提交 d16e0b37 编写于 作者: I Ilya Lavrenov

disabled IPP functions that slower than OpenCV

上级 e1b5a547
......@@ -423,7 +423,7 @@ Mat& Mat::setTo(InputArray _value, InputArray _mask)
int cn = channels(), depth0 = depth();
if (!mask.empty() && (dims <= 2 || (isContinuous() && mask.isContinuous())) &&
(depth0 == CV_8U || depth0 == CV_16U || depth0 == CV_16S || depth0 == CV_32S || depth0 == CV_32F) &&
(/*depth0 == CV_8U ||*/ depth0 == CV_16U || depth0 == CV_16S || depth0 == CV_32S || depth0 == CV_32F) &&
(cn == 1 || cn == 3 || cn == 4))
{
uchar _buf[32];
......@@ -442,9 +442,9 @@ Mat& Mat::setTo(InputArray _value, InputArray _mask)
if (cn == 1)
{
if (depth0 == CV_8U)
/*if (depth0 == CV_8U)
status = ippiSet_8u_C1MR(*(Ipp8u *)buf, (Ipp8u *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_16U)
else*/ if (depth0 == CV_16U)
status = ippiSet_16u_C1MR(*(Ipp16u *)buf, (Ipp16u *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_16S)
status = ippiSet_16s_C1MR(*(Ipp16s *)buf, (Ipp16s *)data, dstep, roisize, mask.data, mstep);
......@@ -468,9 +468,9 @@ Mat& Mat::setTo(InputArray _value, InputArray _mask)
{ \
if (cn == ippcn) \
{ \
if (depth0 == CV_8U) \
/*if (depth0 == CV_8U) \
IPP_SET(8u, ippcn); \
else if (depth0 == CV_16U) \
else*/ if (depth0 == CV_16U) \
IPP_SET(16u, ippcn); \
else if (depth0 == CV_16S) \
IPP_SET(16s, ippcn); \
......
......@@ -237,7 +237,7 @@ float cubeRoot( float value )
static void Magnitude_32f(const float* x, const float* y, float* mag, int len)
{
#if defined(HAVE_IPP)
#if defined HAVE_IPP && 0
IppStatus status = ippsMagnitude_32f(x, y, mag, len);
if (status >= 0)
return;
......
......@@ -2798,7 +2798,7 @@ dotProd_(const T* src1, const T* src2, int len)
static double dotProd_8u(const uchar* src1, const uchar* src2, int len)
{
double r = 0;
#if ARITHM_USE_IPP
#if ARITHM_USE_IPP && 0
if (0 <= ippiDotProd_8u64f_C1R(src1, (int)(len*sizeof(src1[0])),
src2, (int)(len*sizeof(src2[0])),
ippiSize(len, 1), &r))
......
......@@ -3617,25 +3617,25 @@ namespace cv
#if IPP_VERSION_X100 > 0
#define USE_IPP_SORT
typedef IppStatus (CV_STDCALL *IppSortFunc)(void *, int);
typedef IppStatus (CV_STDCALL * IppSortFunc)(void *, int);
typedef IppSortFunc IppFlipFunc;
static IppSortFunc getSortFunc(int depth, bool sortDescending)
{
if (!sortDescending)
return depth == CV_8U ? (IppSortFunc)ippsSortAscend_8u_I :
depth == CV_16U ? (IppSortFunc)ippsSortAscend_16u_I :
/*depth == CV_16U ? (IppSortFunc)ippsSortAscend_16u_I :
depth == CV_16S ? (IppSortFunc)ippsSortAscend_16s_I :
depth == CV_32S ? (IppSortFunc)ippsSortAscend_32s_I :
