Attachment 262630 Details for Bug 145366 – Patch

[patch] Patch

bug-145366-20151007223110.patch (text/plain), 65.81 KB, created by peavo on 2015-10-07 13:31:58 PDT

(hide)

Description:

Filename:

MIME Type:

Creator: peavo

Created: 2015-10-07 13:31:58 PDT

Size: 65.81 KB

patch

obsolete

>Index: ChangeLog
>===================================================================
>--- ChangeLog	(revision 190665)
>+++ ChangeLog	(working copy)
>@@ -1,3 +1,14 @@
>+2015-10-07  Per Arne Vollan  <peavo@outlook.com>
>+
>+        FTL is not working on Windows.
>+        https://bugs.webkit.org/show_bug.cgi?id=145366
>+
>+        Reviewed by NOBODY (OOPS!).
>+
>+        Enable FTL.
>+
>+        * Source/cmake/OptionsWin.cmake:
>+
> 2015-10-06  Emanuele Aina  <emanuele.aina@collabora.com>
> 
>         Fix ENABLE_OPENGL=OFF builds
>Index: Source/JavaScriptCore/ChangeLog
>===================================================================
>--- Source/JavaScriptCore/ChangeLog	(revision 190665)
>+++ Source/JavaScriptCore/ChangeLog	(working copy)
>@@ -1,3 +1,94 @@
>+2015-10-07  Per Arne Vollan  <peavo@outlook.com>
>+
>+        FTL is not working on Windows.
>+        https://bugs.webkit.org/show_bug.cgi?id=145366
>+
>+        Reviewed by NOBODY (OOPS!).
>+
>+        The purpose of this patch is to get FTL up and running on Windows.
>+        It mainly contains adjustments concerning Win64 calling conventions,
>+        and fixes for register preservation issues on Windows.
>+        Also, it is important to set a llvm target triple which llvm supports
>+        for Windows.
>+
>+        * CMakeLists.txt: Compile fixes.
>+        * JavaScriptCore.vcxproj/JavaScriptCoreCommon.props: Remove duplicate.
>+        * JavaScriptCore.vcxproj/libllvmForJSC/libllvmForJSC.props:
>+        * assembler/MacroAssemblerX86Common.h:
>+        (JSC::MacroAssemblerX86Common::loadDouble128): Added.
>+        (JSC::MacroAssemblerX86Common::storeDouble128): Added.
>+        * assembler/MacroAssemblerX86_64.h:
>+        (JSC::MacroAssemblerX86_64::callJIT): Added.
>+        (JSC::MacroAssemblerX86_64::callSlowPath): Added.
>+        * assembler/X86Assembler.h:
>+        (JSC::X86Assembler::movups_rm): Added.
>+        (JSC::X86Assembler::movups_mr): Added.
>+        * dfg/DFGOSREntry.cpp:
>+        (JSC::DFG::prepareOSREntry): Save callee saved fp registers.
>+        * dfg/DFGOSRExitCompiler.cpp:
>+        (JSC::DFG::OSRExitCompiler::emitRestoreArguments):
>+        Allocate stack shadow space for parameters, as required by Win64 ABI.
>+        * dfg/DFGOSRExitCompilerCommon.cpp:
>+        (JSC::DFG::handleExitCounts): Ditto.
>+        (JSC::DFG::osrWriteBarrier): Ditto.
>+        * disassembler/LLVMDisassembler.cpp:
>+        (JSC::tryToDisassembleWithLLVM): Set llvm target triple for Win64.
>+        * ftl/FTLCapabilities.cpp:
>+        (JSC::FTL::canCompile): llvm does not support powi on Windows.
>+        * ftl/FTLCompile.cpp: Set llvm target triple for Win64.
>+        (JSC::FTL::mmAllocateDataSection): We have same section names as Linux.
>+        * ftl/FTLInlineCacheSize.cpp:
>+        (JSC::FTL::sizeOfCallVarargs): Adjusted ic size for varargs on Win64.
>+        * ftl/FTLJITCode.h: Define section name prefix.
>+        (JSC::FTL::OutOfLineCodeInfo::OutOfLineCodeInfo): Added move constructor.
>+        * ftl/FTLLowerDFGToLLVM.cpp:
>+        (JSC::FTL::DFG::LowerDFGToLLVM::compileArithPow): llvm does not support powi on Windows.
>+        * ftl/FTLOSRExitCompiler.cpp:
>+        (JSC::FTL::compileStub): Allocate stack shadow space for parameters.
>+        * ftl/FTLSlowPathCall.cpp: Allocate enough space for call frame.
>+        Use new Win64 method to call JIT code.
>+        * ftl/FTLSlowPathCallKey.h: Added argument count member.
>+        (JSC::FTL::SlowPathCallKey::SlowPathCallKey):
>+        (JSC::FTL::SlowPathCallKey::callTarget):
>+        (JSC::FTL::SlowPathCallKey::argumentRegisters):
>+        (JSC::FTL::SlowPathCallKey::offset):
>+        (JSC::FTL::SlowPathCallKey::numArgs):
>+        (JSC::FTL::SlowPathCallKey::withCallTarget):
>+        * ftl/FTLThunks.cpp: Use new Win64 method to call slow path.
>+        (JSC::FTL::osrExitGenerationThunkGenerator):
>+        (JSC::FTL::slowPathCallThunkGenerator):
>+        * ftl/FTLUnwindInfo.cpp:
>+        (JSC::FTL::parseUnwindInfo): Adjust saved registers stack locations.
>+        * jit/AssemblyHelpers.h:
>+        (JSC::AssemblyHelpers::emitSaveCalleeSavesFor): Save callee saved fp registers.
>+        (JSC::AssemblyHelpers::emitRestoreCalleeSavesFor): Restore callee saved fp registers.
>+        (JSC::AssemblyHelpers::debugCall): Allocate stack shadow space for parameters.
>+        * jit/FPRInfo.h: Add callee saved fp registers on Windows.
>+        * jit/GPRInfo.h: Add callee saved fp register count to callee saved register count.
>+        * jit/RegisterSet.cpp: Add callee saved registers for Win64.
>+        (JSC::RegisterSet::calleeSaveRegisters):
>+        (JSC::RegisterSet::vmCalleeSaveRegisters):
>+        (JSC::RegisterSet::llintBaselineCalleeSaveRegisters):
>+        (JSC::RegisterSet::dfgCalleeSaveRegisters):
>+        (JSC::RegisterSet::ftlCalleeSaveRegisters):
>+        * jit/ThunkGenerators.cpp:
>+        (JSC::throwExceptionFromCallSlowPathGenerator): Allocate stack shadow space.
>+        * llint/LLIntData.cpp:
>+        (JSC::LLInt::Data::performAssertions):
>+        * llint/LowLevelInterpreter.asm:
>+        * llvm/InitializeLLVM.cpp:
>+        (JSC::initializeLLVMImpl):
>+        (JSC::initializeLLVM):
>+        * llvm/LLVMAPIFunctions.h: Compile fix, remove llvm api function which has changed
>+        signature.
>+        * llvm/library/LLVMExports.cpp:
>+        (initializeAndGetJSCLLVMAPI): Set correct dll name for Windows.
>+        * llvm/library/LLVMOverrides.cpp:
>+        (abort): Link fix, abort(), etc. is already included in run time library.
>+        * llvm/library/LLVMTrapCallback.h: Compile fix.
>+        * offlineasm/registers.rb: Added two more callee saved fp registers.
>+        * offlineasm/x86.rb: Save and restore callee saved fp registers.
>+
> 2015-10-07  Xabier Rodriguez Calvar  <calvaris@igalia.com> and Youenn Fablet  <youenn.fablet@crf.canon.fr>
> 
>         Automate WebCore JS builtins generation and build system
>Index: Source/JavaScriptCore/CMakeLists.txt
>===================================================================
>--- Source/JavaScriptCore/CMakeLists.txt	(revision 190665)
>+++ Source/JavaScriptCore/CMakeLists.txt	(working copy)
>@@ -843,6 +843,12 @@ list(APPEND JavaScriptCore_SOURCES
> 
> 
> if (ENABLE_FTL_JIT)
>+    # The llvm-config utility used to find llvm libraries does not work properly on Windows, it returns file names with ".a" suffix.
>+    # We therefore set the library dependencies manually here
>+    if (WIN32)
>+        set(LLVM_STATIC_LIBRARIES "LLVMAnalysis.lib;LLVMAsmParser.lib;LLVMAsmPrinter.lib;LLVMBitReader.lib;LLVMBitWriter.lib;LLVMCodeGen.lib;LLVMCore.lib;LLVMDebugInfo.lib;LLVMExecutionEngine.lib;LLVMInstCombine.lib;LLVMInstrumentation.lib;LLVMInterpreter.lib;LLVMipa.lib;LLVMipo.lib;LLVMIRReader.lib;LLVMJIT.lib;LLVMLinker.lib;LLVMLTO.lib;LLVMMC.lib;LLVMMCDisassembler.lib;LLVMMCJIT.lib;LLVMMCParser.lib;LLVMObjCARCOpts.lib;LLVMObject.lib;LLVMOption.lib;LLVMRuntimeDyld.lib;LLVMScalarOpts.lib;LLVMSelectionDAG.lib;LLVMSupport.lib;LLVMTableGen.lib;LLVMTarget.lib;LLVMTransformUtils.lib;LLVMVectorize.lib;LLVMX86AsmParser.lib;LLVMX86AsmPrinter.lib;LLVMX86CodeGen.lib;LLVMX86Desc.lib;LLVMX86Disassembler.lib;LLVMX86Info.lib;LLVMX86Utils.lib;LLVMProfileData.lib")
>+    endif ()
>+
>     if (NOT LLVM_STATIC_LIBRARIES)
>         message(FATAL_ERROR "No LLVM libs found, but FTL is enabled. Stopping build.")
>     endif ()
>@@ -866,7 +872,17 @@ if (ENABLE_FTL_JIT)
> 
>     WEBKIT_WRAP_SOURCELIST(${llvmForJSC_SOURCES})
>     add_library(llvmForJSC SHARED ${llvmForJSC_SOURCES} ${DERIVED_SOURCES_JAVASCRIPTCORE_DIR}/WebKitLLVMLibraryToken.h)
>-    target_link_libraries(llvmForJSC ${LLVM_STATIC_LIBRARIES} "pthread" "dl" -Wl,--version-script=${JAVASCRIPTCORE_DIR}/llvm/library/libllvmForJSC.version)
>+    if (WIN32)
>+        # For FTL to work properly on Windows you'll need to build with llvm revision 241725 or later.
