potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-31 08:13:04 +04:00
Code Packages Releases Wiki Activity

Merge branch 4.x

Browse Source
This commit is contained in:
Alexander Smorkalov
2026-02-11 13:54:39 +03:00
committed by Alexander Smorkalov
parent 1719aa1339 a6c10358a4
commit 82ff8e45e9
468 changed files with 16647 additions and 11284 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/src/arithm_ipp.hpp
+16
View File
@@ -283,24 +283,40 @@ inline IppCmpOp arithm_ipp_convert_cmp(int cmpop)
inline int arithm_ipp_cmp8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2,
uchar* dst, size_t step, int width, int height, int cmpop)
{
// perf regression with AVX512: https://github.com/opencv/opencv/issues/28251
if (ippCPUID_AVX512F&cv::ipp::getIppFeatures())
return 0;
ARITHM_IPP_CMP(ippiCompare_8u_C1R, src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(width, height));
}
inline int arithm_ipp_cmp16u(const ushort* src1, size_t step1, const ushort* src2, size_t step2,
uchar* dst, size_t step, int width, int height, int cmpop)
{
// perf regression with AVX512: https://github.com/opencv/opencv/issues/28251
if (ippCPUID_AVX512F&cv::ipp::getIppFeatures())
return 0;
ARITHM_IPP_CMP(ippiCompare_16u_C1R, src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(width, height));
}
inline int arithm_ipp_cmp16s(const short* src1, size_t step1, const short* src2, size_t step2,
uchar* dst, size_t step, int width, int height, int cmpop)
{
// perf regression with AVX512: https://github.com/opencv/opencv/issues/28251
if (ippCPUID_AVX512F&cv::ipp::getIppFeatures())
return 0;
ARITHM_IPP_CMP(ippiCompare_16s_C1R, src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(width, height));
}
inline int arithm_ipp_cmp32f(const float* src1, size_t step1, const float* src2, size_t step2,
uchar* dst, size_t step, int width, int height, int cmpop)
{
// perf regression with AVX512: https://github.com/opencv/opencv/issues/28251
if (ippCPUID_AVX512F&cv::ipp::getIppFeatures())
return 0;
ARITHM_IPP_CMP(ippiCompare_32f_C1R, src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(width, height));
}
modules/core/src/hal_internal.cpp
+14 -2
View File
@@ -69,10 +69,17 @@
#include <sanitizer/msan_interface.h>
#define CV_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) \
__msan_unpoison(address, size)
#define CV_ANNOTATE_NO_SANITIZE_MEMORY __attribute__((no_sanitize("memory")))
#endif
#endif
#ifndef CV_ANNOTATE_MEMORY_IS_INITIALIZED
#define CV_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) do { } while(0)
#define CV_ANNOTATE_NO_SANITIZE_MEMORY
#endif
#if defined(__APPLE__) && defined(__clang__)
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wdeprecated-declarations"
#endif
//lapack stores matrices in column-major order so transposing is needed everywhere
@@ -108,8 +115,9 @@ set_value(fptype *dst, size_t dst_ld, fptype value, size_t m, size_t n)
dst[i*dst_ld + j] = value;
}
// MSAN can't see that the fortran LAPACK functions initialize `info`
template <typename fptype> static inline int
lapack_LU(fptype* a, size_t a_step, int m, fptype* b, size_t b_step, int n, int* info)
CV_ANNOTATE_NO_SANITIZE_MEMORY lapack_LU(fptype* a, size_t a_step, int m, fptype* b, size_t b_step, int n, int* info)
{
#if defined (ACCELERATE_NEW_LAPACK) && defined (ACCELERATE_LAPACK_ILP64)
cv::AutoBuffer<long> piv_buff(m);
@@ -710,4 +718,8 @@ int lapack_gemm64fc(const double *src1, size_t src1_step, const double *src2, si
return lapack_gemm_c(src1, src1_step, src2, src2_step, alpha, src3, src3_step, beta, dst, dst_step, m, n, k, flags);
}
#endif //HAVE_LAPACK
#if defined(__APPLE__) && defined(__clang__)
#pragma clang diagnostic pop
#endif
#endif //HAVE_LAPACK
modules/core/src/mathfuncs.cpp
+24 -5
View File
@@ -46,6 +46,8 @@
#include <atomic>
#include <limits>
#include <iostream>
#include <algorithm>
#include <cmath>
#include "mathfuncs.hpp"
namespace cv
@@ -1423,12 +1425,21 @@ int cv::solveCubic( InputArray _coeffs, OutputArray _roots )
a3 = coeffs.at<double>(i+3);
}
if( a0 == 0 )
// Fix for Issue #27748: Normalize coefficients to avoid overflow/underflow
// and correctly detect negligible cubic terms.
