Fix argument passing of structs with padding on x86 (#108867)

When passing a struct by value on x86, we copy its value to the stack.
Sometimes this is done one field at a time, if the struct has been
promoted. If there is padding between the fields (to create appropriate
field alignment), then that padding needs to be zeroed.

The bug is a case where there is padding between an earlier `double`
field and a later, larger, SIMD field. We push 4-byte zeros but the code thinks
it is pushing 8-byte zeros. The fix is simply to correctly calculate the number
of zeros to push.

Fixes #106140

Author: BruceForstall

src/coreclr/jit/codegenxarch.cpp

@@ -8239,12 +8239,15 @@ void CodeGen::genPutArgStkFieldList(GenTreePutArgStk* putArgStk)
             unsigned pushSize = genTypeSize(genActualType(fieldType));
             assert((pushSize % 4) == 0);
             adjustment -= pushSize;
+
+            // If there is padding before this argument, zero it out.
+            assert((adjustment % TARGET_POINTER_SIZE) == 0);
             while (adjustment != 0)
             {
-                inst_IV(INS_push, 0);
-                currentOffset -= pushSize;
-                AddStackLevel(pushSize);
-                adjustment -= pushSize;
+                inst_IV(INS_push, 0); // Push TARGET_POINTER_SIZE bytes of zeros.
+                currentOffset -= TARGET_POINTER_SIZE;
+                AddStackLevel(TARGET_POINTER_SIZE);
+                adjustment -= TARGET_POINTER_SIZE;
             }
 
             m_pushStkArg = true;

src/tests/JIT/Regression/JitBlue/Runtime_106140/Runtime_106140.cs

@@ -0,0 +1,32 @@
+// 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.Runtime.Intrinsics;
+using Xunit;
+
+public class Runtime_106140
+{
+    public struct S1
+    {
+        public double F0;
+        public Vector256<int> F1;
+        public S1(bool f3) : this()
+        {
+            F1 = Vector256.Create(1,2,100,4,5,6,7,8);
+        }
+    }
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    public static int t(S1 s)
+    {
+        return s.F1.GetElement(2);
+    }
+
+    [Fact]
+    public static void TestEntryPoint()
+    {
+        Assert.Equal(100, t(new S1(false)));
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_106140/Runtime_106140.csproj

@@ -0,0 +1,8 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <Optimize>True</Optimize>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>