>+        # The llvm project files for 64-bit are generated with cmake:
>+        # C:\llvm> "\Program Files (x86)\CMake\bin\cmake.exe" -G "Visual Studio 14 2015 Win64" .
>+        # When the build is finished, copy C:\llvm\include\llvm and C:\llvm\include\llvm-c to WEBKIT_LIBRARIES\include,
>+        # and copy the libraries from C:\llvm\Release\lib to WEBKIT_LIBRARIES\lib64.
>+        set_target_properties(llvmForJSC PROPERTIES COMPILE_FLAGS "/MD")
>+        target_link_libraries(llvmForJSC ${LLVM_STATIC_LIBRARIES})
>+    else ()
>+        target_link_libraries(llvmForJSC ${LLVM_STATIC_LIBRARIES} "pthread" "dl" -Wl,--version-script=${JAVASCRIPTCORE_DIR}/llvm/library/libllvmForJSC.version)
>+    endif ()
> 
>     # Added extra items for JavaScriptCore
>     list(APPEND JavaScriptCore_SYSTEM_INCLUDE_DIRECTORIES
>@@ -924,10 +940,19 @@ if (ENABLE_FTL_JIT)
>         ftl/FTLValueRange.cpp
> 
>         llvm/InitializeLLVM.cpp
>-        llvm/InitializeLLVMLinux.cpp
>-        llvm/InitializeLLVMPOSIX.cpp
>-        llvm/LLVMAPI.cpp
>     )
>+    if (WIN32)
>+        list(APPEND JavaScriptCore_SOURCES
>+            llvm/InitializeLLVMWin.cpp
>+            llvm/LLVMAPI.cpp
>+        )
>+    else ()
>+        list(APPEND JavaScriptCore_SOURCES
>+            llvm/InitializeLLVMLinux.cpp
>+            llvm/InitializeLLVMPOSIX.cpp
>+            llvm/LLVMAPI.cpp
>+        )
>+    endif ()
> endif ()
> 
> set(HASH_LUT_GENERATOR ${CMAKE_CURRENT_SOURCE_DIR}/create_hash_table)
>Index: Source/JavaScriptCore/JavaScriptCore.vcxproj/JavaScriptCoreCommon.props
>===================================================================
>--- Source/JavaScriptCore/JavaScriptCore.vcxproj/JavaScriptCoreCommon.props	(revision 190665)
>+++ Source/JavaScriptCore/JavaScriptCore.vcxproj/JavaScriptCoreCommon.props	(working copy)
>@@ -5,7 +5,6 @@
>   <PropertyGroup />
>   <ItemDefinitionGroup>
>     <ClCompile>
>-      <DisableSpecificWarnings>4611;%(DisableSpecificWarnings)</DisableSpecificWarnings>
>       <AdditionalIncludeDirectories>..\;..\tools\;..\runtime\;..\llint\;..\jit\;..\disassembler\;..\heap\;..\debugger\;..\assembler\;..\profiler\;..\yarr\;..\interpreter\;..\bytecode\;..\builtins\;..\dfg\;..\bytecompiler\;..\parser\;..\API\;..\ftl\;..\bindings\;..\inspector\;..\wasm\;..\ftl;..\llvm;..\llvm\library;$(ConfigurationBuildDir)\obj$(PlatformArchitecture)\JavaScriptCore\DerivedSources\;$(ConfigurationBuildDir)\include\;$(ConfigurationBuildDir)\include\JavaScriptCore\;$(ConfigurationBuildDir)\include\private\;$(WebKit_Libraries)\include;$(WebKit_Libraries)\include\private;%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
>       <ForcedIncludeFiles>ICUVersion.h;%(ForcedIncludeFiles)</ForcedIncludeFiles>
>       <DisableSpecificWarnings>4611;%(DisableSpecificWarnings)</DisableSpecificWarnings>
>Index: Source/JavaScriptCore/JavaScriptCore.vcxproj/libllvmForJSC/libllvmForJSC.props
>===================================================================
>--- Source/JavaScriptCore/JavaScriptCore.vcxproj/libllvmForJSC/libllvmForJSC.props	(revision 190665)
>+++ Source/JavaScriptCore/JavaScriptCore.vcxproj/libllvmForJSC/libllvmForJSC.props	(working copy)
>@@ -2,16 +2,13 @@
> <Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
>   <ImportGroup Label="PropertySheets" />
>   <PropertyGroup Label="UserMacros" />
>-  <PropertyGroup />
>   <ItemDefinitionGroup>
>     <ClCompile>
>       <AdditionalIncludeDirectories>..\;..\..\;..\..\llvm;$(WebKit_Libraries)\include;$(WebKit_Libraries)\include\private;$(ConfigurationBuildDir)\include\;$(ConfigurationBuildDir)\include\JavaScriptCore\;$(ConfigurationBuildDir)\include\private\;$(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
>     </ClCompile>
>     <Link>
>-      <AdditionalLibraryDirectories>$(WebKit_Libraries)\lib$(PlatformArchitecture)\llvm</AdditionalLibraryDirectories>
>-      <AdditionalDependencies>LLVMAnalysis.lib;LLVMAsmParser.lib;LLVMAsmPrinter.lib;LLVMBitReader.lib;LLVMBitWriter.lib;LLVMCodeGen.lib;LLVMCore.lib;LLVMDebugInfo.lib;LLVMExecutionEngine.lib;LLVMInstCombine.lib;LLVMInstrumentation.lib;LLVMInterpreter.lib;LLVMipa.lib;LLVMipo.lib;LLVMIRReader.lib;LLVMJIT.lib;LLVMLinker.lib;LLVMLTO.lib;LLVMMC.lib;LLVMMCDisassembler.lib;LLVMMCJIT.lib;LLVMMCParser.lib;LLVMObjCARCOpts.lib;LLVMObject.lib;LLVMOption.lib;LLVMRuntimeDyld.lib;LLVMScalarOpts.lib;LLVMSelectionDAG.lib;LLVMSupport.lib;LLVMTableGen.lib;LLVMTarget.lib;LLVMTransformUtils.lib;LLVMVectorize.lib;LLVMX86AsmParser.lib;LLVMX86AsmPrinter.lib;LLVMX86CodeGen.lib;LLVMX86Desc.lib;LLVMX86Disassembler.lib;LLVMX86Info.lib;LLVMX86Utils.lib;%(AdditionalDependencies)</AdditionalDependencies>
>-      <GenerateDebugInformation>true</GenerateDebugInformation>
>+      <AdditionalLibraryDirectories>$(WebKit_Libraries)\lib$(PlatformArchitecture)</AdditionalLibraryDirectories>
>+      <AdditionalDependencies>LLVMAnalysis.lib;LLVMAsmParser.lib;LLVMAsmPrinter.lib;LLVMBitReader.lib;LLVMBitWriter.lib;LLVMCodeGen.lib;LLVMCore.lib;LLVMDebugInfo.lib;LLVMExecutionEngine.lib;LLVMInstCombine.lib;LLVMInstrumentation.lib;LLVMInterpreter.lib;LLVMipa.lib;LLVMipo.lib;LLVMIRReader.lib;LLVMJIT.lib;LLVMLinker.lib;LLVMLTO.lib;LLVMMC.lib;LLVMMCDisassembler.lib;LLVMMCJIT.lib;LLVMMCParser.lib;LLVMObjCARCOpts.lib;LLVMObject.lib;LLVMOption.lib;LLVMRuntimeDyld.lib;LLVMScalarOpts.lib;LLVMSelectionDAG.lib;LLVMSupport.lib;LLVMTableGen.lib;LLVMTarget.lib;LLVMTransformUtils.lib;LLVMVectorize.lib;LLVMX86AsmParser.lib;LLVMX86AsmPrinter.lib;LLVMX86CodeGen.lib;LLVMX86Desc.lib;LLVMX86Disassembler.lib;LLVMX86Info.lib;LLVMX86Utils.lib;LLVMProfileData.lib;%(AdditionalDependencies)</AdditionalDependencies>
>     </Link>
>   </ItemDefinitionGroup>
>-  <ItemGroup />
> </Project>
>Index: Source/JavaScriptCore/assembler/MacroAssemblerX86Common.h
>===================================================================
>--- Source/JavaScriptCore/assembler/MacroAssemblerX86Common.h	(revision 190665)
>+++ Source/JavaScriptCore/assembler/MacroAssemblerX86Common.h	(working copy)
>@@ -712,7 +712,15 @@ public:
>         ASSERT(isSSE2Present());
>         m_assembler.movsd_mr(address.offset, address.base, dest);
>     }
>-    
>+
>+#if CPU(X86_64)
>+    void loadDouble128(ImplicitAddress address, FPRegisterID src)
>+    {
>+        ASSERT(isSSE2Present());
>+        m_assembler.movups_mr(src, address.offset, address.base);
>+    }
>+#endif
>+
>     void loadDouble(BaseIndex address, FPRegisterID dest)
>     {
>         ASSERT(isSSE2Present());
>@@ -729,7 +737,15 @@ public:
>         ASSERT(isSSE2Present());
>         m_assembler.movsd_rm(src, address.offset, address.base);
>     }
>-    
>+
>+#if CPU(X86_64)
>+    void storeDouble128(FPRegisterID src, ImplicitAddress address)
>+    {
>+        ASSERT(isSSE2Present());
>+        m_assembler.movups_rm(src, address.offset, address.base);
>+    }
>+#endif
>+
>     void storeDouble(FPRegisterID src, BaseIndex address)
>     {
>         ASSERT(isSSE2Present());
>Index: Source/JavaScriptCore/assembler/MacroAssemblerX86_64.h
>===================================================================
>--- Source/JavaScriptCore/assembler/MacroAssemblerX86_64.h	(revision 190665)
>+++ Source/JavaScriptCore/assembler/MacroAssemblerX86_64.h	(working copy)
>@@ -186,6 +186,43 @@ public:
>     }
> #endif
> 
>+#if OS(WINDOWS)
>+    Call callJIT()
>+    {
>+        DataLabelPtr label = moveWithPatch(TrustedImmPtr(0), scratchRegister);
>+        Call result = Call(m_assembler.call(scratchRegister), Call::Linkable);
>+
>+        ASSERT_UNUSED(label, differenceBetween(label, result) == REPTACH_OFFSET_CALL_R11);
>+        return result;
>+    }
>+
>+    Call callSlowPath(unsigned numArgs)
>+    {
>+        if (numArgs >= 5) {
>+            // Copy argument 5
>+            load64(Address(X86Registers::esp, 4 * sizeof(int64_t)), scratchRegister);
>+            store64(scratchRegister, Address(X86Registers::esp, -4 * static_cast<int32_t>(sizeof(int64_t))));
>+        }
>+        if (numArgs >= 6) {
>+            // Copy argument 6
>+            load64(Address(X86Registers::esp, 5 * sizeof(int64_t)), scratchRegister);
>+            store64(scratchRegister, Address(X86Registers::esp, -3 * static_cast<int32_t>(sizeof(int64_t))));
>+        }
>+
>+        // We need to allocate the shadow space on the stack for the 4 parameter registers.
