Merge pull request #11470 from seiko2plus:VSXImprovements_2

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

@@ -589,7 +589,7 @@ inline _Tpvec v_rotate_left(const _Tpvec& a, const _Tpvec& b)
     return _Tpvec(vec_sld(a.val, b.val, CV_SHIFT));
 }
 
-#define OPENCV_IMPL_VSX_ROTATE_64(_Tpvec, suffix, rg1, rg2)       \
+#define OPENCV_IMPL_VSX_ROTATE_64_2RG(_Tpvec, suffix, rg1, rg2)   \
 template<int imm>                                                 \
 inline _Tpvec v_rotate_##suffix(const _Tpvec& a, const _Tpvec& b) \
 {                                                                 \
@@ -598,11 +598,13 @@ inline _Tpvec v_rotate_##suffix(const _Tpvec& a, const _Tpvec& b) \
     return imm ? b : a;                                           \
 }
 
-OPENCV_IMPL_VSX_ROTATE_64(v_int64x2,  right, a, b)
-OPENCV_IMPL_VSX_ROTATE_64(v_uint64x2, right, a, b)
+#define OPENCV_IMPL_VSX_ROTATE_64_2RG_LR(_Tpvec)    \
+OPENCV_IMPL_VSX_ROTATE_64_2RG(_Tpvec, left,  b, a)  \
+OPENCV_IMPL_VSX_ROTATE_64_2RG(_Tpvec, right, a, b)
 
-OPENCV_IMPL_VSX_ROTATE_64(v_int64x2,  left, b, a)
-OPENCV_IMPL_VSX_ROTATE_64(v_uint64x2, left, b, a)
+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)
 
 /* Extract */
 template<int s, typename _Tpvec>
@@ -716,26 +718,33 @@ inline int v_signmask(const v_uint64x2& a)
 inline int v_signmask(const v_float64x2& a)
 { return v_signmask(v_reinterpret_as_s64(a)); }
 
-
 template<typename _Tpvec>
 inline bool v_check_all(const _Tpvec& a)
-{ return vec_all_lt(a.val, _Tpvec().val);}
-inline bool v_check_all(const v_uint8x16 &a)
+{ return vec_all_lt(a.val, _Tpvec().val); }
+inline bool v_check_all(const v_uint8x16& a)
 { return v_check_all(v_reinterpret_as_s8(a)); }
-inline bool v_check_all(const v_uint16x8 &a)
+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)
+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_float32x4& a)
 { return v_check_all(v_reinterpret_as_s32(a)); }
+inline bool v_check_all(const v_float64x2& a)
+{ return v_check_all(v_reinterpret_as_s64(a)); }
 
 template<typename _Tpvec>
 inline bool v_check_any(const _Tpvec& a)
-{ return vec_any_lt(a.val, _Tpvec().val);}
-inline bool v_check_any(const v_uint8x16 &a)
+{ return vec_any_lt(a.val, _Tpvec().val); }
+inline bool v_check_any(const v_uint8x16& a)
 { return v_check_any(v_reinterpret_as_s8(a)); }
-inline bool v_check_any(const v_uint16x8 &a)
+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)
+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_float32x4& a)
 { return v_check_any(v_reinterpret_as_s32(a)); }
+inline bool v_check_any(const v_float64x2& a)
+{ return v_check_any(v_reinterpret_as_s64(a)); }
 
 ////////// Other math /////////
 

modules/core/test/test_intrin_utils.hpp

@@ -837,17 +837,28 @@ template<typename R> struct TheTest
         Data<R> resC = v_rotate_right<s>(a);
         Data<R> resD = v_rotate_right<s>(a, b);
 
+        Data<R> resE = v_rotate_left<s>(a);
+        Data<R> resF = v_rotate_left<s>(a, b);
+
         for (int i = 0; i < R::nlanes; ++i)
         {
             if (i + s >= R::nlanes)
             {
                 EXPECT_EQ((LaneType)0, resC[i]);
                 EXPECT_EQ(dataB[i - R::nlanes + s], resD[i]);
+
+                EXPECT_EQ((LaneType)0, resE[i - R::nlanes + s]);
+                EXPECT_EQ(dataB[i], resF[i - R::nlanes + s]);
             }
             else
+            {
                 EXPECT_EQ(dataA[i + s], resC[i]);
-        }
+                EXPECT_EQ(dataA[i + s], resD[i]);
 
+                EXPECT_EQ(dataA[i], resE[i + s]);
+                EXPECT_EQ(dataA[i], resF[i + s]);
+            }
+        }
         return *this;
     }
 

modules/photo/test/test_hdr.cpp

@@ -213,7 +213,7 @@ TEST(Photo_MergeRobertson, regression)
     loadImage(test_path + "merge/robertson.hdr", expected);
     merge->process(images, result, times);
 
-#ifdef __aarch64__
+#if defined(__aarch64__) || defined(__PPC64__)
     const float eps = 6.f;
 #else
     const float eps = 5.f;