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

modules/core/include/opencv2/core/hal/intrin_avx.hpp

@@ -1012,6 +1012,54 @@ OPENCV_HAL_IMPL_AVX_ROTATE_CAST(v_rotate_right, v_float32x8, _mm256_castsi256_ps
 OPENCV_HAL_IMPL_AVX_ROTATE_CAST(v_rotate_left,  v_float64x4, _mm256_castsi256_pd)
 OPENCV_HAL_IMPL_AVX_ROTATE_CAST(v_rotate_right, v_float64x4, _mm256_castsi256_pd)
 
+/** Reverse **/
+inline v_uint8x32 v_reverse(const v_uint8x32 &a)
+{
+    static const __m256i perm = _mm256_setr_epi8(
+            15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0,
+            15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+    __m256i vec = _mm256_shuffle_epi8(a.val, perm);
+    return v_uint8x32(_mm256_permute2x128_si256(vec, vec, 1));
+}
+
+inline v_int8x32 v_reverse(const v_int8x32 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x16 v_reverse(const v_uint16x16 &a)
+{
+    static const __m256i perm = _mm256_setr_epi8(
+            14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1,
+            14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1);
+    __m256i vec = _mm256_shuffle_epi8(a.val, perm);
+    return v_uint16x16(_mm256_permute2x128_si256(vec, vec, 1));
+}
+
+inline v_int16x16 v_reverse(const v_int16x16 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x8 v_reverse(const v_uint32x8 &a)
+{
+    static const __m256i perm = _mm256_setr_epi32(7, 6, 5, 4, 3, 2, 1, 0);
+    return v_uint32x8(_mm256_permutevar8x32_epi32(a.val, perm));
+}
+
+inline v_int32x8 v_reverse(const v_int32x8 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x8 v_reverse(const v_float32x8 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x4 v_reverse(const v_uint64x4 &a)
+{
+    return v_uint64x4(_mm256_permute4x64_epi64(a.val, _MM_SHUFFLE(0, 1, 2, 3)));
+}
+
+inline v_int64x4 v_reverse(const v_int64x4 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+inline v_float64x4 v_reverse(const v_float64x4 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+
 ////////// Reduce and mask /////////
 
 /** Reduce **/

modules/core/include/opencv2/core/hal/intrin_avx512.hpp

@@ -1068,6 +1068,79 @@ OPENCV_HAL_IMPL_AVX512_ROTATE_EC(v_int64x8,    epi64)
 OPENCV_HAL_IMPL_AVX512_ROTATE_EC(v_float32x16, ps)
 OPENCV_HAL_IMPL_AVX512_ROTATE_EC(v_float64x8,  pd)
 
+/** Reverse **/
+inline v_uint8x64 v_reverse(const v_uint8x64 &a)
+{
+#if CV_AVX_512VBMI
+    static const __m512i perm = _mm512_set_epi32(
+            0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f,
+            0x10111213, 0x14151617, 0x18191a1b, 0x1c1d1e1f,
+            0x20212223, 0x24252627, 0x28292a2b, 0x2c2d2e2f,
+            0x30313233, 0x34353637, 0x38393a3b, 0x3c3d3e3f);
+    return v_uint8x64(_mm512_permutexvar_epi8(perm, a.val));
+#else
+    static const __m512i shuf = _mm512_set_epi32(
+            0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f,
+            0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f,
+            0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f,
+            0x00010203, 0x04050607, 0x08090a0b, 0x0c0d0e0f);
+    static const __m512i perm = _mm512_set_epi64(1, 0, 3, 2, 5, 4, 7, 6);
+    __m512i vec = _mm512_shuffle_epi8(a.val, shuf);
+    return v_uint8x64(_mm512_permutexvar_epi64(perm, vec));
+#endif
+}
+
+inline v_int8x64 v_reverse(const v_int8x64 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x32 v_reverse(const v_uint16x32 &a)
+{
+#if CV_AVX_512VBMI
+    static const __m512i perm = _mm512_set_epi32(
+            0x00000001, 0x00020003, 0x00040005, 0x00060007,
+            0x00080009, 0x000a000b, 0x000c000d, 0x000e000f,
+            0x00100011, 0x00120013, 0x00140015, 0x00160017,
+            0x00180019, 0x001a001b, 0x001c001d, 0x001e001f);
+    return v_uint16x32(_mm512_permutexvar_epi16(perm, a.val));
+#else
+    static const __m512i shuf = _mm512_set_epi32(
+            0x01000302, 0x05040706, 0x09080b0a, 0x0d0c0f0e,
+            0x01000302, 0x05040706, 0x09080b0a, 0x0d0c0f0e,
+            0x01000302, 0x05040706, 0x09080b0a, 0x0d0c0f0e,
+            0x01000302, 0x05040706, 0x09080b0a, 0x0d0c0f0e);
+    static const __m512i perm = _mm512_set_epi64(1, 0, 3, 2, 5, 4, 7, 6);
+    __m512i vec = _mm512_shuffle_epi8(a.val, shuf);
+    return v_uint16x32(_mm512_permutexvar_epi64(perm, vec));
+#endif
+}
+
+inline v_int16x32 v_reverse(const v_int16x32 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x16 v_reverse(const v_uint32x16 &a)
+{
+    static const __m512i perm = _mm512_set_epi32(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13,14, 15);
+    return v_uint32x16(_mm512_permutexvar_epi32(perm, a.val));
+}
+
+inline v_int32x16 v_reverse(const v_int32x16 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x16 v_reverse(const v_float32x16 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x8 v_reverse(const v_uint64x8 &a)
+{
+    static const __m512i perm = _mm512_set_epi64(0, 1, 2, 3, 4, 5, 6, 7);
+    return v_uint64x8(_mm512_permutexvar_epi64(perm, a.val));
+}
+
+inline v_int64x8 v_reverse(const v_int64x8 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+inline v_float64x8 v_reverse(const v_float64x8 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+
 ////////// Reduce /////////
 
