Fix passing too small args for PUTARG_STK on macOS arm64 ABI (#68113)

Backport of #68108

src/coreclr/jit/codegenarmarch.cpp

@@ -681,12 +681,6 @@ void CodeGen::genIntrinsic(GenTree* treeNode)
 void CodeGen::genPutArgStk(GenTreePutArgStk* treeNode)
 {
     assert(treeNode->OperIs(GT_PUTARG_STK));
-    GenTree* source = treeNode->gtOp1;
-#if !defined(OSX_ARM64_ABI)
-    var_types targetType = genActualType(source->TypeGet());
-#else
-    var_types targetType = source->TypeGet();
-#endif
     emitter* emit = GetEmitter();
 
     // This is the varNum for our store operations,
@@ -730,11 +724,13 @@ void CodeGen::genPutArgStk(GenTreePutArgStk* treeNode)
         argOffsetMax = compiler->lvaOutgoingArgSpaceSize;
     }
 
-    bool isStruct = (targetType == TYP_STRUCT) || (source->OperGet() == GT_FIELD_LIST);
+    GenTree* source = treeNode->gtGetOp1();
+
+    bool isStruct = source->TypeIs(TYP_STRUCT) || (source->OperGet() == GT_FIELD_LIST);
 
     if (!isStruct) // a normal non-Struct argument
     {
-        if (varTypeIsSIMD(targetType))
+        if (varTypeIsSIMD(source->TypeGet()))
         {
             assert(!source->isContained());
 
@@ -753,31 +749,34 @@ void CodeGen::genPutArgStk(GenTreePutArgStk* treeNode)
             else
 #endif // OSX_ARM64_ABI
             {
-                emitAttr storeAttr = emitTypeSize(targetType);
+                emitAttr storeAttr = emitTypeSize(source->TypeGet());
                 emit->emitIns_S_R(INS_str, storeAttr, srcReg, varNumOut, argOffsetOut);
                 argOffsetOut += EA_SIZE_IN_BYTES(storeAttr);
             }
             assert(argOffsetOut <= argOffsetMax); // We can't write beyond the outgoing arg area
             return;
         }
 
-#if defined(OSX_ARM64_ABI)
+        var_types slotType = genActualType(source);
+#ifdef OSX_ARM64_ABI
+        // Small typed args do not get their own full stack slots, so make
+        // sure we do not overwrite adjacent arguments.
         switch (treeNode->GetStackByteSize())
         {
             case 1:
-                targetType = TYP_BYTE;
+                slotType = TYP_BYTE;
                 break;
             case 2:
-                targetType = TYP_SHORT;
+                slotType = TYP_SHORT;
                 break;
             default:
                 assert(treeNode->GetStackByteSize() >= 4);
                 break;
         }
 #endif
 
-        instruction storeIns  = ins_Store(targetType);
-        emitAttr    storeAttr = emitTypeSize(targetType);
+        instruction storeIns  = ins_Store(slotType);
+        emitAttr    storeAttr = emitTypeSize(slotType);
 
         // If it is contained then source must be the integer constant zero
         if (source->isContained())
@@ -797,7 +796,7 @@ void CodeGen::genPutArgStk(GenTreePutArgStk* treeNode)
             genConsumeReg(source);
             emit->emitIns_S_R(storeIns, storeAttr, source->GetRegNum(), varNumOut, argOffsetOut);
 #ifdef TARGET_ARM
-            if (targetType == TYP_LONG)
+            if (source->TypeIs(TYP_LONG))
             {
                 // This case currently only occurs for double types that are passed as TYP_LONG;
                 // actual long types would have been decomposed by now.

src/tests/JIT/Regression/JitBlue/Runtime_66720/Runtime_66720.cs

@@ -0,0 +1,29 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+
+using System.Runtime.CompilerServices;
+
+public class Runtime_66720
+{
+    public static int Main()
+    {
+        return Test(0);
+    }
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    private static int Test(in short zero)
+    {
+        // Fill arg stack slot with all ones
+        LastArg(0, 0, 0, 0, 0, 0, 0, 0, -1);
+        // Bug was that the last arg passed here would write only 16 bits
+        // instead of 32 bits
+        int last = LastArg(0, 0, 0, 0, 0, 0, 0, 0, zero);
+        return last == 0 ? 100 : -1;
+    }
+
+    [MethodImpl(MethodImplOptions.NoInlining)]
+    private static int LastArg(int a, int b, int c, int d, int e, int f, int g, int h, int i)
+    {
+        return i;
+    }
+}

src/tests/JIT/Regression/JitBlue/Runtime_66720/Runtime_66720.csproj

@@ -0,0 +1,9 @@
+<Project Sdk="Microsoft.NET.Sdk">
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <Optimize>True</Optimize>
+  </PropertyGroup>
+  <ItemGroup>
+    <Compile Include="$(MSBuildProjectName).cs" />
+  </ItemGroup>
+</Project>