Opt, 使用Tls重写ptd_downlevel，提高性能

Directory.Build.props

@@ -18,8 +18,6 @@
       <EnableEnhancedInstructionSet Condition="'$(Platform)' == 'Win32'">NoExtensions</EnableEnhancedInstructionSet>
       <DisableSpecificWarnings>4200;%(DisableSpecificWarnings)</DisableSpecificWarnings>
       <MultiProcessorCompilation>true</MultiProcessorCompilation>
-      <!--最低支持目标为XP 2003时，则还需要禁用threadSafeInit-->
-      <AdditionalOptions Condition="'$(WindowsTargetPlatformMinVersionBuild)' &gt;= '6000'">/Zc:threadSafeInit- %(AdditionalOptions)</AdditionalOptions>
       <!--禁用FH4，因为早期版本 2015 2017都是没有这个功能的，实际上，微软自己也是禁用的。-->
       <AdditionalOptions Condition="'$(Platform)'=='x64'">/d2FH4- %(AdditionalOptions)</AdditionalOptions>
       <!--避免使用tlsGuards，因为VC-LTL能确保不会使用那些动态的。此外避免高版本依赖-->
@@ -33,4 +31,4 @@
     <DirectoryBuildPropsPath>$([MSBuild]::GetPathOfFileAbove('Directory.Build.props', '$(MSBuildThisFileDirectory)../'))</DirectoryBuildPropsPath>
   </PropertyGroup>
   <Import Project="$(DirectoryBuildPropsPath)" Condition="'$(DirectoryBuildPropsPath)' != ''"/>
-</Project>
+</Project>

Nuget/VC-LTL.nuspec

@@ -27,14 +27,10 @@
     <dependencies>
       <group targetFramework="native">
         <dependency id="YY.NuGet.Import.Helper" version="1.0.0.4" />
+        <dependency id="YY-Thunks" version="[1.0.10-Beta8,)"/>
       </group>
       <group targetFramework="net8.0-windows7.0">
-        <!--
-        YY-Thunks 1.0.10-Beta2开始才解决Windows XP、Windows Vista、Windows 7的API缺失问题
-        YY-Thunks 1.0.10-Beta3开始直接支持在.NET Native AOT使用
-        YY-Thunks 1.0.10-Beta4开始解决兼容XP时PE头部的最小系统版本不更新问题
-        -->
-        <dependency id="YY-Thunks" version="[1.0.10-Beta4,)"/>
+        <dependency id="YY-Thunks" version="[1.0.10-Beta8,)"/>
       </group>
     </dependencies>
   </metadata>

Sources/crt/ptd_downlevel.h

@@ -6,30 +6,60 @@
 
 struct ptd_downlevel
 {
-	//所需线程ID
-	unsigned dwThreadId;
+    uintptr_t bInit;
+#if WindowsTargetPlatformMinVersion < WindowsTargetPlatformWindows10_10240
 	void(__cdecl* _thread_local_iph)(
 		wchar_t const*,
 		wchar_t const*,
 		wchar_t const*,
 		unsigned int,
 		uintptr_t
 		);
-#if defined (_M_IA64) || defined (_M_AMD64) || defined (_M_ARM64) || defined (_M_ARM)
-	void* _pForeignException;
+#endif
+
+#if WindowsTargetPlatformMinVersion < WindowsTargetPlatformWindowsXP
+    int         _ProcessingThrow; /* for uncaught_exception */
+#endif
+
+#if WindowsTargetPlatformMinVersion < WindowsTargetPlatformWindows6
+    int _cxxReThrow;
+#endif
+
+#if defined (_M_IA64) || defined (_M_AMD64) || defined (_M_ARM64) || defined (_M_ARM) || defined _M_HYBRID
+#if WindowsTargetPlatformMinVersion < WindowsTargetPlatformWindows6
 	uintptr_t   _ImageBase;
 	uintptr_t _ThrowImageBase;
+    void* _pForeignException;
 #endif
-	int _cxxReThrow;
-#if defined _M_X64 || defined _M_ARM || defined _M_ARM64 || defined _M_HYBRID
-	int _CatchStateInParent;
+
+#if WindowsTargetPlatformMinVersion < __MakeVersion(10, 0, 19041)
+    int _CatchStateInParent;
 #endif
+
+#endif // defined (_M_IA64) || defined (_M_AMD64) || defined (_M_ARM64) || defined (_M_ARM)
+
 #if defined(_M_IX86)
-	int         _ProcessingThrow; /* for uncaught_exception */
+
+#if WindowsTargetPlatformMinVersion < WindowsTargetPlatformWindows6
 	void* _pFrameInfoChain;
 #endif
+#endif // defined(_M_IX86)
 };
 
