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
2024-01-15 17:23:10 +03:00
parent 788c7252dc dad8af6b17
commit c739117a7c
531 changed files with 20900 additions and 12934 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/CMakeLists.txt
+24 -23
View File
@@ -1,27 +1,27 @@
set(the_description "The Core Functionality")
ocv_add_dispatched_file(mathfuncs_core SSE2 AVX AVX2)
ocv_add_dispatched_file(stat SSE4_2 AVX2)
ocv_add_dispatched_file(arithm SSE2 SSE4_1 AVX2 VSX3)
ocv_add_dispatched_file(convert SSE2 AVX2 VSX3)
ocv_add_dispatched_file(convert_scale SSE2 AVX2)
ocv_add_dispatched_file(count_non_zero SSE2 AVX2)
ocv_add_dispatched_file(has_non_zero SSE2 AVX2)
ocv_add_dispatched_file(matmul SSE2 SSE4_1 AVX2 AVX512_SKX NEON_DOTPROD)
ocv_add_dispatched_file(mean SSE2 AVX2)
ocv_add_dispatched_file(merge SSE2 AVX2)
ocv_add_dispatched_file(nan_mask SSE2 AVX2)
ocv_add_dispatched_file(split SSE2 AVX2)
ocv_add_dispatched_file(sum SSE2 AVX2)
ocv_add_dispatched_file(mathfuncs_core SSE2 AVX AVX2 LASX)
ocv_add_dispatched_file(stat SSE4_2 AVX2 LASX)
ocv_add_dispatched_file(arithm SSE2 SSE4_1 AVX2 VSX3 LASX)
ocv_add_dispatched_file(convert SSE2 AVX2 VSX3 LASX)
ocv_add_dispatched_file(convert_scale SSE2 AVX2 LASX)
ocv_add_dispatched_file(count_non_zero SSE2 AVX2 LASX)
ocv_add_dispatched_file(has_non_zero SSE2 AVX2 LASX )
ocv_add_dispatched_file(matmul SSE2 SSE4_1 AVX2 AVX512_SKX NEON_DOTPROD LASX)
ocv_add_dispatched_file(mean SSE2 AVX2 LASX)
ocv_add_dispatched_file(merge SSE2 AVX2 LASX)
ocv_add_dispatched_file(nan_mask SSE2 AVX2 LASX)
ocv_add_dispatched_file(split SSE2 AVX2 LASX)
ocv_add_dispatched_file(sum SSE2 AVX2 LASX)
# dispatching for accuracy tests
ocv_add_dispatched_file_force_all(test_intrin128 TEST SSE2 SSE3 SSSE3 SSE4_1 SSE4_2 AVX FP16 AVX2 AVX512_SKX)
ocv_add_dispatched_file_force_all(test_intrin256 TEST AVX2 AVX512_SKX)
ocv_add_dispatched_file_force_all(test_intrin256 TEST AVX2 AVX512_SKX LASX)
ocv_add_dispatched_file_force_all(test_intrin512 TEST AVX512_SKX)
set(PARALLEL_ENABLE_PLUGINS_DEFAULT ON)
if(EMSCRIPTEN OR IOS OR WINRT)
if(EMSCRIPTEN OR IOS OR XROS OR WINRT)
set(PARALLEL_ENABLE_PLUGINS_DEFAULT OFF)
endif()
# parallel backends configuration
@@ -50,13 +50,6 @@ if(DEFINED WINRT AND NOT DEFINED ENABLE_WINRT_MODE_NATIVE)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /ZW")
endif()
if(HAVE_CUDA)
if(NOT HAVE_opencv_cudev)
message(FATAL_ERROR "CUDA: OpenCV requires enabled 'cudev' module from 'opencv_contrib' repository: https://github.com/opencv/opencv_contrib")
endif()
ocv_warnings_disable(CMAKE_CXX_FLAGS -Wundef -Wenum-compare -Wunused-function -Wshadow)
endif()
if(CV_TRACE AND HAVE_ITT)
add_definitions(-DOPENCV_WITH_ITT=1)
endif()
@@ -154,7 +147,6 @@ elseif(HAVE_CXX11 OR DEFINED OPENCV_ALLOCATOR_STATS_COUNTER_TYPE)
endif()
endif()
if(PARALLEL_ENABLE_PLUGINS)
ocv_append_source_file_compile_definitions(${CMAKE_CURRENT_SOURCE_DIR}/src/parallel/parallel.cpp "PARALLEL_ENABLE_PLUGINS=1")
if(OPENCV_DEBUG_POSTFIX)
@@ -162,6 +154,15 @@ if(PARALLEL_ENABLE_PLUGINS)
endif()
endif()
if(HAVE_CUDA)
if(NOT HAVE_opencv_cudev)
message(FATAL_ERROR "CUDA: OpenCV requires enabled 'cudev' module from 'opencv_contrib' repository: https://github.com/opencv/opencv_contrib")
endif()
if(ENABLE_CUDA_FIRST_CLASS_LANGUAGE)
ocv_module_include_directories(${CUDAToolkit_INCLUDE_DIRS})
endif()
ocv_warnings_disable(CMAKE_CXX_FLAGS -Wundef -Wenum-compare -Wunused-function -Wshadow)
endif()
ocv_create_module(${extra_libs})
modules/core/include/opencv2/core.hpp
+4
View File
@@ -3201,6 +3201,10 @@ public:
* @overload
*/
CV_WRAP void write(FileStorage& fs, const String& name) const;
#if CV_VERSION_MAJOR < 5
/** @deprecated */
void write(const Ptr<FileStorage>& fs, const String& name = String()) const;
#endif
/** @brief Reads algorithm parameters from a file storage
*/
modules/core/include/opencv2/core/bindings_utils.hpp
