffffffffffffff) : 0))

16
modules/core/include/opencv2/core/types_c.h
View File

@@ -113,22 +113,6 @@ bytes of the header. In C++ interface the role of CvArr is played by InputArray
*/
typedef void CvArr;
typedef union Cv32suf
{
int i;
unsigned u;
float f;
}
Cv32suf;
typedef union Cv64suf
{
int64 i;
uint64 u;
double f;
}
Cv64suf;
typedef int CVStatus;
/** @see cv::Error::Code */

8
modules/core/src/kmeans.cpp
View File

@@ -79,7 +79,7 @@ public:
for ( int i = begin; i<end; i++ )
{
tdist2[i] = std::min(normL2Sqr_(data + step*i, data + stepci, dims), dist[i]);
tdist2[i] = std::min(normL2Sqr(data + step*i, data + stepci, dims), dist[i]);
}
}
@@ -114,7 +114,7 @@ static void generateCentersPP(const Mat& _data, Mat& _out_centers,
for( i = 0; i < N; i++ )
{
dist[i] = normL2Sqr_(data + step*i, data + step*centers[0], dims);
dist[i] = normL2Sqr(data + step*i, data + step*centers[0], dims);
sum0 += dist[i];
}
@@ -189,7 +189,7 @@ public:
for( int k = 0; k < K; k++ )
{
const float* center = centers.ptr<float>(k);
const double dist = normL2Sqr_(sample, center, dims);
const double dist = normL2Sqr(sample, center, dims);
if( min_dist > dist )
{
@@ -384,7 +384,7 @@ double cv::kmeans( InputArray _data, int K,
if( labels[i] != max_k )
continue;
sample = data.ptr<float>(i);
double dist = normL2Sqr_(sample, _old_center, dims);
double dist = normL2Sqr(sample, _old_center, dims);
if( max_dist <= dist )
{

