Extended v_check_any/v_check_all universal intrinsics to support 64-bit integer

diff --git a/modules/core/include/opencv2/core/hal/intrin_avx.hpp b/modules/core/include/opencv2/core/hal/intrin_avx.hpp
index 74db46b..c821dc3 100644 (file)
--- a/modules/core/include/opencv2/core/hal/intrin_avx.hpp
+++ b/modules/core/include/opencv2/core/hal/intrin_avx.hpp
@@ -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 /////////
 

diff --git a/modules/core/include/opencv2/core/hal/intrin_cpp.hpp b/modules/core/include/opencv2/core/hal/intrin_cpp.hpp
index d494761..75c39ac 100644 (file)
--- a/modules/core/include/opencv2/core/hal/intrin_cpp.hpp
+++ b/modules/core/include/opencv2/core/hal/intrin_cpp.hpp
@@ -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++ )

diff --git a/modules/core/include/opencv2/core/hal/intrin_neon.hpp b/modules/core/include/opencv2/core/hal/intrin_neon.hpp
index 5617bc2..e5f707c 100644 (file)
--- a/modules/core/include/opencv2/core/hal/intrin_neon.hpp
+++ b/modules/core/include/opencv2/core/hal/intrin_neon.hpp
@@ -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

diff --git a/modules/core/include/opencv2/core/hal/intrin_sse.hpp b/modules/core/include/opencv2/core/hal/intrin_sse.hpp
index e172d45..f661c58 100644 (file)
--- a/modules/core/include/opencv2/core/hal/intrin_sse.hpp
+++ b/modules/core/include/opencv2/core/hal/intrin_sse.hpp
@@ -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_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(_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_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))); }

diff --git a/modules/core/include/opencv2/core/hal/intrin_vsx.hpp b/modules/core/include/opencv2/core/hal/intrin_vsx.hpp
index 5e417a0..4f41021 100644 (file)
--- a/modules/core/include/opencv2/core/hal/intrin_vsx.hpp
+++ b/modules/core/include/opencv2/core/hal/intrin_vsx.hpp
@@ -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)