 /** Reduce **/

modules/core/include/opencv2/core/hal/intrin_cpp.hpp

@@ -112,6 +112,7 @@ These operations allow to reorder or recombine elements in one or multiple vecto
 - Pack: @ref v_pack, @ref v_pack_u, @ref v_pack_b, @ref v_rshr_pack, @ref v_rshr_pack_u,
 @ref v_pack_store, @ref v_pack_u_store, @ref v_rshr_pack_store, @ref v_rshr_pack_u_store
 - Recombine: @ref v_zip, @ref v_recombine, @ref v_combine_low, @ref v_combine_high
+- Reverse: @ref v_reverse
 - Extract: @ref v_extract
 
 
@@ -215,6 +216,7 @@ Regular integers:
 |cvt_flt32          |   |   |   |   |   | x |
 |cvt_flt64          |   |   |   |   |   | x |
 |transpose4x4       |   |   |   |   | x | x |
+|reverse            | x | x | x | x | x | x |
 
 Big integers:
 
@@ -224,6 +226,7 @@ Big integers:
 |add, sub           | x | x |
 |shift              | x | x |
 |logical            | x | x |
+|reverse            | x | x |
 |extract            | x | x |
 |rotate (lanes)     | x | x |
 |cvt_flt64          |   | x |
@@ -250,6 +253,7 @@ Floating point:
 |transpose4x4       | x |   |
 |extract            | x | x |
 |rotate (lanes)     | x | x |
+|reverse            | x | x |
 
  @{ */
 
@@ -1724,6 +1728,23 @@ inline void v_recombine(const v_reg<_Tp, n>& a, const v_reg<_Tp, n>& b,
     }
 }
 
+/** @brief Vector reverse order
+
+Reverse the order of the vector
+Scheme:
+@code
+  REG {A1 ... An} ==> REG {An ... A1}
+@endcode
+For all types. */
+template<typename _Tp, int n>
+inline v_reg<_Tp, n> v_reverse(const v_reg<_Tp, n>& a)
+{
+    v_reg<_Tp, n> c;
+    for( int i = 0; i < n; i++ )
+        c.s[i] = a.s[n-i-1];
+    return c;
+}
+
 /** @brief Vector extract
 
 Scheme:

modules/core/include/opencv2/core/hal/intrin_msa.hpp

@@ -906,6 +906,57 @@ OPENCV_HAL_IMPL_MSA_LOADSTORE_OP(v_int64x2, int64, s64)
 OPENCV_HAL_IMPL_MSA_LOADSTORE_OP(v_float32x4, float, f32)
 OPENCV_HAL_IMPL_MSA_LOADSTORE_OP(v_float64x2, double, f64)
 