-EXTERN_C ptd_downlevel* __fastcall __LTL_get_ptd_downlevel(BOOL bCanAlloc);
+#if WindowsTargetPlatformMinVersion < __MakeVersion(10, 0, 19041)
+__declspec(noinline) inline ptd_downlevel* __fastcall __LTL_get_ptd_downlevel()
+{
+    static thread_local ptd_downlevel s_ptd_downlevel;
+    if (!s_ptd_downlevel.bInit)
+    {
+        s_ptd_downlevel.bInit = 1;
+        __if_exists(ptd_downlevel::_CatchStateInParent)
+        {
+            s_ptd_downlevel._CatchStateInParent = -2;//INVALID_CATCH_SPECIFIC_STATE
+        }
+    }
 
-#endif
+    return &s_ptd_downlevel;
+}
+#endif
+#endif // _LTL_ptd_downlevel

Sources/crt/vcruntime/ehhelpers.h

@@ -14,7 +14,7 @@
 #else
 #include <ptd_downlevel.h>
 #define _pForeignExceptWin6   (*((EHExceptionRecord **)&(((_ptd_msvcrt_win6_shared*)RENAME_BASE_PTD(__vcrt_getptd)())->_pForeignException)))
-#define _pForeignExceptdownlevel  (*((EHExceptionRecord **)&(__LTL_get_ptd_downlevel(TRUE)->_pForeignException)))
+#define _pForeignExceptdownlevel  (*((EHExceptionRecord **)&(__LTL_get_ptd_downlevel()->_pForeignException)))
 #define _pForeignExcept (__LTL_GetOsMinVersion() >= MakeMiniVersion(6, 0) ? _pForeignExceptWin6 : _pForeignExceptdownlevel)
 #endif
 

Sources/crt/vcruntime/frame.cpp

@@ -47,7 +47,7 @@ using namespace FH4;
 #include <ptd_downlevel.h>
 
 #define cxxReThrowWin6   (((_ptd_msvcrt_win6_shared*)RENAME_BASE_PTD(__vcrt_getptd)())->_cxxReThrow)
-#define cxxReThrowdownlevel (__LTL_get_ptd_downlevel(TRUE)->_cxxReThrow)
+#define cxxReThrowdownlevel (__LTL_get_ptd_downlevel()->_cxxReThrow)
 #define cxxReThrow (__LTL_GetOsMinVersion() >= MakeMiniVersion(6,0) ? cxxReThrowWin6 : cxxReThrowdownlevel)
 #endif
 

Sources/crt/vcruntime/risctrnsctrl.cpp

@@ -36,7 +36,7 @@
 #else
 #include <ptd_downlevel.h>
 #define _ImageBaseWin6        (((_ptd_msvcrt_win6_shared*)__vcrt_getptd())->_ImageBase)
-#define _ImageBasedownlevel   (__LTL_get_ptd_downlevel(TRUE)->_ImageBase)
+#define _ImageBasedownlevel   (__LTL_get_ptd_downlevel()->_ImageBase)
 #define _ImageBase (__LTL_GetOsMinVersion() >= MakeMiniVersion(6,0) ? _ImageBaseWin6 : _ImageBasedownlevel)
 #endif
 extern "C" uintptr_t __cdecl _GetImageBase()
@@ -60,7 +60,7 @@ extern "C" void __cdecl _SetImageBase(uintptr_t ImageBaseToRestore)
 #else
 #include <ptd_downlevel.h>
 #define _ThrowImageBaseWin6        (((_ptd_msvcrt_win6_shared*)__vcrt_getptd())->_ThrowImageBase)
-#define _ThrowImageBasedownlevel   (__LTL_get_ptd_downlevel(TRUE)->_ThrowImageBase)
+#define _ThrowImageBasedownlevel   (__LTL_get_ptd_downlevel()->_ThrowImageBase)
 #define _ThrowImageBase (__LTL_GetOsMinVersion() >= MakeMiniVersion(6,0) ? _ThrowImageBaseWin6 : _ThrowImageBasedownlevel)
 #endif
 