+38 -33
View File
@@ -75,20 +75,6 @@ String dumpString(const String& argument)
return cv::format("String: %s", argument.c_str());
}
CV_WRAP static inline
String testOverloadResolution(int value, const Point& point = Point(42, 24))
{
return format("overload (int=%d, point=(x=%d, y=%d))", value, point.x,
point.y);
}
CV_WRAP static inline
String testOverloadResolution(const Rect& rect)
{
return format("overload (rect=(x=%d, y=%d, w=%d, h=%d))", rect.x, rect.y,
rect.width, rect.height);
}
CV_WRAP static inline
String dumpRect(const Rect& argument)
{
@@ -111,6 +97,42 @@ String dumpRotatedRect(const RotatedRect& argument)
argument.size.height, argument.angle);
}
CV_WRAP static inline
String dumpRange(const Range& argument)
{
if (argument == Range::all())
{
return "range: all";
}
else
{
return format("range: (s=%d, e=%d)", argument.start, argument.end);
}
}
CV_EXPORTS_W String dumpVectorOfInt(const std::vector<int>& vec);
CV_EXPORTS_W String dumpVectorOfDouble(const std::vector<double>& vec);
CV_EXPORTS_W String dumpVectorOfRect(const std::vector<Rect>& vec);
//! @cond IGNORED
CV_WRAP static inline
String testOverloadResolution(int value, const Point& point = Point(42, 24))
{
return format("overload (int=%d, point=(x=%d, y=%d))", value, point.x,
point.y);
}
CV_WRAP static inline
String testOverloadResolution(const Rect& rect)
{
return format("overload (rect=(x=%d, y=%d, w=%d, h=%d))", rect.x, rect.y,
rect.width, rect.height);
}
CV_WRAP static inline
RotatedRect testRotatedRect(float x, float y, float w, float h, float angle)
{
@@ -126,19 +148,6 @@ std::vector<RotatedRect> testRotatedRectVector(float x, float y, float w, float
return result;
}
CV_WRAP static inline
String dumpRange(const Range& argument)
{
if (argument == Range::all())
{
return "range: all";
}
else
{
return format("range: (s=%d, e=%d)", argument.start, argument.end);
}
}
CV_WRAP static inline
int testOverwriteNativeMethod(int argument)
{
@@ -151,12 +160,6 @@ String testReservedKeywordConversion(int positional_argument, int lambda = 2, in
return format("arg=%d, lambda=%d, from=%d", positional_argument, lambda, from);
}
CV_EXPORTS_W String dumpVectorOfInt(const std::vector<int>& vec);
CV_EXPORTS_W String dumpVectorOfDouble(const std::vector<double>& vec);
CV_EXPORTS_W String dumpVectorOfRect(const std::vector<Rect>& vec);
CV_WRAP static inline
void generateVectorOfRect(size_t len, CV_OUT std::vector<Rect>& vec)
{
@@ -323,6 +326,8 @@ private:
typedef OriginalClassName::Params OriginalClassName_Params;
} // namespace nested
//! @endcond IGNORED
namespace fs {
CV_EXPORTS_W cv::String getCacheDirectoryForDownloads();
} // namespace fs
modules/core/include/opencv2/core/cv_cpu_helper.h
+42
View File
@@ -441,6 +441,48 @@
#endif
#define __CV_CPU_DISPATCH_CHAIN_NEON_DOTPROD(fn, args, mode, ...) CV_CPU_CALL_NEON_DOTPROD(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
#if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_NEON_FP16
# define CV_TRY_NEON_FP16 1
# define CV_CPU_FORCE_NEON_FP16 1
# define CV_CPU_HAS_SUPPORT_NEON_FP16 1
# define CV_CPU_CALL_NEON_FP16(fn, args) return (cpu_baseline::fn args)
# define CV_CPU_CALL_NEON_FP16_(fn, args) return (opt_NEON_FP16::fn args)
#elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_NEON_FP16
# define CV_TRY_NEON_FP16 1
# define CV_CPU_FORCE_NEON_FP16 0
# define CV_CPU_HAS_SUPPORT_NEON_FP16 (cv::checkHardwareSupport(CV_CPU_NEON_FP16))
# define CV_CPU_CALL_NEON_FP16(fn, args) if (CV_CPU_HAS_SUPPORT_NEON_FP16) return (opt_NEON_FP16::fn args)
# define CV_CPU_CALL_NEON_FP16_(fn, args) if (CV_CPU_HAS_SUPPORT_NEON_FP16) return (opt_NEON_FP16::fn args)
#else
# define CV_TRY_NEON_FP16 0
# define CV_CPU_FORCE_NEON_FP16 0
# define CV_CPU_HAS_SUPPORT_NEON_FP16 0
# define CV_CPU_CALL_NEON_FP16(fn, args)
# define CV_CPU_CALL_NEON_FP16_(fn, args)
#endif