182
modules/core/src/lapack.cpp
View File

@@ -50,168 +50,6 @@
namespace cv
{
/****************************************************************************************\
* LU & Cholesky implementation for small matrices *
\****************************************************************************************/
template<typename _Tp> static inline int
LUImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n)
{
int i, j, k, p = 1;
astep /= sizeof(A[0]);
bstep /= sizeof(b[0]);
for( i = 0; i < m; i++ )
{
k = i;
for( j = i+1; j < m; j++ )
if( std::abs(A[j*astep + i]) > std::abs(A[k*astep + i]) )
k = j;
if( std::abs(A[k*astep + i]) < std::numeric_limits<_Tp>::epsilon() )
return 0;
if( k != i )
{
for( j = i; j < m; j++ )
std::swap(A[i*astep + j], A[k*astep + j]);
if( b )
for( j = 0; j < n; j++ )
std::swap(b[i*bstep + j], b[k*bstep + j]);
p = -p;
}
_Tp d = -1/A[i*astep + i];
for( j = i+1; j < m; j++ )
{
_Tp alpha = A[j*astep + i]*d;
for( k = i+1; k < m; k++ )
A[j*astep + k] += alpha*A[i*astep + k];
if( b )
for( k = 0; k < n; k++ )
b[j*bstep + k] += alpha*b[i*bstep + k];
}
A[i*astep + i] = -d;
}
if( b )
{
for( i = m-1; i >= 0; i-- )
for( j = 0; j < n; j++ )
{
_Tp s = b[i*bstep + j];
for( k = i+1; k < m; k++ )
s -= A[i*astep + k]*b[k*bstep + j];
b[i*bstep + j] = s*A[i*astep + i];
}
}
return p;
}
int LU(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n);
}
int LU(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n);
}
template<typename _Tp> static inline bool
CholImpl(_Tp* A, size_t astep, int m, _Tp* b, size_t bstep, int n)
{
_Tp* L = A;
int i, j, k;
double s;
astep /= sizeof(A[0]);
bstep /= sizeof(b[0]);
for( i = 0; i < m; i++ )
{
for( j = 0; j < i; j++ )
{
s = A[i*astep + j];
for( k = 0; k < j; k++ )
s -= L[i*astep + k]*L[j*astep + k];
L[i*astep + j] = (_Tp)(s*L[j*astep + j]);
}
s = A[i*astep + i];
for( k = 0; k < j; k++ )
{
double t = L[i*astep + k];
s -= t*t;
}
if( s < std::numeric_limits<_Tp>::epsilon() )
return false;
L[i*astep + i] = (_Tp)(1./std::sqrt(s));
}
if( !b )
return true;
// LLt x = b
// 1: L y = b
// 2. Lt x = y
/*
[ L00 ] y0 b0
[ L10 L11 ] y1 = b1
[ L20 L21 L22 ] y2 b2
[ L30 L31 L32 L33 ] y3 b3
[ L00 L10 L20 L30 ] x0 y0
[ L11 L21 L31 ] x1 = y1
[ L22 L32 ] x2 y2
[ L33 ] x3 y3
*/
for( i = 0; i < m; i++ )
{
for( j = 0; j < n; j++ )
{
s = b[i*bstep + j];
for( k = 0; k < i; k++ )
s -= L[i*astep + k]*b[k*bstep + j];
b[i*bstep + j] = (_Tp)(s*L[i*astep + i]);
}
}
for( i = m-1; i >= 0; i-- )
{
for( j = 0; j < n; j++ )
{
s = b[i*bstep + j];
for( k = m-1; k > i; k-- )
s -= L[k*astep + i]*b[k*bstep + j];
b[i*bstep + j] = (_Tp)(s*L[i*astep + i]);
}
}
return true;
}
bool Cholesky(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return CholImpl(A, astep, m, b, bstep, n);
}
bool Cholesky(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return CholImpl(A, astep, m, b, bstep, n);
}
template<typename _Tp> static inline _Tp hypot(_Tp a, _Tp b)
{
a = std::abs(a);
@@ -882,7 +720,7 @@ double cv::determinant( InputArray _mat )
Mat a(rows, rows, CV_32F, (uchar*)buffer);
mat.copyTo(a);
result = LU(a.ptr<float>(), a.step, rows, 0, 0, 0);
result = hal::LU(a.ptr<float>(), a.step, rows, 0, 0, 0);
if( result )
{
for( int i = 0; i < rows; i++ )
@@ -906,7 +744,7 @@ double cv::determinant( InputArray _mat )
Mat a(rows, rows, CV_64F, (uchar*)buffer);
mat.copyTo(a);
result = LU(a.ptr<double>(), a.step, rows, 0, 0, 0);
result = hal::LU(a.ptr<double>(), a.step, rows, 0, 0, 0);
if( result )
{
for( int i = 0; i < rows; i++ )
@@ -1169,13 +1007,13 @@ double cv::invert( InputArray _src, OutputArray _dst, int method )
setIdentity(dst);
if( method == DECOMP_LU && type == CV_32F )
result = LU(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n) != 0;
result = hal::LU(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n) != 0;
else if( method == DECOMP_LU && type == CV_64F )
result = LU(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n) != 0;
result = hal::LU(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n) != 0;
else if( method == DECOMP_CHOLESKY && type == CV_32F )
result = Cholesky(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n);
result = hal::Cholesky(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n);
else
result = Cholesky(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n);
result = hal::Cholesky(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n);
if( !result )
dst = Scalar(0);
@@ -1407,16 +1245,16 @@ bool cv::solve( InputArray _src, InputArray _src2arg, OutputArray _dst, int meth
if( method == DECOMP_LU )
{
if( type == CV_32F )
result = LU(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb) != 0;
result = hal::LU(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb) != 0;
else
result = LU(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb) != 0;
result = hal::LU(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb) != 0;
}
else if( method == DECOMP_CHOLESKY )
{
if( type == CV_32F )
result = Cholesky(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb);
result = hal::Cholesky(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb);
else
result = Cholesky(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb);
result = hal::Cholesky(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb);
}
else
{