@@ -76,12 +76,6 @@ extern "C" void __cdecl _SetThrowImageBase(uintptr_t NewThrowImageBase)
 
 #endif
 
-#if 1
-#if _VCRT_BUILD_FH4 && WindowsTargetPlatformMinVersion >= WindowsTargetPlatformWindows6
-thread_local int VC_LTL_UCRT_CatchStateInParent = INVALID_CATCH_SPECIFIC_STATE;
-#endif // _VCRT_BUILD_FH4 && WindowsTargetPlatformMinVersion >= WindowsTargetPlatformWindows6
-#endif // 1
-
 #if _EH_RELATIVE_FUNCINFO
 #if _VCRT_BUILD_FH4
 //

Sources/crt/vcruntime/trnsctrl.h

@@ -262,14 +262,12 @@ extern "C" _VCRTIMP int*   __cdecl __processing_throw();
 #if _VCRT_BUILD_FH4
 // In the satellite build, __vcrt_getptd uses satellite's PTD
 // In the non-DLL build, __vcrt_getptd uses the main PTD which was updated to have this field
-#if WindowsTargetPlatformMinVersion >= WindowsTargetPlatformWindows6
-extern thread_local int VC_LTL_UCRT_CatchStateInParent;
-#define CatchStateInParent      (VC_LTL_UCRT_CatchStateInParent)
+#if WindowsTargetPlatformMinVersion >= __MakeVersion(10, 0, 19041)
+#define CatchStateInParent (RENAME_BASE_PTD(__vcrt_getptd)()->_CatchStateInParent)
 #else
 #include <ptd_downlevel.h>
-#define CatchStateInParent (__LTL_get_ptd_downlevel(TRUE)->_CatchStateInParent)
+#define CatchStateInParent (__LTL_get_ptd_downlevel()->_CatchStateInParent)
 #endif
-
 #endif
 
 #pragma pack(pop)

Sources/crt/vcruntime/vcruntime_internal.h

@@ -209,10 +209,11 @@ typedef struct RENAME_BASE_PTD(__vcrt_ptd)
     uintptr_t          _ThrowImageBase;
     void*              _pForeignException;
 
-    //2015 没有这个
-    //int                _CatchStateInParent;   // Used to link together the catch funclet with the parent. During dispatch contains state associated
+#if WindowsTargetPlatformMinVersion >= __MakeVersion(10, 0, 19041)
+    int                _CatchStateInParent;   // Used to link together the catch funclet with the parent. During dispatch contains state associated
                                               // with catch in the parent. During unwind represents the current unwind state that is resumed to
                                               // during collided unwind and used to look for handlers of the throwing dtor.
+#endif
     #elif defined _M_IX86
     void*              _pFrameInfoChain;
     #endif

Sources/ucrt/misc/invalid_parameter.cpp

@@ -15,11 +15,9 @@ static _invalid_parameter_handler __acrt_invalid_parameter_handler = nullptr;
 
 #if WindowsTargetPlatformMinVersion >= WindowsTargetPlatformWindows10_10240
 #define _thread_local_iph _get_thread_local_invalid_parameter_handler()
-#elif WindowsTargetPlatformMinVersion >= WindowsTargetPlatformWindows6
-static thread_local _invalid_parameter_handler _thread_local_iph;
 #else
 #include <ptd_downlevel.h>
-#define _thread_local_iph (__LTL_get_ptd_downlevel(TRUE)->_thread_local_iph)
+#define _thread_local_iph (__LTL_get_ptd_downlevel()->_thread_local_iph)
 #endif
 
 #if defined _M_X64 && !defined _UCRT_ENCLAVE_BUILD
@@ -328,4 +326,4 @@ extern "C" _invalid_parameter_handler __cdecl _get_thread_local_invalid_paramete
 }
 