depth == CV_32F ? (IppSortFunc)ippsSortAscend_32f_I :
depth == CV_64F ? (IppSortFunc)ippsSortAscend_64f_I : 0;
depth == CV_64F ? (IppSortFunc)ippsSortAscend_64f_I :*/ 0;
else
return depth == CV_8U ? (IppSortFunc)ippsSortDescend_8u_I :
depth == CV_16U ? (IppSortFunc)ippsSortDescend_16u_I :
/*depth == CV_16U ? (IppSortFunc)ippsSortDescend_16u_I :
depth == CV_16S ? (IppSortFunc)ippsSortDescend_16s_I :
depth == CV_32S ? (IppSortFunc)ippsSortDescend_32s_I :
depth == CV_32F ? (IppSortFunc)ippsSortDescend_32f_I :
depth == CV_64F ? (IppSortFunc)ippsSortDescend_64f_I : 0;
depth == CV_64F ? (IppSortFunc)ippsSortDescend_64f_I :*/ 0;
}
static IppFlipFunc getFlipFunc(int depth)
......@@ -3643,9 +3643,9 @@ static IppFlipFunc getFlipFunc(int depth)
CV_SUPPRESS_DEPRECATED_START
return
depth == CV_8U || depth == CV_8S ? (IppFlipFunc)ippsFlip_8u_I :
depth == CV_16U || depth == CV_16S ? (IppFlipFunc)ippsFlip_16u_I :
/*depth == CV_16U || depth == CV_16S ? (IppFlipFunc)ippsFlip_16u_I :
depth == CV_32S || depth == CV_32F ? (IppFlipFunc)ippsFlip_32f_I :
depth == CV_64F ? (IppFlipFunc)ippsFlip_64f_I : 0;
depth == CV_64F ? (IppFlipFunc)ippsFlip_64f_I : */0;
CV_SUPPRESS_DEPRECATED_END
}
......@@ -3700,7 +3700,8 @@ template<typename T> static void sort_( const Mat& src, Mat& dst, int flags )
#endif
{
#ifdef USE_IPP_SORT
setIppErrorStatus();
if (depth != CV_8U)
setIppErrorStatus();
#endif
std::sort( ptr, ptr + len );
if( sortDescending )
......@@ -3732,7 +3733,7 @@ public:
const _Tp* arr;
};
#ifdef USE_IPP_SORT
#if defined USE_IPP_SORT && 0
typedef IppStatus (CV_STDCALL *IppSortIndexFunc)(void *, int *, int);
......@@ -3779,7 +3780,7 @@ template<typename T> static void sortIdx_( const Mat& src, Mat& dst, int flags )
bptr = (T*)buf;
_iptr = (int*)ibuf;
#ifdef USE_IPP_SORT
#if defined USE_IPP_SORT && 0
int depth = src.depth();
IppSortIndexFunc ippFunc = getSortIndexFunc(depth, sortDescending);
IppFlipFunc ippFlipFunc = getFlipFunc(depth);
......@@ -3803,21 +3804,21 @@ template<typename T> static void sortIdx_( const Mat& src, Mat& dst, int flags )
for( j = 0; j < len; j++ )
iptr[j] = j;
#ifdef USE_IPP_SORT
#if defined USE_IPP_SORT && 0
if (sortRows || !ippFunc || ippFunc(ptr, iptr, len) < 0)
#endif
{
#ifdef USE_IPP_SORT
#if defined USE_IPP_SORT && 0
setIppErrorStatus();
#endif
std::sort( iptr, iptr + len, LessThanIdx<T>(ptr) );
if( sortDescending )
{
#ifdef USE_IPP_SORT
#if defined USE_IPP_SORT && 0
if (!ippFlipFunc || ippFlipFunc(iptr, len) < 0)
#endif
{
#ifdef USE_IPP_SORT
#if defined USE_IPP_SORT && 0
setIppErrorStatus();
#endif
for( j = 0; j < len/2; j++ )
......