+
+/** Reverse **/
+inline v_uint8x16 v_reverse(const v_uint8x16 &a)
+{
+    v_uint8x16 c = v_uint8x16((v16u8)__builtin_msa_vshf_b((v16i8)((v2i64){0x08090A0B0C0D0E0F, 0x0001020304050607}), msa_dupq_n_s8(0), (v16i8)a.val));
+    return c;
+}
+
+inline v_int8x16 v_reverse(const v_int8x16 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x8 v_reverse(const v_uint16x8 &a)
+{
+    v_uint16x8 c = v_uint16x8((v8u16)__builtin_msa_vshf_h((v8i16)((v2i64){0x0004000500060007, 0x0000000100020003}), msa_dupq_n_s16(0), (v8i16)a.val));
+    return c;
+}
+
+inline v_int16x8 v_reverse(const v_int16x8 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x4 v_reverse(const v_uint32x4 &a)
+{
+    v_uint32x4 c;
+    c.val[0] = a.val[3];
+    c.val[1] = a.val[2];
+    c.val[2] = a.val[1];
+    c.val[3] = a.val[0];
+    return c;
+}
+
+inline v_int32x4 v_reverse(const v_int32x4 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x4 v_reverse(const v_float32x4 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x2 v_reverse(const v_uint64x2 &a)
+{
+    v_uint64x2 c;
+    c.val[0] = a.val[1];
+    c.val[1] = a.val[0];
+    return c;
+}
+
+inline v_int64x2 v_reverse(const v_int64x2 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+inline v_float64x2 v_reverse(const v_float64x2 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+
+
 #define OPENCV_HAL_IMPL_MSA_REDUCE_OP_8U(func, cfunc) \
 inline unsigned short v_reduce_##func(const v_uint16x8& a) \
 { \

modules/core/include/opencv2/core/hal/intrin_neon.hpp

@@ -1585,6 +1585,52 @@ OPENCV_HAL_IMPL_NEON_UNPACKS(float32x4, f32)
 OPENCV_HAL_IMPL_NEON_UNPACKS(float64x2, f64)
 #endif
 
+inline v_uint8x16 v_reverse(const v_uint8x16 &a)
+{
+    uint8x16_t vec = vrev64q_u8(a.val);
+    return v_uint8x16(vextq_u8(vec, vec, 8));
+}
+
+inline v_int8x16 v_reverse(const v_int8x16 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x8 v_reverse(const v_uint16x8 &a)
+{
+    uint16x8_t vec = vrev64q_u16(a.val);
+    return v_uint16x8(vextq_u16(vec, vec, 4));
+}
+
+inline v_int16x8 v_reverse(const v_int16x8 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x4 v_reverse(const v_uint32x4 &a)
+{
+    uint32x4_t vec = vrev64q_u32(a.val);
+    return v_uint32x4(vextq_u32(vec, vec, 2));
+}
+
+inline v_int32x4 v_reverse(const v_int32x4 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x4 v_reverse(const v_float32x4 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x2 v_reverse(const v_uint64x2 &a)
+{
+    uint64x2_t vec = a.val;
+    uint64x1_t vec_lo = vget_low_u64(vec);
+    uint64x1_t vec_hi = vget_high_u64(vec);
+    return v_uint64x2(vcombine_u64(vec_hi, vec_lo));
+}
+
+inline v_int64x2 v_reverse(const v_int64x2 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+#if CV_SIMD128_64F
+inline v_float64x2 v_reverse(const v_float64x2 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+#endif
+
 #define OPENCV_HAL_IMPL_NEON_EXTRACT(_Tpvec, suffix) \
 template <int s> \
 inline v_##_Tpvec v_extract(const v_##_Tpvec& a, const v_##_Tpvec& b) \

modules/core/include/opencv2/core/hal/intrin_sse.hpp

@@ -1220,14 +1220,23 @@ inline _Tpvec operator >= (const _Tpvec& a, const _Tpvec& b) \
 OPENCV_HAL_IMPL_SSE_FLT_CMP_OP(v_float32x4, ps)
 OPENCV_HAL_IMPL_SSE_FLT_CMP_OP(v_float64x2, pd)
 
-#define OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(_Tpvec, cast) \
+#if CV_SSE4_1
+#define OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(_Tpvec) \
 inline _Tpvec operator == (const _Tpvec& a, const _Tpvec& b) \
-{ return cast(v_reinterpret_as_f64(a) == v_reinterpret_as_f64(b)); } \
+{ return _Tpvec(_mm_cmpeq_epi64(a.val, b.val)); } \
 inline _Tpvec operator != (const _Tpvec& a, const _Tpvec& b) \
-{ return cast(v_reinterpret_as_f64(a) != v_reinterpret_as_f64(b)); }
+{ return ~(a == b); }
+#else
+#define OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(_Tpvec) \
+inline _Tpvec operator == (const _Tpvec& a, const _Tpvec& b) \
+{ __m128i cmp = _mm_cmpeq_epi32(a.val, b.val); \
+  return _Tpvec(_mm_and_si128(cmp, _mm_shuffle_epi32(cmp, _MM_SHUFFLE(2, 3, 0, 1)))); } \
+inline _Tpvec operator != (const _Tpvec& a, const _Tpvec& b) \
+{ return ~(a == b); }
+#endif
 
-OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(v_uint64x2, v_reinterpret_as_u64)
-OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(v_int64x2, v_reinterpret_as_s64)
+OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(v_uint64x2)
+OPENCV_HAL_IMPL_SSE_64BIT_CMP_OP(v_int64x2)
 
 inline v_float32x4 v_not_nan(const v_float32x4& a)
 { return v_float32x4(_mm_cmpord_ps(a.val, a.val)); }
@@ -1914,6 +1923,59 @@ OPENCV_HAL_IMPL_SSE_UNPACKS(v_int32x4, epi32, OPENCV_HAL_NOP, OPENCV_HAL_NOP)
 OPENCV_HAL_IMPL_SSE_UNPACKS(v_float32x4, ps, _mm_castps_si128, _mm_castsi128_ps)
 OPENCV_HAL_IMPL_SSE_UNPACKS(v_float64x2, pd, _mm_castpd_si128, _mm_castsi128_pd)
 
+inline v_uint8x16 v_reverse(const v_uint8x16 &a)
+{
+#if CV_SSSE3
+    static const __m128i perm = _mm_setr_epi8(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+    return v_uint8x16(_mm_shuffle_epi8(a.val, perm));
+#else
+    uchar CV_DECL_ALIGNED(32) d[16];
+    v_store_aligned(d, a);
+    return v_uint8x16(d[15], d[14], d[13], d[12], d[11], d[10], d[9], d[8], d[7], d[6], d[5], d[4], d[3], d[2], d[1], d[0]);
+#endif
+}
+
+inline v_int8x16 v_reverse(const v_int8x16 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x8 v_reverse(const v_uint16x8 &a)
+{
+#if CV_SSSE3
+    static const __m128i perm = _mm_setr_epi8(14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1);
+    return v_uint16x8(_mm_shuffle_epi8(a.val, perm));
+#else
+    __m128i r = _mm_shuffle_epi32(a.val, _MM_SHUFFLE(0, 1, 2, 3));
+    r = _mm_shufflelo_epi16(r, _MM_SHUFFLE(2, 3, 0, 1));
+    r = _mm_shufflehi_epi16(r, _MM_SHUFFLE(2, 3, 0, 1));
+    return v_uint16x8(r);
+#endif
+}
+
+inline v_int16x8 v_reverse(const v_int16x8 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x4 v_reverse(const v_uint32x4 &a)
+{
+    return v_uint32x4(_mm_shuffle_epi32(a.val, _MM_SHUFFLE(0, 1, 2, 3)));
+}
+
+inline v_int32x4 v_reverse(const v_int32x4 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x4 v_reverse(const v_float32x4 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x2 v_reverse(const v_uint64x2 &a)
+{
+    return v_uint64x2(_mm_shuffle_epi32(a.val, _MM_SHUFFLE(1, 0, 3, 2)));
+}
+
+inline v_int64x2 v_reverse(const v_int64x2 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+inline v_float64x2 v_reverse(const v_float64x2 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+
 template<int s, typename _Tpvec>
 inline _Tpvec v_extract(const _Tpvec& a, const _Tpvec& b)
 {

modules/core/include/opencv2/core/hal/intrin_vsx.hpp

@@ -678,6 +678,53 @@ OPENCV_IMPL_VSX_ROTATE_64_2RG_LR(v_float64x2)
 OPENCV_IMPL_VSX_ROTATE_64_2RG_LR(v_uint64x2)
 OPENCV_IMPL_VSX_ROTATE_64_2RG_LR(v_int64x2)
 
