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

Merge pull request #5810 from mshabunin:hal_interface

Browse Source
This commit is contained in:
Vadim Pisarevsky
2015-12-17 16:48:01 +00:00
parent 5021350082 c04d62db8b
commit 9aeb8c8d5a
65 changed files with 3782 additions and 3437 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1106
modules/core/src/arithm.cpp
View File

File diff suppressed because it is too large Load Diff

607
modules/core/src/arithm_core.hpp Normal file
View File

@@ -0,0 +1,607 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Copyright (C) 2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_ARITHM_CORE_HPP__
#define __OPENCV_ARITHM_CORE_HPP__
#include "arithm_simd.hpp"
namespace cv {
template<typename T1, typename T2=T1, typename T3=T1> struct OpAdd
{
typedef T1 type1;
typedef T2 type2;
typedef T3 rtype;
T3 operator ()(const T1 a, const T2 b) const { return saturate_cast<T3>(a + b); }
};
template<typename T1, typename T2=T1, typename T3=T1> struct OpSub
{
typedef T1 type1;
typedef T2 type2;
typedef T3 rtype;
T3 operator ()(const T1 a, const T2 b) const { return saturate_cast<T3>(a - b); }
};
template<typename T1, typename T2=T1, typename T3=T1> struct OpRSub
{
typedef T1 type1;
typedef T2 type2;
typedef T3 rtype;
T3 operator ()(const T1 a, const T2 b) const { return saturate_cast<T3>(b - a); }
};
template<typename T> struct OpMin
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator ()(const T a, const T b) const { return std::min(a, b); }
};
template<typename T> struct OpMax
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator ()(const T a, const T b) const { return std::max(a, b); }
};
template<typename T> struct OpAbsDiff
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator()(T a, T b) const { return a > b ? a - b : b - a; }
};
template<typename T> struct OpAnd
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator()( T a, T b ) const { return a & b; }
};
template<typename T> struct OpOr
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator()( T a, T b ) const { return a | b; }
};
template<typename T> struct OpXor
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator()( T a, T b ) const { return a ^ b; }
};
template<typename T> struct OpNot
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator()( T a, T ) const { return ~a; }
};
//=============================================================================
template<typename T, class Op, class VOp>
void vBinOp(const T* src1, size_t step1, const T* src2, size_t step2, T* dst, size_t step, int width, int height)
{
#if CV_SSE2 || CV_NEON
VOp vop;
#endif
Op op;
for( ; height--; src1 = (const T *)((const uchar *)src1 + step1),
src2 = (const T *)((const uchar *)src2 + step2),
dst = (T *)((uchar *)dst + step) )
{
int x = 0;
#if CV_NEON || CV_SSE2
#if CV_AVX2
if( USE_AVX2 )
{
for( ; x <= width - 32/(int)sizeof(T); x += 32/sizeof(T) )
{
typename VLoadStore256<T>::reg_type r0 = VLoadStore256<T>::load(src1 + x);
r0 = vop(r0, VLoadStore256<T>::load(src2 + x));
VLoadStore256<T>::store(dst + x, r0);
}
}
#else
#if CV_SSE2
if( USE_SSE2 )
{
#endif // CV_SSE2
for( ; x <= width - 32/(int)sizeof(T); x += 32/sizeof(T) )
{
typename VLoadStore128<T>::reg_type r0 = VLoadStore128<T>::load(src1 + x );
typename VLoadStore128<T>::reg_type r1 = VLoadStore128<T>::load(src1 + x + 16/sizeof(T));
r0 = vop(r0, VLoadStore128<T>::load(src2 + x ));
r1 = vop(r1, VLoadStore128<T>::load(src2 + x + 16/sizeof(T)));
VLoadStore128<T>::store(dst + x , r0);
VLoadStore128<T>::store(dst + x + 16/sizeof(T), r1);
}
#if CV_SSE2
}
#endif // CV_SSE2
#endif // CV_AVX2
#endif // CV_NEON || CV_SSE2
#if CV_AVX2
// nothing
#elif CV_SSE2
if( USE_SSE2 )
{
for( ; x <= width - 8/(int)sizeof(T); x += 8/sizeof(T) )
{
typename VLoadStore64<T>::reg_type r = VLoadStore64<T>::load(src1 + x);
r = vop(r, VLoadStore64<T>::load(src2 + x));
VLoadStore64<T>::store(dst + x, r);
}
}
#endif
#if CV_ENABLE_UNROLLED
for( ; x <= width - 4; x += 4 )
{
T v0 = op(src1[x], src2[x]);
T v1 = op(src1[x+1], src2[x+1]);
dst[x] = v0; dst[x+1] = v1;
v0 = op(src1[x+2], src2[x+2]);
v1 = op(src1[x+3], src2[x+3]);
dst[x+2] = v0; dst[x+3] = v1;
}
#endif
for( ; x < width; x++ )
dst[x] = op(src1[x], src2[x]);
}
}
template<typename T, class Op, class Op32>
void vBinOp32(const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, int width, int height)
{
#if CV_SSE2 || CV_NEON
Op32 op32;
#endif
Op op;
for( ; height--; src1 = (const T *)((const uchar *)src1 + step1),
src2 = (const T *)((const uchar *)src2 + step2),
dst = (T *)((uchar *)dst + step) )
{
int x = 0;
#if CV_AVX2
if( USE_AVX2 )
{
if( (((size_t)src1|(size_t)src2|(size_t)dst)&31) == 0 )
{
for( ; x <= width - 8; x += 8 )
{
typename VLoadStore256Aligned<T>::reg_type r0 = VLoadStore256Aligned<T>::load(src1 + x);
r0 = op32(r0, VLoadStore256Aligned<T>::load(src2 + x));
VLoadStore256Aligned<T>::store(dst + x, r0);
}
}
}
#elif CV_SSE2
if( USE_SSE2 )
{
if( (((size_t)src1|(size_t)src2|(size_t)dst)&15) == 0 )
{
for( ; x <= width - 8; x += 8 )
{
typename VLoadStore128Aligned<T>::reg_type r0 = VLoadStore128Aligned<T>::load(src1 + x );
typename VLoadStore128Aligned<T>::reg_type r1 = VLoadStore128Aligned<T>::load(src1 + x + 4);
r0 = op32(r0, VLoadStore128Aligned<T>::load(src2 + x ));
r1 = op32(r1, VLoadStore128Aligned<T>::load(src2 + x + 4));
VLoadStore128Aligned<T>::store(dst + x , r0);
VLoadStore128Aligned<T>::store(dst + x + 4, r1);
}
}
}
#endif // CV_AVX2
#if CV_NEON || CV_SSE2
#if CV_AVX2
if( USE_AVX2 )
{
for( ; x <= width - 8; x += 8 )
{
typename VLoadStore256<T>::reg_type r0 = VLoadStore256<T>::load(src1 + x);
r0 = op32(r0, VLoadStore256<T>::load(src2 + x));
VLoadStore256<T>::store(dst + x, r0);
}
}
#else
#if CV_SSE2
if( USE_SSE2 )
{
#endif // CV_SSE2
for( ; x <= width - 8; x += 8 )
{
typename VLoadStore128<T>::reg_type r0 = VLoadStore128<T>::load(src1 + x );
typename VLoadStore128<T>::reg_type r1 = VLoadStore128<T>::load(src1 + x + 4);
r0 = op32(r0, VLoadStore128<T>::load(src2 + x ));
r1 = op32(r1, VLoadStore128<T>::load(src2 + x + 4));
VLoadStore128<T>::store(dst + x , r0);
VLoadStore128<T>::store(dst + x + 4, r1);
}
#if CV_SSE2
}
#endif // CV_SSE2
#endif // CV_AVX2
#endif // CV_NEON || CV_SSE2
#if CV_ENABLE_UNROLLED
for( ; x <= width - 4; x += 4 )
{
T v0 = op(src1[x], src2[x]);
T v1 = op(src1[x+1], src2[x+1]);
dst[x] = v0; dst[x+1] = v1;
v0 = op(src1[x+2], src2[x+2]);
v1 = op(src1[x+3], src2[x+3]);
dst[x+2] = v0; dst[x+3] = v1;
}
#endif
for( ; x < width; x++ )
dst[x] = op(src1[x], src2[x]);
}
}
template<typename T, class Op, class Op64>
void vBinOp64(const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, int width, int height)
{
#if CV_SSE2
Op64 op64;
#endif
Op op;
for( ; height--; src1 = (const T *)((const uchar *)src1 + step1),
src2 = (const T *)((const uchar *)src2 + step2),
