Ensure we consistently broadcast the result of simd dot product (#105888)

src/coreclr/jit/lowerxarch.cpp

@@ -6320,8 +6320,9 @@ GenTree* Lowering::LowerHWIntrinsicDot(GenTreeHWIntrinsic* node)
 
                 horizontalAdd = NI_SSE3_HorizontalAdd;
 
-                if (!comp->compOpportunisticallyDependsOn(InstructionSet_SSE3))
+                if ((simdSize == 8) || !comp->compOpportunisticallyDependsOn(InstructionSet_SSE3))
                 {
+                    // We also do this for simdSize == 8 to ensure we broadcast the result as expected
                     shuffle = NI_SSE_Shuffle;
                 }
                 break;
@@ -6372,10 +6373,8 @@ GenTree* Lowering::LowerHWIntrinsicDot(GenTreeHWIntrinsic* node)
 
                 horizontalAdd = NI_SSE3_HorizontalAdd;
 
-                if (!comp->compOpportunisticallyDependsOn(InstructionSet_SSE3))
+                // We need to ensure we broadcast the result as expected
-                {
+                shuffle = NI_SSE2_Shuffle;
-                    shuffle = NI_SSE2_Shuffle;
-                }
                 break;
             }
 

src/coreclr/jit/morph.cpp

@@ -9916,16 +9916,6 @@ GenTree* Compiler::fgOptimizeHWIntrinsic(GenTreeHWIntrinsic* node)
                 break;
             }
 
-#if defined(TARGET_XARCH)
-            if ((simdSize == 8) && !compOpportunisticallyDependsOn(InstructionSet_SSE41))
-            {
-                // When SSE4.1 isn't supported then Vector2 only needs a single horizontal add
-                // which means the result isn't broadcast across the entire vector and we can't
-                // optimize
-                break;
-            }
-#endif // TARGET_XARCH
-
             GenTree* op1      = node->Op(1);
             GenTree* sqrt     = nullptr;
             GenTree* toScalar = nullptr;

src/tests/JIT/Regression/JitBlue/Runtime_99391/Runtime_99391.cs

@@ -0,0 +1,50 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System;
+using System.Runtime.CompilerServices;
+using System.Numerics;
+using Xunit;
+
+public class Runtime_99391
+{
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        Vector2 result2a = Vector2.Normalize(Value2);
+        Assert.Equal(new Vector2(0, 1), result2a);
+
+        Vector2 result2b = Vector2.Normalize(new Vector2(0, 2));
+        Assert.Equal(new Vector2(0, 1), result2b);
+
+        Vector3 result3a = Vector3.Normalize(Value3);
+        Assert.Equal(new Vector3(0, 0, 1), result3a);
+
+        Vector3 result3b = Vector3.Normalize(new Vector3(0, 0, 2));
+        Assert.Equal(new Vector3(0, 0, 1), result3b);
+
+        Vector4 result4a = Vector4.Normalize(Value4);
+        Assert.Equal(new Vector4(0, 0, 0, 1), result4a);
+
+        Vector4 result4b = Vector4.Normalize(new Vector4(0, 0, 0, 2));
+        Assert.Equal(new Vector4(0, 0, 0, 1), result4b);
+    }
+
+    private static Vector2 Value2
+    {
+        [MethodImpl(MethodImplOptions.NoInlining)]
+        get => new Vector2(0, 2);
+    }
+
+    private static Vector3 Value3
+    {
+        [MethodImpl(MethodImplOptions.NoInlining)]
+        get => new Vector3(0, 0, 2);
+    }
+
+    private static Vector4 Value4
+    {
+        [MethodImpl(MethodImplOptions.NoInlining)]
+        get => new Vector4(0, 0, 0, 2);
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_99391/Runtime_99391.csproj

@@ -0,0 +1,11 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+  <ItemGroup>
+    <CLRTestEnvironmentVariable Include="DOTNET_TieredCompilation" Value="0" />
+  </ItemGroup>
+</Project>