#define __CV_CPU_DISPATCH_CHAIN_NEON_FP16(fn, args, mode, ...) CV_CPU_CALL_NEON_FP16(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
#if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_NEON_BF16
# define CV_TRY_NEON_BF16 1
# define CV_CPU_FORCE_NEON_BF16 1
# define CV_CPU_HAS_SUPPORT_NEON_BF16 1
# define CV_CPU_CALL_NEON_BF16(fn, args) return (cpu_baseline::fn args)
# define CV_CPU_CALL_NEON_BF16_(fn, args) return (opt_NEON_BF16::fn args)
#elif !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_DISPATCH_COMPILE_NEON_BF16
# define CV_TRY_NEON_BF16 1
# define CV_CPU_FORCE_NEON_BF16 0
# define CV_CPU_HAS_SUPPORT_NEON_BF16 (cv::checkHardwareSupport(CV_CPU_NEON_BF16))
# define CV_CPU_CALL_NEON_BF16(fn, args) if (CV_CPU_HAS_SUPPORT_NEON_BF16) return (opt_NEON_BF16::fn args)
# define CV_CPU_CALL_NEON_BF16_(fn, args) if (CV_CPU_HAS_SUPPORT_NEON_BF16) return (opt_NEON_BF16::fn args)
#else
# define CV_TRY_NEON_BF16 0
# define CV_CPU_FORCE_NEON_BF16 0
# define CV_CPU_HAS_SUPPORT_NEON_BF16 0
# define CV_CPU_CALL_NEON_BF16(fn, args)
# define CV_CPU_CALL_NEON_BF16_(fn, args)
#endif
#define __CV_CPU_DISPATCH_CHAIN_NEON_BF16(fn, args, mode, ...) CV_CPU_CALL_NEON_BF16(fn, args); __CV_EXPAND(__CV_CPU_DISPATCH_CHAIN_ ## mode(fn, args, __VA_ARGS__))
#if !defined CV_DISABLE_OPTIMIZATION && defined CV_ENABLE_INTRINSICS && defined CV_CPU_COMPILE_MSA
# define CV_TRY_MSA 1
# define CV_CPU_FORCE_MSA 1
modules/core/include/opencv2/core/fast_math.hpp
+1 -1
View File
@@ -68,7 +68,7 @@
// nothing, intrinsics/asm code is not supported
#else
#if ((defined _MSC_VER && defined _M_X64) \
|| (defined __GNUC__ && defined __x86_64__ && defined __SSE2__)) \
|| (defined __GNUC__ && defined __SSE2__)) \
&& !defined(OPENCV_SKIP_INCLUDE_EMMINTRIN_H)
#include <emmintrin.h>
#endif
modules/core/include/opencv2/core/hal/intrin.hpp
-6
View File
@@ -246,12 +246,6 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
#include "opencv2/core/hal/intrin_lsx.hpp"
#elif CV_LASX
#if !defined(CV_FORCE_SIMD128_CPP)
#define CV_FORCE_SIMD128_CPP 1
#endif
#include "opencv2/core/hal/intrin_cpp.hpp"
#else
#include "opencv2/core/hal/intrin_cpp.hpp"
modules/core/include/opencv2/core/hal/intrin_lasx.hpp
+24 -23
View File
@@ -1419,20 +1419,6 @@ inline v_uint32x8 v_popcount(const v_int32x8& a)
inline v_uint64x4 v_popcount(const v_int64x4& a)
{ return v_popcount(v_reinterpret_as_u64(a)); }
/** Mask **/
#define OPENCV_HAL_IMPL_REINTERPRET_INT(ft, tt) \
inline tt reinterpret_int(ft x) { union { ft l; tt i; } v; v.l = x; return v.i; }
OPENCV_HAL_IMPL_REINTERPRET_INT(uchar, schar)
OPENCV_HAL_IMPL_REINTERPRET_INT(schar, schar)
OPENCV_HAL_IMPL_REINTERPRET_INT(ushort, short)
OPENCV_HAL_IMPL_REINTERPRET_INT(short, short)
OPENCV_HAL_IMPL_REINTERPRET_INT(unsigned, int)
OPENCV_HAL_IMPL_REINTERPRET_INT(int, int)
OPENCV_HAL_IMPL_REINTERPRET_INT(float, int)
OPENCV_HAL_IMPL_REINTERPRET_INT(uint64, int64)
OPENCV_HAL_IMPL_REINTERPRET_INT(int64, int64)
OPENCV_HAL_IMPL_REINTERPRET_INT(double, int64)
inline int v_signmask(const v_int8x32& a)
{
__m256i result = __lasx_xvmskltz_b(a.val);
@@ -2151,7 +2137,8 @@ template<int n> inline
void v_rshr_pack_store(uchar* ptr, const v_uint16x16& a)
{
__m256i res = __lasx_xvssrlrni_bu_h(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@@ -2165,7 +2152,8 @@ template<int n> inline
void v_rshr_pack_u_store(uchar* ptr, const v_int16x16& a)
{
__m256i res = __lasx_xvssrarni_bu_h(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@@ -2179,7 +2167,8 @@ template<int n> inline
void v_rshr_pack_store(schar* ptr, const v_int16x16& a)
{
__m256i res = __lasx_xvssrarni_b_h(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
// 32
@@ -2198,7 +2187,8 @@ inline void v_pack_store(short* ptr, const v_int32x8& a)
inline void v_pack_store(ushort* ptr, const v_uint32x8& a)
{