dst = (T *)((uchar *)dst + step) )
{
int x = 0;
#if CV_AVX2
if( USE_AVX2 )
{
if( (((size_t)src1|(size_t)src2|(size_t)dst)&31) == 0 )
{
for( ; x <= width - 4; x += 4 )
{
typename VLoadStore256Aligned<T>::reg_type r0 = VLoadStore256Aligned<T>::load(src1 + x);
r0 = op64(r0, VLoadStore256Aligned<T>::load(src2 + x));
VLoadStore256Aligned<T>::store(dst + x, r0);
}
}
}
#elif CV_SSE2
if( USE_SSE2 )
{
if( (((size_t)src1|(size_t)src2|(size_t)dst)&15) == 0 )
{
for( ; x <= width - 4; x += 4 )
{
typename VLoadStore128Aligned<T>::reg_type r0 = VLoadStore128Aligned<T>::load(src1 + x );
typename VLoadStore128Aligned<T>::reg_type r1 = VLoadStore128Aligned<T>::load(src1 + x + 2);
r0 = op64(r0, VLoadStore128Aligned<T>::load(src2 + x ));
r1 = op64(r1, VLoadStore128Aligned<T>::load(src2 + x + 2));
VLoadStore128Aligned<T>::store(dst + x , r0);
VLoadStore128Aligned<T>::store(dst + x + 2, r1);
}
}
}
#endif
for( ; x <= width - 4; x += 4 )
{
T v0 = op(src1[x], src2[x]);
T v1 = op(src1[x+1], src2[x+1]);
dst[x] = v0; dst[x+1] = v1;
v0 = op(src1[x+2], src2[x+2]);
v1 = op(src1[x+3], src2[x+3]);
dst[x+2] = v0; dst[x+3] = v1;
}
for( ; x < width; x++ )
dst[x] = op(src1[x], src2[x]);
}
}
template<typename T> static void
cmp_(const T* src1, size_t step1, const T* src2, size_t step2,
uchar* dst, size_t step, int width, int height, int code)
{
step1 /= sizeof(src1[0]);
step2 /= sizeof(src2[0]);
if( code == CMP_GE || code == CMP_LT )
{
std::swap(src1, src2);
std::swap(step1, step2);
code = code == CMP_GE ? CMP_LE : CMP_GT;
}
Cmp_SIMD<T> vop(code);
if( code == CMP_GT || code == CMP_LE )
{
int m = code == CMP_GT ? 0 : 255;
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int x = vop(src1, src2, dst, width);
#if CV_ENABLE_UNROLLED
for( ; x <= width - 4; x += 4 )
{
int t0, t1;
t0 = -(src1[x] > src2[x]) ^ m;
t1 = -(src1[x+1] > src2[x+1]) ^ m;
dst[x] = (uchar)t0; dst[x+1] = (uchar)t1;
t0 = -(src1[x+2] > src2[x+2]) ^ m;
t1 = -(src1[x+3] > src2[x+3]) ^ m;
dst[x+2] = (uchar)t0; dst[x+3] = (uchar)t1;
}
#endif
for( ; x < width; x++ )
dst[x] = (uchar)(-(src1[x] > src2[x]) ^ m);
}
}
else if( code == CMP_EQ || code == CMP_NE )
{
int m = code == CMP_EQ ? 0 : 255;
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int x = 0;
#if CV_ENABLE_UNROLLED
for( ; x <= width - 4; x += 4 )
{
int t0, t1;
t0 = -(src1[x] == src2[x]) ^ m;
t1 = -(src1[x+1] == src2[x+1]) ^ m;
dst[x] = (uchar)t0; dst[x+1] = (uchar)t1;
t0 = -(src1[x+2] == src2[x+2]) ^ m;
t1 = -(src1[x+3] == src2[x+3]) ^ m;
dst[x+2] = (uchar)t0; dst[x+3] = (uchar)t1;
}
#endif
for( ; x < width; x++ )
dst[x] = (uchar)(-(src1[x] == src2[x]) ^ m);
}
}
}
template<typename T, typename WT> static void
mul_( const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, int width, int height, WT scale )
{
step1 /= sizeof(src1[0]);
step2 /= sizeof(src2[0]);
step /= sizeof(dst[0]);
Mul_SIMD<T, WT> vop;
if( scale == (WT)1. )
{
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int i = vop(src1, src2, dst, width, scale);
#if CV_ENABLE_UNROLLED
for(; i <= width - 4; i += 4 )
{
T t0;
T t1;
t0 = saturate_cast<T>(src1[i ] * src2[i ]);
t1 = saturate_cast<T>(src1[i+1] * src2[i+1]);
dst[i ] = t0;
dst[i+1] = t1;
t0 = saturate_cast<T>(src1[i+2] * src2[i+2]);
t1 = saturate_cast<T>(src1[i+3] * src2[i+3]);
dst[i+2] = t0;
dst[i+3] = t1;
}
#endif
for( ; i < width; i++ )
dst[i] = saturate_cast<T>(src1[i] * src2[i]);
}
}
else
{
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int i = vop(src1, src2, dst, width, scale);
#if CV_ENABLE_UNROLLED
for(; i <= width - 4; i += 4 )
{
T t0 = saturate_cast<T>(scale*(WT)src1[i]*src2[i]);
T t1 = saturate_cast<T>(scale*(WT)src1[i+1]*src2[i+1]);
dst[i] = t0; dst[i+1] = t1;
t0 = saturate_cast<T>(scale*(WT)src1[i+2]*src2[i+2]);
t1 = saturate_cast<T>(scale*(WT)src1[i+3]*src2[i+3]);
dst[i+2] = t0; dst[i+3] = t1;
}
#endif
for( ; i < width; i++ )
dst[i] = saturate_cast<T>(scale*(WT)src1[i]*src2[i]);
}
}
}
template<typename T> static void
div_i( const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, int width, int height, double scale )
{
step1 /= sizeof(src1[0]);
step2 /= sizeof(src2[0]);
step /= sizeof(dst[0]);
Div_SIMD<T> vop;
float scale_f = (float)scale;
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int i = vop(src1, src2, dst, width, scale);
for( ; i < width; i++ )
{
T num = src1[i], denom = src2[i];
dst[i] = denom != 0 ? saturate_cast<T>(num*scale_f/denom) : (T)0;
}
}
}
template<typename T> static void
div_f( const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, int width, int height, double scale )
{
T scale_f = (T)scale;
step1 /= sizeof(src1[0]);
step2 /= sizeof(src2[0]);
step /= sizeof(dst[0]);
Div_SIMD<T> vop;
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int i = vop(src1, src2, dst, width, scale);
for( ; i < width; i++ )
{
T num = src1[i], denom = src2[i];
dst[i] = denom != 0 ? saturate_cast<T>(num*scale_f/denom) : (T)0;
}
}
}
template<typename T> static void
recip_i( const T*, size_t, const T* src2, size_t step2,
T* dst, size_t step, int width, int height, double scale )
{
step2 /= sizeof(src2[0]);
step /= sizeof(dst[0]);
Recip_SIMD<T> vop;
float scale_f = (float)scale;
for( ; height--; src2 += step2, dst += step )
{
int i = vop(src2, dst, width, scale);
for( ; i < width; i++ )
{
T denom = src2[i];
dst[i] = denom != 0 ? saturate_cast<T>(scale_f/denom) : (T)0;
}
}
}
template<typename T> static void
recip_f( const T*, size_t, const T* src2, size_t step2,
T* dst, size_t step, int width, int height, double scale )
{
T scale_f = (T)scale;
step2 /= sizeof(src2[0]);
step /= sizeof(dst[0]);
Recip_SIMD<T> vop;
for( ; height--; src2 += step2, dst += step )
{
int i = vop(src2, dst, width, scale);
for( ; i < width; i++ )
{
T denom = src2[i];
dst[i] = denom != 0 ? saturate_cast<T>(scale_f/denom) : (T)0;
}
}
}
template<typename T, typename WT> static void
addWeighted_( const T* src1, size_t step1, const T* src2, size_t step2,
T* dst, size_t step, int width, int height, void* _scalars )
{
const double* scalars = (const double*)_scalars;
WT alpha = (WT)scalars[0], beta = (WT)scalars[1], gamma = (WT)scalars[2];
step1 /= sizeof(src1[0]);
step2 /= sizeof(src2[0]);
step /= sizeof(dst[0]);
AddWeighted_SIMD<T, WT> vop;
for( ; height--; src1 += step1, src2 += step2, dst += step )
{
int x = vop(src1, src2, dst, width, alpha, beta, gamma);
#if CV_ENABLE_UNROLLED
for( ; x <= width - 4; x += 4 )
{
T t0 = saturate_cast<T>(src1[x]*alpha + src2[x]*beta + gamma);
T t1 = saturate_cast<T>(src1[x+1]*alpha + src2[x+1]*beta + gamma);
dst[x] = t0; dst[x+1] = t1;
t0 = saturate_cast<T>(src1[x+2]*alpha + src2[x+2]*beta + gamma);
t1 = saturate_cast<T>(src1[x+3]*alpha + src2[x+3]*beta + gamma);
dst[x+2] = t0; dst[x+3] = t1;
}
#endif
for( ; x < width; x++ )
dst[x] = saturate_cast<T>(src1[x]*alpha + src2[x]*beta + gamma);
}
}
} // cv::
#endif // __OPENCV_ARITHM_CORE_HPP__

