potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-31 00:03:03 +04:00
Code Packages Releases Wiki Activity

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

Browse Source
This commit is contained in:
Alexander Alekhin
2019-09-25 18:59:07 +03:00
committed by Alexander Alekhin
parent 9f4f9000bc 7ce9428e96
commit e2a5a6a05c
50 changed files with 5955 additions and 161 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/cv_cpu_dispatch.h
+9
View File
@@ -157,6 +157,11 @@
# define CV_MSA 1
#endif
#ifdef __EMSCRIPTEN__
# define CV_WASM_SIMD 1
# include <wasm_simd128.h>
#endif
#endif // CV_ENABLE_INTRINSICS && !CV_DISABLE_OPTIMIZATION && !__CUDACC__
#if defined CV_CPU_COMPILE_AVX && !defined CV_CPU_BASELINE_COMPILE_AVX
@@ -328,3 +333,7 @@ struct VZeroUpperGuard {
#ifndef CV_MSA
# define CV_MSA 0
#endif
#ifndef CV_WASM_SIMD
# define CV_WASM_SIMD 0
#endif
modules/core/include/opencv2/core/cvdef.h
+1
View File
@@ -695,6 +695,7 @@ __CV_ENUM_FLAGS_BITWISE_XOR_EQ (EnumType, EnumType)
#endif
#define CV_CXX_MOVE_SEMANTICS 1
#define CV_CXX_MOVE(x) std::move(x)
#define CV_CXX_STD_ARRAY 1
#include <array>
#ifndef CV_OVERRIDE
modules/core/include/opencv2/core/hal/intrin.hpp
+4 -1
View File
@@ -168,7 +168,7 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
# undef CV_MSA
#endif
#if CV_SSE2 || CV_NEON || CV_VSX || CV_MSA
#if CV_SSE2 || CV_NEON || CV_VSX || CV_MSA || CV_WASM_SIMD
#define CV__SIMD_FORWARD 128
#include "opencv2/core/hal/intrin_forward.hpp"
#endif
@@ -190,6 +190,9 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
#include "opencv2/core/hal/intrin_msa.hpp"
#elif CV_WASM_SIMD
#include "opencv2/core/hal/intrin_wasm.hpp"
#else
#define CV_SIMD128_CPP 1
modules/core/include/opencv2/core/hal/intrin_avx.hpp
+21 -33
View File
@@ -1241,6 +1241,11 @@ inline int v_signmask(const v_int32x8& a)
inline int v_signmask(const v_uint32x8& a)
{ return v_signmask(v_reinterpret_as_f32(a)); }
inline int v_signmask(const v_int64x4& a)
{ return v_signmask(v_reinterpret_as_f64(a)); }
inline int v_signmask(const v_uint64x4& a)
{ return v_signmask(v_reinterpret_as_f64(a)); }
inline int v_scan_forward(const v_int8x32& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))); }
inline int v_scan_forward(const v_uint8x32& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))); }
inline int v_scan_forward(const v_int16x16& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))) / 2; }
@@ -1253,40 +1258,23 @@ inline int v_scan_forward(const v_uint64x4& a) { return trailingZeros32(v_signma
inline int v_scan_forward(const v_float64x4& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))) / 8; }
/** Checks **/
#define OPENCV_HAL_IMPL_AVX_CHECK(_Tpvec, and_op, allmask) \
inline bool v_check_all(const _Tpvec& a) \
{ \
int mask = v_signmask(v_reinterpret_as_s8(a)); \