>+        sub64(TrustedImm32(8 * sizeof(int64_t)), X86Registers::esp);
>+
>+        DataLabelPtr label = moveWithPatch(TrustedImmPtr(0), scratchRegister);
>+        Call result = Call(m_assembler.call(scratchRegister), Call::Linkable);
>+
>+        add64(TrustedImm32(8 * sizeof(int64_t)), X86Registers::esp);
>+
>+        ASSERT_UNUSED(label, differenceBetween(label, result) == REPTACH_OFFSET_CALL_R11);
>+        return result;
>+    }
>+
>+#endif
>+
>     Call call()
>     {
> #if OS(WINDOWS)
>Index: Source/JavaScriptCore/assembler/X86Assembler.h
>===================================================================
>--- Source/JavaScriptCore/assembler/X86Assembler.h	(revision 190665)
>+++ Source/JavaScriptCore/assembler/X86Assembler.h	(working copy)
>@@ -1836,6 +1836,16 @@ public:
>         m_formatter.prefix(PRE_SSE_66);
>         m_formatter.twoByteOp64(OP2_MOVD_VdEd, (RegisterID)dst, src);
>     }
>+
>+    void movups_rm(XMMRegisterID src, int offset, RegisterID base)
>+    {
>+        m_formatter.twoByteOp(OP2_MOVSD_WsdVsd, (RegisterID)src, base, offset);
>+    }
>+
>+    void movups_mr(XMMRegisterID src, int offset, RegisterID base)
>+    {
>+        m_formatter.twoByteOp(OP2_MOVSD_VsdWsd, (RegisterID)src, base, offset);
>+    }
> #endif
> 
>     void movsd_rr(XMMRegisterID src, XMMRegisterID dst)
>Index: Source/JavaScriptCore/dfg/DFGOSREntry.cpp
>===================================================================
>--- Source/JavaScriptCore/dfg/DFGOSREntry.cpp	(revision 190665)
>+++ Source/JavaScriptCore/dfg/DFGOSREntry.cpp	(working copy)
>@@ -313,7 +313,11 @@ void* prepareOSREntry(ExecState* exec, C
> #if NUMBER_OF_CALLEE_SAVES_REGISTERS > 0
>     RegisterAtOffsetList* registerSaveLocations = codeBlock->calleeSaveRegisters();
>     RegisterAtOffsetList* allCalleeSaves = vm->getAllCalleeSaveRegisterOffsets();
>+#if OS(WINDOWS) && CPU(X86_64)
>+    RegisterSet dontSaveRegisters = RegisterSet::stackRegisters();
>+#else
>     RegisterSet dontSaveRegisters = RegisterSet(RegisterSet::stackRegisters(), RegisterSet::allFPRs());
>+#endif
> 
>     unsigned registerCount = registerSaveLocations->size();
>     for (unsigned i = 0; i < registerCount; i++) {
>Index: Source/JavaScriptCore/dfg/DFGOSRExitCompiler.cpp
>===================================================================
>--- Source/JavaScriptCore/dfg/DFGOSRExitCompiler.cpp	(revision 190665)
>+++ Source/JavaScriptCore/dfg/DFGOSRExitCompiler.cpp	(working copy)
>@@ -101,7 +101,13 @@ void OSRExitCompiler::emitRestoreArgumen
>             RELEASE_ASSERT_NOT_REACHED();
>             break;
>         }
>+#if OS(WINDOWS) && CPU(X86_64)
>+        m_jit.sub64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>         m_jit.call(GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+        m_jit.add64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>         m_jit.storeCell(GPRInfo::returnValueGPR, AssemblyHelpers::addressFor(operand));
>         
>         alreadyAllocatedArguments.add(id, operand);
>Index: Source/JavaScriptCore/dfg/DFGOSRExitCompilerCommon.cpp
>===================================================================
>--- Source/JavaScriptCore/dfg/DFGOSRExitCompilerCommon.cpp	(revision 190665)
>+++ Source/JavaScriptCore/dfg/DFGOSRExitCompilerCommon.cpp	(working copy)
>@@ -101,7 +101,13 @@ void handleExitCounts(CCallHelpers& jit,
>     jit.move(AssemblyHelpers::TrustedImmPtr(&exit), GPRInfo::argumentGPR1);
> #endif
>     jit.move(AssemblyHelpers::TrustedImmPtr(bitwise_cast<void*>(triggerReoptimizationNow)), GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+    jit.sub64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>     jit.call(GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+    jit.add64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>     AssemblyHelpers::Jump doneAdjusting = jit.jump();
>     
>     tooFewFails.link(&jit);
>@@ -257,7 +263,13 @@ static void osrWriteBarrier(CCallHelpers
> 
>     jit.setupArgumentsWithExecState(owner);
>     jit.move(MacroAssembler::TrustedImmPtr(reinterpret_cast<void*>(operationOSRWriteBarrier)), scratch);
>+#if OS(WINDOWS) && CPU(X86_64)
>+    jit.sub64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>     jit.call(scratch);
>+#if OS(WINDOWS) && CPU(X86_64)
>+    jit.add64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
> 
> #if CPU(X86)
>     jit.addPtr(MacroAssembler::TrustedImm32(sizeof(void*) * 3), MacroAssembler::stackPointerRegister);
>Index: Source/JavaScriptCore/disassembler/LLVMDisassembler.cpp
>===================================================================
>--- Source/JavaScriptCore/disassembler/LLVMDisassembler.cpp	(revision 190665)
>+++ Source/JavaScriptCore/disassembler/LLVMDisassembler.cpp	(working copy)
>@@ -31,6 +31,7 @@
> #include "InitializeLLVM.h"
> #include "LLVMAPI.h"
> #include "MacroAssemblerCodeRef.h"
>+#include <wtf/StringExtras.h>
> 
> namespace JSC {
> 
>@@ -80,11 +81,14 @@ bool tryToDisassembleWithLLVM(
>     const MacroAssemblerCodePtr& codePtr, size_t size, const char* prefix, PrintStream& out,
>     InstructionSubsetHint)
> {
>-    initializeLLVM();
>+    if (!initializeLLVM())
>+        return false;
>     
>     const char* triple;
>-#if CPU(X86_64)
>+#if CPU(X86_64) && OS(DARWIN)
>     triple = "x86_64-apple-darwin";
>+#elif CPU(X86_64) && OS(WINDOWS)
>+    triple = "x86_64-pc-win32-elf";
> #elif CPU(X86)
>     triple = "x86-apple-darwin";
> #elif CPU(ARM64)
>Index: Source/JavaScriptCore/ftl/FTLCapabilities.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLCapabilities.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLCapabilities.cpp	(working copy)
>@@ -92,7 +92,6 @@ inline CapabilityLevel canCompile(Node*
>     case ArithAbs:
>     case ArithSin:
>     case ArithCos:
>-    case ArithPow:
>     case ArithRound:
>     case ArithSqrt:
>     case ArithLog:
>@@ -205,6 +204,13 @@ inline CapabilityLevel canCompile(Node*
>     case TypeOf:
>         // These are OK.
>         break;
>+    case ArithPow:
>+#if OS(WINDOWS)
>+        // LLVM does not support powi on Windows.
>+        if (node->child2().useKind() == Int32Use)
>+            return CannotCompile;
>+#endif
>+        break;
>     case Identity:
>         // No backend handles this because it will be optimized out. But we may check
>         // for capabilities before optimization. It would be a deep error to remove this
>Index: Source/JavaScriptCore/ftl/FTLCompile.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLCompile.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLCompile.cpp	(working copy)
>@@ -36,6 +36,7 @@
> #include "DFGGraphSafepoint.h"
> #include "DataView.h"
> #include "Disassembler.h"
>+#include "FTLDataSection.h"
> #include "FTLExitThunkGenerator.h"
> #include "FTLInlineCacheSize.h"
> #include "FTLJITCode.h"
>@@ -44,6 +45,7 @@
> #include "JITStubs.h"
> #include "LLVMAPI.h"
> #include "LinkBuffer.h"
>+#include <wtf/StringExtras.h>
> 
> namespace JSC { namespace FTL {
> 
>@@ -105,7 +107,7 @@ static uint8_t* mmAllocateDataSection(
>         state.dataSectionNames.append(sectionName);
> #if OS(DARWIN)
>         if (!strcmp(sectionName, SECTION_NAME("compact_unwind"))) {
>-#elif OS(LINUX)
>+#elif OS(LINUX) || OS(WINDOWS)
>         if (!strcmp(sectionName, SECTION_NAME("eh_frame"))) {
> #else
> #error "Unrecognized OS"
>@@ -416,7 +418,17 @@ static void fixFunctionBasedOnStackMaps(
>                 // It was optimized out.