......@@ -3977,22 +3977,20 @@ void cv::cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
setIppErrorStatus();
}
else
#endif
if (code == CV_LRGB2Lab && scn == 3 && depth == CV_8U)
if (code == CV_LRGB2Lab && scn == 3 && depth == CV_8U) // slower than OpenCV
{
if (CvtColorIPPLoop(src, dst, IPPReorderGeneralFunctor(ippiSwapChannelsC3RTab[depth],
(ippiGeneralFunc)ippiBGRToLab_8u_C3R, 2, 1, 0, depth)))
return;
setIppErrorStatus();
}
else if (code == CV_LRGB2Lab && scn == 4 && depth == CV_8U)
else if (code == CV_LRGB2Lab && scn == 4 && depth == CV_8U) // slower than OpenCV
{
if (CvtColorIPPLoop(src, dst, IPPReorderGeneralFunctor(ippiSwapChannelsC4C3RTab[depth],
(ippiGeneralFunc)ippiBGRToLab_8u_C3R, 2, 1, 0, depth)))
return;
setIppErrorStatus();
}
#if 0
else if (code == CV_LRGB2Luv && scn == 3)
{
if (CvtColorIPPLoop(src, dst, IPPGeneralFunctor(ippiRGBToLUVTab[depth])))
......@@ -4054,7 +4052,7 @@ void cv::cvtColor( InputArray _src, OutputArray _dst, int code, int dcn )
_dst.create(sz, CV_MAKETYPE(depth, dcn));
dst = _dst.getMat();
#if defined(HAVE_IPP) && 0
#if defined HAVE_IPP && 0
if( code == CV_Lab2LBGR && dcn == 3 && depth == CV_8U)
{
if( CvtColorIPPLoop(src, dst, IPPGeneralFunctor((ippiGeneralFunc)ippiLabToBGR_8u_C3R)) )
......
......@@ -526,7 +526,7 @@ void cv::cornerHarris( InputArray _src, OutputArray _dst, int blockSize, int ksi
_dst.create( src.size(), CV_32FC1 );
Mat dst = _dst.getMat();
#if IPP_VERSION_X100 >= 801
#if IPP_VERSION_X100 >= 801 && 0
int type = src.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
int borderTypeNI = borderType & ~BORDER_ISOLATED;
bool isolated = (borderType & BORDER_ISOLATED) != 0;
......@@ -563,10 +563,8 @@ void cv::cornerHarris( InputArray _src, OutputArray _dst, int blockSize, int ksi
if (status >= 0)
return;
setIppErrorStatus();
}
else
setIppErrorStatus();
setIppErrorStatus();
}
#endif
......
......@@ -2085,10 +2085,8 @@ static bool ocl_resize( InputArray _src, OutputArray _dst, Size dsize,
cn);
k.create("resizeSampler", ocl::imgproc::resize_oclsrc, compileOpts);
if(k.empty())
{
if (k.empty())
useSampler = false;
}
else
{
// Convert the input into an OpenCL image type, using normalized channel data types
......@@ -2397,6 +2395,12 @@ void cv::resize( InputArray _src, OutputArray _dst, Size dsize,
double scale_x = 1./inv_scale_x, scale_y = 1./inv_scale_y;
int k, sx, sy, dx, dy;
int iscale_x = saturate_cast<int>(scale_x);
int iscale_y = saturate_cast<int>(scale_y);
bool is_area_fast = std::abs(scale_x - iscale_x) < DBL_EPSILON &&
std::abs(scale_y - iscale_y) < DBL_EPSILON;
#if IPP_VERSION_X100 >= 701
#define IPP_RESIZE_EPS 1e-10
......@@ -2404,7 +2408,8 @@ void cv::resize( InputArray _src, OutputArray _dst, Size dsize,
double ey = fabs((double)dsize.height / src.rows - inv_scale_y) / inv_scale_y;
if ( ((ex < IPP_RESIZE_EPS && ey < IPP_RESIZE_EPS && depth != CV_64F) || (ex == 0 && ey == 0 && depth == CV_64F)) &&
(interpolation == INTER_LINEAR || interpolation == INTER_CUBIC))
(interpolation == INTER_LINEAR || interpolation == INTER_CUBIC) &&
!(interpolation == INTER_LINEAR && is_area_fast && iscale_x == 2 && iscale_y == 2 && depth == CV_8U))
{
int mode = -1;
if (interpolation == INTER_LINEAR && src.rows >= 2 && src.cols >= 2)
......@@ -2435,12 +2440,6 @@ void cv::resize( InputArray _src, OutputArray _dst, Size dsize,
}
{
int iscale_x = saturate_cast<int>(scale_x);
int iscale_y = saturate_cast<int>(scale_y);
bool is_area_fast = std::abs(scale_x - iscale_x) < DBL_EPSILON &&
std::abs(scale_y - iscale_y) < DBL_EPSILON;
// in case of scale_x && scale_y is equal to 2
// INTER_AREA (fast) also is equal to INTER_LINEAR
if( interpolation == INTER_LINEAR && is_area_fast && iscale_x == 2 && iscale_y == 2 )
......