2025
modules/core/src/arithm_simd.hpp Normal file
View File

File diff suppressed because it is too large Load Diff

228
modules/core/src/hal_replacement.hpp Normal file
View File

@@ -0,0 +1,228 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
// Copyright (C) 2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_CORE_HAL_REPLACEMENT_HPP__
#define __OPENCV_CORE_HAL_REPLACEMENT_HPP__
#include "opencv2/core/hal/interface.h"
inline int hal_ni_add8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_add8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_add16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_add16s(const short*, size_t, const short*, size_t, short*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_add32s(const int*, size_t, const int*, size_t, int*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_add32f(const float*, size_t, const float*, size_t, float*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_add64f(const double*, size_t, const double*, size_t, double*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub16s(const short*, size_t, const short*, size_t, short*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub32s(const int*, size_t, const int*, size_t, int*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub32f(const float*, size_t, const float*, size_t, float*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_sub64f(const double*, size_t, const double*, size_t, double*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max16s(const short*, size_t, const short*, size_t, short*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max32s(const int*, size_t, const int*, size_t, int*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max32f(const float*, size_t, const float*, size_t, float*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_max64f(const double*, size_t, const double*, size_t, double*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min16s(const short*, size_t, const short*, size_t, short*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min32s(const int*, size_t, const int*, size_t, int*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min32f(const float*, size_t, const float*, size_t, float*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_min64f(const double*, size_t, const double*, size_t, double*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff16s(const short*, size_t, const short*, size_t, short*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff32s(const int*, size_t, const int*, size_t, int*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff32f(const float*, size_t, const float*, size_t, float*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_absdiff64f(const double*, size_t, const double*, size_t, double*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_and8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_or8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_xor8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_not8u(const uchar*, size_t, uchar*, size_t, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
#define cv_hal_add8u hal_ni_add8u
#define cv_hal_add8s hal_ni_add8s
#define cv_hal_add16u hal_ni_add16u
#define cv_hal_add16s hal_ni_add16s
#define cv_hal_add32s hal_ni_add32s
#define cv_hal_add32f hal_ni_add32f
#define cv_hal_add64f hal_ni_add64f
#define cv_hal_sub8u hal_ni_sub8u
#define cv_hal_sub8s hal_ni_sub8s
#define cv_hal_sub16u hal_ni_sub16u
#define cv_hal_sub16s hal_ni_sub16s
#define cv_hal_sub32s hal_ni_sub32s
#define cv_hal_sub32f hal_ni_sub32f
#define cv_hal_sub64f hal_ni_sub64f
#define cv_hal_max8u hal_ni_max8u
#define cv_hal_max8s hal_ni_max8s
#define cv_hal_max16u hal_ni_max16u
#define cv_hal_max16s hal_ni_max16s
#define cv_hal_max32s hal_ni_max32s
#define cv_hal_max32f hal_ni_max32f
#define cv_hal_max64f hal_ni_max64f
#define cv_hal_min8u hal_ni_min8u
#define cv_hal_min8s hal_ni_min8s
#define cv_hal_min16u hal_ni_min16u
#define cv_hal_min16s hal_ni_min16s
#define cv_hal_min32s hal_ni_min32s
#define cv_hal_min32f hal_ni_min32f
#define cv_hal_min64f hal_ni_min64f
#define cv_hal_absdiff8u hal_ni_absdiff8u
#define cv_hal_absdiff8s hal_ni_absdiff8s
#define cv_hal_absdiff16u hal_ni_absdiff16u
#define cv_hal_absdiff16s hal_ni_absdiff16s
#define cv_hal_absdiff32s hal_ni_absdiff32s
#define cv_hal_absdiff32f hal_ni_absdiff32f
#define cv_hal_absdiff64f hal_ni_absdiff64f
#define cv_hal_and8u hal_ni_and8u
#define cv_hal_or8u hal_ni_or8u
#define cv_hal_xor8u hal_ni_xor8u
#define cv_hal_not8u hal_ni_not8u
inline int hal_ni_cmp8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_cmp8s(const schar*, size_t, const schar*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_cmp16u(const ushort*, size_t, const ushort*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_cmp16s(const short*, size_t, const short*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_cmp32s(const int*, size_t, const int*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_cmp32f(const float*, size_t, const float*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_cmp64f(const double*, size_t, const double*, size_t, uchar*, size_t, int, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
#define cv_hal_cmp8u hal_ni_cmp8u
#define cv_hal_cmp8s hal_ni_cmp8s
#define cv_hal_cmp16u hal_ni_cmp16u
#define cv_hal_cmp16s hal_ni_cmp16s
#define cv_hal_cmp32s hal_ni_cmp32s
#define cv_hal_cmp32f hal_ni_cmp32f
#define cv_hal_cmp64f hal_ni_cmp64f
inline int hal_ni_mul8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_mul8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_mul16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_mul16s(const short*, size_t, const short*, size_t, short*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_mul32s(const int*, size_t, const int*, size_t, int*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_mul32f(const float*, size_t, const float*, size_t, float*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_mul64f(const double*, size_t, const double*, size_t, double*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div16s(const short*, size_t, const short*, size_t, short*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div32s(const int*, size_t, const int*, size_t, int*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div32f(const float*, size_t, const float*, size_t, float*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_div64f(const double*, size_t, const double*, size_t, double*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip16s(const short*, size_t, const short*, size_t, short*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip32s(const int*, size_t, const int*, size_t, int*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip32f(const float*, size_t, const float*, size_t, float*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_recip64f(const double*, size_t, const double*, size_t, double*, size_t, int, int, double) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
#define cv_hal_mul8u hal_ni_mul8u
#define cv_hal_mul8s hal_ni_mul8s
#define cv_hal_mul16u hal_ni_mul16u
#define cv_hal_mul16s hal_ni_mul16s
#define cv_hal_mul32s hal_ni_mul32s
#define cv_hal_mul32f hal_ni_mul32f
#define cv_hal_mul64f hal_ni_mul64f
#define cv_hal_div8u hal_ni_div8u
#define cv_hal_div8s hal_ni_div8s
#define cv_hal_div16u hal_ni_div16u
#define cv_hal_div16s hal_ni_div16s
#define cv_hal_div32s hal_ni_div32s
#define cv_hal_div32f hal_ni_div32f
#define cv_hal_div64f hal_ni_div64f
#define cv_hal_recip8u hal_ni_recip8u
#define cv_hal_recip8s hal_ni_recip8s
#define cv_hal_recip16u hal_ni_recip16u
#define cv_hal_recip16s hal_ni_recip16s
#define cv_hal_recip32s hal_ni_recip32s
#define cv_hal_recip32f hal_ni_recip32f
#define cv_hal_recip64f hal_ni_recip64f
inline int hal_ni_addWeighted8u(const uchar*, size_t, const uchar*, size_t, uchar*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_addWeighted8s(const schar*, size_t, const schar*, size_t, schar*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_addWeighted16u(const ushort*, size_t, const ushort*, size_t, ushort*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_addWeighted16s(const short*, size_t, const short*, size_t, short*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_addWeighted32s(const int*, size_t, const int*, size_t, int*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_addWeighted32f(const float*, size_t, const float*, size_t, float*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_addWeighted64f(const double*, size_t, const double*, size_t, double*, size_t, int, int, const double*) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
#define cv_hal_addWeighted8u hal_ni_addWeighted8u
#define cv_hal_addWeighted8s hal_ni_addWeighted8s
#define cv_hal_addWeighted16u hal_ni_addWeighted16u
#define cv_hal_addWeighted16s hal_ni_addWeighted16s
#define cv_hal_addWeighted32s hal_ni_addWeighted32s
#define cv_hal_addWeighted32f hal_ni_addWeighted32f
#define cv_hal_addWeighted64f hal_ni_addWeighted64f
inline int hal_ni_split8u(const uchar*, uchar**, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_split16u(const ushort*, ushort**, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_split32s(const int*, int**, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_split64s(const int64*, int64**, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
#define cv_hal_split8u hal_ni_split8u
#define cv_hal_split16u hal_ni_split16u
#define cv_hal_split32s hal_ni_split32s
#define cv_hal_split64s hal_ni_split64s
inline int hal_ni_merge8u(const uchar**, uchar*, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_merge16u(const ushort**, ushort*, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_merge32s(const int**, int*, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
inline int hal_ni_merge64s(const int64**, int64*, int, int) { return CV_HAL_ERROR_NOT_IMPLEMENTED; }
#define cv_hal_merge8u hal_ni_merge8u
#define cv_hal_merge16u hal_ni_merge16u
#define cv_hal_merge32s hal_ni_merge32s
#define cv_hal_merge64s hal_ni_merge64s
#include "custom_hal.hpp"
#endif

