NativeAOT: Add win-x86 support (#99372)

eng/Subsets.props

@@ -119,7 +119,9 @@
 
   <PropertyGroup>
     <!-- CLR NativeAot only builds in a subset of the matrix -->
-    <NativeAotSupported Condition="('$(TargetOS)' == 'windows' or '$(TargetOS)' == 'linux' or '$(TargetOS)' == 'osx' or '$(TargetOS)' == 'maccatalyst' or '$(TargetOS)' == 'iossimulator' or '$(TargetOS)' == 'ios' or '$(TargetOS)' == 'tvossimulator' or '$(TargetOS)' == 'tvos' or '$(TargetOS)' == 'freebsd') and ('$(TargetArchitecture)' == 'x64' or '$(TargetArchitecture)' == 'arm64' or '$(TargetArchitecture)' == 'arm')">true</NativeAotSupported>
+    <_NativeAotSupportedOS Condition="'$(TargetOS)' == 'windows' or '$(TargetOS)' == 'linux' or '$(TargetOS)' == 'osx' or '$(TargetOS)' == 'maccatalyst' or '$(TargetOS)' == 'iossimulator' or '$(TargetOS)' == 'ios' or '$(TargetOS)' == 'tvossimulator' or '$(TargetOS)' == 'tvos' or '$(TargetOS)' == 'freebsd'">true</_NativeAotSupportedOS>
+    <_NativeAotSupportedArch Condition="'$(TargetArchitecture)' == 'x64' or '$(TargetArchitecture)' == 'arm64' or '$(TargetArchitecture)' == 'arm' or ('$(TargetOS)' == 'windows' and '$(TargetArchitecture)' == 'x86')">true</_NativeAotSupportedArch>
+    <NativeAotSupported Condition="'$(_NativeAotSupportedOS)' == 'true' and $(_NativeAotSupportedArch) == 'true'">true</NativeAotSupported>
 
     <!-- If we're building clr.nativeaotlibs and not building the CLR runtime, compile libraries against NativeAOT CoreLib -->
     <UseNativeAotCoreLib Condition="'$(TestNativeAot)' == 'true' or ($(_subset.Contains('+clr.nativeaotlibs+')) and !$(_subset.Contains('+clr.native+')) and !$(_subset.Contains('+clr.runtime+')))">true</UseNativeAotCoreLib>

src/coreclr/CMakeLists.txt

@@ -147,7 +147,7 @@ add_subdirectory(tools/aot/jitinterface)
 
 if(NOT CLR_CROSS_COMPONENTS_BUILD)
   # NativeAOT only buildable for a subset of CoreCLR-supported configurations
-  if(CLR_CMAKE_HOST_ARCH_ARM64 OR CLR_CMAKE_HOST_ARCH_AMD64 OR CLR_CMAKE_HOST_ARCH_ARM)
+  if(CLR_CMAKE_HOST_ARCH_ARM64 OR CLR_CMAKE_HOST_ARCH_AMD64 OR CLR_CMAKE_HOST_ARCH_ARM OR (CLR_CMAKE_HOST_ARCH_I386 AND CLR_CMAKE_HOST_WIN32))
     add_subdirectory(nativeaot)
   endif()
 endif(NOT CLR_CROSS_COMPONENTS_BUILD)

src/coreclr/nativeaot/Runtime.Base/src/System/Runtime/CompilerServices/Unsafe.cs

@@ -129,5 +129,15 @@ public static void CopyBlock(void* destination, void* source, uint byteCount)
         {
             throw new PlatformNotSupportedException();
         }
+
+        /// <summary>
+        /// Copies bytes from the source address to the destination address.
+        /// </summary>
+        [Intrinsic]
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static void CopyBlock(ref byte destination, ref readonly byte source, uint byteCount)
+        {
+            throw new PlatformNotSupportedException();
+        }
     }
 }

src/coreclr/nativeaot/Runtime.Base/src/System/Runtime/InternalCalls.cs

@@ -156,10 +156,6 @@ internal static int RhEndNoGCRegion()
         [RuntimeImport(Redhawk.BaseName, "RhpGcSafeZeroMemory")]
         internal static extern unsafe ref byte RhpGcSafeZeroMemory(ref byte dmem, nuint size);
 