>                 continue;
>             }
>-            
>+
>+            if (verboseCompilationEnabled()) {
>+                for (unsigned i = 0; i < iter->value.size(); ++i) {
>+                    StackMaps::Record& record = iter->value[i];
>+
>+                    RegisterSet usedRegisters = usedRegistersFor(record);
>+
>+                    dataLog("Used registers ", usedRegisters, "\n");
>+                }
>+            }
>+
>             info.m_thunkAddress = linkBuffer->locationOf(info.m_thunkLabel);
>             exit.m_patchableCodeOffset = linkBuffer->offsetOf(info.m_thunkJump);
>             
>@@ -461,7 +473,10 @@ static void fixFunctionBasedOnStackMaps(
>                 StackMaps::Record& record = iter->value[i];
>             
>                 RegisterSet usedRegisters = usedRegistersFor(record);
>-                
>+
>+                if (verboseCompilationEnabled())
>+                    dataLog("Used registers ", usedRegisters, "\n");
>+
>                 GPRReg result = record.locations[0].directGPR();
>                 GPRReg base = record.locations[1].directGPR();
>                 
>@@ -499,7 +514,10 @@ static void fixFunctionBasedOnStackMaps(
>                 StackMaps::Record& record = iter->value[i];
>                 
>                 RegisterSet usedRegisters = usedRegistersFor(record);
>-                
>+
>+                if (verboseCompilationEnabled())
>+                    dataLog("Used registers ", usedRegisters, "\n");
>+
>                 GPRReg base = record.locations[0].directGPR();
>                 GPRReg value = record.locations[1].directGPR();
>                 
>@@ -594,6 +612,20 @@ static void fixFunctionBasedOnStackMaps(
>         JSCall& call = state.jsCalls[i];
> 
>         CCallHelpers fastPathJIT(&vm, codeBlock);
>+
>+        if (verboseCompilationEnabled()) {
>+            dataLog("Handling JSCall stackmap #", call.stackmapID(), "\n");
>+
>+            auto iter = recordMap.find(call.stackmapID());
>+            if (iter != recordMap.end()) {
>+                for (unsigned i = 0; i < iter->value.size(); ++i) {
>+                    StackMaps::Record& record = iter->value[i];
>+                    RegisterSet usedRegisters = usedRegistersFor(record);
>+
>+                    dataLog("Used registers = ", usedRegisters, "\n");
>+                }
>+            }
>+        }
>         call.emit(fastPathJIT, state.jitCode->stackmaps.stackSizeForLocals());
> 
>         char* startOfIC = bitwise_cast<char*>(generatedFunction) + call.m_instructionOffset;
>@@ -712,6 +744,11 @@ void compile(State& state, Safepoint::Re
>     
>         LLVMExecutionEngineRef engine;
>         
>+#if OS(WINDOWS) && CPU(X86_64)
>+        // On Windows, we can also use the ELF format.
>+        // Currently, this seems to be the only triple which is supported for Windows 64-bit.
>+        llvm->SetTarget(state.module, "x86_64-pc-win32-elf");
>+#else
>         if (isARM64()) {
> #if OS(DARWIN)
>             llvm->SetTarget(state.module, "arm64-apple-ios");
>@@ -721,6 +758,7 @@ void compile(State& state, Safepoint::Re
> #error "Unrecognized OS"
> #endif
>         }
>+#endif
> 
>         if (llvm->CreateMCJITCompilerForModule(&engine, state.module, &options, sizeof(options), &error)) {
>             dataLog("FATAL: Could not create LLVM execution engine: ", error, "\n");
>Index: Source/JavaScriptCore/ftl/FTLInlineCacheSize.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLInlineCacheSize.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLInlineCacheSize.cpp	(working copy)
>@@ -77,6 +77,8 @@ size_t sizeOfCallVarargs()
> {
> #if CPU(ARM64)
>     return 332;
>+#elif OS(WINDOWS)
>+    return 279;
> #else
>     return 275;
> #endif
>Index: Source/JavaScriptCore/ftl/FTLJITCode.h
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLJITCode.h	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLJITCode.h	(working copy)
>@@ -39,7 +39,7 @@
> 
> #if OS(DARWIN)
> #define SECTION_NAME_PREFIX "__"
>-#elif OS(LINUX)
>+#elif OS(LINUX) || OS(WINDOWS)
> #define SECTION_NAME_PREFIX "."
> #else
> #error "Unsupported platform"
>Index: Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLLowerDFGToLLVM.cpp	(working copy)
>@@ -53,7 +53,9 @@
> #include "VirtualRegister.h"
> #include "Watchdog.h"
> #include <atomic>
>+#if !OS(WINDOWS)
> #include <dlfcn.h>
>+#endif
> #include <llvm/InitializeLLVM.h>
> #include <unordered_set>
> #include <wtf/ProcessID.h>
>@@ -1832,21 +1834,32 @@ private:
>     {
>         // FIXME: investigate llvm.powi to better understand its performance characteristics.
>         // It might be better to have the inline loop in DFG too.
>-        if (m_node->child2().useKind() == Int32Use)
>+        if (m_node->child2().useKind() == Int32Use) {
>+#if OS(WINDOWS)
>+            // LLVM does not support powi on Windows.
>+            LValue int32Exponent = lowInt32(m_node->child2());
>+            LValue doubleExponent = m_out.intToDouble(int32Exponent);
>+            setDouble(m_out.doublePow(lowDouble(m_node->child1()), doubleExponent));
>+#else
>             setDouble(m_out.doublePowi(lowDouble(m_node->child1()), lowInt32(m_node->child2())));
>-        else {
>+#endif
>+        } else {
>             LValue base = lowDouble(m_node->child1());
>             LValue exponent = lowDouble(m_node->child2());
> 
>+#if !OS(WINDOWS)
>             LBasicBlock integerExponentIsSmallBlock = FTL_NEW_BLOCK(m_out, ("ArithPow test integer exponent is small."));
>             LBasicBlock integerExponentPowBlock = FTL_NEW_BLOCK(m_out, ("ArithPow pow(double, (int)double)."));
>             LBasicBlock doubleExponentPowBlockEntry = FTL_NEW_BLOCK(m_out, ("ArithPow pow(double, double)."));
>+#endif
>             LBasicBlock nanExceptionExponentIsInfinity = FTL_NEW_BLOCK(m_out, ("ArithPow NaN Exception, check exponent is infinity."));
>             LBasicBlock nanExceptionBaseIsOne = FTL_NEW_BLOCK(m_out, ("ArithPow NaN Exception, check base is one."));
>             LBasicBlock powBlock = FTL_NEW_BLOCK(m_out, ("ArithPow regular pow"));
>             LBasicBlock nanExceptionResultIsNaN = FTL_NEW_BLOCK(m_out, ("ArithPow NaN Exception, result is NaN."));
>             LBasicBlock continuation = FTL_NEW_BLOCK(m_out, ("ArithPow continuation"));
> 
>+#if !OS(WINDOWS)
>+            // LLVM does not support powi on Windows.
>             LValue integerExponent = m_out.fpToInt32(exponent);
>             LValue integerExponentConvertedToDouble = m_out.intToDouble(integerExponent);
>             LValue exponentIsInteger = m_out.doubleEqual(exponent, integerExponentConvertedToDouble);
>@@ -1857,11 +1870,13 @@ private:
>             m_out.branch(integerExponentBelow1000, usually(integerExponentPowBlock), rarely(doubleExponentPowBlockEntry));
> 
>             m_out.appendTo(integerExponentPowBlock, doubleExponentPowBlockEntry);
>+
>             ValueFromBlock powDoubleIntResult = m_out.anchor(m_out.doublePowi(base, integerExponent));
>             m_out.jump(continuation);
> 
>             // If y is NaN, the result is NaN.
>             m_out.appendTo(doubleExponentPowBlockEntry, nanExceptionExponentIsInfinity);
>+#endif
>             LValue exponentIsNaN;
>             if (provenType(m_node->child2()) & SpecDoubleNaN)
>                 exponentIsNaN = m_out.doubleNotEqualOrUnordered(exponent, exponent);
>@@ -1871,7 +1886,11 @@ private:
> 
>             // If abs(x) is 1 and y is +infinity, the result is NaN.
>             // If abs(x) is 1 and y is -infinity, the result is NaN.
>+#if !OS(WINDOWS)
>             m_out.appendTo(nanExceptionExponentIsInfinity, nanExceptionBaseIsOne);
>+#else
>+            LBasicBlock lastNext = m_out.appendTo(nanExceptionExponentIsInfinity, nanExceptionBaseIsOne);
>+#endif
>             LValue absoluteExponent = m_out.doubleAbs(exponent);
>             LValue absoluteExponentIsInfinity = m_out.doubleEqual(absoluteExponent, m_out.constDouble(std::numeric_limits<double>::infinity()));
>             m_out.branch(absoluteExponentIsInfinity, rarely(nanExceptionBaseIsOne), usually(powBlock));
>@@ -1890,7 +1909,12 @@ private:
>             m_out.jump(continuation);
> 
>             m_out.appendTo(continuation, lastNext);
>+
>+#if !OS(WINDOWS)
>             setDouble(m_out.phi(m_out.doubleType, powDoubleIntResult, powResult, pureNan));
>+#else
>+            setDouble(m_out.phi(m_out.doubleType, powResult, pureNan));
>+#endif
>         }
>     }
> 
>Index: Source/JavaScriptCore/ftl/FTLOSRExitCompiler.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLOSRExitCompiler.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLOSRExitCompiler.cpp	(working copy)
>@@ -328,7 +328,13 @@ static void compileStub(
>                 CCallHelpers::TrustedImmPtr(materialization),
>                 CCallHelpers::TrustedImmPtr(materializationArguments));
>             jit.move(CCallHelpers::TrustedImmPtr(bitwise_cast<void*>(operationMaterializeObjectInOSR)), GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+            jit.sub64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>             jit.call(GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+            jit.add64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>             jit.storePtr(GPRInfo::returnValueGPR, materializationToPointer.get(materialization));
> 
>             // Let everyone know that we're done.