__m256i res = __lasx_xvssrlrni_hu_w(a.val, a.val, 0);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
inline void v_pack_u_store(ushort* ptr, const v_int32x8& a)
@@ -2212,7 +2202,8 @@ template<int n> inline
void v_rshr_pack_store(ushort* ptr, const v_uint32x8& a)
{
__m256i res = __lasx_xvssrlrni_hu_w(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@@ -2223,7 +2214,8 @@ template<int n> inline
void v_rshr_pack_u_store(ushort* ptr, const v_int32x8& a)
{
__m256i res = __lasx_xvssrarni_hu_w(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
@@ -2234,7 +2226,8 @@ template<int n> inline
void v_rshr_pack_store(short* ptr, const v_int32x8& a)
{
__m256i res = __lasx_xvssrarni_h_w(a.val, a.val, n);
__lsx_vst(_v256_extract_low(_v256_shuffle_odd_64(res)), ptr, 0);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
// 64
@@ -2263,7 +2256,11 @@ v_uint32x8 v_rshr_pack(const v_uint64x4& a, const v_uint64x4& b)
template<int n> inline
void v_rshr_pack_store(unsigned* ptr, const v_uint64x4& a)
{ __lsx_vst(_v256_shuffle_odd_64(__lasx_xvsrlrni_w_d(a.val, a.val, n)), ptr, 0); }
{
__m256i res = __lasx_xvsrlrni_w_d(a.val, a.val, n);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
template<int n> inline
v_int32x8 v_rshr_pack(const v_int64x4& a, const v_int64x4& b)
@@ -2271,7 +2268,11 @@ v_int32x8 v_rshr_pack(const v_int64x4& a, const v_int64x4& b)
template<int n> inline
void v_rshr_pack_store(int* ptr, const v_int64x4& a)
{ __lsx_vst(_v256_shuffle_odd_64(__lasx_xvsrarni_w_d(a.val, a.val, n)), ptr, 0); }
{
__m256i res = __lasx_xvsrarni_w_d(a.val, a.val, n);
__lasx_xvstelm_d(res, ptr, 0, 0);
__lasx_xvstelm_d(res, ptr, 8, 2);
}
// pack boolean
inline v_uint8x32 v_pack_b(const v_uint16x16& a, const v_uint16x16& b)
modules/core/include/opencv2/core/hal/intrin_neon.hpp
+3 -5
View File
@@ -2007,11 +2007,9 @@ inline v_int32x4 v_round(const v_float32x4& a)
#else
inline v_int32x4 v_round(const v_float32x4& a)
{
static const int32x4_t v_sign = vdupq_n_s32(1 << 31),
v_05 = vreinterpretq_s32_f32(vdupq_n_f32(0.5f));
int32x4_t v_addition = vorrq_s32(v_05, vandq_s32(v_sign, vreinterpretq_s32_f32(a.val)));
return v_int32x4(vcvtq_s32_f32(vaddq_f32(a.val, vreinterpretq_f32_s32(v_addition))));
// See https://github.com/opencv/opencv/pull/24271#issuecomment-1867318007
float32x4_t delta = vdupq_n_f32(12582912.0f);
return v_int32x4(vcvtq_s32_f32(vsubq_f32(vaddq_f32(a.val, delta), delta)));
}
#endif
inline v_int32x4 v_floor(const v_float32x4& a)
modules/core/include/opencv2/core/hal/intrin_rvv.hpp
+4 -1
View File
@@ -32,6 +32,8 @@
namespace cv
{
//! @cond IGNORED
CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
#define CV_SIMD128 1
@@ -3336,7 +3338,8 @@ inline void v_cleanup() {}
CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
//! @endcond
}
} // namespace cv
#endif
modules/core/include/opencv2/core/hal/intrin_rvv071.hpp
+1 -1
View File
@@ -2536,5 +2536,5 @@ CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
//! @endcond
}
} // namespace cv
#endif
modules/core/include/opencv2/core/hal/intrin_rvv_scalable.hpp
+61 -30
View File
@@ -38,6 +38,9 @@
namespace cv
{
//! @cond IGNORED
CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
#define CV_SIMD_SCALABLE 1
@@ -445,29 +448,7 @@ OPENCV_HAL_IMPL_RVV_LOADSTORE_OP(v_float64, vfloat64m1_t, double, VTraits<v_floa
#define OPENCV_HAL_IMPL_RVV_LUT(_Tpvec, _Tp, suffix) \
inline _Tpvec v_lut(const _Tp* tab, const int* idx) \
{ \
vuint32##suffix##_t vidx = vmul(vreinterpret_u32##suffix(vle32_v_i32##suffix(idx, VTraits<_Tpvec>::vlanes())), sizeof(_Tp), VTraits<_Tpvec>::vlanes()); \
return vloxei32(tab, vidx, VTraits<_Tpvec>::vlanes()); \
} \
inline _Tpvec v_lut_pairs(const _Tp* tab, const int* idx) \
{ \
std::vector<uint> idx_; \
for (int i = 0; i < VTraits<v_int16>::vlanes(); ++i) { \
idx_.push_back(idx[i]); \
idx_.push_back(idx[i]+1); \
} \
vuint32##suffix##_t vidx = vmul(vle32_v_u32##suffix(idx_.data(), VTraits<_Tpvec>::vlanes()), sizeof(_Tp), VTraits<_Tpvec>::vlanes()); \