-        [MethodImplAttribute(MethodImplOptions.InternalCall)]
-        [RuntimeImport(Redhawk.BaseName, "memmove")]
-        internal static extern unsafe void* memmove(byte* dmem, byte* smem, nuint size);
-
         [MethodImplAttribute(MethodImplOptions.InternalCall)]
         [RuntimeImport(Redhawk.BaseName, "RhBulkMoveWithWriteBarrier")]
         internal static extern unsafe void RhBulkMoveWithWriteBarrier(ref byte dmem, ref byte smem, nuint size);

src/coreclr/nativeaot/Runtime.Base/src/System/Runtime/RuntimeExports.cs

@@ -114,9 +114,7 @@ public static unsafe object RhBox(MethodTable* pEEType, ref byte data)
             }
             else
             {
-                fixed (byte* pFields = &result.GetRawData())
-                fixed (byte* pData = &dataAdjustedForNullable)
-                    InternalCalls.memmove(pFields, pData, pEEType->ValueTypeSize);
+                Unsafe.CopyBlock(ref result.GetRawData(), ref dataAdjustedForNullable, pEEType->ValueTypeSize);
             }
 
             return result;
@@ -271,9 +269,7 @@ public static unsafe void RhUnbox(object? obj, ref byte data, MethodTable* pUnbo
             else
             {
                 // Copy the boxed fields into the new location.
-                fixed (byte *pData = &data)
-                    fixed (byte* pFields = &fields)
-                        InternalCalls.memmove(pData, pFields, pEEType->ValueTypeSize);
+                Unsafe.CopyBlock(ref data, ref fields, pEEType->ValueTypeSize);
             }
         }
 

src/coreclr/nativeaot/Runtime/CommonMacros.h

@@ -17,7 +17,7 @@
 #define STDCALL
 #endif
 
-#define REDHAWK_CALLCONV FASTCALL
+#define F_CALL_CONV FASTCALL
 #define QCALLTYPE
 
 #ifdef _MSC_VER
@@ -219,6 +219,42 @@ typedef uint8_t CODE_LOCATION;
     _Pragma(FCALL_XSTRINGIFY(comment (linker, FCALL_DECL_ALTNAME(FCALL_METHOD_NAME_((__VA_ARGS__)), FCALL_ARGHELPER_STACKSIZE(__VA_ARGS__)))))
 #define FCIMPL_RENAME(_rettype, ...) \
     _Pragma(FCALL_XSTRINGIFY(comment (linker, FCALL_IMPL_ALTNAME(FCALL_METHOD_NAME_((__VA_ARGS__)), FCALL_ARGHELPER_STACKSIZE(__VA_ARGS__)))))
+#define FCIMPL_RENAME_ARGSIZE(_rettype, _method, _argSize) \
+    _Pragma(FCALL_XSTRINGIFY(comment (linker, FCALL_XSTRINGIFY(/alternatename:_method=@_method##_FCall@_argSize))))
+
+#define FCIMPL1_F(_rettype, _method, a) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 4) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (a) \
+    {
+#define FCIMPL1_D(_rettype, _method, a) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 8) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (a) \
+    {
+#define FCIMPL1_L FCIMPL1_D
+#define FCIMPL2_FF(_rettype, _method, a, b) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 8) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (b, a) \
+    {
+#define FCIMPL2_DD(_rettype, _method, a, b) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 16) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (b, a) \
+    {
+#define FCIMPL2_FI(_rettype, _method, a, b) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 8) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (a, b) \
+    {
+#define FCIMPL2_DI(_rettype, _method, a, b) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 12) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (a, b) \
+    {
+#define FCIMPL3_FFF(_rettype, _method, a, b, c) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 12) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (c, b, a) \
+    {
+#define FCIMPL3_DDD(_rettype, _method, a, b, c) \
+    FCIMPL_RENAME_ARGSIZE(_rettype, _method, 24) \
+    EXTERN_C _rettype F_CALL_CONV _method##_FCall (c, b, a) \
+    {
 
 #else
 
@@ -228,11 +264,37 @@ typedef uint8_t CODE_LOCATION;
 #define FCALL_METHOD_ARGS(dummy, ...) (__VA_ARGS__)
 #define FCALL_METHOD_ARGS_(tuple) FCALL_METHOD_ARGS tuple
 