>@@ -359,7 +365,13 @@ static void compileStub(
>             CCallHelpers::TrustedImmPtr(materializationToPointer.get(materialization)),
>             CCallHelpers::TrustedImmPtr(materializationArguments));
>         jit.move(CCallHelpers::TrustedImmPtr(bitwise_cast<void*>(operationPopulateObjectInOSR)), GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+        jit.sub64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>         jit.call(GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+        jit.add64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>     }
> 
>     // Save all state from wherever the exit data tells us it was, into the appropriate place in
>Index: Source/JavaScriptCore/ftl/FTLSlowPathCall.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLSlowPathCall.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLSlowPathCall.cpp	(working copy)
>@@ -32,6 +32,7 @@
> #include "FTLState.h"
> #include "GPRInfo.h"
> #include "JSCInlines.h"
>+#include "MaxFrameExtentForSlowPathCall.h"
> 
> namespace JSC { namespace FTL {
> 
>@@ -66,6 +67,10 @@ public:
>         m_offsetToSavingArea =
>             (std::max(m_numArgs, NUMBER_OF_ARGUMENT_REGISTERS) - NUMBER_OF_ARGUMENT_REGISTERS) * wordSize;
>         
>+#if OS(WINDOWS)
>+        m_offsetToSavingArea = std::max(m_offsetToSavingArea, maxFrameExtentForSlowPathCall);
>+#endif
>+
>         for (unsigned i = std::min(NUMBER_OF_ARGUMENT_REGISTERS, numArgs); i--;)
>             m_argumentRegisters.set(GPRInfo::toArgumentRegister(i));
>         m_callingConventionRegisters.merge(m_argumentRegisters);
>@@ -132,12 +137,16 @@ public:
>     
>     SlowPathCallKey keyWithTarget(void* callTarget) const
>     {
>-        return SlowPathCallKey(usedRegisters(), callTarget, m_argumentRegisters, offset());
>+        return SlowPathCallKey(usedRegisters(), callTarget, m_argumentRegisters, offset(), m_numArgs);
>     }
>     
>     MacroAssembler::Call makeCall(void* callTarget, MacroAssembler::JumpList* exceptionTarget)
>     {
>+#if OS(WINDOWS)
>+        MacroAssembler::Call result = m_jit.callJIT();
>+#else
>         MacroAssembler::Call result = m_jit.call();
>+#endif
>         m_state.finalizer->slowPathCalls.append(SlowPathCall(
>             result, keyWithTarget(callTarget)));
>         if (exceptionTarget)
>Index: Source/JavaScriptCore/ftl/FTLSlowPathCallKey.h
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLSlowPathCallKey.h	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLSlowPathCallKey.h	(working copy)
>@@ -47,16 +47,18 @@ public:
>     SlowPathCallKey()
>         : m_callTarget(0)
>         , m_offset(0)
>+        , m_numArgs(0)
>     {
>     }
>     
>     SlowPathCallKey(
>         const RegisterSet& set, void* callTarget, const RegisterSet& argumentRegisters,
>-        ptrdiff_t offset)
>-        : m_usedRegisters(set)
>-        , m_callTarget(callTarget)
>-        , m_argumentRegisters(argumentRegisters)
>-        , m_offset(offset)
>+        ptrdiff_t offset, unsigned numArgs)
>+            : m_usedRegisters(set)
>+            , m_callTarget(callTarget)
>+            , m_argumentRegisters(argumentRegisters)
>+            , m_offset(offset)
>+            , m_numArgs(numArgs)
>     {
>     }
>     
>@@ -64,10 +66,11 @@ public:
>     void* callTarget() const { return m_callTarget; }
>     const RegisterSet& argumentRegisters() const { return m_argumentRegisters; }
>     ptrdiff_t offset() const { return m_offset; }
>-    
>+    unsigned numArgs() const { return m_numArgs; }
>+
>     SlowPathCallKey withCallTarget(void* callTarget)
>     {
>-        return SlowPathCallKey(usedRegisters(), callTarget, argumentRegisters(), offset());
>+        return SlowPathCallKey(usedRegisters(), callTarget, argumentRegisters(), offset(), numArgs());
>     }
>     
>     void dump(PrintStream&) const;
>@@ -79,6 +82,7 @@ public:
>         : m_usedRegisters(RegisterSet::EmptyValue)
>         , m_callTarget(0)
>         , m_offset(0)
>+        , m_numArgs(0)
>     {
>     }
>     
>@@ -86,6 +90,7 @@ public:
>         : m_usedRegisters(RegisterSet::DeletedValue)
>         , m_callTarget(0)
>         , m_offset(0)
>+        , m_numArgs(0)
>     {
>     }
>     
>@@ -108,6 +113,7 @@ private:
>     void* m_callTarget;
>     RegisterSet m_argumentRegisters;
>     ptrdiff_t m_offset;
>+    unsigned m_numArgs;
> };
> 
> struct SlowPathCallKeyHash {
>Index: Source/JavaScriptCore/ftl/FTLThunks.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLThunks.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLThunks.cpp	(working copy)
>@@ -73,7 +73,11 @@ MacroAssemblerCodeRef osrExitGenerationT
>     jit.peek(
>         GPRInfo::argumentGPR1,
>         (stackMisalignment - MacroAssembler::pushToSaveByteOffset()) / sizeof(void*));
>+#if OS(WINDOWS)
>+    MacroAssembler::Call functionCall = jit.callSlowPath(2);
>+#else
>     MacroAssembler::Call functionCall = jit.call();
>+#endif
>     
>     // At this point we want to make a tail call to what was returned to us in the
>     // returnValueGPR. But at the same time as we do this, we must restore all registers.
>@@ -167,7 +171,11 @@ MacroAssemblerCodeRef slowPathCallThunkG
>     
>     registerClobberCheck(jit, key.argumentRegisters());
>     
>+#if OS(WINDOWS)
>+    AssemblyHelpers::Call call = jit.callSlowPath(key.numArgs());
>+#else
>     AssemblyHelpers::Call call = jit.call();
>+#endif
> 
>     jit.loadPtr(AssemblyHelpers::Address(MacroAssembler::stackPointerRegister, key.offset()), GPRInfo::nonPreservedNonReturnGPR);
>     jit.restoreReturnAddressBeforeReturn(GPRInfo::nonPreservedNonReturnGPR);
>Index: Source/JavaScriptCore/ftl/FTLUnwindInfo.cpp
>===================================================================
>--- Source/JavaScriptCore/ftl/FTLUnwindInfo.cpp	(revision 190665)
>+++ Source/JavaScriptCore/ftl/FTLUnwindInfo.cpp	(working copy)
>@@ -115,7 +115,7 @@ struct CompactUnwind {
>     void* personality;
>     void* lsda;
> };
>-#elif OS(LINUX)
>+#elif OS(LINUX) || OS(WINDOWS)
> // DWARF unwind instructions
> enum {
>     DW_CFA_nop                 = 0x0,
>@@ -162,14 +162,32 @@ enum {
> 
> // 64-bit x86_64 registers
> enum {
>-    UNW_X86_64_rbx =  3,
>-    UNW_X86_64_rbp =  6,
>+    UNW_X86_64_rbx = 3,
>+    UNW_X86_64_rsi = 4,
>+    UNW_X86_64_rdi = 5,
>+    UNW_X86_64_rbp = 6,
>     UNW_X86_64_r12 = 12,
>     UNW_X86_64_r13 = 13,
>     UNW_X86_64_r14 = 14,
>     UNW_X86_64_r15 = 15
> };
> 
>+#if OS(WINDOWS)
>+// Windows has callee saved fp registers
>+enum {
>+    UNW_X86_64_xmm6 = 23,
>+    UNW_X86_64_xmm7 = 24,
>+    UNW_X86_64_xmm8 = 25,
>+    UNW_X86_64_xmm9 = 26,
>+    UNW_X86_64_xmm10 = 27,
>+    UNW_X86_64_xmm11 = 28,
>+    UNW_X86_64_xmm12 = 29,
>+    UNW_X86_64_xmm13 = 30,
>+    UNW_X86_64_xmm14 = 31,
>+    UNW_X86_64_xmm15 = 32
>+};
>+#endif
>+
> enum {
>     DW_X86_64_RET_addr = 16
> };
>@@ -394,7 +412,11 @@ struct FDE_Info {
> #if CPU(ARM64)
> enum { MaxRegisterNumber = 120 };
> #elif CPU(X86_64)
>+#if OS(WINDOWS)
>+enum { MaxRegisterNumber = 33 };
>+#else
> enum { MaxRegisterNumber = 17 };
>+#endif // OS(WINDOWS)
> #else
> #error "Unrecognized architecture"
> #endif
>@@ -644,7 +666,7 @@ static void parseInstructions(uintptr_t
> static void parseFDEInstructions(const FDE_Info& fdeInfo, const CIE_Info& cieInfo, PrologInfo* results)
> {
>     // clear results
>-    bzero(results, sizeof(PrologInfo));
>+    memset(results, 0, sizeof(PrologInfo));
> 
>     // parse CIE then FDE instructions
>     parseInstructions(cieInfo.cieInstructions, cieInfo.cieStart + cieInfo.cieLength, cieInfo, results);
>@@ -761,7 +783,109 @@ std::unique_ptr<RegisterAtOffsetList> pa
> #else
> #error "Unrecognized architecture"
> #endif
>+#elif OS(WINDOWS)
>+    FDE_Info fdeInfo;
>+    CIE_Info cieInfo;
>+    PrologInfo prolog;
>+
>+    findFDE((uintptr_t)generatedFunction, (uintptr_t)section, size, &fdeInfo, &cieInfo);
>+    parseFDEInstructions(fdeInfo, cieInfo, &prolog);
>+
>+    RELEASE_ASSERT(prolog.cfaRegister == UNW_X86_64_rbp);
>+    RELEASE_ASSERT(prolog.cfaRegisterOffset == 16);
>+    RELEASE_ASSERT(prolog.savedRegisters[UNW_X86_64_rbp].saved);
>+    RELEASE_ASSERT(prolog.savedRegisters[UNW_X86_64_rbp].offset == -prolog.cfaRegisterOffset);
>+
>+    // On Windows, LLVM does not generate correct stack locations for the saved registers, except for rbp.
>+    // We should fix this in LLVM, but in the meanwhile we have to make adjustments here.
>+    // rbp has offset 0 relative to rbp, and the register pushed after rbp, should have offset -8, and so on.
>+    // We find this register by finding the register with the largest offet (furthest down on the stack).
>+    // We can then find the required adjustment, so that this register ends up at offset -8.
>+    // This adjustment is then applied to all registers, except rbp, which is already correct.
