From a65e50377de0c08c715d08b93ac5c2202e546150 Mon Sep 17 00:00:00 2001
From: Dmitry Pentin <riotbr3aker@gmail.com>
Date: Tue, 25 May 2021 14:45:26 +0300
Subject: [PATCH] Added MultiplySubstract method to the HwIntrinsics

---
 .../Common/Helpers/SimdUtils.HwIntrinsics.cs  | 26 +++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs b/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
index 4faf577fd9..00c0d89f02 100644
--- a/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
+++ b/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
@@ -532,6 +532,7 @@ namespace SixLabors.ImageSharp
             /// <summary>
             /// Performs a multiplication and an addition of the <see cref="Vector256{T}"/>.
             /// </summary>
+            /// <remarks>ret = (vm0 * vm1) + va</remarks>
             /// <param name="va">The vector to add to the intermediate result.</param>
             /// <param name="vm0">The first vector to multiply.</param>
             /// <param name="vm1">The second vector to multiply.</param>
@@ -552,6 +553,31 @@ namespace SixLabors.ImageSharp
                 }
             }
 
+            /// <summary>
+            /// Performs a multiplication and a substraction of the <see cref="Vector256{T}"/>.
+            /// </summary>
+            /// <remarks>ret = (vm0 * vm1) - vs</remarks>
+            /// <param name="vs">The vector to substract from the intermediate result.</param>
+            /// <param name="vm0">The first vector to multiply.</param>
+            /// <param name="vm1">The second vector to multiply.</param>
+            /// <returns>The <see cref="Vector256{T}"/>.</returns>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public static Vector256<float> MultiplySubstract(
+                in Vector256<float> vs,
+                in Vector256<float> vm0,
+                in Vector256<float> vm1)
+            {
+                if (Fma.IsSupported)
+                {
+                    return Fma.MultiplySubtract(vm1, vm0, vs);
+                }
+                else
+                {
+                    return Avx.Subtract(Avx.Multiply(vm0, vm1), vs);
+                }
+            }
+
+
             /// <summary>
             /// <see cref="ByteToNormalizedFloat"/> as many elements as possible, slicing them down (keeping the remainder).
             /// </summary>