return vloxei32(tab, vidx, VTraits<_Tpvec>::vlanes()); \
} \
inline _Tpvec v_lut_quads(const _Tp* tab, const int* idx) \
{ \
std::vector<uint> idx_; \
for (int i = 0; i < VTraits<v_int32>::vlanes(); ++i) { \
idx_.push_back(idx[i]); \
idx_.push_back(idx[i]+1); \
idx_.push_back(idx[i]+2); \
idx_.push_back(idx[i]+3); \
} \
vuint32##suffix##_t vidx = vmul(vle32_v_u32##suffix(idx_.data(), VTraits<_Tpvec>::vlanes()), sizeof(_Tp), VTraits<_Tpvec>::vlanes()); \
auto vidx = vmul(vreinterpret_u32##suffix(vle32_v_i32##suffix(idx, VTraits<_Tpvec>::vlanes())), sizeof(_Tp), VTraits<_Tpvec>::vlanes()); \
return vloxei32(tab, vidx, VTraits<_Tpvec>::vlanes()); \
}
OPENCV_HAL_IMPL_RVV_LUT(v_int8, schar, m4)
@@ -479,6 +460,55 @@ OPENCV_HAL_IMPL_RVV_LUT(v_float32, float, m1)
OPENCV_HAL_IMPL_RVV_LUT(v_float64, double, mf2)
#endif
#define OPENCV_HAL_IMPL_RVV_LUT_PAIRS(_Tpvec, _Tp, suffix1, suffix2, v_trunc) \
inline _Tpvec v_lut_pairs(const _Tp* tab, const int* idx) \
{ \
auto v0 = vle32_v_u32##suffix1((unsigned*)idx, VTraits<_Tpvec>::vlanes()/2); \
auto v1 = vadd(v0, 1, VTraits<_Tpvec>::vlanes()/2); \
auto w0 = vwcvtu_x(v0, VTraits<_Tpvec>::vlanes()/2); \
auto w1 = vwcvtu_x(v1, VTraits<_Tpvec>::vlanes()/2); \
auto sh1 = vslide1up(v_trunc(vreinterpret_u32##suffix2(w1)),0, VTraits<_Tpvec>::vlanes()); \
auto vid = vor(sh1, v_trunc(vreinterpret_u32##suffix2(w0)), VTraits<_Tpvec>::vlanes()); \
auto vidx = vmul(vid, sizeof(_Tp), VTraits<_Tpvec>::vlanes()); \
return vloxei32(tab, vidx, VTraits<_Tpvec>::vlanes()); \
}
OPENCV_HAL_IMPL_RVV_LUT_PAIRS(v_int8, schar, m2, m4, OPENCV_HAL_NOP)
OPENCV_HAL_IMPL_RVV_LUT_PAIRS(v_int16, short, m1, m2, OPENCV_HAL_NOP)
OPENCV_HAL_IMPL_RVV_LUT_PAIRS(v_int32, int, mf2, m1, OPENCV_HAL_NOP)
OPENCV_HAL_IMPL_RVV_LUT_PAIRS(v_float32, float, mf2, m1, OPENCV_HAL_NOP)
OPENCV_HAL_IMPL_RVV_LUT_PAIRS(v_int64, int64_t, mf2, m1, vlmul_trunc_u32mf2)
#if CV_SIMD_SCALABLE_64F
OPENCV_HAL_IMPL_RVV_LUT_PAIRS(v_float64, double, mf2, m1, vlmul_trunc_u32mf2)
#endif
#define OPENCV_HAL_IMPL_RVV_LUT_QUADS(_Tpvec, _Tp, suffix0, suffix1, suffix2, v_trunc) \
inline _Tpvec v_lut_quads(const _Tp* tab, const int* idx) \
{ \
auto v0 = vle32_v_u32##suffix0((unsigned*)idx, VTraits<_Tpvec>::vlanes()/4); \
auto v1 = vadd(v0, 1, VTraits<_Tpvec>::vlanes()/4); \
auto v2 = vadd(v0, 2, VTraits<_Tpvec>::vlanes()/4); \
auto v3 = vadd(v0, 3, VTraits<_Tpvec>::vlanes()/4); \
auto w0 = vwcvtu_x(v0, VTraits<_Tpvec>::vlanes()/4); \
auto w1 = vwcvtu_x(v1, VTraits<_Tpvec>::vlanes()/4); \
auto w2 = vwcvtu_x(v2, VTraits<_Tpvec>::vlanes()/4); \
auto w3 = vwcvtu_x(v3, VTraits<_Tpvec>::vlanes()/4); \
auto sh2 = vslide1up(vreinterpret_u32##suffix1(w2),0, VTraits<_Tpvec>::vlanes()/2); \
auto sh3 = vslide1up(vreinterpret_u32##suffix1(w3),0, VTraits<_Tpvec>::vlanes()/2); \
auto vid0 = vor(sh2, vreinterpret_u32##suffix1(w0), VTraits<_Tpvec>::vlanes()/2); \
auto vid1 = vor(sh3, vreinterpret_u32##suffix1(w1), VTraits<_Tpvec>::vlanes()/2); \
auto wid0 = vwcvtu_x(v_trunc(vid0), VTraits<_Tpvec>::vlanes()/2); \
auto wid1 = vwcvtu_x(v_trunc(vid1), VTraits<_Tpvec>::vlanes()/2); \
auto shwid1 = vslide1up(vreinterpret_u32##suffix2(wid1),0, VTraits<_Tpvec>::vlanes()); \
auto vid = vor(shwid1, vreinterpret_u32##suffix2(wid0), VTraits<_Tpvec>::vlanes()); \
auto vidx = vmul(vid, sizeof(_Tp), VTraits<_Tpvec>::vlanes()); \
return vloxei32(tab, vidx, VTraits<_Tpvec>::vlanes()); \
}
OPENCV_HAL_IMPL_RVV_LUT_QUADS(v_int8, schar, m1, m2, m4, OPENCV_HAL_NOP)
OPENCV_HAL_IMPL_RVV_LUT_QUADS(v_int16, short, mf2 , m1, m2, OPENCV_HAL_NOP)
OPENCV_HAL_IMPL_RVV_LUT_QUADS(v_int32, int, mf2, m1, m1, vlmul_trunc_u32mf2)
OPENCV_HAL_IMPL_RVV_LUT_QUADS(v_float32, float, mf2, m1, m1, vlmul_trunc_u32mf2)
#define OPENCV_HAL_IMPL_RVV_LUT_VEC(_Tpvec, _Tp) \
inline _Tpvec v_lut(const _Tp* tab, const v_int32& vidx) \
{ \