+/* Reverse */
+inline v_uint8x16 v_reverse(const v_uint8x16 &a)
+{
+    static const vec_uchar16 perm = {15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0};
+    vec_uchar16 vec = (vec_uchar16)a.val;
+    return v_uint8x16(vec_perm(vec, vec, perm));
+}
+
+inline v_int8x16 v_reverse(const v_int8x16 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x8 v_reverse(const v_uint16x8 &a)
+{
+    static const vec_uchar16 perm = {14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1};
+    vec_uchar16 vec = (vec_uchar16)a.val;
+    return v_reinterpret_as_u16(v_uint8x16(vec_perm(vec, vec, perm)));
+}
+
+inline v_int16x8 v_reverse(const v_int16x8 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x4 v_reverse(const v_uint32x4 &a)
+{
+    static const vec_uchar16 perm = {12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3};
+    vec_uchar16 vec = (vec_uchar16)a.val;
+    return v_reinterpret_as_u32(v_uint8x16(vec_perm(vec, vec, perm)));
+}
+
+inline v_int32x4 v_reverse(const v_int32x4 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x4 v_reverse(const v_float32x4 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x2 v_reverse(const v_uint64x2 &a)
+{
+    static const vec_uchar16 perm = {8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7};
+    vec_uchar16 vec = (vec_uchar16)a.val;
+    return v_reinterpret_as_u64(v_uint8x16(vec_perm(vec, vec, perm)));
+}
+
+inline v_int64x2 v_reverse(const v_int64x2 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+inline v_float64x2 v_reverse(const v_float64x2 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+
 /* Extract */
 template<int s, typename _Tpvec>
 inline _Tpvec v_extract(const _Tpvec& a, const _Tpvec& b)

modules/core/include/opencv2/core/hal/intrin_wasm.hpp

@@ -21,6 +21,18 @@ namespace cv
 
 CV_CPU_OPTIMIZATION_HAL_NAMESPACE_BEGIN
 
+#if (__EMSCRIPTEN_major__ * 1000000 + __EMSCRIPTEN_minor__ * 1000 + __EMSCRIPTEN_tiny__) < (1038046)
+// handle renames: https://github.com/emscripten-core/emscripten/pull/9440 (https://github.com/emscripten-core/emscripten/commit/755d5b46cb84d0aa120c10981b11d05646c29673)
+#define wasm_i32x4_trunc_saturate_f32x4 wasm_trunc_saturate_i32x4_f32x4
+#define wasm_u32x4_trunc_saturate_f32x4 wasm_trunc_saturate_u32x4_f32x4
+#define wasm_i64x2_trunc_saturate_f64x2 wasm_trunc_saturate_i64x2_f64x2
+#define wasm_u64x2_trunc_saturate_f64x2 wasm_trunc_saturate_u64x2_f64x2
+#define wasm_f32x4_convert_i32x4 wasm_convert_f32x4_i32x4
+#define wasm_f32x4_convert_u32x4 wasm_convert_f32x4_u32x4
+#define wasm_f64x2_convert_i64x2 wasm_convert_f64x2_i64x2
+#define wasm_f64x2_convert_u64x2 wasm_convert_f64x2_u64x2
+#endif // COMPATIBILITY: <1.38.46
+
 ///////// Types ///////////
 
 struct v_uint8x16
@@ -3111,6 +3123,38 @@ OPENCV_HAL_IMPL_WASM_LOADSTORE_INT_OP(v_float32x4, float)
 OPENCV_HAL_IMPL_WASM_LOADSTORE_INT_OP(v_float64x2, double)
 
 
+/** Reverse **/
+inline v_uint8x16 v_reverse(const v_uint8x16 &a)
+{ return v_uint8x16(wasm_v8x16_shuffle(a.val, a.val, 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)); }
+
+inline v_int8x16 v_reverse(const v_int8x16 &a)
+{ return v_reinterpret_as_s8(v_reverse(v_reinterpret_as_u8(a))); }
+
+inline v_uint16x8 v_reverse(const v_uint16x8 &a)
+{ return v_uint16x8(wasm_v8x16_shuffle(a.val, a.val, 14, 15, 12, 13, 10, 11, 8, 9, 6, 7, 4, 5, 2, 3, 0, 1)); }
+
+inline v_int16x8 v_reverse(const v_int16x8 &a)
+{ return v_reinterpret_as_s16(v_reverse(v_reinterpret_as_u16(a))); }
+
+inline v_uint32x4 v_reverse(const v_uint32x4 &a)
+{ return v_uint32x4(wasm_v8x16_shuffle(a.val, a.val, 12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3)); }
+
+inline v_int32x4 v_reverse(const v_int32x4 &a)
+{ return v_reinterpret_as_s32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_float32x4 v_reverse(const v_float32x4 &a)
+{ return v_reinterpret_as_f32(v_reverse(v_reinterpret_as_u32(a))); }
+
+inline v_uint64x2 v_reverse(const v_uint64x2 &a)
+{ return v_uint64x2(wasm_v8x16_shuffle(a.val, a.val, 8, 9, 10, 11, 12, 13, 14, 15, 0, 1, 2, 3, 4, 5, 6, 7)); }
+
+inline v_int64x2 v_reverse(const v_int64x2 &a)
+{ return v_reinterpret_as_s64(v_reverse(v_reinterpret_as_u64(a))); }
+
+inline v_float64x2 v_reverse(const v_float64x2 &a)
+{ return v_reinterpret_as_f64(v_reverse(v_reinterpret_as_u64(a))); }
+
+
 #define OPENCV_HAL_IMPL_WASM_REDUCE_OP_4_SUM(_Tpvec, scalartype, regtype, suffix, esuffix) \
 inline scalartype v_reduce_sum(const _Tpvec& a) \
 { \
@@ -3400,25 +3444,25 @@ inline _Tpvec v_extract(const _Tpvec& a, const _Tpvec& b)
 inline v_int32x4 v_round(const v_float32x4& a)
 {
     v128_t h = wasm_f32x4_splat(0.5);
-    return v_int32x4(wasm_trunc_saturate_i32x4_f32x4(wasm_f32x4_add(a.val, h)));
+    return v_int32x4(wasm_i32x4_trunc_saturate_f32x4(wasm_f32x4_add(a.val, h)));
 }
 