 _LCRT_DEFINE_IAT_SYMBOL(_get_thread_local_invalid_parameter_handler);
-#endif
+#endif

config/config.props

@@ -132,8 +132,10 @@
       <PreprocessorDefinitions>_Build_By_LTL=1;_LTL_Core_Version=$(LTL_CoreVersion);%(PreprocessorDefinitions)</PreprocessorDefinitions>
 
       <PreprocessorDefinitions Condition="'$(SupportWinXP)'=='true'">_ATL_XP_TARGETING=1;%(PreprocessorDefinitions)</PreprocessorDefinitions>
-      <!--当兼容XP时，对于非 exe 的需要禁用线程安全初始化。避免XP中在DllMain中使用TLS而崩溃-->
+      <!--YY-Thunks 1.0.10-Beta8开始已经为Windows XP添加了完整的Tls支持，所以无需在`/Zc:threadSafeInit-`。
+      当兼容XP时，对于非 exe 的需要禁用线程安全初始化。避免XP中在DllMain中使用TLS而崩溃
       <AdditionalOptions Condition="'$(SupportWinXP)'=='true' and '$(ConfigurationType)'!='Application' ">/Zc:threadSafeInit- %(AdditionalOptions)</AdditionalOptions>
+      -->
     </ClCompile>
     <Link Condition="'$(SupportWinXP)'=='true'">
       <MinimumRequiredVersion Condition=" '$(PlatformShortName)'=='x86' ">5.01</MinimumRequiredVersion>

ucrtbase.msvcrt/localeconv.cpp

@@ -87,14 +87,8 @@ extern "C" lconv* __cdecl localeconv(void)
 
     auto _Page = ___lc_codepage_func();
 
-#if WindowsTargetPlatformMinVersion < WindowsTargetPlatformWindows6
-    //低于Vista，那么无法安全的使用 thread_local
-    static lconv_with_static_buffer ThreadBuffer[0x4000];
-
-    auto& Current = ThreadBuffer[(GetCurrentThreadId() >> 1) % __crt_countof(ThreadBuffer)];
-#else
+    // 由YY-Thunks为Windows XP提供完整的thread_local支持
     static thread_local lconv_with_static_buffer Current;
-#endif
 
     if (Current.decimal_point != pOrglconv->decimal_point)
     {

ucrtbase.msvcrt/ucrtbase.msvcrt.vcxproj

@@ -658,9 +658,6 @@
     </Lib>
   </ItemDefinitionGroup>
   <ItemGroup>
-    <ClCompile Include="..\Sources\crt\ptd_downlevel.cpp">
-      <ExcludedFromBuild Condition="'$(WindowsTargetPlatformMinVersionBuild)' &gt;= '6000'">true</ExcludedFromBuild>
-    </ClCompile>
     <ClCompile Include="..\Sources\ltl\debug_heap.cpp">
       <ExcludedFromBuild Condition="'$(UseDebugLibraries)' != 'true'">true</ExcludedFromBuild>
     </ClCompile>
@@ -1296,6 +1293,7 @@
     <ClCompile Include="_file.cpp" />
   </ItemGroup>
   <ItemGroup>
+    <ClInclude Include="..\Sources\crt\ptd_downlevel.h" />
     <ClInclude Include="msvcrt_build_type.h" />
     <ClInclude Include="msvcrt_loadhelper.h" />
   </ItemGroup>

ucrtbase.msvcrt/ucrtbase.msvcrt.vcxproj.filters

@@ -765,9 +765,6 @@
     <ClCompile Include="..\Sources\ucrt\stdlib\rand_s.cpp">
       <Filter>源文件\WinXP</Filter>
     </ClCompile>
-    <ClCompile Include="..\Sources\crt\ptd_downlevel.cpp">
-      <Filter>头文件\ltl</Filter>
-    </ClCompile>
     <ClCompile Include="..\Sources\ucrt\locale\GetStringTypeA.cpp">
       <Filter>源文件\objs</Filter>
     </ClCompile>
@@ -806,5 +803,8 @@
     <ClInclude Include="msvcrt_build_type.h">
       <Filter>头文件\ltl</Filter>
     </ClInclude>
+    <ClInclude Include="..\Sources\crt\ptd_downlevel.h">
+      <Filter>头文件\ltl</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>