@@ -509,7 +539,6 @@ inline v_uint32 v_lut_pairs(const unsigned* tab, const int* idx) { return v_rein
inline v_uint32 v_lut_quads(const unsigned* tab, const int* idx) { return v_reinterpret_as_u32(v_lut_quads((int*)tab, idx)); }
inline v_uint64 v_lut(const uint64* tab, const int* idx) { return v_reinterpret_as_u64(v_lut((const int64_t *)tab, idx)); }
inline v_uint64 v_lut_pairs(const uint64* tab, const int* idx) { return v_reinterpret_as_u64(v_lut_pairs((const int64_t *)tab, idx)); }
inline v_uint64 v_lut_quads(const uint64* tab, const int* idx) { return v_reinterpret_as_u64(v_lut_quads((const int64_t*)tab, idx)); }
////////////// Pack boolean ////////////////////
inline v_uint8 v_pack_b(const v_uint16& a, const v_uint16& b)
@@ -1390,23 +1419,23 @@ OPENCV_HAL_IMPL_RVV_REVERSE(v_float64, 64)
#define OPENCV_HAL_IMPL_RVV_EXPAND(_Tp, _Tpwvec, _Tpwvec_m2, _Tpvec, width, suffix, suffix2, cvt) \
inline void v_expand(const _Tpvec& a, _Tpwvec& b0, _Tpwvec& b1) \
{ \
_Tpwvec_m2 temp = cvt(a, vsetvlmax_e##width##m1()); \
_Tpwvec_m2 temp = cvt(a, VTraits<_Tpvec>::vlanes()); \
b0 = vget_##suffix##m1(temp, 0); \
b1 = vget_##suffix##m1(temp, 1); \
} \
inline _Tpwvec v_expand_low(const _Tpvec& a) \
{ \
_Tpwvec_m2 temp = cvt(a, vsetvlmax_e##width##m1()); \
_Tpwvec_m2 temp = cvt(a, VTraits<_Tpvec>::vlanes()); \
return vget_##suffix##m1(temp, 0); \
} \
inline _Tpwvec v_expand_high(const _Tpvec& a) \
{ \
_Tpwvec_m2 temp = cvt(a, vsetvlmax_e##width##m1()); \
_Tpwvec_m2 temp = cvt(a, VTraits<_Tpvec>::vlanes()); \
return vget_##suffix##m1(temp, 1); \
} \
inline _Tpwvec v_load_expand(const _Tp* ptr) \
{ \
return cvt(vle##width##_v_##suffix2##mf2(ptr, vsetvlmax_e##width##m1()), vsetvlmax_e##width##m1()); \
return cvt(vle##width##_v_##suffix2##mf2(ptr, VTraits<_Tpvec>::vlanes()), VTraits<_Tpvec>::vlanes()); \
}
OPENCV_HAL_IMPL_RVV_EXPAND(uchar, v_uint16, vuint16m2_t, v_uint8, 8, u16, u8, vwcvtu_x)
@@ -1759,8 +1788,8 @@ inline int v_scan_forward(const v_float64& a)
// mask: {0,0,0,1, ...} -> {T,T,T,F, ...}
#define OPENCV_HAL_IMPL_RVV_PACK_TRIPLETS(_Tpvec, v_trunc) \
inline _Tpvec v_pack_triplets(const _Tpvec& vec) { \
size_t vl = vsetvlmax_e8m1(); \
vuint32m1_t one = vmv_v_x_u32m1(1, vl/4); \
size_t vl = __cv_rvv_e8m1_nlanes; \
vuint32m1_t one = vmv_v_x_u32m1(1, __cv_rvv_e32m1_nlanes); \
vuint8m1_t zero = vmv_v_x_u8m1(0, vl); \
vuint8m1_t mask = vreinterpret_u8m1(one); \
return vcompress(vmseq(v_trunc(vslideup(zero, mask, 3, vl)), 0, vl), vec, vec, VTraits<_Tpvec>::vlanes()); \
@@ -2126,6 +2155,8 @@ inline void v_cleanup() {}
CV_CPU_OPTIMIZATION_HAL_NAMESPACE_END
//! @endcond
} //namespace cv
#endif //OPENCV_HAL_INTRIN_RVV_SCALABLE_HPP
modules/core/include/opencv2/core/matx.hpp
+1 -1
View File
@@ -781,7 +781,7 @@ inline Matx<_Tp, m, n>::operator Matx<T2, m, n>() const
template<typename _Tp, int m, int n> template<int m1, int n1> inline
Matx<_Tp, m1, n1> Matx<_Tp, m, n>::reshape() const
{
CV_StaticAssert(m1*n1 == m*n, "Input and destnarion matrices must have the same number of elements");
CV_StaticAssert(m1*n1 == m*n, "Input and destination matrices must have the same number of elements");
return (const Matx<_Tp, m1, n1>&)*this;
}
modules/core/perf/perf_arithm.cpp
+1 -2
View File
@@ -1,6 +1,6 @@
#include "perf_precomp.hpp"
#include "opencv2/core/softfloat.hpp"
#include <numeric>
#include "opencv2/core/softfloat.hpp"
namespace opencv_test
{
@@ -452,7 +452,6 @@ INSTANTIATE_TEST_CASE_P(/*nothing*/ , BinaryOpTest,
)
);
///////////// PatchNaNs ////////////////////////
template<typename _Tp>
modules/core/src/arithm.cpp
+1 -1
View File
@@ -1646,7 +1646,7 @@ static bool ocl_inRange( InputArray _src, InputArray _lowerb,
if (kercn % cn != 0)
kercn = cn;
int colsPerWI = kercn / cn;
String opts = format("%s-D cn=%d -D srcT=%s -D srcT1=%s -D dstT=%s -D kercn=%d -D depth=%d%s -D colsPerWI=%d",