 inline v_int32x4 v_floor(const v_float32x4& a)
 {
-    v128_t a1 = wasm_trunc_saturate_i32x4_f32x4(a.val);
-    v128_t mask = wasm_f32x4_lt(a.val, wasm_convert_f32x4_i32x4(a1));
+    v128_t a1 = wasm_i32x4_trunc_saturate_f32x4(a.val);
+    v128_t mask = wasm_f32x4_lt(a.val, wasm_f32x4_convert_i32x4(a1));
     return v_int32x4(wasm_i32x4_add(a1, mask));
 }
 
 inline v_int32x4 v_ceil(const v_float32x4& a)
 {
-    v128_t a1 = wasm_trunc_saturate_i32x4_f32x4(a.val);
-    v128_t mask = wasm_f32x4_gt(a.val, wasm_convert_f32x4_i32x4(a1));
+    v128_t a1 = wasm_i32x4_trunc_saturate_f32x4(a.val);
+    v128_t mask = wasm_f32x4_gt(a.val, wasm_f32x4_convert_i32x4(a1));
     return v_int32x4(wasm_i32x4_sub(a1, mask));
 }
 
 inline v_int32x4 v_trunc(const v_float32x4& a)
-{ return v_int32x4(wasm_trunc_saturate_i32x4_f32x4(a.val)); }
+{ return v_int32x4(wasm_i32x4_trunc_saturate_f32x4(a.val)); }
 
 #define OPENCV_HAL_IMPL_WASM_MATH_FUNC(func, cfunc, _Tpvec, _Tpnvec, _Tp, _Tpn) \
 inline _Tpnvec func(const _Tpvec& a) \
@@ -3924,7 +3968,7 @@ OPENCV_HAL_IMPL_WASM_LOADSTORE_INTERLEAVE(v_float64x2, double, f64, v_uint64x2,
 
 inline v_float32x4 v_cvt_f32(const v_int32x4& a)
 {
-    return v_float32x4(wasm_convert_f32x4_i32x4(a.val));
+    return v_float32x4(wasm_f32x4_convert_i32x4(a.val));
 }
 
 inline v_float32x4 v_cvt_f32(const v_float64x2& a)
@@ -3943,7 +3987,7 @@ inline v_float64x2 v_cvt_f64(const v_int32x4& a)
 {
 #ifdef __wasm_unimplemented_simd128__
     v128_t p = v128_cvti32x4_i64x2(a.val);
-    return v_float64x2(wasm_convert_f64x2_i64x2(p));
+    return v_float64x2(wasm_f64x2_convert_i64x2(p));
 #else
     fallback::v_int32x4 a_(a);
     return fallback::v_cvt_f64(a_);
@@ -3954,7 +3998,7 @@ inline v_float64x2 v_cvt_f64_high(const v_int32x4& a)
 {
 #ifdef __wasm_unimplemented_simd128__
     v128_t p = v128_cvti32x4_i64x2_high(a.val);
-    return v_float64x2(wasm_convert_f64x2_i64x2(p));
+    return v_float64x2(wasm_f64x2_convert_i64x2(p));
 #else
     fallback::v_int32x4 a_(a);
     return fallback::v_cvt_f64_high(a_);
@@ -3976,7 +4020,7 @@ inline v_float64x2 v_cvt_f64_high(const v_float32x4& a)
 inline v_float64x2 v_cvt_f64(const v_int64x2& a)
 {
 #ifdef __wasm_unimplemented_simd128__
-    return v_float64x2(wasm_convert_f64x2_i64x2(a.val));
+    return v_float64x2(wasm_f64x2_convert_i64x2(a.val));
 #else
     fallback::v_int64x2 a_(a);
     return fallback::v_cvt_f64(a_);

modules/core/src/alloc.cpp

@@ -112,6 +112,13 @@ bool isAlignedAllocationEnabled()
     }
     return useMemalign;
 }
+// do not use variable directly, details: https://github.com/opencv/opencv/issues/15691
+static const bool g_force_initialization_memalign_flag
+#if defined __GNUC__
+    __attribute__((unused))
+#endif
+    = isAlignedAllocationEnabled();
+
 #endif
 