+#define FCIMPL1_F(_rettype, _method, a) \
+    EXTERN_C _rettype F_CALL_CONV _method (a) \
+    {
+#define FCIMPL1_D(_rettype, _method, a) \
+    EXTERN_C _rettype F_CALL_CONV _method (a) \
+    {
+#define FCIMPL1_L FCIMPL1_D
+#define FCIMPL2_FF(_rettype, _method, a, b) \
+    EXTERN_C _rettype F_CALL_CONV _method (a, b) \
+    {
+#define FCIMPL2_DD(_rettype, _method, a, b) \
+    EXTERN_C _rettype F_CALL_CONV _method (a, b) \
+    {
+#define FCIMPL2_FI(_rettype, _method, a, b) \
+    EXTERN_C _rettype F_CALL_CONV _method (a, b) \
+    {
+#define FCIMPL2_DI(_rettype, _method, a, b) \
+    EXTERN_C _rettype F_CALL_CONV _method (a, b) \
+    {
+#define FCIMPL3_FFF(_rettype, _method, a, b, c) \
+    EXTERN_C _rettype F_CALL_CONV _method (a, b, c) \
+    {
+#define FCIMPL3_DDD(_rettype, _method, a, b, c) \
+    EXTERN_C _rettype F_CALL_CONV _method (a, b, c) \
+    {
+
 #endif
 
 #define FCDECL_(_rettype, ...) \
     FCDECL_RENAME(_rettype, __VA_ARGS__) \
-    EXTERN_C _rettype REDHAWK_CALLCONV FCALL_METHOD_NAME_((__VA_ARGS__)) FCALL_METHOD_ARGS_((__VA_ARGS__))
+    EXTERN_C _rettype F_CALL_CONV FCALL_METHOD_NAME_((__VA_ARGS__)) FCALL_METHOD_ARGS_((__VA_ARGS__))
 #define FCDECL0(_rettype, _method) FCDECL_(_rettype, _method)
 #define FCDECL1(_rettype, _method, a) FCDECL_(_rettype, _method, a)
 #define FCDECL2(_rettype, _method, a, b) FCDECL_(_rettype, _method, a, b)
@@ -242,7 +304,7 @@ typedef uint8_t CODE_LOCATION;
 
 #define FCIMPL_(_rettype, ...) \
     FCIMPL_RENAME(_rettype, __VA_ARGS__) \
-    EXTERN_C _rettype REDHAWK_CALLCONV FCALL_METHOD_NAME_((__VA_ARGS__)) FCALL_METHOD_ARGS_((__VA_ARGS__)) \
+    EXTERN_C _rettype F_CALL_CONV FCALL_METHOD_NAME_((__VA_ARGS__)) FCALL_METHOD_ARGS_((__VA_ARGS__)) \
     {
 #define FCIMPL0(_rettype, _method) FCIMPL_(_rettype, _method)
 #define FCIMPL1(_rettype, _method, a) FCIMPL_(_rettype, _method, a)

