From 20a0d846768bb7662fc19cb6ae88648b5b3a0810 Mon Sep 17 00:00:00 2001
From: Dmitry Pentin <riotbr3aker@gmail.com>
Date: Thu, 10 Jun 2021 05:09:53 +0300
Subject: [PATCH] Moved jpeg matrix scaler to jpeg converter

---
 .../Common/Helpers/SimdUtils.HwIntrinsics.cs  | 19 -------------
 .../Encoder/RgbToYCbCrConverterVectorized.cs  | 27 ++++++++++++++++---
 2 files changed, 24 insertions(+), 22 deletions(-)

diff --git a/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs b/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
index caeb694a99..b530a37e77 100644
--- a/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
+++ b/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
@@ -577,25 +577,6 @@ namespace SixLabors.ImageSharp
                 }
             }
 
-            /// <summary>
-            /// Scales 8x8 matrix to 4x2 using 2x2 average
-            /// </summary>
-            /// <param name="v">Input matrix consisting of 4 256bit vectors, first row: (v[0], v[2]), second row: (v[1], v[3])</param>
-            /// <returns>256bit vector containing upper and lower scaled parts of the input matrix</returns>
-            [MethodImpl(MethodImplOptions.AggressiveInlining)]
-            public static Vector256<float> Scale16x2_8x1(ReadOnlySpan<Vector256<float>> v)
-            {
-                DebugGuard.IsTrue(v.Length == 4, "Input span must consist of 4 elements");
-
-                var f025 = Vector256.Create(0.25f);
-
-                Vector256<float> left = Avx.Add(v[0], v[2]);
-                Vector256<float> right = Avx.Add(v[1], v[3]);
-                Vector256<float> avg2x2 = Avx.Multiply(Avx.HorizontalAdd(left, right), f025);
-
-                return Avx2.Permute4x64(avg2x2.AsDouble(), 0b11_01_10_00).AsSingle();
-            }
-
             /// <summary>
             /// <see cref="ByteToNormalizedFloat"/> as many elements as possible, slicing them down (keeping the remainder).
             /// </summary>
diff --git a/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrConverterVectorized.cs b/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrConverterVectorized.cs
index 56da8acc71..1b7df596c9 100644
--- a/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrConverterVectorized.cs
+++ b/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrConverterVectorized.cs
@@ -221,9 +221,9 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
                     bDataLanes[j] = b;
                 }
 
-                r = SimdUtils.HwIntrinsics.Scale16x2_8x1(rDataLanes);
-                g = SimdUtils.HwIntrinsics.Scale16x2_8x1(gDataLanes);
-                b = SimdUtils.HwIntrinsics.Scale16x2_8x1(bDataLanes);
+                r = Scale16x2_8x1(rDataLanes);
+                g = Scale16x2_8x1(gDataLanes);
+                b = Scale16x2_8x1(bDataLanes);
 
                 // 128F + ((-0.168736F * r) - (0.331264F * g) + (0.5F * b))
                 Unsafe.Add(ref destCbRef, i) = Avx.Add(f128, SimdUtils.HwIntrinsics.MultiplyAdd(SimdUtils.HwIntrinsics.MultiplyAdd(Avx.Multiply(f05, b), fn0331264, g), fn0168736, r));
@@ -233,5 +233,26 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
             }
 #endif
         }
+
+#if SUPPORTS_RUNTIME_INTRINSICS 
+        /// <summary>
+        /// Scales 16x2 matrix to 8x1 using 2x2 average
+        /// </summary>
+        /// <param name="v">Input matrix consisting of 4 256bit vectors</param>
+        /// <returns>256bit vector containing upper and lower scaled parts of the input matrix</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        internal static Vector256<float> Scale16x2_8x1(ReadOnlySpan<Vector256<float>> v)
+        {
+            DebugGuard.IsTrue(v.Length == 4, "Input span must consist of 4 elements");
+
+            var f025 = Vector256.Create(0.25f);
+
+            Vector256<float> left = Avx.Add(v[0], v[2]);
+            Vector256<float> right = Avx.Add(v[1], v[3]);
+            Vector256<float> avg2x2 = Avx.Multiply(Avx.HorizontalAdd(left, right), f025);
+
+            return Avx2.Permute4x64(avg2x2.AsDouble(), 0b11_01_10_00).AsSingle();
+        }
     }
+#endif
 }