>+
>+    int baseOffset = INT_MIN;
>+
>+    for (int i = 0; i < MaxRegisterNumber; ++i) {
>+        if (prolog.savedRegisters[i].saved && i != UNW_X86_64_rbp && i != DW_X86_64_RET_addr) {
>+            if (baseOffset < prolog.savedRegisters[i].offset)
>+                baseOffset = prolog.savedRegisters[i].offset;
>+        }
>+    }
>+
>+    baseOffset += 8;
>+
>+    for (int i = 0; i < MaxRegisterNumber; ++i) {
>+        if (prolog.savedRegisters[i].saved) {
>+
>+            Reg reg;
>+            int offset = prolog.savedRegisters[i].offset - baseOffset;
> 
>+            switch (i) {
>+            case UNW_X86_64_rbx:
>+                reg = X86Registers::ebx;
>+                break;
>+            case UNW_X86_64_rsi:
>+                reg = X86Registers::esi;
>+                break;
>+            case UNW_X86_64_rdi:
>+                reg = X86Registers::edi;
>+                break;
>+            case UNW_X86_64_r12:
>+                reg = X86Registers::r12;
>+                break;
>+            case UNW_X86_64_r13:
>+                reg = X86Registers::r13;
>+                break;
>+            case UNW_X86_64_r14:
>+                reg = X86Registers::r14;
>+                break;
>+            case UNW_X86_64_r15:
>+                reg = X86Registers::r15;
>+                break;
>+            case UNW_X86_64_rbp:
>+                reg = X86Registers::ebp;
>+                offset = 0;
>+                break;
>+            case DW_X86_64_RET_addr:
>+                break;
>+            case UNW_X86_64_xmm6:
>+                reg = X86Registers::xmm6;
>+                break;
>+            case UNW_X86_64_xmm7:
>+                reg = X86Registers::xmm7;
>+                break;
>+            case UNW_X86_64_xmm8:
>+                reg = X86Registers::xmm8;
>+                break;
>+            case UNW_X86_64_xmm9:
>+                reg = X86Registers::xmm9;
>+                break;
>+            case UNW_X86_64_xmm10:
>+                reg = X86Registers::xmm10;
>+                break;
>+            case UNW_X86_64_xmm11:
>+                reg = X86Registers::xmm11;
>+                break;
>+            case UNW_X86_64_xmm12:
>+                reg = X86Registers::xmm12;
>+                break;
>+            case UNW_X86_64_xmm13:
>+                reg = X86Registers::xmm13;
>+                break;
>+            case UNW_X86_64_xmm14:
>+                reg = X86Registers::xmm14;
>+                break;
>+            case UNW_X86_64_xmm15:
>+                reg = X86Registers::xmm15;
>+                break;
>+            default:
>+                RELEASE_ASSERT_NOT_REACHED(); // non-standard register being saved in prolog
>+            }
>+
>+            if (reg.isSet())
>+                registerOffsets->append(RegisterAtOffset(reg, offset));
>+        }
>+    }
> #elif OS(LINUX)
>     FDE_Info fdeInfo;
>     CIE_Info cieInfo;
>Index: Source/JavaScriptCore/jit/AssemblyHelpers.h
>===================================================================
>--- Source/JavaScriptCore/jit/AssemblyHelpers.h	(revision 190665)
>+++ Source/JavaScriptCore/jit/AssemblyHelpers.h	(working copy)
>@@ -182,14 +182,21 @@ public:
>         ASSERT(codeBlock);
> 
>         RegisterAtOffsetList* calleeSaves = codeBlock->calleeSaveRegisters();
>+#if OS(WINDOWS) && CPU(X86_64)
>+        RegisterSet dontSaveRegisters = RegisterSet::stackRegisters();
>+#else
>         RegisterSet dontSaveRegisters = RegisterSet(RegisterSet::stackRegisters(), RegisterSet::allFPRs());
>+#endif
>         unsigned registerCount = calleeSaves->size();
> 
>         for (unsigned i = 0; i < registerCount; i++) {
>             RegisterAtOffset entry = calleeSaves->at(i);
>             if (dontSaveRegisters.get(entry.reg()))
>                 continue;
>-            storePtr(entry.reg().gpr(), Address(framePointerRegister, offsetVirtualRegister.offsetInBytes() + entry.offset()));
>+            if (entry.reg().isGPR())
>+                storePtr(entry.reg().gpr(), Address(framePointerRegister, offsetVirtualRegister.offsetInBytes() + entry.offset()));
>+            else
>+                storeDouble(entry.reg().fpr(), Address(framePointerRegister, offsetVirtualRegister.offsetInBytes() + entry.offset()));            
>         }
>     }
> 
>@@ -198,14 +205,21 @@ public:
>         ASSERT(codeBlock);
> 
>         RegisterAtOffsetList* calleeSaves = codeBlock->calleeSaveRegisters();
>+#if OS(WINDOWS) && CPU(X86_64)
>+        RegisterSet dontRestoreRegisters = RegisterSet::stackRegisters();
>+#else
>         RegisterSet dontRestoreRegisters = RegisterSet(RegisterSet::stackRegisters(), RegisterSet::allFPRs());
>+#endif
>         unsigned registerCount = calleeSaves->size();
>         
>         for (unsigned i = 0; i < registerCount; i++) {
>             RegisterAtOffset entry = calleeSaves->at(i);
>             if (dontRestoreRegisters.get(entry.reg()))
>                 continue;
>-            loadPtr(Address(framePointerRegister, entry.offset()), entry.reg().gpr());
>+            if (entry.reg().isGPR())
>+                loadPtr(Address(framePointerRegister, entry.offset()), entry.reg().gpr());
>+            else
>+                loadDouble(Address(framePointerRegister, entry.offset()), entry.reg().fpr());
>         }
>     }
> 
>@@ -894,7 +908,13 @@ public:
> #error "JIT not supported on this platform."
> #endif
>         move(TrustedImmPtr(reinterpret_cast<void*>(function)), scratch);
>+#if OS(WINDOWS) && CPU(X86_64)
>+        sub64(TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>         call(scratch);
>+#if OS(WINDOWS) && CPU(X86_64)
>+        add64(TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
> 
>         move(TrustedImmPtr(scratchBuffer->activeLengthPtr()), GPRInfo::regT0);
>         storePtr(TrustedImmPtr(0), GPRInfo::regT0);
>Index: Source/JavaScriptCore/jit/FPRInfo.h
>===================================================================
>--- Source/JavaScriptCore/jit/FPRInfo.h	(revision 190665)
>+++ Source/JavaScriptCore/jit/FPRInfo.h	(working copy)
>@@ -61,6 +61,19 @@ public:
>     // so we'll copy to xmm0 for sanity!
>     static const FPRReg returnValueFPR = X86Registers::xmm0; // fpRegT0
> 
>+#if CPU(X86_64) && OS(WINDOWS)
>+    static const FPRReg fpRegCS0 = X86Registers::xmm6;
>+    static const FPRReg fpRegCS1 = X86Registers::xmm7;
>+    static const FPRReg fpRegCS2 = X86Registers::xmm8;
>+    static const FPRReg fpRegCS3 = X86Registers::xmm9;
>+    static const FPRReg fpRegCS4 = X86Registers::xmm10;
>+    static const FPRReg fpRegCS5 = X86Registers::xmm11;
>+    static const FPRReg fpRegCS6 = X86Registers::xmm12;
>+    static const FPRReg fpRegCS7 = X86Registers::xmm13;
>+    static const FPRReg fpRegCS8 = X86Registers::xmm14;
>+    static const FPRReg fpRegCS9 = X86Registers::xmm15;
>+#endif
>+
>     // FPRReg mapping is direct, the machine regsiter numbers can
>     // be used directly as indices into the FPR RegisterBank.
>     COMPILE_ASSERT(X86Registers::xmm0 == 0, xmm0_is_0);
>Index: Source/JavaScriptCore/jit/GPRInfo.h
>===================================================================
>--- Source/JavaScriptCore/jit/GPRInfo.h	(revision 190665)
>+++ Source/JavaScriptCore/jit/GPRInfo.h	(working copy)
>@@ -385,7 +385,7 @@ public:
> #define NUMBER_OF_CALLEE_SAVES_REGISTERS 5u
> #else
> #define NUMBER_OF_ARGUMENT_REGISTERS 4u
>-#define NUMBER_OF_CALLEE_SAVES_REGISTERS 7u
>+#define NUMBER_OF_CALLEE_SAVES_REGISTERS 18u
> #endif
> 
> class GPRInfo {
>Index: Source/JavaScriptCore/jit/RegisterSet.cpp
>===================================================================
>--- Source/JavaScriptCore/jit/RegisterSet.cpp	(revision 190665)
>+++ Source/JavaScriptCore/jit/RegisterSet.cpp	(working copy)
>@@ -79,6 +79,26 @@ RegisterSet RegisterSet::calleeSaveRegis
>     result.set(X86Registers::ebp);
>     result.set(X86Registers::edi);
>     result.set(X86Registers::esi);
>+#elif CPU(X86_64) && OS(WINDOWS)
>+    // See https://msdn.microsoft.com/en-us/library/9z1stfyw.aspx
>+    result.set(X86Registers::ebx);
>+    result.set(X86Registers::esi);
>+    result.set(X86Registers::edi);
>+    result.set(X86Registers::ebp);
>+    result.set(X86Registers::r12);
>+    result.set(X86Registers::r13);
>+    result.set(X86Registers::r14);
>+    result.set(X86Registers::r15);
>+    result.set(X86Registers::xmm6);
>+    result.set(X86Registers::xmm7);
>+    result.set(X86Registers::xmm8);
>+    result.set(X86Registers::xmm9);
>+    result.set(X86Registers::xmm10);
>+    result.set(X86Registers::xmm11);
>+    result.set(X86Registers::xmm12);
>+    result.set(X86Registers::xmm13);
>+    result.set(X86Registers::xmm14);
>+    result.set(X86Registers::xmm15);
> #elif CPU(X86_64)
>     result.set(X86Registers::ebx);
>     result.set(X86Registers::ebp);
>@@ -139,6 +159,16 @@ RegisterSet RegisterSet::vmCalleeSaveReg
> #if OS(WINDOWS)
>     result.set(GPRInfo::regCS5);
>     result.set(GPRInfo::regCS6);
>+    result.set(FPRInfo::fpRegCS0);
>+    result.set(FPRInfo::fpRegCS1);
>+    result.set(FPRInfo::fpRegCS2);
>+    result.set(FPRInfo::fpRegCS3);
>+    result.set(FPRInfo::fpRegCS4);
>+    result.set(FPRInfo::fpRegCS5);
>+    result.set(FPRInfo::fpRegCS6);
>+    result.set(FPRInfo::fpRegCS7);
>+    result.set(FPRInfo::fpRegCS8);
>+    result.set(FPRInfo::fpRegCS9);
> #endif
> #elif CPU(ARM64)
>     result.set(GPRInfo::regCS0);
>@@ -175,11 +205,25 @@ RegisterSet RegisterSet::llintBaselineCa
>     result.set(GPRInfo::regCS3);
>     result.set(GPRInfo::regCS4);
> #else
>+    result.set(GPRInfo::regCS0);
>+    result.set(GPRInfo::regCS1);
>+    result.set(GPRInfo::regCS2);
>+    result.set(GPRInfo::regCS3);
>     result.set(GPRInfo::regCS4);
>     ASSERT(GPRInfo::regCS5 == GPRInfo::tagTypeNumberRegister);
>     ASSERT(GPRInfo::regCS6 == GPRInfo::tagMaskRegister);
>     result.set(GPRInfo::regCS5);
>     result.set(GPRInfo::regCS6);
>+    result.set(FPRInfo::fpRegCS0);
>+    result.set(FPRInfo::fpRegCS1);
>+    result.set(FPRInfo::fpRegCS2);
>+    result.set(FPRInfo::fpRegCS3);
>+    result.set(FPRInfo::fpRegCS4);
>+    result.set(FPRInfo::fpRegCS5);
>+    result.set(FPRInfo::fpRegCS6);
>+    result.set(FPRInfo::fpRegCS7);
>+    result.set(FPRInfo::fpRegCS8);
>+    result.set(FPRInfo::fpRegCS9);
> #endif
> #elif CPU(ARM_THUMB2)
> #elif CPU(ARM_TRADITIONAL)
>@@ -217,6 +261,16 @@ RegisterSet RegisterSet::dfgCalleeSaveRe
>     ASSERT(GPRInfo::regCS6 == GPRInfo::tagMaskRegister);
>     result.set(GPRInfo::regCS5);
>     result.set(GPRInfo::regCS6);
>+    result.set(FPRInfo::fpRegCS0);
>+    result.set(FPRInfo::fpRegCS1);
>+    result.set(FPRInfo::fpRegCS2);
>+    result.set(FPRInfo::fpRegCS3);
>+    result.set(FPRInfo::fpRegCS4);
>+    result.set(FPRInfo::fpRegCS5);
>+    result.set(FPRInfo::fpRegCS6);
>+    result.set(FPRInfo::fpRegCS7);
>+    result.set(FPRInfo::fpRegCS8);
>+    result.set(FPRInfo::fpRegCS9);
> #endif
> #elif CPU(ARM_THUMB2)
> #elif CPU(ARM_TRADITIONAL)
>@@ -237,7 +291,7 @@ RegisterSet RegisterSet::ftlCalleeSaveRe
> {
>     RegisterSet result;
> #if ENABLE(FTL_JIT)
>-#if CPU(X86_64) && !OS(WINDOWS)
>+#if CPU(X86_64)
>     result.set(GPRInfo::regCS0);
>     result.set(GPRInfo::regCS1);
>     result.set(GPRInfo::regCS2);
>@@ -245,6 +299,20 @@ RegisterSet RegisterSet::ftlCalleeSaveRe
>     ASSERT(GPRInfo::regCS4 == GPRInfo::tagMaskRegister);
>     result.set(GPRInfo::regCS3);
>     result.set(GPRInfo::regCS4);
>+#if OS(WINDOWS)
>+    result.set(GPRInfo::regCS5);
>+    result.set(GPRInfo::regCS6);
>+    result.set(FPRInfo::fpRegCS0);
>+    result.set(FPRInfo::fpRegCS1);
>+    result.set(FPRInfo::fpRegCS2);
>+    result.set(FPRInfo::fpRegCS3);
>+    result.set(FPRInfo::fpRegCS4);
>+    result.set(FPRInfo::fpRegCS5);
>+    result.set(FPRInfo::fpRegCS6);
>+    result.set(FPRInfo::fpRegCS7);
>+    result.set(FPRInfo::fpRegCS8);
>+    result.set(FPRInfo::fpRegCS9);
>+#endif
> #elif CPU(ARM64)
>     // LLVM might save and use all ARM64 callee saves specified in the ABI.