src/coreclr/nativeaot/Runtime/EHHelpers.cpp

@@ -309,7 +309,23 @@ EXTERN_C void* RhpRethrow2   = NULL;
 #endif
 
 EXTERN_C CODE_LOCATION RhpAssignRefAVLocation;
+#if defined(HOST_X86)
+EXTERN_C CODE_LOCATION RhpAssignRefEAXAVLocation;
+EXTERN_C CODE_LOCATION RhpAssignRefECXAVLocation;
+EXTERN_C CODE_LOCATION RhpAssignRefEBXAVLocation;
+EXTERN_C CODE_LOCATION RhpAssignRefESIAVLocation;
+EXTERN_C CODE_LOCATION RhpAssignRefEDIAVLocation;
+EXTERN_C CODE_LOCATION RhpAssignRefEBPAVLocation;
+#endif
 EXTERN_C CODE_LOCATION RhpCheckedAssignRefAVLocation;
+#if defined(HOST_X86)
+EXTERN_C CODE_LOCATION RhpCheckedAssignRefEAXAVLocation;
+EXTERN_C CODE_LOCATION RhpCheckedAssignRefECXAVLocation;
+EXTERN_C CODE_LOCATION RhpCheckedAssignRefEBXAVLocation;
+EXTERN_C CODE_LOCATION RhpCheckedAssignRefESIAVLocation;
+EXTERN_C CODE_LOCATION RhpCheckedAssignRefEDIAVLocation;
+EXTERN_C CODE_LOCATION RhpCheckedAssignRefEBPAVLocation;
+#endif
 EXTERN_C CODE_LOCATION RhpCheckedLockCmpXchgAVLocation;
 EXTERN_C CODE_LOCATION RhpCheckedXchgAVLocation;
 #if !defined(HOST_AMD64) && !defined(HOST_ARM64)
@@ -335,13 +351,31 @@ static bool InWriteBarrierHelper(uintptr_t faultingIP)
     static uintptr_t writeBarrierAVLocations[] =
     {
         (uintptr_t)&RhpAssignRefAVLocation,
+#if defined(HOST_X86)
+        (uintptr_t)&RhpAssignRefEAXAVLocation,
+        (uintptr_t)&RhpAssignRefECXAVLocation,
+        (uintptr_t)&RhpAssignRefEBXAVLocation,
+        (uintptr_t)&RhpAssignRefESIAVLocation,
+        (uintptr_t)&RhpAssignRefEDIAVLocation,
+        (uintptr_t)&RhpAssignRefEBPAVLocation,
+#endif
         (uintptr_t)&RhpCheckedAssignRefAVLocation,
+#if defined(HOST_X86)
+        (uintptr_t)&RhpCheckedAssignRefEAXAVLocation,
+        (uintptr_t)&RhpCheckedAssignRefECXAVLocation,
+        (uintptr_t)&RhpCheckedAssignRefEBXAVLocation,
+        (uintptr_t)&RhpCheckedAssignRefESIAVLocation,
+        (uintptr_t)&RhpCheckedAssignRefEDIAVLocation,
+        (uintptr_t)&RhpCheckedAssignRefEBPAVLocation,
+#endif
         (uintptr_t)&RhpCheckedLockCmpXchgAVLocation,
         (uintptr_t)&RhpCheckedXchgAVLocation,
 #if !defined(HOST_AMD64) && !defined(HOST_ARM64)
+#if !defined(HOST_X86)
         (uintptr_t)&RhpLockCmpXchg8AVLocation,
         (uintptr_t)&RhpLockCmpXchg16AVLocation,
         (uintptr_t)&RhpLockCmpXchg32AVLocation,
+#endif
         (uintptr_t)&RhpLockCmpXchg64AVLocation,
 #endif
         (uintptr_t)&RhpByRefAssignRefAVLocation1,

src/coreclr/nativeaot/Runtime/GCHelpers.cpp

@@ -418,12 +418,11 @@ FCIMPL0(int64_t, RhGetTotalAllocatedBytes)
 }
 FCIMPLEND
 
-FCIMPL2(void, RhEnumerateConfigurationValues, void* configurationContext, ConfigurationValueFunc callback)
+EXTERN_C void QCALLTYPE RhEnumerateConfigurationValues(void* configurationContext, ConfigurationValueFunc callback)
 {
     IGCHeap* pHeap = GCHeapUtilities::GetGCHeap();
     pHeap->EnumerateConfigurationValues(configurationContext, callback);
 }
-FCIMPLEND
 
 GCHeapHardLimitInfo g_gcHeapHardLimitInfo;
 bool g_gcHeapHardLimitInfoSpecified = false;
@@ -563,7 +562,7 @@ static Object* GcAllocInternal(MethodTable* pEEType, uint32_t uFlags, uintptr_t
 //  numElements     -  number of array elements
 //  pTransitionFrame-  transition frame to make stack crawlable
 // Returns a pointer to the object allocated or NULL on failure.
-EXTERN_C void* REDHAWK_CALLCONV RhpGcAlloc(MethodTable* pEEType, uint32_t uFlags, uintptr_t numElements, PInvokeTransitionFrame* pTransitionFrame)
+EXTERN_C void* F_CALL_CONV RhpGcAlloc(MethodTable* pEEType, uint32_t uFlags, uintptr_t numElements, PInvokeTransitionFrame* pTransitionFrame)
 {
     Thread* pThread = ThreadStore::GetCurrentThread();
 
@@ -702,7 +701,7 @@ EXTERN_C UInt32_BOOL g_fGcStressStarted;
 UInt32_BOOL g_fGcStressStarted = UInt32_FALSE; // UInt32_BOOL because asm code reads it
 
 // static
-EXTERN_C void REDHAWK_CALLCONV RhpStressGc()
+EXTERN_C void F_CALL_CONV RhpStressGc()
 {
     // The GarbageCollect operation below may trash the last win32 error. We save the error here so that it can be
     // restored after the GC operation;