return and_op(mask, allmask) == allmask; \
} \
inline bool v_check_any(const _Tpvec& a) \
{ \
int mask = v_signmask(v_reinterpret_as_s8(a)); \
return and_op(mask, allmask) != 0; \
}
OPENCV_HAL_IMPL_AVX_CHECK(v_uint8x32, OPENCV_HAL_1ST, -1)
OPENCV_HAL_IMPL_AVX_CHECK(v_int8x32, OPENCV_HAL_1ST, -1)
OPENCV_HAL_IMPL_AVX_CHECK(v_uint16x16, OPENCV_HAL_AND, (int)0xaaaaaaaa)
OPENCV_HAL_IMPL_AVX_CHECK(v_int16x16, OPENCV_HAL_AND, (int)0xaaaaaaaa)
OPENCV_HAL_IMPL_AVX_CHECK(v_uint32x8, OPENCV_HAL_AND, (int)0x88888888)
OPENCV_HAL_IMPL_AVX_CHECK(v_int32x8, OPENCV_HAL_AND, (int)0x88888888)
#define OPENCV_HAL_IMPL_AVX_CHECK_FLT(_Tpvec, allmask) \
inline bool v_check_all(const _Tpvec& a) \
{ \
int mask = v_signmask(a); \
return mask == allmask; \
} \
inline bool v_check_any(const _Tpvec& a) \
{ \
int mask = v_signmask(a); \
return mask != 0; \
}
OPENCV_HAL_IMPL_AVX_CHECK_FLT(v_float32x8, 255)
OPENCV_HAL_IMPL_AVX_CHECK_FLT(v_float64x4, 15)
#define OPENCV_HAL_IMPL_AVX_CHECK(_Tpvec, allmask) \
inline bool v_check_all(const _Tpvec& a) { return v_signmask(a) == allmask; } \
inline bool v_check_any(const _Tpvec& a) { return v_signmask(a) != 0; }
OPENCV_HAL_IMPL_AVX_CHECK(v_uint8x32, -1)
OPENCV_HAL_IMPL_AVX_CHECK(v_int8x32, -1)
OPENCV_HAL_IMPL_AVX_CHECK(v_uint32x8, 255)
OPENCV_HAL_IMPL_AVX_CHECK(v_int32x8, 255)
OPENCV_HAL_IMPL_AVX_CHECK(v_uint64x4, 15)
OPENCV_HAL_IMPL_AVX_CHECK(v_int64x4, 15)
OPENCV_HAL_IMPL_AVX_CHECK(v_float32x8, 255)
OPENCV_HAL_IMPL_AVX_CHECK(v_float64x4, 15)
#define OPENCV_HAL_IMPL_AVX_CHECK_SHORT(_Tpvec) \
inline bool v_check_all(const _Tpvec& a) { return (v_signmask(v_reinterpret_as_s8(a)) & 0xaaaaaaaa) == 0xaaaaaaaa; } \
inline bool v_check_any(const _Tpvec& a) { return (v_signmask(v_reinterpret_as_s8(a)) & 0xaaaaaaaa) != 0; }
OPENCV_HAL_IMPL_AVX_CHECK_SHORT(v_uint16x16)
OPENCV_HAL_IMPL_AVX_CHECK_SHORT(v_int16x16)
////////// Other math /////////
modules/core/include/opencv2/core/hal/intrin_cpp.hpp
+3 -3
View File
@@ -1080,7 +1080,7 @@ Example:
v_int32x4 r; // set to {-1, -1, 1, 1}
int mask = v_signmask(r); // mask = 3 <== 00000000 00000000 00000000 00000011
@endcode
For all types except 64-bit. */
*/
template<typename _Tp, int n> inline int v_signmask(const v_reg<_Tp, n>& a)
{
int mask = 0;
@@ -1109,7 +1109,7 @@ template <typename _Tp, int n> inline int v_scan_forward(const v_reg<_Tp, n>& a)
/** @brief Check if all packed values are less than zero
Unsigned values will be casted to signed: `uchar 254 => char -2`.
For all types except 64-bit. */
*/
template<typename _Tp, int n> inline bool v_check_all(const v_reg<_Tp, n>& a)
{
for( int i = 0; i < n; i++ )
@@ -1121,7 +1121,7 @@ template<typename _Tp, int n> inline bool v_check_all(const v_reg<_Tp, n>& a)
/** @brief Check if any of packed values is less than zero
Unsigned values will be casted to signed: `uchar 254 => char -2`.
