fix for cv::Mat::convertTo with scale

diff --git a/modules/core/src/convert.cpp b/modules/core/src/convert.cpp
index 4855371..67adfde 100644 (file)
--- a/modules/core/src/convert.cpp
+++ b/modules/core/src/convert.cpp
@@ -1507,8 +1507,8 @@ struct cvtScale_SIMD<uchar, uchar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1_u8(dst + x, vqmovn_u16(v_dst));
         }
 
@@ -1530,8 +1530,8 @@ struct cvtScale_SIMD<uchar, schar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1_s8(dst + x, vqmovn_s16(v_dst));
         }
 
@@ -1553,8 +1553,8 @@ struct cvtScale_SIMD<uchar, ushort, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1q_u16(dst + x, v_dst);
         }
 
@@ -1576,8 +1576,8 @@ struct cvtScale_SIMD<uchar, short, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1q_s16(dst + x, v_dst);
         }
 
@@ -1599,8 +1599,8 @@ struct cvtScale_SIMD<uchar, int, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            vst1q_s32(dst + x, vcvtq_s32_f32(v_dst1));
-            vst1q_s32(dst + x + 4, vcvtq_s32_f32(v_dst2));
+            vst1q_s32(dst + x, cv_vrndq_s32_f32(v_dst1));
+            vst1q_s32(dst + x + 4, cv_vrndq_s32_f32(v_dst2));
         }
 
         return x;
@@ -1642,8 +1642,8 @@ struct cvtScale_SIMD<schar, uchar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1_u8(dst + x, vqmovn_u16(v_dst));
         }
 
@@ -1665,8 +1665,8 @@ struct cvtScale_SIMD<schar, schar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1_s8(dst + x, vqmovn_s16(v_dst));
         }
 
@@ -1688,8 +1688,8 @@ struct cvtScale_SIMD<schar, ushort, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1q_u16(dst + x, v_dst);
         }
 
@@ -1711,8 +1711,8 @@ struct cvtScale_SIMD<schar, short, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1q_s16(dst + x, v_dst);
         }
 
@@ -1734,8 +1734,8 @@ struct cvtScale_SIMD<schar, int, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            vst1q_s32(dst + x, vcvtq_s32_f32(v_dst1));
-            vst1q_s32(dst + x + 4, vcvtq_s32_f32(v_dst2));
+            vst1q_s32(dst + x, cv_vrndq_s32_f32(v_dst1));
+            vst1q_s32(dst + x + 4, cv_vrndq_s32_f32(v_dst2));
         }
 
         return x;
@@ -1777,8 +1777,8 @@ struct cvtScale_SIMD<ushort, uchar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1_u8(dst + x, vqmovn_u16(v_dst));
         }
 
@@ -1800,8 +1800,8 @@ struct cvtScale_SIMD<ushort, schar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1_s8(dst + x, vqmovn_s16(v_dst));
         }
 
@@ -1823,8 +1823,8 @@ struct cvtScale_SIMD<ushort, ushort, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1q_u16(dst + x, v_dst);
         }
 
@@ -1846,8 +1846,8 @@ struct cvtScale_SIMD<ushort, short, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1q_s16(dst + x, v_dst);
         }
 
@@ -1869,8 +1869,8 @@ struct cvtScale_SIMD<ushort, int, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_low_u16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_src))), v_scale), v_shift);
 
-            vst1q_s32(dst + x, vcvtq_s32_f32(v_dst1));
-            vst1q_s32(dst + x + 4, vcvtq_s32_f32(v_dst2));
+            vst1q_s32(dst + x, cv_vrndq_s32_f32(v_dst1));
+            vst1q_s32(dst + x + 4, cv_vrndq_s32_f32(v_dst2));
         }
 
         return x;
@@ -1912,8 +1912,8 @@ struct cvtScale_SIMD<short, uchar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1_u8(dst + x, vqmovn_u16(v_dst));
         }
 
@@ -1935,8 +1935,8 @@ struct cvtScale_SIMD<short, schar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1_s8(dst + x, vqmovn_s16(v_dst));
         }
 
@@ -1958,8 +1958,8 @@ struct cvtScale_SIMD<short, ushort, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(v_src))), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(v_src))), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1q_u16(dst + x, v_dst);
         }
 
@@ -2001,8 +2001,8 @@ struct cvtScale_SIMD<int, uchar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x)), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x + 4)), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1_u8(dst + x, vqmovn_u16(v_dst));
         }
 
@@ -2023,8 +2023,8 @@ struct cvtScale_SIMD<int, schar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x)), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x + 4)), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1_s8(dst + x, vqmovn_s16(v_dst));
         }
 
@@ -2045,8 +2045,8 @@ struct cvtScale_SIMD<int, ushort, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x)), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x + 4)), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1q_u16(dst + x, v_dst);
         }
 
@@ -2067,8 +2067,8 @@ struct cvtScale_SIMD<int, short, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x)), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vcvtq_f32_s32(vld1q_s32(src + x + 4)), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                            vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1q_s16(dst + x, v_dst);
         }
 
@@ -2091,8 +2091,8 @@ struct cvtScale_SIMD<float, uchar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vld1q_f32(src + x), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vld1q_f32(src + x + 4), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1_u8(dst + x, vqmovn_u16(v_dst));
         }
 
@@ -2113,8 +2113,8 @@ struct cvtScale_SIMD<float, schar, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vld1q_f32(src + x), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vld1q_f32(src + x + 4), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                           vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                           vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1_s8(dst + x, vqmovn_s16(v_dst));
         }
 
@@ -2135,8 +2135,8 @@ struct cvtScale_SIMD<float, ushort, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vld1q_f32(src + x), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vld1q_f32(src + x + 4), v_scale), v_shift);
 
-            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(vcvtq_u32_f32(v_dst1)),
-                                            vqmovn_u32(vcvtq_u32_f32(v_dst2)));
+            uint16x8_t v_dst = vcombine_u16(vqmovn_u32(cv_vrndq_u32_f32(v_dst1)),
+                                            vqmovn_u32(cv_vrndq_u32_f32(v_dst2)));
             vst1q_u16(dst + x, v_dst);
         }
 
@@ -2157,8 +2157,8 @@ struct cvtScale_SIMD<float, short, float>
             float32x4_t v_dst1 = vaddq_f32(vmulq_f32(vld1q_f32(src + x), v_scale), v_shift);
             float32x4_t v_dst2 = vaddq_f32(vmulq_f32(vld1q_f32(src + x + 4), v_scale), v_shift);
 
-            int16x8_t v_dst = vcombine_s16(vqmovn_s32(vcvtq_s32_f32(v_dst1)),
-                                            vqmovn_s32(vcvtq_s32_f32(v_dst2)));
+            int16x8_t v_dst = vcombine_s16(vqmovn_s32(cv_vrndq_s32_f32(v_dst1)),
+                                            vqmovn_s32(cv_vrndq_s32_f32(v_dst2)));
             vst1q_s16(dst + x, v_dst);
         }
 
@@ -2175,7 +2175,7 @@ struct cvtScale_SIMD<float, int, float>
         float32x4_t v_shift = vdupq_n_f32(shift), v_scale = vdupq_n_f32(scale);
 
         for ( ; x <= width - 4; x += 4)
-            vst1q_s32(dst + x, vcvtq_s32_f32(vaddq_f32(vmulq_f32(vld1q_f32(src + x), v_scale), v_shift)));
+            vst1q_s32(dst + x, cv_vrndq_s32_f32(vaddq_f32(vmulq_f32(vld1q_f32(src + x), v_scale), v_shift)));
 
         return x;
     }