potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 07:43:03 +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
2019-06-05 19:11:22 +03:00
parent e329c84d5e 51631b90af
commit f3de2b4be7
155 changed files with 5856 additions and 1800 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/hal/intrin.hpp
+41 -1
View File
@@ -180,6 +180,18 @@ using namespace CV_CPU_OPTIMIZATION_HAL_NAMESPACE;
#endif
// AVX512 can be used together with SSE2 and AVX2, so
// we define those sets of intrinsics at once.
// For some of AVX512 intrinsics get v512_ prefix instead of v_, e.g. v512_load() vs v_load().
// Wide intrinsics will be mapped to v512_ counterparts in this case(e.g. vx_load() => v512_load())
#if CV_AVX512_SKX
#define CV__SIMD_FORWARD 512
#include "opencv2/core/hal/intrin_forward.hpp"
#include "opencv2/core/hal/intrin_avx512.hpp"
#endif
//! @cond IGNORED
namespace cv {
@@ -321,13 +333,41 @@ template<typename _Tp> struct V_RegTraits
CV_DEF_REG_TRAITS(v256, v_float64x4, double, f64, v_float64x4, void, void, v_int64x4, v_int32x8);
#endif
#if CV_SIMD512
CV_DEF_REG_TRAITS(v512, v_uint8x64, uchar, u8, v_uint8x64, v_uint16x32, v_uint32x16, v_int8x64, void);
CV_DEF_REG_TRAITS(v512, v_int8x64, schar, s8, v_uint8x64, v_int16x32, v_int32x16, v_int8x64, void);
CV_DEF_REG_TRAITS(v512, v_uint16x32, ushort, u16, v_uint16x32, v_uint32x16, v_uint64x8, v_int16x32, void);
CV_DEF_REG_TRAITS(v512, v_int16x32, short, s16, v_uint16x32, v_int32x16, v_int64x8, v_int16x32, void);
CV_DEF_REG_TRAITS(v512, v_uint32x16, unsigned, u32, v_uint32x16, v_uint64x8, void, v_int32x16, void);
CV_DEF_REG_TRAITS(v512, v_int32x16, int, s32, v_uint32x16, v_int64x8, void, v_int32x16, void);
CV_DEF_REG_TRAITS(v512, v_float32x16, float, f32, v_float32x16, v_float64x8, void, v_int32x16, v_int32x16);
CV_DEF_REG_TRAITS(v512, v_uint64x8, uint64, u64, v_uint64x8, void, void, v_int64x8, void);
CV_DEF_REG_TRAITS(v512, v_int64x8, int64, s64, v_uint64x8, void, void, v_int64x8, void);
CV_DEF_REG_TRAITS(v512, v_float64x8, double, f64, v_float64x8, void, void, v_int64x8, v_int32x16);
#endif
#if CV_SIMD512 && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 512)
#define CV__SIMD_NAMESPACE simd512
namespace CV__SIMD_NAMESPACE {
#define CV_SIMD 1
#define CV_SIMD_64F CV_SIMD512_64F
#define CV_SIMD_FP16 CV_SIMD512_FP16
#define CV_SIMD_WIDTH 64
// TODO typedef v_uint8 / v_int32 / etc types here
typedef v_uint8x64 v_uint8;
typedef v_int8x64 v_int8;
typedef v_uint16x32 v_uint16;
typedef v_int16x32 v_int16;
typedef v_uint32x16 v_uint32;
typedef v_int32x16 v_int32;
typedef v_uint64x8 v_uint64;
typedef v_int64x8 v_int64;
typedef v_float32x16 v_float32;
CV_INTRIN_DEFINE_WIDE_INTRIN_ALL_TYPES(v512)
#if CV_SIMD512_64F
typedef v_float64x8 v_float64;
CV_INTRIN_DEFINE_WIDE_INTRIN(double, v_float64, f64, v512, load)
#endif
inline void vx_cleanup() { v512_cleanup(); }
} // namespace
using namespace CV__SIMD_NAMESPACE;
#elif CV_SIMD256 && (!defined(CV__SIMD_FORCE_WIDTH) || CV__SIMD_FORCE_WIDTH == 256)
modules/core/include/opencv2/core/hal/intrin_avx.hpp
+1 -1
View File
@@ -1083,7 +1083,7 @@ OPENCV_HAL_IMPL_AVX_REDUCE_8(v_int32x8, int, max, _mm_max_epi32)
__m128 v1 = _v256_extract_high(a.val); \
v0 = intrin(v0, v1); \
v0 = intrin(v0, _mm_permute_ps(v0, _MM_SHUFFLE(0, 0, 3, 2))); \
v0 = intrin(v0, _mm_permute_ps(v0, _MM_SHUFFLE(0, 0, 0, 3))); \
v0 = intrin(v0, _mm_permute_ps(v0, _MM_SHUFFLE(0, 0, 0, 1))); \
return _mm_cvtss_f32(v0); \
}
modules/core/include/opencv2/core/hal/intrin_avx512.hpp
+2743
View File
File diff suppressed because it is too large Load Diff
modules/core/include/opencv2/core/hal/intrin_forward.hpp
+26 -3
View File
@@ -14,9 +14,32 @@ namespace cv
CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
/** Types **/
#if CV__SIMD_FORWARD == 512
// [todo] 512
#error "AVX512 Not implemented yet"
#if CV__SIMD_FORWARD == 1024
// [todo] 1024
#error "1024-long ops not implemented yet"
#elif CV__SIMD_FORWARD == 512
// 512
#define __CV_VX(fun) v512_##fun
#define __CV_V_UINT8 v_uint8x64
#define __CV_V_INT8 v_int8x64
#define __CV_V_UINT16 v_uint16x32
#define __CV_V_INT16 v_int16x32
#define __CV_V_UINT32 v_uint32x16
#define __CV_V_INT32 v_int32x16
#define __CV_V_UINT64 v_uint64x8
#define __CV_V_INT64 v_int64x8
#define __CV_V_FLOAT32 v_float32x16
#define __CV_V_FLOAT64 v_float64x8
struct v_uint8x64;
struct v_int8x64;
struct v_uint16x32;
struct v_int16x32;
struct v_uint32x16;
struct v_int32x16;
struct v_uint64x8;
struct v_int64x8;
struct v_float32x16;
struct v_float64x8;
#elif CV__SIMD_FORWARD == 256
// 256
#define __CV_VX(fun) v256_##fun