For all types except 64-bit. */
*/
template<typename _Tp, int n> inline bool v_check_any(const v_reg<_Tp, n>& a)
{
for( int i = 0; i < n; i++ )
modules/core/include/opencv2/core/hal/intrin_neon.hpp
+14 -6
View File
@@ -1139,9 +1139,17 @@ inline bool v_check_any(const v_##_Tpvec& a) \
OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(uint8x16, u8, 7)
OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(uint16x8, u16, 15)
OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(uint32x4, u32, 31)
#if CV_SIMD128_64F
OPENCV_HAL_IMPL_NEON_CHECK_ALLANY(uint64x2, u64, 63)
#endif
inline bool v_check_all(const v_uint64x2& a)
{
uint64x2_t v0 = vshrq_n_u64(a.val, 63);
return (vgetq_lane_u64(v0, 0) & vgetq_lane_u64(v0, 1)) == 1;
}
inline bool v_check_any(const v_uint64x2& a)
{
uint64x2_t v0 = vshrq_n_u64(a.val, 63);
return (vgetq_lane_u64(v0, 0) | vgetq_lane_u64(v0, 1)) != 0;
}
inline bool v_check_all(const v_int8x16& a)
{ return v_check_all(v_reinterpret_as_u8(a)); }
@@ -1161,13 +1169,13 @@ inline bool v_check_any(const v_int32x4& a)
inline bool v_check_any(const v_float32x4& a)
{ return v_check_any(v_reinterpret_as_u32(a)); }
#if CV_SIMD128_64F
inline bool v_check_all(const v_int64x2& a)
{ return v_check_all(v_reinterpret_as_u64(a)); }
inline bool v_check_all(const v_float64x2& a)
{ return v_check_all(v_reinterpret_as_u64(a)); }
inline bool v_check_any(const v_int64x2& a)
{ return v_check_any(v_reinterpret_as_u64(a)); }
#if CV_SIMD128_64F
inline bool v_check_all(const v_float64x2& a)
{ return v_check_all(v_reinterpret_as_u64(a)); }
inline bool v_check_any(const v_float64x2& a)
{ return v_check_any(v_reinterpret_as_u64(a)); }
#endif
modules/core/include/opencv2/core/hal/intrin_sse.hpp
+18 -24
View File
@@ -1591,31 +1591,25 @@ inline v_uint32x4 v_popcount(const v_int32x4& a)
inline v_uint64x2 v_popcount(const v_int64x2& a)
{ return v_popcount(v_reinterpret_as_u64(a)); }
#define OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(_Tpvec, suffix, pack_op, and_op, signmask, allmask) \
inline int v_signmask(const _Tpvec& a) \
{ \
return and_op(_mm_movemask_##suffix(pack_op(a.val)), signmask); \
} \
inline bool v_check_all(const _Tpvec& a) \
{ return and_op(_mm_movemask_##suffix(a.val), allmask) == allmask; } \
inline bool v_check_any(const _Tpvec& a) \
{ return and_op(_mm_movemask_##suffix(a.val), allmask) != 0; }
#define OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(_Tpvec, suffix, cast_op, allmask) \
inline int v_signmask(const _Tpvec& a) { return _mm_movemask_##suffix(cast_op(a.val)); } \
inline bool v_check_all(const _Tpvec& a) { return _mm_movemask_##suffix(cast_op(a.val)) == allmask; } \
inline bool v_check_any(const _Tpvec& a) { return _mm_movemask_##suffix(cast_op(a.val)) != 0; }
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_uint8x16, epi8, OPENCV_HAL_NOP, 65535)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_int8x16, epi8, OPENCV_HAL_NOP, 65535)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_uint32x4, ps, _mm_castsi128_ps, 15)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_int32x4, ps, _mm_castsi128_ps, 15)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_uint64x2, pd, _mm_castsi128_pd, 3)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_int64x2, pd, _mm_castsi128_pd, 3)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_float32x4, ps, OPENCV_HAL_NOP, 15)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_float64x2, pd, OPENCV_HAL_NOP, 3)
#define OPENCV_HAL_PACKS(a) _mm_packs_epi16(a, a)
inline __m128i v_packq_epi32(__m128i a)
{
__m128i b = _mm_packs_epi32(a, a);
return _mm_packs_epi16(b, b);
}
