added IPP dot product functions. Only 32f data type supported for now (there are accuracy issues in 8u/16s/16u/32s functions which will be fixed in IPP 7.0.3)
This commit is contained in:
Vladimir Dudnik
parent
17dc1e1340
commit
4f83a06358
@ -42,6 +42,10 @@
|
||||
|
||||
#include "precomp.hpp"
|
||||
|
||||
#ifdef HAVE_IPP
|
||||
#include "ippversion.h"
|
||||
#endif
|
||||
|
||||
namespace cv
|
||||
{
|
||||
|
||||
@ -2629,11 +2633,183 @@ void mulTransposed( const Mat& src, Mat& dst, bool ata,
|
||||
* Dot Product *
|
||||
\****************************************************************************************/
|
||||
|
||||
#if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
|
||||
|
||||
static double ippDotProd8u(const Mat& srcmat1, const Mat& srcmat2)
|
||||
{
|
||||
int nchan = srcmat1.channels();
|
||||
Ipp64f sum[4] = { 0.0 };
|
||||
IppiSize roi = { srcmat1.cols, srcmat1.rows };
|
||||
|
||||
switch(nchan)
|
||||
{
|
||||
case 1:
|
||||
ippiDotProd_8u64f_C1R((const Ipp8u*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp8u*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 3:
|
||||
ippiDotProd_8u64f_C3R((const Ipp8u*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp8u*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
ippiDotProd_8u64f_C4R((const Ipp8u*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp8u*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
}
|
||||
|
||||
for(int c = 1; c < nchan; c++)
|
||||
{
|
||||
sum[0] += sum[c];
|
||||
}
|
||||
|
||||
return sum[0];
|
||||
} // ippDotProd8u()
|
||||
|
||||
|
||||
static double ippDotProd16u(const Mat& srcmat1, const Mat& srcmat2)
|
||||
{
|
||||
int nchan = srcmat1.channels();
|
||||
Ipp64f sum[4] = { 0.0 };
|
||||
IppiSize roi = { srcmat1.cols, srcmat1.rows };
|
||||
|
||||
switch(nchan)
|
||||
{
|
||||
case 1:
|
||||
ippiDotProd_16u64f_C1R((const Ipp16u*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp16u*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 3:
|
||||
ippiDotProd_16u64f_C3R((const Ipp16u*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp16u*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
ippiDotProd_16u64f_C4R((const Ipp16u*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp16u*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
}
|
||||
|
||||
for(int c = 1; c < nchan; c++)
|
||||
{
|
||||
sum[0] += sum[c];
|
||||
}
|
||||
|
||||
return sum[0];
|
||||
} // ippDotProd16u()
|
||||
|
||||
|
||||
static double ippDotProd16s(const Mat& srcmat1, const Mat& srcmat2)
|
||||
{
|
||||
int nchan = srcmat1.channels();
|
||||
Ipp64f sum[4] = { 0.0 };
|
||||
IppiSize roi = { srcmat1.cols, srcmat1.rows };
|
||||
|
||||
switch(nchan)
|
||||
{
|
||||
case 1:
|
||||
ippiDotProd_16s64f_C1R((const Ipp16s*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp16s*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 3:
|
||||
ippiDotProd_16s64f_C3R((const Ipp16s*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp16s*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
ippiDotProd_16s64f_C4R((const Ipp16s*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp16s*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
}
|
||||
|
||||
for(int c = 1; c < nchan; c++)
|
||||
{
|
||||
sum[0] += sum[c];
|
||||
}
|
||||
|
||||
return sum[0];
|
||||
} // ippDotProd16s()
|
||||
|
||||
|
||||
static double ippDotProd32s(const Mat& srcmat1, const Mat& srcmat2)
|
||||
{
|
||||
int nchan = srcmat1.channels();
|
||||
Ipp64f sum[4] = { 0.0 };
|
||||
IppiSize roi = { srcmat1.cols, srcmat1.rows };
|
||||
|
||||
switch(nchan)
|
||||
{
|
||||
case 1:
|
||||
ippiDotProd_32s64f_C1R((const Ipp32s*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp32s*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 3:
|
||||
ippiDotProd_32s64f_C3R((const Ipp32s*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp32s*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
ippiDotProd_32s64f_C4R((const Ipp32s*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp32s*)srcmat2.data, (int)srcmat2.step, roi, sum);