 #ifdef OPENCV_ALLOC_ENABLE_STATISTICS

modules/core/src/copy.cpp

@@ -711,6 +711,13 @@ static bool ipp_flip(Mat &src, Mat &dst, int flip_mode)
 #ifdef HAVE_IPP_IW
     CV_INSTRUMENT_REGION_IPP();
 
+    // Details: https://github.com/opencv/opencv/issues/12943
+    if (flip_mode <= 0 /* swap rows */
+        && cv::ipp::getIppTopFeatures() != ippCPUID_SSE42
+        && (int64_t)(src.total()) * src.elemSize() >= CV_BIG_INT(0x80000000)/*2Gb*/
+    )
+        return false;
+
     IppiAxis ippMode;
     if(flip_mode < 0)
         ippMode = ippAxsBoth;

modules/core/src/count_non_zero.simd.hpp

@@ -179,7 +179,25 @@ static int countNonZero32f( const float* src, int len )
 
 static int countNonZero64f( const double* src, int len )
 {
-    return countNonZero_(src, len);
+    int nz = 0, i = 0;
+#if CV_SIMD_64F
+    v_int64 sum1 = vx_setzero_s64();
+    v_int64 sum2 = vx_setzero_s64();
+    v_float64 zero = vx_setzero_f64();
+    int step = v_float64::nlanes * 2;
+    int len0 = len & -step;
+
+    for(i = 0; i < len0; i += step )
+        {
+        sum1 += v_reinterpret_as_s64(vx_load(&src[i]) == zero);
+        sum2 += v_reinterpret_as_s64(vx_load(&src[i + step / 2]) == zero);
+        }
+
+    // N.B the value is incremented by -1 (0xF...F) for each value
+    nz = i + (int)v_reduce_sum(sum1 + sum2);
+    v_cleanup();
+#endif
+    return nz + countNonZero_(src + i, len - i);
 }
 
 CountNonZeroFunc getCountNonZeroTab(int depth)

modules/core/test/test_intrin_utils.hpp

@@ -1115,6 +1115,22 @@ template<typename R> struct TheTest
         return *this;
     }
 
+    TheTest & test_reverse()
+    {
+        Data<R> dataA;
+        R a = dataA;
+
+        Data<R> resB = v_reverse(a);
+
+        for (int i = 0; i < R::nlanes; ++i)
+        {
+            SCOPED_TRACE(cv::format("i=%d", i));
+            EXPECT_EQ(dataA[R::nlanes - i - 1], resB[i]);
+        }
+
+        return *this;
+    }
+
     template<int s>
     TheTest & test_extract()
     {
@@ -1426,6 +1442,50 @@ template<typename R> struct TheTest
         return *this;
     }
 #endif
+
+#if CV_SIMD_64F
+    TheTest & test_cmp64()
+    {
+        Data<R> dataA, dataB;
+        R a = dataA, b = dataB;
+
+        for (int i = 0; i < R::nlanes; ++i)
+        {
+            dataA[i] = dataB[i];
+        }
+        dataA[0]++;
+
+        a = dataA, b = dataB;
+
+        Data<R> resC = (a == b);
+        Data<R> resD = (a != b);
+
+        for (int i = 0; i < R::nlanes; ++i)
+        {
+            SCOPED_TRACE(cv::format("i=%d", i));
+            EXPECT_EQ(dataA[i] == dataB[i], resC[i] != 0);
+            EXPECT_EQ(dataA[i] != dataB[i], resD[i] != 0);
+        }
+
+        for (int i = 0; i < R::nlanes; ++i)
+        {
+            dataA[i] = dataB[i] = (LaneType)-1;
+        }
+
+        a = dataA, b = dataB;
+
+        resC = (a == b);
+        resD = (a != b);
+
+        for (int i = 0; i < R::nlanes; ++i)
+        {
+            SCOPED_TRACE(cv::format("i=%d", i));
+            EXPECT_EQ(dataA[i] == dataB[i], resC[i] != 0);
+            EXPECT_EQ(dataA[i] != dataB[i], resD[i] != 0);
+        }
+        return *this;
+    }
+#endif
 };
 