double max_coeff = std::max({std::abs(a0), std::abs(a1), std::abs(a2), std::abs(a3)});
// If max_coeff is effectively zero, the equation is 0=0
if (max_coeff < std::numeric_limits<double>::epsilon())
return -1;
// Check if the cubic term is negligible relative to the largest coefficient
if( std::abs(a0) < std::numeric_limits<double>::epsilon() * max_coeff )
{
if( a1 == 0 )
if( std::abs(a1) < std::numeric_limits<double>::epsilon() * max_coeff )
{
if( a2 == 0 ) // constant
n = a3 == 0 ? -1 : 0;
if( std::abs(a2) < std::numeric_limits<double>::epsilon() * max_coeff )
n = std::abs(a3) < std::numeric_limits<double>::epsilon() * max_coeff ? -1 : 0;
else
{
// linear equation
@@ -1445,7 +1456,7 @@ int cv::solveCubic( InputArray _coeffs, OutputArray _roots )
d = std::sqrt(d);
double q1 = (-a2 + d) * 0.5;
double q2 = (a2 + d) * -0.5;
if( fabs(q1) > fabs(q2) )
if( std::abs(q1) > std::abs(q2) )
{
x0 = q1 / a1;
x1 = a3 / q1;
@@ -1462,6 +1473,13 @@ int cv::solveCubic( InputArray _coeffs, OutputArray _roots )
else
{
// cubic equation
// Normalize coefficients in-place to prevent overflow/instability
double scale = 1.0 / max_coeff;
a0 *= scale;
a1 *= scale;
a2 *= scale;
a3 *= scale;
a0 = 1./a0;
a1 *= a0;
a2 *= a0;
@@ -1470,6 +1488,7 @@ int cv::solveCubic( InputArray _coeffs, OutputArray _roots )
double Q = (a1 * a1 - 3 * a2) * (1./9);
double R = (a1 * (2 * a1 * a1 - 9 * a2) + 27 * a3) * (1./54);
double Qcubed = Q * Q * Q;
/*
Here we expand expression `Qcubed - R * R` for `d` variable
to reduce common terms `a1^6 / 729` and `-a1^4 * a2 / 81`
modules/core/src/opencl/runtime/opencl_core.cpp
+1 -1
View File
@@ -93,7 +93,7 @@ static void* AppleCLGetProcAddress(const char* name)
handle = dlopen(path.c_str(), RTLD_LAZY | RTLD_GLOBAL);
if (handle == NULL)
{
if (path != NULL && path != defaultPath)
if (!path.empty() && path != defaultPath)
fprintf(stderr, ERROR_MSG_CANT_LOAD);
}
else if (dlsym(handle, OPENCL_FUNC_TO_CHECK_1_1) == NULL)
modules/core/src/parallel.cpp
-4
View File
@@ -465,10 +465,6 @@ namespace {
static inline int _initMaxThreads()
{
int maxThreads = omp_get_max_threads();
if (!utils::getConfigurationParameterBool("OPENCV_FOR_OPENMP_DYNAMIC_DISABLE", false))
{
omp_set_dynamic(1);
}
return maxThreads;
}
static int numThreadsMax = _initMaxThreads();
modules/core/src/system.cpp
+3
View File
@@ -179,6 +179,7 @@ const uint64_t AT_HWCAP = NT_GNU_HWCAP;
#define _WIN32_WINNT 0x0400 // http://msdn.microsoft.com/en-us/library/ms686857(VS.85).aspx
#endif
#include <windows.h>
#include <combaseapi.h>
#if (_WIN32_WINNT >= 0x0602)
#include <synchapi.h>
#endif
@@ -478,6 +479,7 @@ struct HWFeatures
g_hwFeatureNames[CPU_NEON_DOTPROD] = "NEON_DOTPROD";
g_hwFeatureNames[CPU_NEON_FP16] = "NEON_FP16";
g_hwFeatureNames[CPU_NEON_BF16] = "NEON_BF16";
g_hwFeatureNames[CPU_SVE] = "SVE";
g_hwFeatureNames[CPU_VSX] = "VSX";
g_hwFeatureNames[CPU_VSX3] = "VSX3";
@@ -641,6 +643,7 @@ struct HWFeatures
{
have[CV_CPU_NEON_DOTPROD] = (auxv.a_un.a_val & (1 << 20)) != 0; // HWCAP_ASIMDDP
have[CV_CPU_NEON_FP16] = (auxv.a_un.a_val & (1 << 10)) != 0; // HWCAP_ASIMDHP
have[CV_CPU_SVE] = (auxv.a_un.a_val & (1 << 22)) != 0; // HWCAP_SVE
}
#if defined(AT_HWCAP2)
else if (auxv.a_type == AT_HWCAP2)