>     result.set(GPRInfo::regCS0);
>Index: Source/JavaScriptCore/jit/ThunkGenerators.cpp
>===================================================================
>--- Source/JavaScriptCore/jit/ThunkGenerators.cpp	(revision 190665)
>+++ Source/JavaScriptCore/jit/ThunkGenerators.cpp	(working copy)
>@@ -71,7 +71,13 @@ MacroAssemblerCodeRef throwExceptionFrom
>     jit.setupArguments(CCallHelpers::TrustedImmPtr(vm), GPRInfo::callFrameRegister);
>     jit.move(CCallHelpers::TrustedImmPtr(bitwise_cast<void*>(lookupExceptionHandler)), GPRInfo::nonArgGPR0);
>     emitPointerValidation(jit, GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+    jit.sub64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>     jit.call(GPRInfo::nonArgGPR0);
>+#if OS(WINDOWS) && CPU(X86_64)
>+    jit.add64(CCallHelpers::TrustedImm32(4 * sizeof(int64_t)), X86Registers::esp);
>+#endif
>     jit.jumpToExceptionHandler();
> 
>     LinkBuffer patchBuffer(*vm, jit, GLOBAL_THUNK_ID);
>Index: Source/JavaScriptCore/llint/LLIntData.cpp
>===================================================================
>--- Source/JavaScriptCore/llint/LLIntData.cpp	(revision 190665)
>+++ Source/JavaScriptCore/llint/LLIntData.cpp	(working copy)
>@@ -138,7 +138,7 @@ void Data::performAssertions(VM& vm)
> #elif (CPU(X86_64) && !OS(WINDOWS))  || CPU(ARM64)
>     ASSERT(CodeBlock::llintBaselineCalleeSaveSpaceAsVirtualRegisters() == 3);
> #elif (CPU(X86_64) && OS(WINDOWS))
>-    ASSERT(CodeBlock::llintBaselineCalleeSaveSpaceAsVirtualRegisters() == 3);
>+    ASSERT(CodeBlock::llintBaselineCalleeSaveSpaceAsVirtualRegisters() == 17);
> #endif
>     
>     ASSERT(StringType == 6);
>Index: Source/JavaScriptCore/llint/LowLevelInterpreter.asm
>===================================================================
>--- Source/JavaScriptCore/llint/LowLevelInterpreter.asm	(revision 190665)
>+++ Source/JavaScriptCore/llint/LowLevelInterpreter.asm	(working copy)
>@@ -234,8 +234,10 @@ elsif X86_64_WIN
>     const maxFrameExtentForSlowPathCall = 64
> end
> 
>-if X86_64 or X86_64_WIN or ARM64
>+if X86_64 or ARM64
>     const CalleeSaveSpaceAsVirtualRegisters = 3
>+elsif X86_64_WIN
>+    const CalleeSaveSpaceAsVirtualRegisters = 17
> else
>     const CalleeSaveSpaceAsVirtualRegisters = 0
> end
>@@ -545,9 +547,23 @@ macro preserveCalleeSavesUsedByLLInt()
>         storep csr3, -16[cfr]
>         storep csr2, -24[cfr]
>     elsif X86_64_WIN
>-        storep csr6, -8[cfr]
>-        storep csr5, -16[cfr]
>-        storep csr4, -24[cfr]
>+        stored csfr9, -8[cfr]
>+        stored csfr8, -16[cfr]
>+        stored csfr7, -24[cfr]
>+        stored csfr6, -32[cfr]
>+        stored csfr5, -40[cfr]
>+        stored csfr4, -48[cfr]
>+        stored csfr3, -56[cfr]
>+        stored csfr2, -64[cfr]
>+        stored csfr1, -72[cfr]
>+        stored csfr0, -80[cfr]
>+        storep csr6, -88[cfr]
>+        storep csr5, -96[cfr]
>+        storep csr4, -104[cfr]
>+        storep csr3, -112[cfr]
>+        storep csr2, -120[cfr]
>+        storep csr1, -128[cfr]
>+        storep csr0, -136[cfr]
>     end
> end
> 
>@@ -567,9 +583,23 @@ macro restoreCalleeSavesUsedByLLInt()
>         loadp -16[cfr], csr3
>         loadp -8[cfr], csr4
>     elsif X86_64_WIN
>-        loadp -24[cfr], csr4
>-        loadp -16[cfr], csr5
>-        loadp -8[cfr], csr6
>+        loadp -136[cfr], csr0
>+        loadp -128[cfr], csr1
>+        loadp -120[cfr], csr2
>+        loadp -112[cfr], csr3
>+        loadp -104[cfr], csr4
>+        loadp -96[cfr], csr5
>+        loadp -88[cfr], csr6
>+        loadd -80[cfr], csfr0
>+        loadd -72[cfr], csfr1
>+        loadd -64[cfr], csfr2
>+        loadd -56[cfr], csfr3
>+        loadd -48[cfr], csfr4
>+        loadd -40[cfr], csfr5
>+        loadd -32[cfr], csfr6
>+        loadd -24[cfr], csfr7
>+        loadd -16[cfr], csfr8
>+        loadd -8[cfr], csfr9
>     end
> end
> 
>@@ -609,6 +639,16 @@ macro copyCalleeSavesToVMCalleeSavesBuff
>             storep csr4, 32[temp]
>             storep csr5, 40[temp]
>             storep csr6, 48[temp]
>+            stored csfr0, 56[temp]
>+            stored csfr1, 64[temp]
>+            stored csfr2, 72[temp]
>+            stored csfr3, 80[temp]
>+            stored csfr4, 88[temp]
>+            stored csfr5, 96[temp]
>+            stored csfr6, 104[temp]
>+            stored csfr7, 112[temp]
>+            stored csfr8, 120[temp]
>+            stored csfr9, 128[temp]
>         end
>     end
> end
>@@ -649,6 +689,16 @@ macro restoreCalleeSavesFromVMCalleeSave
>             loadp 32[temp], csr4
>             loadp 40[temp], csr5
>             loadp 48[temp], csr6
>+            loadd 56[temp], csfr0
>+            loadd 64[temp], csfr1
>+            loadd 72[temp], csfr2
>+            loadd 80[temp], csfr3
>+            loadd 88[temp], csfr4
>+            loadd 96[temp], csfr5
>+            loadd 104[temp], csfr6
>+            loadd 112[temp], csfr7
>+            loadd 120[temp], csfr8
>+            loadd 128[temp], csfr9
>         end
>     end
> end
>Index: Source/JavaScriptCore/llvm/InitializeLLVM.cpp
>===================================================================
>--- Source/JavaScriptCore/llvm/InitializeLLVM.cpp	(revision 190665)
>+++ Source/JavaScriptCore/llvm/InitializeLLVM.cpp	(working copy)
>@@ -30,13 +30,11 @@
> 
> #include "LLVMAPI.h"
> #include "Options.h"
>-#include <pthread.h>
>+#include <mutex>
> #include <wtf/DataLog.h>
> 
> namespace JSC {
> 
>-static pthread_once_t initializeLLVMOnceKey = PTHREAD_ONCE_INIT;
>-
> static void initializeLLVMImpl()
> {
>     const bool verbose =
>@@ -67,7 +65,9 @@ static void initializeLLVMImpl()
> 
> bool initializeLLVM()
> {
>-    pthread_once(&initializeLLVMOnceKey, initializeLLVMImpl);
>+    static std::once_flag initializeLLVMOnceKey;
>+
>+    std::call_once(initializeLLVMOnceKey, initializeLLVMImpl);
>     return !!llvm;
> }
> 
>Index: Source/JavaScriptCore/llvm/LLVMAPIFunctions.h
>===================================================================
>--- Source/JavaScriptCore/llvm/LLVMAPIFunctions.h	(revision 190665)
>+++ Source/JavaScriptCore/llvm/LLVMAPIFunctions.h	(working copy)
>@@ -412,7 +412,6 @@
>     macro(LLVMValueRef, BuildSwitch, (LLVMBuilderRef, LLVMValueRef V, LLVMBasicBlockRef Else, unsigned NumCases)) \
>     macro(LLVMValueRef, BuildIndirectBr, (LLVMBuilderRef B, LLVMValueRef Addr, unsigned NumDests)) \
>     macro(LLVMValueRef, BuildInvoke, (LLVMBuilderRef, LLVMValueRef Fn, LLVMValueRef *Args, unsigned NumArgs, LLVMBasicBlockRef Then, LLVMBasicBlockRef Catch, const char *Name)) \
>-    macro(LLVMValueRef, BuildLandingPad, (LLVMBuilderRef B, LLVMTypeRef Ty, LLVMValueRef PersFn, unsigned NumClauses, const char *Name)) \
>     macro(LLVMValueRef, BuildResume, (LLVMBuilderRef B, LLVMValueRef Exn)) \
>     macro(LLVMValueRef, BuildUnreachable, (LLVMBuilderRef)) \
>     macro(void, AddCase, (LLVMValueRef Switch, LLVMValueRef OnVal, LLVMBasicBlockRef Dest)) \
>Index: Source/JavaScriptCore/llvm/library/LLVMExports.cpp
>===================================================================
>--- Source/JavaScriptCore/llvm/library/LLVMExports.cpp	(revision 190665)
>+++ Source/JavaScriptCore/llvm/library/LLVMExports.cpp	(working copy)
>@@ -117,10 +117,16 @@ extern "C" JSC::LLVMAPI* initializeAndGe
>     *enableFastISel = false;
> #endif
> 
>+#if OS(WINDOWS)
>+    const char* dll = "libllvmForJSC.dll";