1489
modules/core/src/mathfuncs.cpp
View File

File diff suppressed because it is too large Load Diff

166
modules/core/src/stat.cpp
View File

@@ -2416,140 +2416,6 @@ void cv::minMaxLoc( InputArray _img, double* minVal, double* maxVal,
namespace cv
{
float normL2Sqr_(const float* a, const float* b, int n)
{
int j = 0; float d = 0.f;
#if CV_SSE
if( USE_SSE2 )
{
float CV_DECL_ALIGNED(16) buf[4];
__m128 d0 = _mm_setzero_ps(), d1 = _mm_setzero_ps();
for( ; j <= n - 8; j += 8 )
{
__m128 t0 = _mm_sub_ps(_mm_loadu_ps(a + j), _mm_loadu_ps(b + j));
__m128 t1 = _mm_sub_ps(_mm_loadu_ps(a + j + 4), _mm_loadu_ps(b + j + 4));
d0 = _mm_add_ps(d0, _mm_mul_ps(t0, t0));
d1 = _mm_add_ps(d1, _mm_mul_ps(t1, t1));
}
_mm_store_ps(buf, _mm_add_ps(d0, d1));
d = buf[0] + buf[1] + buf[2] + buf[3];
}
else
#endif
{
for( ; j <= n - 4; j += 4 )
{
float t0 = a[j] - b[j], t1 = a[j+1] - b[j+1], t2 = a[j+2] - b[j+2], t3 = a[j+3] - b[j+3];
d += t0*t0 + t1*t1 + t2*t2 + t3*t3;
}
}
for( ; j < n; j++ )
{
float t = a[j] - b[j];
d += t*t;
}
return d;
}
float normL1_(const float* a, const float* b, int n)
{
int j = 0; float d = 0.f;
#if CV_SSE
if( USE_SSE2 )
{
float CV_DECL_ALIGNED(16) buf[4];
static const int CV_DECL_ALIGNED(16) absbuf[4] = {0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff};
__m128 d0 = _mm_setzero_ps(), d1 = _mm_setzero_ps();
__m128 absmask = _mm_load_ps((const float*)absbuf);
for( ; j <= n - 8; j += 8 )
{
__m128 t0 = _mm_sub_ps(_mm_loadu_ps(a + j), _mm_loadu_ps(b + j));
__m128 t1 = _mm_sub_ps(_mm_loadu_ps(a + j + 4), _mm_loadu_ps(b + j + 4));
d0 = _mm_add_ps(d0, _mm_and_ps(t0, absmask));
d1 = _mm_add_ps(d1, _mm_and_ps(t1, absmask));
}
_mm_store_ps(buf, _mm_add_ps(d0, d1));
d = buf[0] + buf[1] + buf[2] + buf[3];
}
else
#elif CV_NEON
float32x4_t v_sum = vdupq_n_f32(0.0f);
for ( ; j <= n - 4; j += 4)
v_sum = vaddq_f32(v_sum, vabdq_f32(vld1q_f32(a + j), vld1q_f32(b + j)));
float CV_DECL_ALIGNED(16) buf[4];
vst1q_f32(buf, v_sum);
d = buf[0] + buf[1] + buf[2] + buf[3];
#endif
{
for( ; j <= n - 4; j += 4 )
{
d += std::abs(a[j] - b[j]) + std::abs(a[j+1] - b[j+1]) +
std::abs(a[j+2] - b[j+2]) + std::abs(a[j+3] - b[j+3]);
}
}
for( ; j < n; j++ )
d += std::abs(a[j] - b[j]);
return d;
}
int normL1_(const uchar* a, const uchar* b, int n)
{
int j = 0, d = 0;
#if CV_SSE
if( USE_SSE2 )
{
__m128i d0 = _mm_setzero_si128();
for( ; j <= n - 16; j += 16 )
{
__m128i t0 = _mm_loadu_si128((const __m128i*)(a + j));
__m128i t1 = _mm_loadu_si128((const __m128i*)(b + j));
d0 = _mm_add_epi32(d0, _mm_sad_epu8(t0, t1));
}
for( ; j <= n - 4; j += 4 )
{
__m128i t0 = _mm_cvtsi32_si128(*(const int*)(a + j));
__m128i t1 = _mm_cvtsi32_si128(*(const int*)(b + j));
d0 = _mm_add_epi32(d0, _mm_sad_epu8(t0, t1));
}
d = _mm_cvtsi128_si32(_mm_add_epi32(d0, _mm_unpackhi_epi64(d0, d0)));
}
else
#elif CV_NEON
uint32x4_t v_sum = vdupq_n_u32(0.0f);
for ( ; j <= n - 16; j += 16)
{
uint8x16_t v_dst = vabdq_u8(vld1q_u8(a + j), vld1q_u8(b + j));
uint16x8_t v_low = vmovl_u8(vget_low_u8(v_dst)), v_high = vmovl_u8(vget_high_u8(v_dst));
v_sum = vaddq_u32(v_sum, vaddl_u16(vget_low_u16(v_low), vget_low_u16(v_high)));
v_sum = vaddq_u32(v_sum, vaddl_u16(vget_high_u16(v_low), vget_high_u16(v_high)));
}
uint CV_DECL_ALIGNED(16) buf[4];
vst1q_u32(buf, v_sum);
d = buf[0] + buf[1] + buf[2] + buf[3];
#endif
{
