third_party/opencv
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #12877 from maver1:3.4

Browse Source 
* Updated ICV packages and IPP integration

* core(test): minMaxIdx IPP regression test

* core(ipp): workaround minMaxIdx problem

* core(ipp): workaround meanStdDev() CV_32FC3 buffer overrun

* Returned semicolon after CV_INSTRUMENT_REGION_IPP()
This commit is contained in:
maver1
2018-10-24 15:02:53 +03:00
committed by Alexander Alekhin
parent c8fc8d210b
commit e397434cb6
17 changed files with 201 additions and 297 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/base.hpp
+5 -1
View File
@@ -767,9 +767,13 @@ CV_EXPORTS_W void setUseIPP(bool flag);
CV_EXPORTS_W String getIppVersion();
// IPP Not-Exact mode. This function may force use of IPP then both IPP and OpenCV provide proper results
// but have internal accuracy differences which have to much direct or indirect impact on accuracy tests.
// but have internal accuracy differences which have too much direct or indirect impact on accuracy tests.
CV_EXPORTS_W bool useIPP_NotExact();
CV_EXPORTS_W void setUseIPP_NotExact(bool flag);
#if OPENCV_ABI_COMPATIBILITY < 400
CV_EXPORTS_W bool useIPP_NE();
CV_EXPORTS_W void setUseIPP_NE(bool flag);
#endif
} // ipp
modules/core/include/opencv2/core/private.hpp
+3 -1
View File
@@ -194,8 +194,8 @@ T* allocSingleton(size_t count = 1) { return static_cast<T*>(allocSingletonBuffe
#define IPP_DISABLE_LAB_RGB 1 // breaks OCL accuracy tests
#define IPP_DISABLE_RGB_XYZ 1 // big accuracy difference
#define IPP_DISABLE_XYZ_RGB 1 // big accuracy difference
#define IPP_DISABLE_HAAR 1 // improper integration/results
#define IPP_DISABLE_HOUGH 1 // improper integration/results
#define IPP_DISABLE_FILTER2D_BIG_MASK 1 // different results on masks > 7x7
#define IPP_DISABLE_GAUSSIANBLUR_PARALLEL 1 // not supported (2017u2 / 2017u3)
@@ -229,7 +229,9 @@ T* allocSingleton(size_t count = 1) { return static_cast<T*>(allocSingletonBuffe
# pragma GCC diagnostic ignored "-Wsuggest-override"
# endif
#include "iw++/iw.hpp"
# ifdef HAVE_IPP_IW_LL
#include "iw/iw_ll.h"
# endif
# if defined(__OPENCV_BUILD) && defined(__GNUC__) && __GNUC__ >= 5
# pragma GCC diagnostic pop
# endif
modules/core/src/channels.cpp
+2 -2
View File
@@ -341,7 +341,7 @@ namespace cv
{
static bool ipp_extractChannel(const Mat &src, Mat &dst, int channel)
{
#ifdef HAVE_IPP_IW
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
int srcChannels = src.channels();
@@ -379,7 +379,7 @@ static bool ipp_extractChannel(const Mat &src, Mat &dst, int channel)
static bool ipp_insertChannel(const Mat &src, Mat &dst, int channel)
{
#ifdef HAVE_IPP_IW
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
int srcChannels = src.channels();
modules/core/src/copy.cpp
+3 -3
View File
@@ -328,7 +328,7 @@ void Mat::copyTo( OutputArray _dst ) const
#ifdef HAVE_IPP
static bool ipp_copyTo(const Mat &src, Mat &dst, const Mat &mask)
{
#ifdef HAVE_IPP_IW
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
if(mask.channels() > 1 || mask.depth() != CV_8U)
@@ -463,7 +463,7 @@ Mat& Mat::operator = (const Scalar& s)
#ifdef HAVE_IPP
static bool ipp_Mat_setTo_Mat(Mat &dst, Mat &_val, Mat &mask)
{
#ifdef HAVE_IPP_IW
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
if(mask.empty())
@@ -1152,7 +1152,7 @@ namespace cv {
static bool ipp_copyMakeBorder( Mat &_src, Mat &_dst, int top, int bottom,
int left, int right, int _borderType, const Scalar& value )
{
#if defined HAVE_IPP_IW && !IPP_DISABLE_PERF_COPYMAKE
#if defined HAVE_IPP_IW_LL && !IPP_DISABLE_PERF_COPYMAKE
CV_INSTRUMENT_REGION_IPP();
::ipp::IwiBorderSize borderSize(left, top, right, bottom);
modules/core/src/mean.cpp
+5
View File
@@ -674,6 +674,11 @@ static bool ipp_meanStdDev(Mat& src, OutputArray _mean, OutputArray _sdv, Mat& m
if (cn > 1)
return false;
#endif
#if IPP_VERSION_X100 < 201901
// IPP_DISABLE: 32f C3C functions can read outside of allocated memory
if (cn > 1 && src.depth() == CV_32F)
return false;
#endif
size_t total_size = src.total();
int rows = src.size[0], cols = rows ? (int)(total_size/rows) : 0;