28
modules/core/src/lapack.cpp
View File

@@ -52,22 +52,22 @@ namespace cv
int LU(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return hal::LU(A, astep, m, b, bstep, n);
return hal::LU32f(A, astep, m, b, bstep, n);
}
int LU(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return hal::LU(A, astep, m, b, bstep, n);
return hal::LU64f(A, astep, m, b, bstep, n);
}
bool Cholesky(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return hal::Cholesky(A, astep, m, b, bstep, n);
return hal::Cholesky32f(A, astep, m, b, bstep, n);
}
bool Cholesky(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return hal::Cholesky(A, astep, m, b, bstep, n);
return hal::Cholesky64f(A, astep, m, b, bstep, n);
}
template<typename _Tp> static inline _Tp hypot(_Tp a, _Tp b)
@@ -740,7 +740,7 @@ double cv::determinant( InputArray _mat )
Mat a(rows, rows, CV_32F, (uchar*)buffer);
mat.copyTo(a);
result = hal::LU(a.ptr<float>(), a.step, rows, 0, 0, 0);
result = hal::LU32f(a.ptr<float>(), a.step, rows, 0, 0, 0);
if( result )
{
for( int i = 0; i < rows; i++ )
@@ -764,7 +764,7 @@ double cv::determinant( InputArray _mat )
Mat a(rows, rows, CV_64F, (uchar*)buffer);
mat.copyTo(a);
result = hal::LU(a.ptr<double>(), a.step, rows, 0, 0, 0);
result = hal::LU64f(a.ptr<double>(), a.step, rows, 0, 0, 0);
if( result )
{
for( int i = 0; i < rows; i++ )
@@ -1027,13 +1027,13 @@ double cv::invert( InputArray _src, OutputArray _dst, int method )
setIdentity(dst);
if( method == DECOMP_LU && type == CV_32F )
result = hal::LU(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n) != 0;
result = hal::LU32f(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n) != 0;
else if( method == DECOMP_LU && type == CV_64F )
result = hal::LU(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n) != 0;
result = hal::LU64f(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n) != 0;
else if( method == DECOMP_CHOLESKY && type == CV_32F )
result = hal::Cholesky(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n);
result = hal::Cholesky32f(src1.ptr<float>(), src1.step, n, dst.ptr<float>(), dst.step, n);
else
result = hal::Cholesky(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n);
result = hal::Cholesky64f(src1.ptr<double>(), src1.step, n, dst.ptr<double>(), dst.step, n);
if( !result )
dst = Scalar(0);
@@ -1265,16 +1265,16 @@ bool cv::solve( InputArray _src, InputArray _src2arg, OutputArray _dst, int meth
if( method == DECOMP_LU )
{
if( type == CV_32F )
result = hal::LU(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb) != 0;
result = hal::LU32f(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb) != 0;
else
result = hal::LU(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb) != 0;
result = hal::LU64f(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb) != 0;
}
else if( method == DECOMP_CHOLESKY )
{
if( type == CV_32F )
result = hal::Cholesky(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb);
result = hal::Cholesky32f(a.ptr<float>(), a.step, n, dst.ptr<float>(), dst.step, nb);
else
result = hal::Cholesky(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb);
result = hal::Cholesky64f(a.ptr<double>(), a.step, n, dst.ptr<double>(), dst.step, nb);
}
else
{

32
modules/core/src/mathfuncs.cpp
View File

@@ -191,13 +191,13 @@ void magnitude( InputArray src1, InputArray src2, OutputArray dst )
{
const float *x = (const float*)ptrs[0], *y = (const float*)ptrs[1];
float *mag = (float*)ptrs[2];
hal::magnitude( x, y, mag, len );
hal::magnitude32f( x, y, mag, len );
}
else
{
const double *x = (const double*)ptrs[0], *y = (const double*)ptrs[1];
double *mag = (double*)ptrs[2];
hal::magnitude( x, y, mag, len );
hal::magnitude64f( x, y, mag, len );
}
}
}
@@ -374,7 +374,7 @@ void cartToPolar( InputArray src1, InputArray src2,
{
const float *x = (const float*)ptrs[0], *y = (const float*)ptrs[1];
float *mag = (float*)ptrs[2], *angle = (float*)ptrs[3];
hal::magnitude( x, y, mag, len );
hal::magnitude32f( x, y, mag, len );
hal::fastAtan2( y, x, angle, len, angleInDegrees );
}
else
@@ -382,7 +382,7 @@ void cartToPolar( InputArray src1, InputArray src2,
const double *x = (const double*)ptrs[0], *y = (const double*)ptrs[1];
double *angle = (double*)ptrs[3];
hal::magnitude(x, y, (double*)ptrs[2], len);
hal::magnitude64f(x, y, (double*)ptrs[2], len);
k = 0;
#if CV_SSE2
@@ -760,7 +760,7 @@ static void Exp_32f_ipp(const float *x, float *y, int n)
}
setIppErrorStatus();
}
hal::exp(x, y, n);
hal::exp32f(x, y, n);
}
static void Exp_64f_ipp(const double *x, double *y, int n)
@@ -774,14 +774,14 @@ static void Exp_64f_ipp(const double *x, double *y, int n)
}
setIppErrorStatus();
}
hal::exp(x, y, n);
hal::exp64f(x, y, n);
}
#define Exp_32f Exp_32f_ipp
#define Exp_64f Exp_64f_ipp
#else
#define Exp_32f hal::exp
#define Exp_64f hal::exp
#define Exp_32f hal::exp32f
#define Exp_64f hal::exp64f
#endif
@@ -828,7 +828,7 @@ static void Log_32f_ipp(const float *x, float *y, int n)
}
setIppErrorStatus();
}
hal::log(x, y, n);
hal::log32f(x, y, n);
}
static void Log_64f_ipp(const double *x, double *y, int n)
@@ -842,14 +842,14 @@ static void Log_64f_ipp(const double *x, double *y, int n)
}
setIppErrorStatus();
}
hal::log(x, y, n);
hal::log64f(x, y, n);
}
#define Log_32f Log_32f_ipp
#define Log_64f Log_64f_ipp
#else
#define Log_32f hal::log
#define Log_64f hal::log
#define Log_32f hal::log32f
#define Log_64f hal::log64f
#endif
void log( InputArray _src, OutputArray _dst )
@@ -1356,10 +1356,10 @@ static bool ocl_pow(InputArray _src, double power, OutputArray _dst,
#endif
static void InvSqrt_32f(const float* src, float* dst, int n) { hal::invSqrt(src, dst, n); }
static void InvSqrt_64f(const double* src, double* dst, int n) { hal::invSqrt(src, dst, n); }
static void Sqrt_32f(const float* src, float* dst, int n) { hal::sqrt(src, dst, n); }
static void Sqrt_64f(const double* src, double* dst, int n) { hal::sqrt(src, dst, n); }
static void InvSqrt_32f(const float* src, float* dst, int n) { hal::invSqrt32f(src, dst, n); }
static void InvSqrt_64f(const double* src, double* dst, int n) { hal::invSqrt64f(src, dst, n); }
static void Sqrt_32f(const float* src, float* dst, int n) { hal::sqrt32f(src, dst, n); }
static void Sqrt_64f(const double* src, double* dst, int n) { hal::sqrt64f(src, dst, n); }
void pow( InputArray _src, double power, OutputArray _dst )
{

1460
modules/core/src/mathfuncs_core.cpp Normal file
View File

File diff suppressed because it is too large Load Diff

231
modules/core/src/matrix_decomp.cpp Normal file
View File

@@ -0,0 +1,231 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009-2011, Willow Garage Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
namespace cv { namespace hal {
/****************************************************************************************\
* 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, _Tp eps)
{
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]) < eps )
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 LU32f(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n, FLT_EPSILON*10);
}
int LU64f(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n, DBL_EPSILON*100);
}
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 Cholesky32f(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return CholImpl(A, astep, m, b, bstep, n);
}
bool Cholesky64f(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return CholImpl(A, astep, m, b, bstep, n);
}
//=============================================================================
// for compatibility with 3.0
int LU(float* A, size_t astep, int m, float* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n, FLT_EPSILON*10);
}
int LU(double* A, size_t astep, int m, double* b, size_t bstep, int n)
{
return LUImpl(A, astep, m, b, bstep, n, DBL_EPSILON*100);
}
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);
}
}}