for( ; j <= n - 4; j += 4 )
{
d += std::abs(a[j] - b[j]) + std::abs(a[j+1] - b[j+1]) +
std::abs(a[j+2] - b[j+2]) + std::abs(a[j+3] - b[j+3]);
}
}
for( ; j < n; j++ )
d += std::abs(a[j] - b[j]);
return d;
}
template<typename T, typename ST> int
normInf_(const T* src, const uchar* mask, ST* _result, int len, int cn)
{
@@ -2564,7 +2430,7 @@ normInf_(const T* src, const uchar* mask, ST* _result, int len, int cn)
if( mask[i] )
{
for( int k = 0; k < cn; k++ )
result = std::max(result, ST(std::abs(src[k])));
result = std::max(result, ST(cv_abs(src[k])));
}
}
*_result = result;
@@ -2585,7 +2451,7 @@ normL1_(const T* src, const uchar* mask, ST* _result, int len, int cn)
if( mask[i] )
{
for( int k = 0; k < cn; k++ )
result += std::abs(src[k]);
result += cv_abs(src[k]);
}
}
*_result = result;
@@ -2684,9 +2550,7 @@ normDiffL2_(const T* src1, const T* src2, const uchar* mask, ST* _result, int le
Hamming::ResultType Hamming::operator()( const unsigned char* a, const unsigned char* b, int size ) const
{
int result = 0;
cv::hal::normHamming(a, b, size, result);
return result;
return cv::hal::normHamming(a, b, size);
}
#define CV_DEF_NORM_FUNC(L, suffix, type, ntype) \
@@ -3037,16 +2901,12 @@ double cv::norm( InputArray _src, int normType, InputArray _mask )
if( normType == NORM_HAMMING )
{
int result = 0;
cv::hal::normHamming(data, (int)len, result);
return result;
return hal::normHamming(data, (int)len);
}
if( normType == NORM_HAMMING2 )
{
int result = 0;
hal::normHamming(data, (int)len, 2, result);
return result;
return hal::normHamming(data, (int)len, 2);
}
}
}
@@ -3072,9 +2932,7 @@ double cv::norm( InputArray _src, int normType, InputArray _mask )
for( size_t i = 0; i < it.nplanes; i++, ++it )
{
int one = 0;
cv::hal::normHamming(ptrs[0], total, cellSize, one);
result += one;
result += hal::normHamming(ptrs[0], total, cellSize);
}
return result;
@@ -3558,9 +3416,7 @@ double cv::norm( InputArray _src1, InputArray _src2, int normType, InputArray _m
for( size_t i = 0; i < it.nplanes; i++, ++it )
{
int one = 0;
hal::normHamming(ptrs[0], ptrs[1], total, cellSize, one);
result += one;
result += hal::normHamming(ptrs[0], ptrs[1], total, cellSize);
}
return result;
@@ -3698,7 +3554,7 @@ static void batchDistHamming(const uchar* src1, const uchar* src2, size_t step2,
if( !mask )
{
for( int i = 0; i < nvecs; i++ )
hal::normHamming(src1, src2 + step2*i, len, dist[i]);
dist[i] = hal::normHamming(src1, src2 + step2*i, len);
}
else
{
@@ -3706,7 +3562,7 @@ static void batchDistHamming(const uchar* src1, const uchar* src2, size_t step2,
for( int i = 0; i < nvecs; i++ )
{
if (mask[i])
hal::normHamming(src1, src2 + step2*i, len, dist[i]);
dist[i] = hal::normHamming(src1, src2 + step2*i, len);
else
dist[i] = val0;
}
@@ -3720,7 +3576,7 @@ static void batchDistHamming2(const uchar* src1, const uchar* src2, size_t step2
if( !mask )
{
for( int i = 0; i < nvecs; i++ )
hal::normHamming(src1, src2 + step2*i, len, 2, dist[i]);
dist[i] = hal::normHamming(src1, src2 + step2*i, len, 2);
}
else
{
@@ -3728,7 +3584,7 @@ static void batchDistHamming2(const uchar* src1, const uchar* src2, size_t step2
for( int i = 0; i < nvecs; i++ )
{
if (mask[i])
hal::normHamming(src1, src2 + step2*i, len, 2, dist[i]);
dist[i] = hal::normHamming(src1, src2 + step2*i, len, 2);
else
dist[i] = val0;
}