1
0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 07:43:03 +04:00

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

This commit is contained in:
Alexander Alekhin
2020-12-04 18:25:32 +00:00
51 changed files with 2880 additions and 112 deletions
+2 -11
View File
@@ -2397,20 +2397,11 @@ public:
UMat(const UMat& m, const Rect& roi);
UMat(const UMat& m, const Range* ranges);
UMat(const UMat& m, const std::vector<Range>& ranges);
// FIXIT copyData=false is not implemented, drop this in favor of cv::Mat (OpenCV 5.0)
//! builds matrix from std::vector with or without copying the data
template<typename _Tp> explicit UMat(const std::vector<_Tp>& vec, bool copyData=false);
//! builds matrix from cv::Vec; the data is copied by default
template<typename _Tp, int n> explicit UMat(const Vec<_Tp, n>& vec, bool copyData=true);
//! builds matrix from cv::Matx; the data is copied by default
template<typename _Tp, int m, int n> explicit UMat(const Matx<_Tp, m, n>& mtx, bool copyData=true);
//! builds matrix from a 2D point
template<typename _Tp> explicit UMat(const Point_<_Tp>& pt, bool copyData=true);
//! builds matrix from a 3D point
template<typename _Tp> explicit UMat(const Point3_<_Tp>& pt, bool copyData=true);
//! builds matrix from comma initializer
template<typename _Tp> explicit UMat(const MatCommaInitializer_<_Tp>& commaInitializer);
//! destructor - calls release()
~UMat();
//! assignment operators
+1 -3
View File
@@ -62,11 +62,9 @@ static bool ipp_countNonZero( Mat &src, int &res )
{
CV_INSTRUMENT_REGION_IPP();
#if defined __APPLE__ || (defined _MSC_VER && defined _M_IX86)
// see https://github.com/opencv/opencv/issues/17453
if (src.dims <= 2 && src.step > 520000)
if (src.dims <= 2 && src.step > 520000 && cv::ipp::getIppTopFeatures() == ippCPUID_SSE42)
return false;
#endif
#if IPP_VERSION_X100 < 201801
// Poor performance of SSE42
+2 -2
View File
@@ -152,10 +152,10 @@ float normL2Sqr_(const float* a, const float* b, int n)
{
v_float32 t0 = vx_load(a + j) - vx_load(b + j);
v_float32 t1 = vx_load(a + j + v_float32::nlanes) - vx_load(b + j + v_float32::nlanes);
v_float32 t2 = vx_load(a + j + 2 * v_float32::nlanes) - vx_load(b + j + 2 * v_float32::nlanes);
v_float32 t3 = vx_load(a + j + 3 * v_float32::nlanes) - vx_load(b + j + 3 * v_float32::nlanes);
v_d0 = v_muladd(t0, t0, v_d0);
v_float32 t2 = vx_load(a + j + 2 * v_float32::nlanes) - vx_load(b + j + 2 * v_float32::nlanes);
v_d1 = v_muladd(t1, t1, v_d1);
v_float32 t3 = vx_load(a + j + 3 * v_float32::nlanes) - vx_load(b + j + 3 * v_float32::nlanes);
v_d2 = v_muladd(t2, t2, v_d2);
v_d3 = v_muladd(t3, t3, v_d3);
}
+9 -6
View File
@@ -2365,6 +2365,13 @@ public:
ippTopFeatures = ippCPUID_SSE42;
pIppLibInfo = ippiGetLibVersion();
// workaround: https://github.com/opencv/opencv/issues/12959
std::string ippName(pIppLibInfo->Name ? pIppLibInfo->Name : "");
if (ippName.find("SSE4.2") != std::string::npos)
{
ippTopFeatures = ippCPUID_SSE42;
}
}
public:
@@ -2396,16 +2403,12 @@ unsigned long long getIppFeatures()
#endif
}
unsigned long long getIppTopFeatures();
#ifdef HAVE_IPP
unsigned long long getIppTopFeatures()
{
#ifdef HAVE_IPP
return getIPPSingleton().ippTopFeatures;
#else
return 0;
#endif
}
#endif
void setIppStatus(int status, const char * const _funcname, const char * const _filename, int _line)
{