convertScaleAbs for s8, f64

modules/core/src/convert.cpp

@@ -1123,6 +1123,48 @@ struct cvtScaleAbs_SIMD<uchar, uchar, float>
     }
 };
 
+template <>
+struct cvtScaleAbs_SIMD<schar, uchar, float>
+{
+    int operator () (const schar * src, uchar * dst, int width,
+                     float scale, float shift) const
+    {
+        int x = 0;
+
+        if (USE_SSE2)
+        {
+            __m128 v_scale = _mm_set1_ps(scale), v_shift = _mm_set1_ps(shift),
+                v_zero_f = _mm_setzero_ps();
+            __m128i v_zero_i = _mm_setzero_si128();
+
+            for ( ; x <= width - 16; x += 16)
+            {
+                __m128i v_src = _mm_loadu_si128((const __m128i *)(src + x));
+                __m128i v_src_12 = _mm_srai_epi16(_mm_unpacklo_epi8(v_zero_i, v_src), 8),
+                        v_src_34 = _mm_srai_epi16(_mm_unpackhi_epi8(v_zero_i, v_src), 8);
+                __m128 v_dst1 = _mm_add_ps(_mm_mul_ps(_mm_cvtepi32_ps(
+                    _mm_srai_epi32(_mm_unpacklo_epi16(v_zero_i, v_src_12), 16)), v_scale), v_shift);
+                v_dst1 = _mm_max_ps(_mm_sub_ps(v_zero_f, v_dst1), v_dst1);
+                __m128 v_dst2 = _mm_add_ps(_mm_mul_ps(_mm_cvtepi32_ps(
+                    _mm_srai_epi32(_mm_unpackhi_epi16(v_zero_i, v_src_12), 16)), v_scale), v_shift);
+                v_dst2 = _mm_max_ps(_mm_sub_ps(v_zero_f, v_dst2), v_dst2);
+                __m128 v_dst3 = _mm_add_ps(_mm_mul_ps(_mm_cvtepi32_ps(
+                    _mm_srai_epi32(_mm_unpacklo_epi16(v_zero_i, v_src_34), 16)), v_scale), v_shift);
+                v_dst3 = _mm_max_ps(_mm_sub_ps(v_zero_f, v_dst3), v_dst3);
+                __m128 v_dst4 = _mm_add_ps(_mm_mul_ps(_mm_cvtepi32_ps(
+                    _mm_srai_epi32(_mm_unpackhi_epi16(v_zero_i, v_src_34), 16)), v_scale), v_shift);
+                v_dst4 = _mm_max_ps(_mm_sub_ps(v_zero_f, v_dst4), v_dst4);
+
+                __m128i v_dst_i = _mm_packus_epi16(_mm_packs_epi32(_mm_cvtps_epi32(v_dst1), _mm_cvtps_epi32(v_dst2)),
+                                                   _mm_packs_epi32(_mm_cvtps_epi32(v_dst3), _mm_cvtps_epi32(v_dst4)));
+                _mm_storeu_si128((__m128i *)(dst + x), v_dst_i);
+            }
+        }
+
+        return x;
+    }
+};
+
 template <>
 struct cvtScaleAbs_SIMD<ushort, uchar, float>
 {
@@ -1242,6 +1284,44 @@ struct cvtScaleAbs_SIMD<float, uchar, float>
     }
 };
 
+template <>
+struct cvtScaleAbs_SIMD<double, uchar, float>
+{
+    int operator () (const double * src, uchar * dst, int width,
+                     float scale, float shift) const
+    {
+        int x = 0;
+
+        if (USE_SSE2)
+        {
+            __m128 v_scale = _mm_set1_ps(scale), v_shift = _mm_set1_ps(shift),
+                v_zero_f = _mm_setzero_ps();
+            __m128i v_zero_i = _mm_setzero_si128();
+
+            for ( ; x <= width - 8; x += 8)
+            {
+                __m128 v_src1 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(src + x)),
+                                              _mm_cvtpd_ps(_mm_loadu_pd(src + x + 2)));
+                __m128 v_src2 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(src + x + 4)),
+                                              _mm_cvtpd_ps(_mm_loadu_pd(src + x + 6)));
+
+                __m128 v_dst1 = _mm_add_ps(_mm_mul_ps(v_src1, v_scale), v_shift);
+                v_dst1 = _mm_max_ps(_mm_sub_ps(v_zero_f, v_dst1), v_dst1);
+
+                __m128 v_dst2 = _mm_add_ps(_mm_mul_ps(v_src2, v_scale), v_shift);
+                v_dst2 = _mm_max_ps(_mm_sub_ps(v_zero_f, v_dst2), v_dst2);
+
+                __m128i v_dst_i = _mm_packs_epi32(_mm_cvtps_epi32(v_dst1),
+                                                  _mm_cvtps_epi32(v_dst2));
+
+                _mm_storel_epi64((__m128i *)(dst + x), _mm_packus_epi16(v_dst_i, v_zero_i));
+            }
+        }
+
+        return x;
+    }
+};
+
 #elif CV_NEON
 
 template <>

modules/core/src/mathfuncs.cpp

@@ -622,7 +622,7 @@ void phase( InputArray src1, InputArray src2, OutputArray dst, bool angleInDegre
                 {
                     __m128 v_src = _mm_loadu_ps(buf[0] + k);
                     _mm_storeu_pd(angle + k, _mm_cvtps_pd(v_src));
-                    _mm_storeu_pd(angle + k, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
+                    _mm_storeu_pd(angle + k + 2, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
                 }
 #endif
 
@@ -753,7 +753,7 @@ void cartToPolar( InputArray src1, InputArray src2,
                 {
                     __m128 v_src = _mm_loadu_ps(buf[0] + k);
                     _mm_storeu_pd(angle + k, _mm_cvtps_pd(v_src));
-                    _mm_storeu_pd(angle + k, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
+                    _mm_storeu_pd(angle + k + 2, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
                 }
 #endif
 

modules/core/test/ocl/test_arithm.cpp

@@ -1577,7 +1577,7 @@ PARAM_TEST_CASE(ConvertScaleAbs, MatDepth, Channels, bool)
 
         Size roiSize = randomSize(1, MAX_VALUE);
         Border srcBorder = randomBorder(0, use_roi ? MAX_VALUE : 0);
-        randomSubMat(src, src_roi, roiSize, srcBorder, stype, 2, 11); // FIXIT: Test with minV, maxV
+        randomSubMat(src, src_roi, roiSize, srcBorder, stype, -11, 11); // FIXIT: Test with minV, maxV
 
         Border dstBorder = randomBorder(0, use_roi ? MAX_VALUE : 0);
         randomSubMat(dst, dst_roi, roiSize, dstBorder, dtype, 5, 16);