412
modules/core/src/merge.cpp Normal file
View File

@@ -0,0 +1,412 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009-2011, Willow Garage Inc., all rights reserved.
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
namespace cv { namespace hal {
#if CV_NEON
template<typename T> struct VMerge2;
template<typename T> struct VMerge3;
template<typename T> struct VMerge4;
#define MERGE2_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template<> \
struct name<data_type>{ \
void operator()(const data_type* src0, const data_type* src1, \
data_type* dst){ \
reg_type r; \
r.val[0] = load_func(src0); \
r.val[1] = load_func(src1); \
store_func(dst, r); \
} \
}
#define MERGE3_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template<> \
struct name<data_type>{ \
void operator()(const data_type* src0, const data_type* src1, \
const data_type* src2, data_type* dst){ \
reg_type r; \
r.val[0] = load_func(src0); \
r.val[1] = load_func(src1); \
r.val[2] = load_func(src2); \
store_func(dst, r); \
} \
}
#define MERGE4_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template<> \
struct name<data_type>{ \
void operator()(const data_type* src0, const data_type* src1, \
const data_type* src2, const data_type* src3, \
data_type* dst){ \
reg_type r; \
r.val[0] = load_func(src0); \
r.val[1] = load_func(src1); \
r.val[2] = load_func(src2); \
r.val[3] = load_func(src3); \
store_func(dst, r); \
} \
}
MERGE2_KERNEL_TEMPLATE(VMerge2, uchar , uint8x16x2_t, vld1q_u8 , vst2q_u8 );
MERGE2_KERNEL_TEMPLATE(VMerge2, ushort, uint16x8x2_t, vld1q_u16, vst2q_u16);
MERGE2_KERNEL_TEMPLATE(VMerge2, int , int32x4x2_t, vld1q_s32, vst2q_s32);
MERGE2_KERNEL_TEMPLATE(VMerge2, int64 , int64x1x2_t, vld1_s64 , vst2_s64 );
MERGE3_KERNEL_TEMPLATE(VMerge3, uchar , uint8x16x3_t, vld1q_u8 , vst3q_u8 );
MERGE3_KERNEL_TEMPLATE(VMerge3, ushort, uint16x8x3_t, vld1q_u16, vst3q_u16);
MERGE3_KERNEL_TEMPLATE(VMerge3, int , int32x4x3_t, vld1q_s32, vst3q_s32);
MERGE3_KERNEL_TEMPLATE(VMerge3, int64 , int64x1x3_t, vld1_s64 , vst3_s64 );
MERGE4_KERNEL_TEMPLATE(VMerge4, uchar , uint8x16x4_t, vld1q_u8 , vst4q_u8 );
MERGE4_KERNEL_TEMPLATE(VMerge4, ushort, uint16x8x4_t, vld1q_u16, vst4q_u16);
MERGE4_KERNEL_TEMPLATE(VMerge4, int , int32x4x4_t, vld1q_s32, vst4q_s32);
MERGE4_KERNEL_TEMPLATE(VMerge4, int64 , int64x1x4_t, vld1_s64 , vst4_s64 );
#elif CV_SSE2
template <typename T>
struct VMerge2
{
VMerge2() : support(false) { }
void operator()(const T *, const T *, T *) const { }
bool support;
};
template <typename T>
struct VMerge3
{
VMerge3() : support(false) { }
void operator()(const T *, const T *, const T *, T *) const { }
bool support;
};
template <typename T>
struct VMerge4
{
VMerge4() : support(false) { }
void operator()(const T *, const T *, const T *, const T *, T *) const { }
bool support;
};
#define MERGE2_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor, se) \
template <> \
struct VMerge2<data_type> \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VMerge2() \
{ \
support = checkHardwareSupport(se); \
} \
\
void operator()(const data_type * src0, const data_type * src1, \
data_type * dst) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((const cast_type *)(src0)); \
reg_type v_src1 = _mm_loadu_##flavor((const cast_type *)(src0 + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((const cast_type *)(src1)); \
reg_type v_src3 = _mm_loadu_##flavor((const cast_type *)(src1 + ELEMS_IN_VEC)); \
\
_mm_interleave(v_src0, v_src1, v_src2, v_src3); \
\
_mm_storeu_##flavor((cast_type *)(dst), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 2), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 3), v_src3); \
} \
\
bool support; \
}
#define MERGE3_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor, se) \
template <> \
struct VMerge3<data_type> \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VMerge3() \
{ \
support = checkHardwareSupport(se); \
} \
\
void operator()(const data_type * src0, const data_type * src1, const data_type * src2,\
data_type * dst) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((const cast_type *)(src0)); \
reg_type v_src1 = _mm_loadu_##flavor((const cast_type *)(src0 + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((const cast_type *)(src1)); \
reg_type v_src3 = _mm_loadu_##flavor((const cast_type *)(src1 + ELEMS_IN_VEC)); \
reg_type v_src4 = _mm_loadu_##flavor((const cast_type *)(src2)); \
reg_type v_src5 = _mm_loadu_##flavor((const cast_type *)(src2 + ELEMS_IN_VEC)); \
\
_mm_interleave(v_src0, v_src1, v_src2, \
v_src3, v_src4, v_src5); \
\
_mm_storeu_##flavor((cast_type *)(dst), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 2), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 3), v_src3); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 4), v_src4); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 5), v_src5); \
} \
\
bool support; \
}
#define MERGE4_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor, se) \
template <> \
struct VMerge4<data_type> \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VMerge4() \
{ \
support = checkHardwareSupport(se); \
} \
\
void operator()(const data_type * src0, const data_type * src1, \
const data_type * src2, const data_type * src3, \
data_type * dst) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((const cast_type *)(src0)); \
reg_type v_src1 = _mm_loadu_##flavor((const cast_type *)(src0 + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((const cast_type *)(src1)); \
reg_type v_src3 = _mm_loadu_##flavor((const cast_type *)(src1 + ELEMS_IN_VEC)); \
reg_type v_src4 = _mm_loadu_##flavor((const cast_type *)(src2)); \
reg_type v_src5 = _mm_loadu_##flavor((const cast_type *)(src2 + ELEMS_IN_VEC)); \
reg_type v_src6 = _mm_loadu_##flavor((const cast_type *)(src3)); \
reg_type v_src7 = _mm_loadu_##flavor((const cast_type *)(src3 + ELEMS_IN_VEC)); \
\
_mm_interleave(v_src0, v_src1, v_src2, v_src3, \
v_src4, v_src5, v_src6, v_src7); \
\
_mm_storeu_##flavor((cast_type *)(dst), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 2), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 3), v_src3); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 4), v_src4); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 5), v_src5); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 6), v_src6); \
_mm_storeu_##flavor((cast_type *)(dst + ELEMS_IN_VEC * 7), v_src7); \
} \
\
bool support; \
}
MERGE2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128, CV_CPU_SSE2);
MERGE3_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128, CV_CPU_SSE2);
MERGE4_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128, CV_CPU_SSE2);
#if CV_SSE4_1
MERGE2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128, CV_CPU_SSE4_1);
MERGE3_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128, CV_CPU_SSE4_1);
MERGE4_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128, CV_CPU_SSE4_1);
#endif
MERGE2_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps, CV_CPU_SSE2);
MERGE3_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps, CV_CPU_SSE2);
MERGE4_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps, CV_CPU_SSE2);
#endif
template<typename T> static void
merge_( const T** src, T* dst, int len, int cn )
{
int k = cn % 4 ? cn % 4 : 4;
int i, j;
if( k == 1 )
{
const T* src0 = src[0];
for( i = j = 0; i < len; i++, j += cn )
dst[j] = src0[i];
}
else if( k == 2 )
{
const T *src0 = src[0], *src1 = src[1];
i = j = 0;
#if CV_NEON
if(cn == 2)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 2 * inc_i;
VMerge2<T> vmerge;
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vmerge(src0 + i, src1 + i, dst + j);
}
#elif CV_SSE2
if(cn == 2)
{
int inc_i = 32/sizeof(T);
int inc_j = 2 * inc_i;
VMerge2<T> vmerge;
if (vmerge.support)
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vmerge(src0 + i, src1 + i, dst + j);
}
#endif
for( ; i < len; i++, j += cn )
{
dst[j] = src0[i];
dst[j+1] = src1[i];
}
}
else if( k == 3 )
{
const T *src0 = src[0], *src1 = src[1], *src2 = src[2];
i = j = 0;
#if CV_NEON
if(cn == 3)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 3 * inc_i;
VMerge3<T> vmerge;
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vmerge(src0 + i, src1 + i, src2 + i, dst + j);
}
#elif CV_SSE2
if(cn == 3)
{
int inc_i = 32/sizeof(T);
int inc_j = 3 * inc_i;
VMerge3<T> vmerge;
if (vmerge.support)
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vmerge(src0 + i, src1 + i, src2 + i, dst + j);
}
#endif
for( ; i < len; i++, j += cn )
{
dst[j] = src0[i];
dst[j+1] = src1[i];
dst[j+2] = src2[i];
}
}
else
{
const T *src0 = src[0], *src1 = src[1], *src2 = src[2], *src3 = src[3];
i = j = 0;
#if CV_NEON
if(cn == 4)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 4 * inc_i;
VMerge4<T> vmerge;
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vmerge(src0 + i, src1 + i, src2 + i, src3 + i, dst + j);
}
#elif CV_SSE2
if(cn == 4)
{
int inc_i = 32/sizeof(T);
int inc_j = 4 * inc_i;
VMerge4<T> vmerge;
if (vmerge.support)
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vmerge(src0 + i, src1 + i, src2 + i, src3 + i, dst + j);
}
#endif
for( ; i < len; i++, j += cn )
{
dst[j] = src0[i]; dst[j+1] = src1[i];
dst[j+2] = src2[i]; dst[j+3] = src3[i];
}
}
for( ; k < cn; k += 4 )
{
const T *src0 = src[k], *src1 = src[k+1], *src2 = src[k+2], *src3 = src[k+3];
for( i = 0, j = k; i < len; i++, j += cn )
{
dst[j] = src0[i]; dst[j+1] = src1[i];
dst[j+2] = src2[i]; dst[j+3] = src3[i];
}
}
}
void merge8u(const uchar** src, uchar* dst, int len, int cn )
{
CALL_HAL(merge8u, cv_hal_merge8u, src, dst, len, cn)
merge_(src, dst, len, cn);
}
void merge16u(const ushort** src, ushort* dst, int len, int cn )
{
CALL_HAL(merge16u, cv_hal_merge16u, src, dst, len, cn)
merge_(src, dst, len, cn);
}
void merge32s(const int** src, int* dst, int len, int cn )
{
CALL_HAL(merge32s, cv_hal_merge32s, src, dst, len, cn)
merge_(src, dst, len, cn);
}
void merge64s(const int64** src, int64* dst, int len, int cn )
{
CALL_HAL(merge64s, cv_hal_merge64s, src, dst, len, cn)
merge_(src, dst, len, cn);
}
}}