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_uint8x16, epi8, OPENCV_HAL_NOP, OPENCV_HAL_1ST, 65535, 65535)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_int8x16, epi8, OPENCV_HAL_NOP, OPENCV_HAL_1ST, 65535, 65535)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_uint16x8, epi8, OPENCV_HAL_PACKS, OPENCV_HAL_AND, 255, (int)0xaaaa)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_int16x8, epi8, OPENCV_HAL_PACKS, OPENCV_HAL_AND, 255, (int)0xaaaa)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_uint32x4, epi8, v_packq_epi32, OPENCV_HAL_AND, 15, (int)0x8888)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_int32x4, epi8, v_packq_epi32, OPENCV_HAL_AND, 15, (int)0x8888)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_float32x4, ps, OPENCV_HAL_NOP, OPENCV_HAL_1ST, 15, 15)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS(v_float64x2, pd, OPENCV_HAL_NOP, OPENCV_HAL_1ST, 3, 3)
#define OPENCV_HAL_IMPL_SSE_CHECK_SIGNS_SHORT(_Tpvec) \
inline int v_signmask(const _Tpvec& a) { return _mm_movemask_epi8(_mm_packs_epi16(a.val, a.val)) & 255; } \
inline bool v_check_all(const _Tpvec& a) { return (_mm_movemask_epi8(a.val) & 0xaaaa) == 0xaaaa; } \
inline bool v_check_any(const _Tpvec& a) { return (_mm_movemask_epi8(a.val) & 0xaaaa) != 0; }
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS_SHORT(v_uint16x8)
OPENCV_HAL_IMPL_SSE_CHECK_SIGNS_SHORT(v_int16x8)
inline int v_scan_forward(const v_int8x16& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))); }
inline int v_scan_forward(const v_uint8x16& a) { return trailingZeros32(v_signmask(v_reinterpret_as_s8(a))); }
modules/core/include/opencv2/core/hal/intrin_vsx.hpp
+4
View File
@@ -899,6 +899,8 @@ inline bool v_check_all(const v_uint16x8& a)
{ return v_check_all(v_reinterpret_as_s16(a)); }
inline bool v_check_all(const v_uint32x4& a)
{ return v_check_all(v_reinterpret_as_s32(a)); }
inline bool v_check_all(const v_uint64x2& a)
{ return v_check_all(v_reinterpret_as_s64(a)); }
inline bool v_check_all(const v_float32x4& a)
{ return v_check_all(v_reinterpret_as_s32(a)); }
inline bool v_check_all(const v_float64x2& a)
@@ -913,6 +915,8 @@ inline bool v_check_any(const v_uint16x8& a)
{ return v_check_any(v_reinterpret_as_s16(a)); }
inline bool v_check_any(const v_uint32x4& a)
{ return v_check_any(v_reinterpret_as_s32(a)); }
inline bool v_check_any(const v_uint64x2& a)
{ return v_check_any(v_reinterpret_as_s64(a)); }
inline bool v_check_any(const v_float32x4& a)
{ return v_check_any(v_reinterpret_as_s32(a)); }
inline bool v_check_any(const v_float64x2& a)
modules/core/include/opencv2/core/hal/intrin_wasm.hpp
+4025
View File
File diff suppressed because it is too large Load Diff
modules/core/src/mathfuncs_core.simd.hpp
+5 -5
View File
@@ -31,6 +31,11 @@ using namespace cv;
namespace {
static const float atan2_p1 = 0.9997878412794807f*(float)(180/CV_PI);
static const float atan2_p3 = -0.3258083974640975f*(float)(180/CV_PI);
static const float atan2_p5 = 0.1555786518463281f*(float)(180/CV_PI);
static const float atan2_p7 = -0.04432655554792128f*(float)(180/CV_PI);
#ifdef __EMSCRIPTEN__
static inline float atan_f32(float y, float x)
{
@@ -42,11 +47,6 @@ static inline float atan_f32(float y, float x)
return a; // range [0; 360)
}
#else
static const float atan2_p1 = 0.9997878412794807f*(float)(180/CV_PI);
static const float atan2_p3 = -0.3258083974640975f*(float)(180/CV_PI);
static const float atan2_p5 = 0.1555786518463281f*(float)(180/CV_PI);
static const float atan2_p7 = -0.04432655554792128f*(float)(180/CV_PI);
static inline float atan_f32(float y, float x)
{
float ax = std::abs(x), ay = std::abs(y);
modules/core/src/ocl.cpp
+14 -3
View File
@@ -267,6 +267,9 @@ static const String getBuildExtraOptions()
return param_buildExtraOptions;
}
static const bool CV_OPENCL_ENABLE_MEM_USE_HOST_PTR = utils::getConfigurationParameterBool("OPENCV_OPENCL_ENABLE_MEM_USE_HOST_PTR", true);