String opts = format("%s-D CN=%d -D SRC_T=%s -D SRC_T1=%s -D DST_T=%s -D KERCN=%d -D DEPTH=%d%s -D COLS_PER_WI=%d",
haveScalar ? "-D HAVE_SCALAR " : "", cn, ocl::typeToStr(CV_MAKE_TYPE(sdepth, kercn)),
ocl::typeToStr(sdepth), ocl::typeToStr(CV_8UC(colsPerWI)), kercn, sdepth,
doubleSupport ? " -D DOUBLE_SUPPORT" : "", colsPerWI);
modules/core/src/arithm.simd.hpp
+1 -1
View File
@@ -861,7 +861,7 @@ static void cmp_loop_nosimd(const double* src1, size_t step1, const double* src2
}
// todo: try to avoid define dispatcher functions using macros with these such cases
DEFINE_SIMD_ALL(cmp)
DEFINE_SIMD_ALL(cmp, void)
//=========================================================================
// scaling helpers for single and dual source
modules/core/src/opencl/inrange.cl
+24 -24
View File
@@ -52,7 +52,7 @@
__kernel void inrange(__global const uchar * src1ptr, int src1_step, int src1_offset,
__global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
#ifdef HAVE_SCALAR
__global const srcT1 * src2, __global const srcT1 * src3,
__global const SRC_T1 * src2, __global const SRC_T1 * src3,
#else
__global const uchar * src2ptr, int src2_step, int src2_offset,
__global const uchar * src3ptr, int src3_step, int src3_offset,
@@ -64,56 +64,56 @@ __kernel void inrange(__global const uchar * src1ptr, int src1_step, int src1_of
if (x < dst_cols)
{
int src1_index = mad24(y0, src1_step, mad24(x, (int)sizeof(srcT1) * kercn, src1_offset));
int dst_index = mad24(y0, dst_step, mad24(x, colsPerWI, dst_offset));
int src1_index = mad24(y0, src1_step, mad24(x, (int)sizeof(SRC_T1) * KERCN, src1_offset));
int dst_index = mad24(y0, dst_step, mad24(x, COLS_PER_WI, dst_offset));
#ifndef HAVE_SCALAR
int src2_index = mad24(y0, src2_step, mad24(x, (int)sizeof(srcT1) * kercn, src2_offset));
int src3_index = mad24(y0, src3_step, mad24(x, (int)sizeof(srcT1) * kercn, src3_offset));
int src2_index = mad24(y0, src2_step, mad24(x, (int)sizeof(SRC_T1) * KERCN, src2_offset));
int src3_index = mad24(y0, src3_step, mad24(x, (int)sizeof(SRC_T1) * KERCN, src3_offset));
#endif
for (int y = y0, y1 = min(dst_rows, y0 + rowsPerWI); y < y1; ++y, src1_index += src1_step, dst_index += dst_step)
{
#if kercn >= cn && kercn == 4 && depth <= 4 && !defined HAVE_SCALAR
srcT src1 = *(__global const srcT *)(src1ptr + src1_index);
srcT src2 = *(__global const srcT *)(src2ptr + src2_index);
srcT src3 = *(__global const srcT *)(src3ptr + src3_index);
__global dstT * dst = (__global dstT *)(dstptr + dst_index);
#if cn == 1
dst[0] = src2 > src1 || src3 < src1 ? (dstT)(0) : (dstT)(255);
#elif cn == 2
dst[0] = (dstT)(src2.xy > src1.xy || src3.xy < src1.xy ||
src2.zw > src1.zw || src3.zw < src1.zw ? (dstT)(0) : (dstT)(255);
#elif cn == 4
dst[0] = (dstT)(src2.x > src1.x || src3.x < src1.x ||
#if KERCN >= CN && KERCN == 4 && DEPTH <= 4 && !defined HAVE_SCALAR
SRC_T src1 = *(__global const SRC_T *)(src1ptr + src1_index);
SRC_T src2 = *(__global const SRC_T *)(src2ptr + src2_index);
SRC_T src3 = *(__global const SRC_T *)(src3ptr + src3_index);
__global DST_T * dst = (__global DST_T *)(dstptr + dst_index);
#if CN == 1
dst[0] = src2 > src1 || src3 < src1 ? (DST_T)(0) : (DST_T)(255);
#elif CN == 2
dst[0] = (DST_T)(src2.xy > src1.xy || src3.xy < src1.xy ||
src2.zw > src1.zw || src3.zw < src1.zw ? (DST_T)(0) : (DST_T)(255);
#elif CN == 4
dst[0] = (DST_T)(src2.x > src1.x || src3.x < src1.x ||
src2.y > src1.y || src3.y < src1.y ||
src2.z > src1.z || src3.z < src1.z ||
src2.w > src1.w || src3.w < src1.w ? 0 : 255);
#endif
#else
__global const srcT1 * src1 = (__global const srcT1 *)(src1ptr + src1_index);
__global const SRC_T1 * src1 = (__global const SRC_T1 *)(src1ptr + src1_index);
__global uchar * dst = dstptr + dst_index;
#ifndef HAVE_SCALAR