68
modules/core/src/precomp.hpp
View File

@@ -58,8 +58,6 @@
#include "opencv2/core/ocl.hpp"
#endif
#include "opencv2/hal.hpp"
#include <assert.h>
#include <ctype.h>
#include <float.h>
@@ -69,6 +67,27 @@
#include <stdlib.h>
#include <string.h>
#include <algorithm>
#include <cmath>
#include <cstdlib>
#include <limits>
#include <float.h>
#include <cstring>
#include <cassert>
#define USE_SSE2 (cv::checkHardwareSupport(CV_CPU_SSE))
#define USE_SSE4_2 (cv::checkHardwareSupport(CV_CPU_SSE4_2))
#define USE_AVX (cv::checkHardwareSupport(CV_CPU_AVX))
#define USE_AVX2 (cv::checkHardwareSupport(CV_CPU_AVX2))
#include "opencv2/core/hal/hal.hpp"
#include "opencv2/core/hal/intrin.hpp"
#include "opencv2/core/sse_utils.hpp"
#include "opencv2/core/neon_utils.hpp"
#include "arithm_core.hpp"
#include "hal_replacement.hpp"
#ifdef HAVE_TEGRA_OPTIMIZATION
#include "opencv2/core/core_tegra.hpp"
#else
@@ -78,6 +97,34 @@
namespace cv
{
// -128.f ... 255.f
extern const float g_8x32fTab[];
#define CV_8TO32F(x) cv::g_8x32fTab[(x)+128]
extern const ushort g_8x16uSqrTab[];
#define CV_SQR_8U(x) cv::g_8x16uSqrTab[(x)+255]
extern const uchar g_Saturate8u[];
#define CV_FAST_CAST_8U(t) (assert(-256 <= (t) && (t) <= 512), cv::g_Saturate8u[(t)+256])
#define CV_MIN_8U(a,b) ((a) - CV_FAST_CAST_8U((a) - (b)))
#define CV_MAX_8U(a,b) ((a) + CV_FAST_CAST_8U((b) - (a)))
template<> inline uchar OpAdd<uchar>::operator ()(uchar a, uchar b) const
{ return CV_FAST_CAST_8U(a + b); }
template<> inline uchar OpSub<uchar>::operator ()(uchar a, uchar b) const
{ return CV_FAST_CAST_8U(a - b); }
template<> inline short OpAbsDiff<short>::operator ()(short a, short b) const
{ return saturate_cast<short>(std::abs(a - b)); }
template<> inline schar OpAbsDiff<schar>::operator ()(schar a, schar b) const
{ return saturate_cast<schar>(std::abs(a - b)); }
template<> inline uchar OpMin<uchar>::operator ()(uchar a, uchar b) const { return CV_MIN_8U(a, b); }
template<> inline uchar OpMax<uchar>::operator ()(uchar a, uchar b) const { return CV_MAX_8U(a, b); }
typedef void (*BinaryFunc)(const uchar* src1, size_t step1,
const uchar* src2, size_t step2,
uchar* dst, size_t step, Size sz,
@@ -100,21 +147,6 @@ BinaryFunc getCopyMaskFunc(size_t esz);
/* maximal average node_count/hash_size ratio beyond which hash table is resized */
#define CV_SPARSE_HASH_RATIO 3
// -128.f ... 255.f
extern const float g_8x32fTab[];
#define CV_8TO32F(x) cv::g_8x32fTab[(x)+128]
extern const ushort g_8x16uSqrTab[];
#define CV_SQR_8U(x) cv::g_8x16uSqrTab[(x)+255]
extern const uchar g_Saturate8u[];
#define CV_FAST_CAST_8U(t) (assert(-256 <= (t) && (t) <= 512), cv::g_Saturate8u[(t)+256])
#define CV_MIN_8U(a,b) ((a) - CV_FAST_CAST_8U((a) - (b)))
#define CV_MAX_8U(a,b) ((a) + CV_FAST_CAST_8U((b) - (a)))
#if defined WIN32 || defined _WIN32
void deleteThreadAllocData();
#endif
@@ -282,6 +314,4 @@ cv::Mutex& getInitializationMutex();
}
#include "opencv2/hal/intrin.hpp"
#endif /*_CXCORE_INTERNAL_H_*/