static const size_t CV_OPENCL_ALIGNMENT_MEM_USE_HOST_PTR = utils::getConfigurationParameterSizeT("OPENCV_OPENCL_ALIGNMENT_MEM_USE_HOST_PTR", 4);
#endif // HAVE_OPENCL
struct UMat2D
@@ -4671,6 +4674,9 @@ public:
bool allocate(UMatData* u, AccessFlag accessFlags, UMatUsageFlags usageFlags) const CV_OVERRIDE
{
#ifndef HAVE_OPENCL
return false;
#else
if(!u)
return false;
@@ -4746,8 +4752,12 @@ public:
#endif
{
tempUMatFlags = UMatData::TEMP_UMAT;
if (u->origdata == cv::alignPtr(u->origdata, 4) // There are OpenCL runtime issues for less aligned data
&& !(u->originalUMatData && u->originalUMatData->handle) // Avoid sharing of host memory between OpenCL buffers
if (CV_OPENCL_ENABLE_MEM_USE_HOST_PTR
// There are OpenCL runtime issues for less aligned data
&& (CV_OPENCL_ALIGNMENT_MEM_USE_HOST_PTR != 0
&& u->origdata == cv::alignPtr(u->origdata, (int)CV_OPENCL_ALIGNMENT_MEM_USE_HOST_PTR))
// Avoid sharing of host memory between OpenCL buffers
&& !(u->originalUMatData && u->originalUMatData->handle)
)
{
handle = clCreateBuffer(ctx_handle, CL_MEM_USE_HOST_PTR|createFlags,
@@ -4777,6 +4787,7 @@ public:
u->markHostCopyObsolete(true);
opencl_allocator_stats.onAllocate(u->size);
return true;
#endif // HAVE_OPENCL
}
/*void sync(UMatData* u) const
@@ -4905,7 +4916,7 @@ public:
(CL_MAP_READ | CL_MAP_WRITE),
0, u->size, 0, 0, 0, &retval);
CV_OCL_CHECK_RESULT(retval, cv::format("clEnqueueMapBuffer(handle=%p, sz=%lld) => %p", (void*)u->handle, (long long int)u->size, data).c_str());
CV_Assert(u->origdata == data);
CV_Assert(u->origdata == data && "Details: https://github.com/opencv/opencv/issues/6293");
if (u->originalUMatData)
{
CV_Assert(u->originalUMatData->data == data);
modules/core/src/parallel.cpp
+2 -2
View File
@@ -54,7 +54,7 @@
#endif
#if defined __linux__ || defined __APPLE__ || defined __GLIBC__ \
|| defined __HAIKU__
|| defined __HAIKU__ || defined __EMSCRIPTEN__
#include <unistd.h>
#include <stdio.h>
#include <sys/types.h>
@@ -808,7 +808,7 @@ int cv::getNumberOfCPUs(void)
#elif defined __ANDROID__
static int ncpus = getNumberOfCPUsImpl();
return ncpus;
#elif defined __linux__ || defined __GLIBC__ || defined __HAIKU__
#elif defined __linux__ || defined __GLIBC__ || defined __HAIKU__ || defined __EMSCRIPTEN__
return (int)sysconf( _SC_NPROCESSORS_ONLN );
#elif defined __APPLE__
int numCPU=0;
modules/core/test/test_io.cpp
+12 -12
View File
@@ -707,10 +707,10 @@ static void test_filestorage_basic(int write_flags, const char* suffix_name, boo
EXPECT_EQ(_em_in.depth(), _em_out.depth());
EXPECT_TRUE(_em_in.empty());
EXPECT_EQ(_2d_in.rows , _2d_out.rows);
EXPECT_EQ(_2d_in.cols , _2d_out.cols);
EXPECT_EQ(_2d_in.dims , _2d_out.dims);
EXPECT_EQ(_2d_in.depth(), _2d_out.depth());
ASSERT_EQ(_2d_in.rows , _2d_out.rows);
ASSERT_EQ(_2d_in.cols , _2d_out.cols);
ASSERT_EQ(_2d_in.dims , _2d_out.dims);
ASSERT_EQ(_2d_in.depth(), _2d_out.depth());
errors = 0;
for(int i = 0; i < _2d_out.rows; ++i)
@@ -731,16 +731,16 @@ static void test_filestorage_basic(int write_flags, const char* suffix_name, boo
}
}
EXPECT_EQ(_nd_in.rows , _nd_out.rows);
EXPECT_EQ(_nd_in.cols , _nd_out.cols);
EXPECT_EQ(_nd_in.dims , _nd_out.dims);
EXPECT_EQ(_nd_in.depth(), _nd_out.depth());
ASSERT_EQ(_nd_in.rows , _nd_out.rows);
ASSERT_EQ(_nd_in.cols , _nd_out.cols);
ASSERT_EQ(_nd_in.dims , _nd_out.dims);
ASSERT_EQ(_nd_in.depth(), _nd_out.depth());
EXPECT_EQ(0, cv::norm(_nd_in, _nd_out, NORM_INF));
EXPECT_EQ(_rd_in.rows , _rd_out.rows);
EXPECT_EQ(_rd_in.cols , _rd_out.cols);
EXPECT_EQ(_rd_in.dims , _rd_out.dims);
EXPECT_EQ(_rd_in.depth(), _rd_out.depth());
ASSERT_EQ(_rd_in.rows , _rd_out.rows);
ASSERT_EQ(_rd_in.cols , _rd_out.cols);
ASSERT_EQ(_rd_in.dims , _rd_out.dims);
ASSERT_EQ(_rd_in.depth(), _rd_out.depth());
EXPECT_EQ(0, cv::norm(_rd_in, _rd_out, NORM_INF));
}
}