[JIT] ARM32/ARM64 - Fixed zero/sign-extension for GT_STORE_LCL_VAR …

…in codegen (#84716)

* Try to zero-extend if it is possible

* Added test

* Fix arm32 as well

* Remove branch check for arm32. Added more specific check for arm64.

* Remove unwanted code

* Trying to fix build

* use general register check

* Update Runtime_84693.cs

src/coreclr/jit/codegenarm.cpp

@@ -1082,6 +1082,8 @@ void CodeGen::genCodeForStoreLclVar(GenTreeLclVar* tree)
         LclVarDsc* varDsc     = compiler->lvaGetDesc(varNum);
         var_types  targetType = varDsc->GetRegisterType(tree);
 
+        emitter* emit = GetEmitter();
+
         if (targetType == TYP_LONG)
         {
             genStoreLongLclVar(tree);
@@ -1114,13 +1116,23 @@ void CodeGen::genCodeForStoreLclVar(GenTreeLclVar* tree)
                 instruction ins  = ins_StoreFromSrc(dataReg, targetType);
                 emitAttr    attr = emitTypeSize(targetType);
 
-                emitter* emit = GetEmitter();
                 emit->emitIns_S_R(ins, attr, dataReg, varNum, /* offset */ 0);
             }
             else // store into register (i.e move into register)
             {
                 // Assign into targetReg when dataReg (from op1) is not the same register
-                inst_Mov(targetType, targetReg, dataReg, /* canSkip */ true);
+                // Only zero/sign extend if we are using general registers.
+                if (varTypeIsIntegral(targetType) && emit->isGeneralRegister(targetReg) &&
+                    emit->isGeneralRegister(dataReg))
+                {
+                    // We use 'emitActualTypeSize' as the instructions require 4BYTE.
+                    inst_Mov_Extend(targetType, /* srcInReg */ true, targetReg, dataReg, /* canSkip */ true,
+                                    emitActualTypeSize(targetType));
+                }
+                else
+                {
+                    inst_Mov(targetType, targetReg, dataReg, /* canSkip */ true);
+                }
             }
 
             genUpdateLifeStore(tree, targetReg, varDsc);

src/coreclr/jit/codegenarm64.cpp

@@ -2940,7 +2940,17 @@ void CodeGen::genCodeForStoreLclVar(GenTreeLclVar* lclNode)
         else // store into register (i.e move into register)
         {
             // Assign into targetReg when dataReg (from op1) is not the same register
-            inst_Mov(targetType, targetReg, dataReg, /* canSkip */ true);
+            // Only zero/sign extend if we are using general registers.
+            if (varTypeIsIntegral(targetType) && emit->isGeneralRegister(targetReg) && emit->isGeneralRegister(dataReg))
+            {
+                // We use 'emitActualTypeSize' as the instructions require 8BYTE or 4BYTE.
+                inst_Mov_Extend(targetType, /* srcInReg */ true, targetReg, dataReg, /* canSkip */ true,
+                                emitActualTypeSize(targetType));
+            }
+            else
+            {
+                inst_Mov(targetType, targetReg, dataReg, /* canSkip */ true);
+            }
         }
         genUpdateLifeStore(lclNode, targetReg, varDsc);
     }

src/coreclr/jit/emitarm64.cpp

@@ -4185,7 +4185,7 @@ void emitter::emitIns_Mov(
 
         case INS_sxtw:
         {
-            assert(size == EA_8BYTE);
+            assert((size == EA_8BYTE) || (size == EA_4BYTE));
             FALLTHROUGH;
         }
 

src/tests/JIT/Regression/JitBlue/Runtime_84693/Runtime_84693.cs

@@ -0,0 +1,40 @@
+// 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.Collections;
+using System.Collections.Generic;
+using System.Runtime.CompilerServices;
+using Xunit;
+
+public class Test
+{
+    // This is trying to verify that we zero-extend from the result of "(byte)(-s_2)".
+	public class Program
+	{
+		public static short s_2;
+
+		[MethodImpl(MethodImplOptions.NoInlining)]
+		public static void Consume(int x) {}
+
+		[MethodImpl(MethodImplOptions.NoInlining)]
+		public static int M8(byte arg0)
+		{
+			s_2 = 1;
+			arg0 = (byte)(-s_2);
+			var vr1 = arg0 & arg0;
+			Consume(vr1);
+			return vr1;
+		}
+	}
+
+	[Fact]
+	public static int TestEntryPoint() {
+		var result = Test.Program.M8(1);
+		if (result != 255)
+		{
+			return 0;
+		}
+		return 100;
+	}
+}

src/tests/JIT/Regression/JitBlue/Runtime_84693/Runtime_84693.csproj

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