|
||||
break;
|
||||
}
|
||||
|
||||
for(int c = 1; c < nchan; c++)
|
||||
{
|
||||
sum[0] += sum[c];
|
||||
}
|
||||
|
||||
return sum[0];
|
||||
} // ippDotProd32s()
|
||||
|
||||
|
||||
static double ippDotProd32f(const Mat& srcmat1, const Mat& srcmat2)
|
||||
{
|
||||
int nchan = srcmat1.channels();
|
||||
Ipp64f sum[4] = { 0.0 };
|
||||
IppiSize roi = { srcmat1.cols, srcmat1.rows };
|
||||
|
||||
switch(nchan)
|
||||
{
|
||||
case 1:
|
||||
ippiDotProd_32f64f_C1R((const Ipp32f*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp32f*)srcmat2.data, (int)srcmat2.step, roi, sum, ippAlgHintAccurate);
|
||||
break;
|
||||
|
||||
case 3:
|
||||
ippiDotProd_32f64f_C3R((const Ipp32f*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp32f*)srcmat2.data, (int)srcmat2.step, roi, sum, ippAlgHintAccurate);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
ippiDotProd_32f64f_C4R((const Ipp32f*)srcmat1.data, (int)srcmat1.step,
|
||||
(const Ipp32f*)srcmat2.data, (int)srcmat2.step, roi, sum, ippAlgHintAccurate);
|
||||
break;
|
||||
}
|
||||
|
||||
for(int c = 1; c < nchan; c++)
|
||||
{
|
||||
sum[0] += sum[c];
|
||||
}
|
||||
|
||||
return sum[0];
|
||||
} // ippDotProd32f()
|
||||
|
||||
#endif
|
||||
|
||||
template<typename T, typename WT, typename ST> static double
|
||||
dotprod_( const Mat& srcmat1, const Mat& srcmat2 )
|
||||
{
|
||||
const T *src1 = (const T*)srcmat1.data, *src2 = (const T*)srcmat2.data;
|
||||
size_t step1 = srcmat1.step/sizeof(src1[0]), step2 = srcmat2.step/sizeof(src2[0]);
|
||||
const T* src1 = (const T*)srcmat1.data;
|
||||
const T* src2 = (const T*)srcmat2.data;
|
||||
|
||||
size_t step1 = srcmat1.step/sizeof(src1[0]);
|
||||
size_t step2 = srcmat2.step/sizeof(src2[0]);
|
||||
|
||||
ST sum = 0;
|
||||
Size size = getContinuousSize( srcmat1, srcmat2, srcmat1.channels() );
|
||||
|
||||
@ -2642,6 +2818,7 @@ dotprod_( const Mat& srcmat1, const Mat& srcmat2 )
|
||||
WT t = 0;
|
||||
for( ; size.height--; src1 += step1, src2 += step2 )
|
||||
t += (WT)src1[0]*src2[0];
|
||||
|
||||
sum += t;
|
||||
}
|
||||
else
|
||||
@ -2652,17 +2829,19 @@ dotprod_( const Mat& srcmat1, const Mat& srcmat2 )
|
||||
WT t = 0;
|
||||
for( i = 0; i <= size.width - 4; i += 4 )
|
||||
{
|
||||
sum += (WT)src1[i]*src2[i] +
|
||||
(WT)src1[i+1]*src2[i+1] +
|
||||
(WT)src1[i+2]*src2[i+2] +
|
||||
(WT)src1[i+3]*src2[i+3];
|
||||
sum += (WT)src1[i ]*src2[i ] +
|
||||
(WT)src1[i+1]*src2[i+1] +
|
||||
(WT)src1[i+2]*src2[i+2] +
|
||||
(WT)src1[i+3]*src2[i+3];
|
||||
}
|
||||
|
||||
for( ; i < size.width; i++ )
|
||||
t += (WT)src1[i]*src2[i];
|
||||
|
||||
sum += t;
|
||||
}
|
||||
}
|
||||
|
||||
return (double)sum;
|
||||
}
|
||||
|
||||
@ -2670,16 +2849,43 @@ typedef double (*DotProductFunc)(const Mat& src1, const Mat& src2);
|
||||
|
||||
double Mat::dot(const Mat& mat) const
|
||||
{
|
||||
static DotProductFunc tab[] = {
|
||||
dotprod_<uchar, int, int64>, 0,
|
||||
#if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
|
||||
static DotProductFunc ipptab[] =
|
||||
{
|
||||
dotprod_<uchar, int, int64>,
|
||||
0,
|
||||
dotprod_<ushort, double, double>,
|
||||
dotprod_<short, double, double>,
|
||||
dotprod_<int, double, double>,
|
||||
ippDotProd32f,
|
||||
dotprod_<double, double, double>,
|
||||
0
|
||||
};
|
||||
#endif
|
||||
|
||||
static DotProductFunc tab[] =
|
||||
{
|
||||
dotprod_<uchar, int, int64>,
|
||||
0,
|
||||
dotprod_<ushort, double, double>,
|
||||
dotprod_<short, double, double>,
|
||||
dotprod_<int, double, double>,
|
||||
dotprod_<float, double, double>,
|
||||
dotprod_<double, double, double>, 0 };
|
||||
dotprod_<double, double, double>,
|
||||
0
|
||||
};
|
||||
|
||||
DotProductFunc func = tab[depth()];
|
||||
|
||||
#if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
|
||||
if((*this).channels() != 2)
|
||||
{
|
||||
func = ipptab[depth()];
|
||||
}
|
||||
#endif
|
||||
|
||||
CV_Assert( mat.type() == type() && mat.size() == size() && func != 0 );
|
||||
|
||||
return func( *this, mat );
|
||||
}
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user