__global const srcT1 * src2 = (__global const srcT1 *)(src2ptr + src2_index);
__global const srcT1 * src3 = (__global const srcT1 *)(src3ptr + src3_index);
__global const SRC_T1 * src2 = (__global const SRC_T1 *)(src2ptr + src2_index);
__global const SRC_T1 * src3 = (__global const SRC_T1 *)(src3ptr + src3_index);
#endif
#pragma unroll
for (int px = 0; px < colsPerWI; ++px, src1 += cn
for (int px = 0; px < COLS_PER_WI; ++px, src1 += CN
#ifndef HAVE_SCALAR
, src2 += cn, src3 += cn
, src2 += CN, src3 += CN
#endif
)
{
dst[px] = 255;
for (int c = 0; c < cn; ++c)
for (int c = 0; c < CN; ++c)
if (src2[c] > src1[c] || src3[c] < src1[c])
{
dst[px] = 0;
break;
}
}
#endif // kercn >= cn
#endif // KERCN >= CN
#ifndef HAVE_SCALAR
src2_index += src2_step;
src3_index += src3_step;
modules/core/src/system.cpp
+10 -5
View File
@@ -154,6 +154,12 @@ void* allocSingletonNewBuffer(size_t size) { return malloc(size); }
# endif
#endif
#if defined __loongarch64
#include "sys/auxv.h"
#define LA_HWCAP_LSX (1<<4)
#define LA_HWCAP_LASX (1<<5)
#endif
#if defined _WIN32 || defined WINCE
#ifndef _WIN32_WINNT // This is needed for the declaration of TryEnterCriticalSection in winbase.h with Visual Studio 2005 (and older?)
#define _WIN32_WINNT 0x0400 // http://msdn.microsoft.com/en-us/library/ms686857(VS.85).aspx
@@ -704,12 +710,11 @@ struct HWFeatures
have[CV_CPU_RVV] = true;
#endif
#if defined __loongarch_sx
have[CV_CPU_LSX] = true;
#endif
#if defined __loongarch64 && defined __linux__
int flag = (int)getauxval(AT_HWCAP);
#if defined __loongarch_asx
have[CV_CPU_LASX] = true;
have[CV_CPU_LSX] = (flag & LA_HWCAP_LSX) != 0;
have[CV_CPU_LASX] = (flag & LA_HWCAP_LASX) != 0;
#endif
bool skip_baseline_check = false;
modules/core/test/ocl/test_image2d.cpp
+1 -1
View File
@@ -83,7 +83,7 @@ TEST(Image2D, turnOffOpenCL)
}
else
std::cout << "CV_8UC1 is not supported for OpenCL images. Test skipped." << std::endl;
// reset state to the previous one
cv::ocl::setUseOpenCL(useOCL);
}
modules/core/test/test_operations.cpp
+50
View File
@@ -1571,4 +1571,54 @@ TEST(Core_Arithm, scalar_handling_19599) // https://github.com/opencv/opencv/is
EXPECT_EQ(1, c.rows);
}
// https://github.com/opencv/opencv/issues/24163
typedef tuple<perf::MatDepth,int,int,int> Arith_Regression24163Param;
typedef testing::TestWithParam<Arith_Regression24163Param> Core_Arith_Regression24163;
#if defined __riscv
TEST_P(Core_Arith_Regression24163, DISABLED_test_for_ties_to_even)
#else
TEST_P(Core_Arith_Regression24163, test_for_ties_to_even)
#endif
{
const int matDepth = get<0>(GetParam());
const int matHeight= get<1>(GetParam());
const int matWidth = 3; // Fixed
const int alpha = get<2>(GetParam());
const int beta = get<3>(GetParam());
// If alpha and/or beta are negative, and matDepth is unsigned, test is passed.
if( ( (alpha < 0) || (beta < 0) )
&&
( (matDepth != CV_8S) && (matDepth != CV_16S) && (matDepth != CV_32S) ) )
{
throw SkipTestException( cv::format("Test is skipped(matDepth is not signed, alpha = %d, beta = %d)", alpha, beta) );
}
const int matType = CV_MAKE_TYPE(matDepth, 1);
const Size matSize(matWidth, matHeight);
const Mat src1(matSize, matType, Scalar(alpha,alpha,alpha,alpha));
const Mat src2(matSize, matType, Scalar(beta, beta, beta, beta));
const Mat result = ( src1 + src2 ) / 2;
// Expected that default is FE_TONEAREST(Ties to Even).
const int mean = lrint( static_cast<double>(alpha + beta) / 2.0 );
const Mat expected(matSize, matType, Scalar(mean,mean,mean,mean));
// Compare result and extected.
ASSERT_EQ(expected.size(), result.size());
EXPECT_EQ(0, cvtest::norm(expected, result, NORM_INF)) <<
"result=" << std::endl << result << std::endl <<
"expected=" << std::endl << expected;
}
INSTANTIATE_TEST_CASE_P(/* */, Core_Arith_Regression24163,
testing::Combine(
testing::Values(perf::MatDepth(CV_8U), CV_8S, CV_16U, CV_16S, CV_32S), // MatType
testing::Values( 3, 4, 5, 6), // MatHeight
testing::Values(-2,-1, 0, 1, 2), // src1
testing::Values( -1, 0, 1 ) // src2
)
);
}} // namespace