428
modules/core/src/split.cpp Normal file
View File

@@ -0,0 +1,428 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009-2011, Willow Garage Inc., all rights reserved.
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
namespace cv { namespace hal {
#if CV_NEON
template<typename T> struct VSplit2;
template<typename T> struct VSplit3;
template<typename T> struct VSplit4;
#define SPLIT2_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template<> \
struct name<data_type> \
{ \
void operator()(const data_type* src, data_type* dst0, \
data_type* dst1) const \
{ \
reg_type r = load_func(src); \
store_func(dst0, r.val[0]); \
store_func(dst1, r.val[1]); \
} \
}
#define SPLIT3_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template<> \
struct name<data_type> \
{ \
void operator()(const data_type* src, data_type* dst0, data_type* dst1, \
data_type* dst2) const \
{ \
reg_type r = load_func(src); \
store_func(dst0, r.val[0]); \
store_func(dst1, r.val[1]); \
store_func(dst2, r.val[2]); \
} \
}
#define SPLIT4_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template<> \
struct name<data_type> \
{ \
void operator()(const data_type* src, data_type* dst0, data_type* dst1, \
data_type* dst2, data_type* dst3) const \
{ \
reg_type r = load_func(src); \
store_func(dst0, r.val[0]); \
store_func(dst1, r.val[1]); \
store_func(dst2, r.val[2]); \
store_func(dst3, r.val[3]); \
} \
}
SPLIT2_KERNEL_TEMPLATE(VSplit2, uchar , uint8x16x2_t, vld2q_u8 , vst1q_u8 );
SPLIT2_KERNEL_TEMPLATE(VSplit2, ushort, uint16x8x2_t, vld2q_u16, vst1q_u16);
SPLIT2_KERNEL_TEMPLATE(VSplit2, int , int32x4x2_t, vld2q_s32, vst1q_s32);
SPLIT2_KERNEL_TEMPLATE(VSplit2, int64 , int64x1x2_t, vld2_s64 , vst1_s64 );
SPLIT3_KERNEL_TEMPLATE(VSplit3, uchar , uint8x16x3_t, vld3q_u8 , vst1q_u8 );
SPLIT3_KERNEL_TEMPLATE(VSplit3, ushort, uint16x8x3_t, vld3q_u16, vst1q_u16);
SPLIT3_KERNEL_TEMPLATE(VSplit3, int , int32x4x3_t, vld3q_s32, vst1q_s32);
SPLIT3_KERNEL_TEMPLATE(VSplit3, int64 , int64x1x3_t, vld3_s64 , vst1_s64 );
SPLIT4_KERNEL_TEMPLATE(VSplit4, uchar , uint8x16x4_t, vld4q_u8 , vst1q_u8 );
SPLIT4_KERNEL_TEMPLATE(VSplit4, ushort, uint16x8x4_t, vld4q_u16, vst1q_u16);
SPLIT4_KERNEL_TEMPLATE(VSplit4, int , int32x4x4_t, vld4q_s32, vst1q_s32);
SPLIT4_KERNEL_TEMPLATE(VSplit4, int64 , int64x1x4_t, vld4_s64 , vst1_s64 );
#elif CV_SSE2
template <typename T>
struct VSplit2
{
VSplit2() : support(false) { }
void operator()(const T *, T *, T *) const { }
bool support;
};
template <typename T>
struct VSplit3
{
VSplit3() : support(false) { }
void operator()(const T *, T *, T *, T *) const { }
bool support;
};
template <typename T>
struct VSplit4
{
VSplit4() : support(false) { }
void operator()(const T *, T *, T *, T *, T *) const { }
bool support;
};
#define SPLIT2_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_deinterleave, flavor) \
template <> \
struct VSplit2<data_type> \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VSplit2() \
{ \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, \
data_type * dst0, data_type * dst1) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((cast_type const *)(src)); \
reg_type v_src1 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 2)); \
reg_type v_src3 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 3)); \
\
_mm_deinterleave(v_src0, v_src1, v_src2, v_src3); \
\
_mm_storeu_##flavor((cast_type *)(dst0), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst0 + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst1), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst1 + ELEMS_IN_VEC), v_src3); \
} \
\
bool support; \
}
#define SPLIT3_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_deinterleave, flavor) \
template <> \
struct VSplit3<data_type> \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VSplit3() \
{ \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, \
data_type * dst0, data_type * dst1, data_type * dst2) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((cast_type const *)(src)); \
reg_type v_src1 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 2)); \
reg_type v_src3 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 3)); \
reg_type v_src4 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 4)); \
reg_type v_src5 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 5)); \
\
_mm_deinterleave(v_src0, v_src1, v_src2, \
v_src3, v_src4, v_src5); \
\
_mm_storeu_##flavor((cast_type *)(dst0), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst0 + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst1), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst1 + ELEMS_IN_VEC), v_src3); \
_mm_storeu_##flavor((cast_type *)(dst2), v_src4); \
_mm_storeu_##flavor((cast_type *)(dst2 + ELEMS_IN_VEC), v_src5); \
} \
\
bool support; \
}
#define SPLIT4_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_deinterleave, flavor) \
template <> \
struct VSplit4<data_type> \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VSplit4() \
{ \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, data_type * dst0, data_type * dst1, \
data_type * dst2, data_type * dst3) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((cast_type const *)(src)); \
reg_type v_src1 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 2)); \
reg_type v_src3 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 3)); \
reg_type v_src4 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 4)); \
reg_type v_src5 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 5)); \
reg_type v_src6 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 6)); \
reg_type v_src7 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 7)); \
\
_mm_deinterleave(v_src0, v_src1, v_src2, v_src3, \
v_src4, v_src5, v_src6, v_src7); \
\
_mm_storeu_##flavor((cast_type *)(dst0), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst0 + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst1), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst1 + ELEMS_IN_VEC), v_src3); \
_mm_storeu_##flavor((cast_type *)(dst2), v_src4); \
_mm_storeu_##flavor((cast_type *)(dst2 + ELEMS_IN_VEC), v_src5); \
_mm_storeu_##flavor((cast_type *)(dst3), v_src6); \
_mm_storeu_##flavor((cast_type *)(dst3 + ELEMS_IN_VEC), v_src7); \
} \
\
bool support; \
}
SPLIT2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
SPLIT2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
SPLIT2_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
SPLIT3_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
SPLIT3_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
SPLIT3_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
SPLIT4_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
SPLIT4_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
SPLIT4_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
#endif
template<typename T> static void
split_( const T* src, T** dst, int len, int cn )
{
int k = cn % 4 ? cn % 4 : 4;
int i, j;
if( k == 1 )
{
T* dst0 = dst[0];
if(cn == 1)
{
memcpy(dst0, src, len * sizeof(T));
}
else
{
for( i = 0, j = 0 ; i < len; i++, j += cn )
dst0[i] = src[j];
}
}
else if( k == 2 )
{
T *dst0 = dst[0], *dst1 = dst[1];
i = j = 0;
#if CV_NEON
if(cn == 2)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 2 * inc_i;
VSplit2<T> vsplit;
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i);
}
#elif CV_SSE2
if (cn == 2)
{
int inc_i = 32/sizeof(T);
int inc_j = 2 * inc_i;
VSplit2<T> vsplit;
if (vsplit.support)
{
for( ; i <= len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i);
}
}
#endif
for( ; i < len; i++, j += cn )
{
dst0[i] = src[j];
dst1[i] = src[j+1];
}
}
else if( k == 3 )
{
T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2];
i = j = 0;
#if CV_NEON
if(cn == 3)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 3 * inc_i;
VSplit3<T> vsplit;
for( ; i <= len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i, dst2 + i);
}
#elif CV_SSE2
if (cn == 3)
{
int inc_i = 32/sizeof(T);
int inc_j = 3 * inc_i;
VSplit3<T> vsplit;
if (vsplit.support)
{
for( ; i <= len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i, dst2 + i);
}
}
#endif
for( ; i < len; i++, j += cn )
{
dst0[i] = src[j];
dst1[i] = src[j+1];
dst2[i] = src[j+2];
}
}
else
{
T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2], *dst3 = dst[3];
i = j = 0;
#if CV_NEON
if(cn == 4)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 4 * inc_i;
VSplit4<T> vsplit;
for( ; i <= len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i, dst2 + i, dst3 + i);
}
#elif CV_SSE2
if (cn == 4)
{
int inc_i = 32/sizeof(T);
int inc_j = 4 * inc_i;
VSplit4<T> vsplit;
if (vsplit.support)
{
for( ; i <= len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i, dst2 + i, dst3 + i);
}
}
#endif
for( ; i < len; i++, j += cn )
{
dst0[i] = src[j]; dst1[i] = src[j+1];
dst2[i] = src[j+2]; dst3[i] = src[j+3];
}
}
for( ; k < cn; k += 4 )
{
T *dst0 = dst[k], *dst1 = dst[k+1], *dst2 = dst[k+2], *dst3 = dst[k+3];
for( i = 0, j = k; i < len; i++, j += cn )
{
dst0[i] = src[j]; dst1[i] = src[j+1];
dst2[i] = src[j+2]; dst3[i] = src[j+3];
}
}
}
void split8u(const uchar* src, uchar** dst, int len, int cn )
{
CALL_HAL(split8u, cv_hal_split8u, src,dst, len, cn)
split_(src, dst, len, cn);
}
void split16u(const ushort* src, ushort** dst, int len, int cn )
{
CALL_HAL(split16u, cv_hal_split16u, src,dst, len, cn)
split_(src, dst, len, cn);
}
void split32s(const int* src, int** dst, int len, int cn )
{
CALL_HAL(split32s, cv_hal_split32s, src,dst, len, cn)
split_(src, dst, len, cn);
}
void split64s(const int64* src, int64** dst, int len, int cn )
{
CALL_HAL(split64s, cv_hal_split64s, src,dst, len, cn)
split_(src, dst, len, cn);
}
}}

263
modules/core/src/stat.cpp
View File

@@ -3996,3 +3996,266 @@ cvNorm( const void* imgA, const void* imgB, int normType, const void* maskarr )
return !maskarr ? cv::norm(a, b, normType) : cv::norm(a, b, normType, mask);
}
namespace cv { namespace hal {
static const uchar popCountTable[] =
{
0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4, 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, 4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8
};
static const uchar popCountTable2[] =
{
0, 1, 1, 1, 1, 2, 2, 2, 1, 2, 2, 2, 1, 2, 2, 2, 1, 2, 2, 2, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3,
1, 2, 2, 2, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 1, 2, 2, 2, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3,
1, 2, 2, 2, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4,
2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4,
1, 2, 2, 2, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4,
2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4,
1, 2, 2, 2, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4,
2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4, 2, 3, 3, 3, 3, 4, 4, 4, 3, 4, 4, 4, 3, 4, 4, 4
};
static const uchar popCountTable4[] =
{
0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
};
int normHamming(const uchar* a, int n)
{
int i = 0;
int result = 0;
#if CV_NEON
{
uint32x4_t bits = vmovq_n_u32(0);
for (; i <= n - 16; i += 16) {
uint8x16_t A_vec = vld1q_u8 (a + i);
uint8x16_t bitsSet = vcntq_u8 (A_vec);
uint16x8_t bitSet8 = vpaddlq_u8 (bitsSet);
uint32x4_t bitSet4 = vpaddlq_u16 (bitSet8);
bits = vaddq_u32(bits, bitSet4);
}
uint64x2_t bitSet2 = vpaddlq_u32 (bits);
result = vgetq_lane_s32 (vreinterpretq_s32_u64(bitSet2),0);
result += vgetq_lane_s32 (vreinterpretq_s32_u64(bitSet2),2);
}
#endif
for( ; i <= n - 4; i += 4 )
result += popCountTable[a[i]] + popCountTable[a[i+1]] +
popCountTable[a[i+2]] + popCountTable[a[i+3]];
for( ; i < n; i++ )
result += popCountTable[a[i]];
return result;
}
int normHamming(const uchar* a, const uchar* b, int n)
{
int i = 0;
int result = 0;
#if CV_NEON
{
uint32x4_t bits = vmovq_n_u32(0);
for (; i <= n - 16; i += 16) {
uint8x16_t A_vec = vld1q_u8 (a + i);
uint8x16_t B_vec = vld1q_u8 (b + i);
uint8x16_t AxorB = veorq_u8 (A_vec, B_vec);
uint8x16_t bitsSet = vcntq_u8 (AxorB);
uint16x8_t bitSet8 = vpaddlq_u8 (bitsSet);
uint32x4_t bitSet4 = vpaddlq_u16 (bitSet8);
bits = vaddq_u32(bits, bitSet4);
}
uint64x2_t bitSet2 = vpaddlq_u32 (bits);
result = vgetq_lane_s32 (vreinterpretq_s32_u64(bitSet2),0);
result += vgetq_lane_s32 (vreinterpretq_s32_u64(bitSet2),2);
}
#endif
for( ; i <= n - 4; i += 4 )
result += popCountTable[a[i] ^ b[i]] + popCountTable[a[i+1] ^ b[i+1]] +
popCountTable[a[i+2] ^ b[i+2]] + popCountTable[a[i+3] ^ b[i+3]];
for( ; i < n; i++ )
result += popCountTable[a[i] ^ b[i]];
return result;
}
int normHamming(const uchar* a, int n, int cellSize)
{
if( cellSize == 1 )
return normHamming(a, n);
const uchar* tab = 0;
if( cellSize == 2 )
tab = popCountTable2;
else if( cellSize == 4 )
tab = popCountTable4;
else
return -1;
int i = 0;
int result = 0;
#if CV_ENABLE_UNROLLED
for( ; i <= n - 4; i += 4 )
result += tab[a[i]] + tab[a[i+1]] + tab[a[i+2]] + tab[a[i+3]];
#endif
for( ; i < n; i++ )
result += tab[a[i]];
return result;
}
int normHamming(const uchar* a, const uchar* b, int n, int cellSize)
{
if( cellSize == 1 )
return normHamming(a, b, n);
const uchar* tab = 0;
if( cellSize == 2 )
tab = popCountTable2;
else if( cellSize == 4 )
tab = popCountTable4;
else
return -1;
int i = 0;
int result = 0;
#if CV_ENABLE_UNROLLED
for( ; i <= n - 4; i += 4 )
result += tab[a[i] ^ b[i]] + tab[a[i+1] ^ b[i+1]] +
tab[a[i+2] ^ b[i+2]] + tab[a[i+3] ^ b[i+3]];
#endif
for( ; i < n; i++ )
result += tab[a[i] ^ b[i]];
return result;
}
float normL2Sqr_(const float* a, const float* b, int n)
{
int j = 0; float d = 0.f;
#if CV_SSE
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];
#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
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];
#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
__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)));
#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;
}
}} //cv::hal