>+#else
>+    const char* dll = "llvmForJSC.dylib";
>+#endif
>+
>     if (*enableFastISel)
>-        initCommandLine("llvmForJSC.dylib", "-enable-misched=false", "-regalloc=basic");
>+        initCommandLine(dll, "-enable-misched=false", "-regalloc=basic");
>     else
>-        initCommandLine("llvmForJSC.dylib", "-enable-patchpoint-liveness=true");
>+        initCommandLine(dll, "-enable-patchpoint-liveness=true");
>     
>     JSC::LLVMAPI* result = new JSC::LLVMAPI;
>     
>Index: Source/JavaScriptCore/llvm/library/LLVMOverrides.cpp
>===================================================================
>--- Source/JavaScriptCore/llvm/library/LLVMOverrides.cpp	(revision 190665)
>+++ Source/JavaScriptCore/llvm/library/LLVMOverrides.cpp	(working copy)
>@@ -34,6 +34,7 @@ extern "C" int __cxa_atexit() { return 0
> 
> void (*g_llvmTrapCallback)(const char* message, ...) NO_RETURN;
> 
>+#if !OS(WINDOWS)
> // If LLVM tries to raise signals, abort, or fail an assertion, then let
> // WebKit handle it instead of trapping.
> extern "C" int raise(int signal);
>@@ -55,5 +56,6 @@ extern "C" void abort(void)
> {
>     g_llvmTrapCallback("abort() called.");
> }
>+#endif // !OS(WINDOWS)
> 
> #endif // HAVE(LLVM)
>Index: Source/JavaScriptCore/llvm/library/LLVMTrapCallback.h
>===================================================================
>--- Source/JavaScriptCore/llvm/library/LLVMTrapCallback.h	(revision 190665)
>+++ Source/JavaScriptCore/llvm/library/LLVMTrapCallback.h	(working copy)
>@@ -28,6 +28,12 @@
> 
> #include <wtf/Assertions.h>
> 
>+#if OS(WINDOWS)
>+// On Windows, NO_RETURN can only be applied to function declarations or definitions.
>+#undef NO_RETURN
>+#define NO_RETURN
>+#endif
>+
> extern void (*g_llvmTrapCallback)(const char* message, ...) NO_RETURN;
> 
> #endif // LLVMTrapCallback_h
>Index: Source/JavaScriptCore/offlineasm/registers.rb
>===================================================================
>--- Source/JavaScriptCore/offlineasm/registers.rb	(revision 190665)
>+++ Source/JavaScriptCore/offlineasm/registers.rb	(working copy)
>@@ -74,6 +74,8 @@ FPRS =
>      "csfr5",
>      "csfr6",
>      "csfr7",
>+     "csfr8",
>+     "csfr9",
>      "fr"
>     ]
> 
>Index: Source/JavaScriptCore/offlineasm/x86.rb
>===================================================================
>--- Source/JavaScriptCore/offlineasm/x86.rb	(revision 190665)
>+++ Source/JavaScriptCore/offlineasm/x86.rb	(working copy)
>@@ -359,7 +359,32 @@ class FPRegisterID
>         when "ft5"
>             register("xmm5")
>         else
>-            raise "Bad register #{name} for X86 at #{codeOriginString}"
>+            if isX64 and isWin
>+                case name
>+                when "csfr0"
>+                    register("xmm6")
>+                when "csfr1"
>+                    register("xmm7")
>+                when "csfr2"
>+                    register("xmm8")
>+                when "csfr3"
>+                    register("xmm9")
>+                when "csfr4"
>+                    register("xmm10")
>+                when "csfr5"
>+                    register("xmm11")
>+                when "csfr6"
>+                    register("xmm12")
>+                when "csfr7"
>+                    register("xmm13")
>+                when "csfr8"
>+                    register("xmm14")
>+                when "csfr9"
>+                    register("xmm15")
>+                end
>+            else
>+                raise "Bad register #{name} for X86 at #{codeOriginString}"
>+            end
>         end
>     end
>     def x87DefaultStackPosition
>Index: Source/WTF/ChangeLog
>===================================================================
>--- Source/WTF/ChangeLog	(revision 190665)
>+++ Source/WTF/ChangeLog	(working copy)
>@@ -1,3 +1,14 @@
>+2015-10-07  Per Arne Vollan  <peavo@outlook.com>
>+
>+        FTL is not working on Windows.
>+        https://bugs.webkit.org/show_bug.cgi?id=145366
>+
>+        Reviewed by NOBODY (OOPS!).
>+
>+        Enable FTL on Windows.
>+
>+        * wtf/Platform.h:
>+
> 2015-10-06  Daniel Bates  <dbates@webkit.org>
> 
>         Enable XSLT when building WebKit for iOS using the public iOS SDK
>Index: Source/WTF/wtf/Platform.h
>===================================================================
>--- Source/WTF/wtf/Platform.h	(revision 190665)
>+++ Source/WTF/wtf/Platform.h	(working copy)
>@@ -711,6 +711,11 @@
> #define ENABLE_FTL_JIT 1
> #endif
> 
>+#if PLATFORM(WIN) && CPU(X86_64)
>+#define HAVE_LLVM 1
>+#define ENABLE_FTL_JIT 1
>+#endif
>+
> /* The FTL *does not* work on 32-bit platforms. Disable it even if someone asked us to enable it. */
> #if USE(JSVALUE32_64)
> #undef ENABLE_FTL_JIT
>Index: Source/cmake/OptionsWin.cmake
>===================================================================
>--- Source/cmake/OptionsWin.cmake	(revision 190665)
>+++ Source/cmake/OptionsWin.cmake	(working copy)
>@@ -36,6 +36,9 @@ WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_
> WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_ES6_MODULES PUBLIC OFF)
> WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_ES6_TEMPLATE_LITERAL_SYNTAX PUBLIC ON)
> WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_FILE_SYSTEM PUBLIC OFF)
>+if (WTF_CPU_X86_64)
>+WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_FTL_JIT PRIVATE ON)
>+endif ()
> WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_FULLSCREEN_API PUBLIC ON)
> WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_GAMEPAD PUBLIC OFF)
> WEBKIT_OPTION_DEFAULT_PORT_VALUE(ENABLE_GEOLOCATION PUBLIC ON)
>Index: Tools/ChangeLog
>===================================================================
>--- Tools/ChangeLog	(revision 190665)
>+++ Tools/ChangeLog	(working copy)
>@@ -1,3 +1,15 @@
>+2015-10-07  Per Arne Vollan  <peavo@outlook.com>
>+
>+        FTL is not working on Windows.
>+        https://bugs.webkit.org/show_bug.cgi?id=145366
>+
>+        Reviewed by NOBODY (OOPS!).
>+
>+        Generate project with 64-bit configuration for VS2015 on Win64.
>+
>+        * Scripts/webkitdirs.pm:
>+        (generateBuildSystemFromCMakeProject):
>+
> 2015-10-06  Michael Catanzaro  <mcatanzaro@igalia.com>
> 
>         [GTK] Add autocleanups
>Index: Tools/Scripts/webkitdirs.pm
>===================================================================
>--- Tools/Scripts/webkitdirs.pm	(revision 190665)
>+++ Tools/Scripts/webkitdirs.pm	(working copy)
>@@ -1915,6 +1915,12 @@ sub generateBuildSystemFromCMakeProject
>         $ENV{'CXXFLAGS'} = "-march=pentium4 -msse2 -mfpmath=sse " . ($ENV{'CXXFLAGS'} || "");
>     }
> 
>+    # Build with VS2015 for Windows x64.
>+    if (isWin64() && !canUseNinja()) {
>+        push @args, "-G";
>+        push @args, "\"Visual Studio 14 2015 Win64\"";
>+    }
>+
>     # We call system("cmake @args") instead of system("cmake", @args) so that @args is
>     # parsed for shell metacharacters.
>     my $wrapper = join(" ", jhbuildWrapperPrefixIfNeeded()) . " ";

Actions: View | Formatted Diff | Diff

Attachments on bug 145366: 253682 | 253683 | 253732 | 254498 | 254499 | 254949 | 255287 | 255288 | 255293 | 255630 | 256090 | 260695 | 260699 | 260703 | 260864 | 261971 | 261972 | 261978 | 262334 | 262379 | 262404 | 262630 | 263482 | 263517 | 264235 | 264249 | 265292 | 265295 | 267558 | 267560 | 279685 | 279774 | 281349 | 468205