modules/core/src/merge.cpp
+1 -1
View File
@@ -228,7 +228,7 @@ static MergeFunc getMergeFunc(int depth)
namespace cv {
static bool ipp_merge(const Mat* mv, Mat& dst, int channels)
{
#ifdef HAVE_IPP_IW
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
if(channels != 3 && channels != 4)
modules/core/src/minmax.cpp
+6
View File
@@ -8,6 +8,8 @@
#include "opencv2/core/openvx/ovx_defs.hpp"
#include "stat.hpp"
#define IPP_DISABLE_MINMAXIDX_MANY_ROWS 1 // see Core_MinMaxIdx.rows_overflow test
/****************************************************************************************\
* minMaxLoc *
\****************************************************************************************/
@@ -624,6 +626,10 @@ static bool ipp_minMaxIdx(Mat &src, double* _minVal, double* _maxVal, int* _minI
if(src.dims <= 2)
{
IppiSize size = ippiSize(src.size());
#if defined(_WIN32) && !defined(_WIN64) && IPP_VERSION_X100 == 201900 && IPP_DISABLE_MINMAXIDX_MANY_ROWS
if (size.height > 65536)
return false; // test: Core_MinMaxIdx.rows_overflow
#endif
size.width *= src.channels();
status = ippMinMaxFun(src.ptr(), (int)src.step, size, dataType, pMinVal, pMaxVal, pMinIdx, pMaxIdx, (Ipp8u*)mask.ptr(), (int)mask.step);
modules/core/src/split.cpp
+1 -1
View File
@@ -236,7 +236,7 @@ static SplitFunc getSplitFunc(int depth)
namespace cv {
static bool ipp_split(const Mat& src, Mat* mv, int channels)
{
#ifdef HAVE_IPP_IW
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
if(channels != 3 && channels != 4)
modules/core/src/system.cpp
+21 -4
View File
@@ -2078,7 +2078,12 @@ public:
cv::String env = pIppEnv;
if(env.size())
{
#if IPP_VERSION_X100 >= 201703
#if IPP_VERSION_X100 >= 201900
const Ipp64u minorFeatures = ippCPUID_MOVBE|ippCPUID_AES|ippCPUID_CLMUL|ippCPUID_ABR|ippCPUID_RDRAND|ippCPUID_F16C|
ippCPUID_ADCOX|ippCPUID_RDSEED|ippCPUID_PREFETCHW|ippCPUID_SHA|ippCPUID_MPX|ippCPUID_AVX512CD|ippCPUID_AVX512ER|
ippCPUID_AVX512PF|ippCPUID_AVX512BW|ippCPUID_AVX512DQ|ippCPUID_AVX512VL|ippCPUID_AVX512VBMI|ippCPUID_AVX512_4FMADDPS|
ippCPUID_AVX512_4VNNIW|ippCPUID_AVX512IFMA;
#elif IPP_VERSION_X100 >= 201703
const Ipp64u minorFeatures = ippCPUID_MOVBE|ippCPUID_AES|ippCPUID_CLMUL|ippCPUID_ABR|ippCPUID_RDRAND|ippCPUID_F16C|
ippCPUID_ADCOX|ippCPUID_RDSEED|ippCPUID_PREFETCHW|ippCPUID_SHA|ippCPUID_MPX|ippCPUID_AVX512CD|ippCPUID_AVX512ER|
ippCPUID_AVX512PF|ippCPUID_AVX512BW|ippCPUID_AVX512DQ|ippCPUID_AVX512VL|ippCPUID_AVX512VBMI;
@@ -2279,7 +2284,7 @@ void setUseIPP(bool flag)
#endif
}
bool useIPP_NE()
bool useIPP_NotExact()
{
#ifdef HAVE_IPP
CoreTLSData* data = getCoreTlsData().get();
@@ -2293,17 +2298,29 @@ bool useIPP_NE()
#endif
}
void setUseIPP_NE(bool flag)
void setUseIPP_NotExact(bool flag)
{
CoreTLSData* data = getCoreTlsData().get();
#ifdef HAVE_IPP
data->useIPP_NE = (getIPPSingleton().useIPP_NE)?flag:false;
data->useIPP_NE = flag;
#else
CV_UNUSED(flag);
data->useIPP_NE = false;
#endif
}
#if OPENCV_ABI_COMPATIBILITY < 400
bool useIPP_NE()
{
return useIPP_NotExact();
}
void setUseIPP_NE(bool flag)
{
setUseIPP_NotExact(flag);
}
#endif
} // namespace ipp
} // namespace cv
modules/core/test/test_arithm.cpp
+26
View File
@@ -2298,4 +2298,30 @@ TEST(UMat_Core_DivideRules, type_32f) { testDivide<float, true>(); }
TEST(Core_DivideRules, type_64f) { testDivide<double, false>(); }
TEST(UMat_Core_DivideRules, type_64f) { testDivide<double, true>(); }
TEST(Core_MinMaxIdx, rows_overflow)
{
const int N = 65536 + 1;
const int M = 1;
{
setRNGSeed(123);
Mat m(N, M, CV_32FC1);
randu(m, -100, 100);
double minVal = 0, maxVal = 0;
int minIdx[CV_MAX_DIM] = { 0 }, maxIdx[CV_MAX_DIM] = { 0 };
cv::minMaxIdx(m, &minVal, &maxVal, minIdx, maxIdx);
double minVal0 = 0, maxVal0 = 0;
int minIdx0[CV_MAX_DIM] = { 0 }, maxIdx0[CV_MAX_DIM] = { 0 };
cv::ipp::setUseIPP(false);
cv::minMaxIdx(m, &minVal0, &maxVal0, minIdx0, maxIdx0);
cv::ipp::setUseIPP(true);
EXPECT_FALSE(fabs(minVal0 - minVal) > 1e-6 || fabs(maxVal0 - maxVal) > 1e-6) << "NxM=" << N << "x" << M <<
" min=" << minVal0 << " vs " << minVal <<
" max=" << maxVal0 << " vs " << maxVal;
}
}
}} // namespace