......@@ -462,7 +462,7 @@ cv::Moments cv::moments( InputArray _src, bool binary )
if( cn > 1 )
CV_Error( CV_StsBadArg, "Invalid image type (must be single-channel)" );
#if IPP_VERSION_X100 >= 801
#if IPP_VERSION_X100 >= 801 && 0
if (!binary)
{
IppiSize roi = { mat.cols, mat.rows };
......
......@@ -508,7 +508,7 @@ void cv::pyrDown( InputArray _src, OutputArray _dst, const Size& _dsz, int borde
return;
#endif
#if IPP_VERSION_X100 >= 801
#if IPP_VERSION_X100 >= 801 && 0
bool isolated = (borderType & BORDER_ISOLATED) != 0;
int borderTypeNI = borderType & ~BORDER_ISOLATED;
if (borderTypeNI == BORDER_DEFAULT && (!src.isSubmatrix() || isolated) && dsz == Size((src.cols + 1)/2, (src.rows + 1)/2))
......@@ -577,7 +577,7 @@ void cv::pyrUp( InputArray _src, OutputArray _dst, const Size& _dsz, int borderT
return;
#endif
#if IPP_VERSION_X100 >= 801
#if IPP_VERSION_X100 >= 801 && 0
bool isolated = (borderType & BORDER_ISOLATED) != 0;
int borderTypeNI = borderType & ~BORDER_ISOLATED;
if (borderTypeNI == BORDER_DEFAULT && (!src.isSubmatrix() || isolated) && dsz == Size(src.cols*2, src.rows*2))
......@@ -648,7 +648,7 @@ void cv::buildPyramid( InputArray _src, OutputArrayOfArrays _dst, int maxlevel,
int i=1;
#if IPP_VERSION_X100 >= 801
#if IPP_VERSION_X100 >= 801 && 0
bool isolated = (borderType & BORDER_ISOLATED) != 0;
int borderTypeNI = borderType & ~BORDER_ISOLATED;
if (borderTypeNI == BORDER_DEFAULT && (!src.isSubmatrix() || isolated))
......@@ -668,17 +668,20 @@ void cv::buildPyramid( InputArray _src, OutputArrayOfArrays _dst, int maxlevel,
pyrInitAllocFunc = (ippiPyramidLayerDownInitAlloc) ippiPyramidLayerDownInitAlloc_8u_C1R;
pyrDownFunc = (ippiPyramidLayerDown) ippiPyramidLayerDown_8u_C1R;
pyrFreeFunc = (ippiPyramidLayerDownFree) ippiPyramidLayerDownFree_8u_C1R;
} else if (type == CV_8UC3)
}
else if (type == CV_8UC3)
{
pyrInitAllocFunc = (ippiPyramidLayerDownInitAlloc) ippiPyramidLayerDownInitAlloc_8u_C3R;
pyrDownFunc = (ippiPyramidLayerDown) ippiPyramidLayerDown_8u_C3R;
pyrFreeFunc = (ippiPyramidLayerDownFree) ippiPyramidLayerDownFree_8u_C3R;
} else if (type == CV_32FC1)
}
else if (type == CV_32FC1)
{
pyrInitAllocFunc = (ippiPyramidLayerDownInitAlloc) ippiPyramidLayerDownInitAlloc_32f_C1R;
pyrDownFunc = (ippiPyramidLayerDown) ippiPyramidLayerDown_32f_C1R;
pyrFreeFunc = (ippiPyramidLayerDownFree) ippiPyramidLayerDownFree_32f_C1R;
} else if (type == CV_32FC3)
}
else if (type == CV_32FC3)
{
pyrInitAllocFunc = (ippiPyramidLayerDownInitAlloc) ippiPyramidLayerDownInitAlloc_32f_C3R;
pyrDownFunc = (ippiPyramidLayerDown) ippiPyramidLayerDown_32f_C3R;
......@@ -722,10 +725,10 @@ void cv::buildPyramid( InputArray _src, OutputArrayOfArrays _dst, int maxlevel,
}
}
pyrFreeFunc(gPyr->pState);
} else
{
setIppErrorStatus();
}
else
setIppErrorStatus();
ippiPyramidFree(gPyr);
}
}
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册