188
modules/core/src/system.cpp
View File

@@ -86,6 +86,45 @@ Mutex* __initialization_mutex_initializer = &getInitializationMutex();
#undef max
#undef abs
#include <tchar.h>
#if defined _MSC_VER
#if _MSC_VER >= 1400
#include <intrin.h>
#elif defined _M_IX86
static void __cpuid(int* cpuid_data, int)
{
__asm
{
push ebx
push edi
mov edi, cpuid_data
mov eax, 1
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
pop edi
pop ebx
}
}
static void __cpuidex(int* cpuid_data, int, int)
{
__asm
{
push edi
mov edi, cpuid_data
mov eax, 7
mov ecx, 0
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
pop edi
}
}
#endif
#endif
#ifdef WINRT
#include <wrl/client.h>
@@ -198,15 +237,154 @@ void Exception::formatMessage()
msg = format("%s:%d: error: (%d) %s\n", file.c_str(), line, code, err.c_str());
}
struct HWFeatures
{
enum { MAX_FEATURE = CV_HARDWARE_MAX_FEATURE };
HWFeatures(void)
{
memset( have, 0, sizeof(have) );
x86_family = 0;
}
static HWFeatures initialize(void)
{
HWFeatures f;
int cpuid_data[4] = { 0, 0, 0, 0 };
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuid(cpuid_data, 1);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $1, %%eax\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $1,%%eax\n\t"
"cpuid\n\t"
"popl %%ebx\n\t"
: "=a"(cpuid_data[0]), "=c"(cpuid_data[2]), "=d"(cpuid_data[3])
:
: "cc"
);
#endif
#endif
f.x86_family = (cpuid_data[0] >> 8) & 15;
if( f.x86_family >= 6 )
{
f.have[CV_CPU_MMX] = (cpuid_data[3] & (1 << 23)) != 0;
f.have[CV_CPU_SSE] = (cpuid_data[3] & (1<<25)) != 0;
f.have[CV_CPU_SSE2] = (cpuid_data[3] & (1<<26)) != 0;
f.have[CV_CPU_SSE3] = (cpuid_data[2] & (1<<0)) != 0;
f.have[CV_CPU_SSSE3] = (cpuid_data[2] & (1<<9)) != 0;
f.have[CV_CPU_FMA3] = (cpuid_data[2] & (1<<12)) != 0;
f.have[CV_CPU_SSE4_1] = (cpuid_data[2] & (1<<19)) != 0;
f.have[CV_CPU_SSE4_2] = (cpuid_data[2] & (1<<20)) != 0;
f.have[CV_CPU_POPCNT] = (cpuid_data[2] & (1<<23)) != 0;
f.have[CV_CPU_AVX] = (((cpuid_data[2] & (1<<28)) != 0)&&((cpuid_data[2] & (1<<27)) != 0));//OS uses XSAVE_XRSTORE and CPU support AVX
// make the second call to the cpuid command in order to get
// information about extended features like AVX2
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuidex(cpuid_data, 7, 0);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $7, %%eax\n\t"
"movl $0, %%ecx\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $7,%%eax\n\t"
"movl $0,%%ecx\n\t"
"cpuid\n\t"
"movl %%ebx, %0\n\t"
"popl %%ebx\n\t"
: "=r"(cpuid_data[1]), "=c"(cpuid_data[2])
:
: "cc"
);
#endif
#endif
f.have[CV_CPU_AVX2] = (cpuid_data[1] & (1<<5)) != 0;
f.have[CV_CPU_AVX_512F] = (cpuid_data[1] & (1<<16)) != 0;
f.have[CV_CPU_AVX_512DQ] = (cpuid_data[1] & (1<<17)) != 0;
f.have[CV_CPU_AVX_512IFMA512] = (cpuid_data[1] & (1<<21)) != 0;
f.have[CV_CPU_AVX_512PF] = (cpuid_data[1] & (1<<26)) != 0;
f.have[CV_CPU_AVX_512ER] = (cpuid_data[1] & (1<<27)) != 0;
f.have[CV_CPU_AVX_512CD] = (cpuid_data[1] & (1<<28)) != 0;
f.have[CV_CPU_AVX_512BW] = (cpuid_data[1] & (1<<30)) != 0;
f.have[CV_CPU_AVX_512VL] = (cpuid_data[1] & (1<<31)) != 0;
f.have[CV_CPU_AVX_512VBMI] = (cpuid_data[2] & (1<<1)) != 0;
}
#if defined ANDROID || defined __linux__
#ifdef __aarch64__
f.have[CV_CPU_NEON] = true;
#else
int cpufile = open("/proc/self/auxv", O_RDONLY);
if (cpufile >= 0)
{
Elf32_auxv_t auxv;
const size_t size_auxv_t = sizeof(auxv);
while ((size_t)read(cpufile, &auxv, size_auxv_t) == size_auxv_t)
{
if (auxv.a_type == AT_HWCAP)
{
f.have[CV_CPU_NEON] = (auxv.a_un.a_val & 4096) != 0;
break;
}
}
close(cpufile);
}
#endif
#elif (defined __clang__ || defined __APPLE__) && (defined __ARM_NEON__ || (defined __ARM_NEON && defined __aarch64__))
f.have[CV_CPU_NEON] = true;
#endif
return f;
}
int x86_family;
bool have[MAX_FEATURE+1];
};
static HWFeatures featuresEnabled = HWFeatures::initialize(), featuresDisabled = HWFeatures();
static HWFeatures* currentFeatures = &featuresEnabled;
bool checkHardwareSupport(int feature)
{
CV_DbgAssert( 0 <= feature && feature <= CV_HARDWARE_MAX_FEATURE );
return cv::hal::checkHardwareSupport(feature);
return currentFeatures->have[feature];
}
volatile bool useOptimizedFlag = true;
void setUseOptimized( bool flag )
{
cv::hal::setUseOptimized(flag);
useOptimizedFlag = flag;
currentFeatures = flag ? &featuresEnabled : &featuresDisabled;
ipp::setUseIPP(flag);
#ifdef HAVE_OPENCL
@@ -219,7 +397,7 @@ void setUseOptimized( bool flag )
bool useOptimized(void)
{
return cv::hal::useOptimized();
return useOptimizedFlag;
}
int64 getTickCount(void)
@@ -499,12 +677,12 @@ redirectError( CvErrorCallback errCallback, void* userdata, void** prevUserdata)
CV_IMPL int cvCheckHardwareSupport(int feature)
{
CV_DbgAssert( 0 <= feature && feature <= CV_HARDWARE_MAX_FEATURE );
return cv::hal::checkHardwareSupport(feature);
return cv::currentFeatures->have[feature];
}
CV_IMPL int cvUseOptimized( int flag )
{
int prevMode = cv::useOptimized();
int prevMode = cv::useOptimizedFlag;
cv::setUseOptimized( flag != 0 );
return prevMode;
}