 
@@ -1459,6 +1519,7 @@ void test_hal_intrin_uint8()
         .test_pack_u<1>().test_pack_u<2>().test_pack_u<3>().test_pack_u<8>()
         .test_pack_b()
         .test_unpack()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<8>().test_extract<15>()
         .test_rotate<0>().test_rotate<1>().test_rotate<8>().test_rotate<15>()
         ;
@@ -1497,6 +1558,7 @@ void test_hal_intrin_int8()
         .test_popcount()
         .test_pack<1>().test_pack<2>().test_pack<3>().test_pack<8>()
         .test_unpack()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<8>().test_extract<15>()
         .test_rotate<0>().test_rotate<1>().test_rotate<8>().test_rotate<15>()
         ;
@@ -1529,6 +1591,7 @@ void test_hal_intrin_uint16()
         .test_pack<1>().test_pack<2>().test_pack<7>().test_pack<16>()
         .test_pack_u<1>().test_pack_u<2>().test_pack_u<7>().test_pack_u<16>()
         .test_unpack()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<4>().test_extract<7>()
         .test_rotate<0>().test_rotate<1>().test_rotate<4>().test_rotate<7>()
         ;
@@ -1561,6 +1624,7 @@ void test_hal_intrin_int16()
         .test_popcount()
         .test_pack<1>().test_pack<2>().test_pack<7>().test_pack<16>()
         .test_unpack()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<4>().test_extract<7>()
         .test_rotate<0>().test_rotate<1>().test_rotate<4>().test_rotate<7>()
         ;
@@ -1590,6 +1654,7 @@ void test_hal_intrin_uint32()
         .test_popcount()
         .test_pack<1>().test_pack<2>().test_pack<15>().test_pack<32>()
         .test_unpack()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<2>().test_extract<3>()
         .test_rotate<0>().test_rotate<1>().test_rotate<2>().test_rotate<3>()
         .test_transpose()
@@ -1619,6 +1684,7 @@ void test_hal_intrin_int32()
         .test_mask()
         .test_pack<1>().test_pack<2>().test_pack<15>().test_pack<32>()
         .test_unpack()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<2>().test_extract<3>()
         .test_rotate<0>().test_rotate<1>().test_rotate<2>().test_rotate<3>()
         .test_float_cvt32()
@@ -1635,8 +1701,12 @@ void test_hal_intrin_uint64()
     TheTest<v_uint64>()
         .test_loadstore()
         .test_addsub()
+#if CV_SIMD_64F
+        .test_cmp64()
+#endif
         .test_shift<1>().test_shift<8>()
         .test_logic()
+        .test_reverse()
         .test_extract<0>().test_extract<1>()
         .test_rotate<0>().test_rotate<1>()
         ;
@@ -1648,8 +1718,12 @@ void test_hal_intrin_int64()
     TheTest<v_int64>()
         .test_loadstore()
         .test_addsub()
+#if CV_SIMD_64F
+        .test_cmp64()
+#endif
         .test_shift<1>().test_shift<8>()
         .test_logic()
+        .test_reverse()
         .test_extract<0>().test_extract<1>()
         .test_rotate<0>().test_rotate<1>()
         .test_cvt64_double()
@@ -1680,6 +1754,7 @@ void test_hal_intrin_float32()
         .test_matmul()
         .test_transpose()
         .test_reduce_sum4()
+        .test_reverse()
         .test_extract<0>().test_extract<1>().test_extract<2>().test_extract<3>()
         .test_rotate<0>().test_rotate<1>().test_rotate<2>().test_rotate<3>()
         ;
@@ -1709,6 +1784,7 @@ void test_hal_intrin_float64()
         .test_unpack()
         .test_float_math()
         .test_float_cvt32()
+        .test_reverse()
         .test_extract<0>().test_extract<1>()
         .test_rotate<0>().test_rotate<1>()
         ;

modules/core/test/test_mat.cpp

@@ -2025,4 +2025,17 @@ TEST(Core_Eigen, eigen2cv_check_Mat_type)
 }
 #endif // HAVE_EIGEN
 
+TEST(Mat, regression_12943)  // memory usage: ~4.5 Gb
+{
+    applyTestTag(CV_TEST_TAG_MEMORY_6GB);
+
+    const int width = 0x8000;
+    const int height = 0x10001;
+
+    cv::Mat src(height, width, CV_8UC1, Scalar::all(128));
+
+    cv::Mat dst;
+    cv::flip(src, dst, 0);
+}
+
 }} // namespace