Docs: Update for release FEX-2312.1

This flag breaks FEX heavily for now.
glibc 2.38 started using this flag as an optimization for posix_spawn.
It will fall back to a "non-optimized" implementation if the clone
syscall returns EINVAL. For now do this while we investigate a more
proper implementation.

Should be backported to 2312.1.

CMakeLists.txt

@@ -293,10 +293,11 @@ if(ENABLE_WERROR OR ENABLE_STRICT_WERROR)
   endif()
 endif()
 
+set(FEX_TUNE_COMPILE_FLAGS)
 if (NOT TUNE_ARCH STREQUAL "generic")
   check_cxx_compiler_flag("-march=${TUNE_ARCH}" COMPILER_SUPPORTS_ARCH_TYPE)
   if(COMPILER_SUPPORTS_ARCH_TYPE)
-    add_compile_options("-march=${TUNE_ARCH}")
+    list(APPEND FEX_TUNE_COMPILE_FLAGS "-march=${TUNE_ARCH}")
   else()
     message(FATAL_ERROR "Trying to compile arch type '${TUNE_ARCH}' but the compiler doesn't support this")
   endif()
@@ -309,7 +310,7 @@ if (TUNE_CPU STREQUAL "native")
       # Clang can not currently check for native Apple M1 type in hypervisor. Currently disabled
       check_cxx_compiler_flag("-mcpu=native" COMPILER_SUPPORTS_CPU_TYPE)
       if(COMPILER_SUPPORTS_CPU_TYPE)
-        add_compile_options("-mcpu=native")
+        list(APPEND FEX_TUNE_COMPILE_FLAGS "-mcpu=native")
       endif()
     else()
       # Due to an oversight in llvm, it declares any reasonably new Kryo CPU to only be ARMv8.0
@@ -323,19 +324,19 @@ if (TUNE_CPU STREQUAL "native")
 
       check_cxx_compiler_flag("-mcpu=${AARCH64_CPU}" COMPILER_SUPPORTS_CPU_TYPE)
       if(COMPILER_SUPPORTS_CPU_TYPE)
-        add_compile_options("-mcpu=${AARCH64_CPU}")
+        list(APPEND FEX_TUNE_COMPILE_FLAGS "-mcpu=${AARCH64_CPU}")
       endif()
     endif()
   else()
     check_cxx_compiler_flag("-march=native" COMPILER_SUPPORTS_MARCH_NATIVE)
     if(COMPILER_SUPPORTS_MARCH_NATIVE)
-      add_compile_options("-march=native")
+      list(APPEND FEX_TUNE_COMPILE_FLAGS "-march=native")
     endif()
   endif()
 else()
   check_cxx_compiler_flag("-mcpu=${TUNE_CPU}" COMPILER_SUPPORTS_CPU_TYPE)
   if(COMPILER_SUPPORTS_CPU_TYPE)
-    add_compile_options("-mcpu=${TUNE_CPU}")
+    list(APPEND FEX_TUNE_COMPILE_FLAGS "-mcpu=${TUNE_CPU}")
   else()
     message(FATAL_ERROR "Trying to compile cpu type '${TUNE_CPU}' but the compiler doesn't support this")
   endif()

FEXCore/CMakeLists.txt

@@ -38,7 +38,12 @@ check_cxx_source_compiles(
   HAS_CLANG_PRESERVE_ALL)
 unset(CMAKE_REQUIRED_FLAGS)
 if (HAS_CLANG_PRESERVE_ALL)
-  message(STATUS "Has clang::preserve_all")
+  if (MINGW_BUILD)
+    message(STATUS "Ignoring broken clang::preserve_all support")
+    set(HAS_CLANG_PRESERVE_ALL FALSE)
+  else()
+    message(STATUS "Has clang::preserve_all")
+  endif()
 endif ()
 
 if (EXISTS ${CMAKE_CURRENT_DIR}/External/vixl/)

FEXCore/Scripts/json_ir_generator.py

@@ -652,7 +652,7 @@ def print_ir_allocator_helpers():
 
             # Save NZCV if needed before clobbering NZCV
             if op.ImplicitFlagClobber:
-                output_file.write("\t\tSaveNZCV();")
+                output_file.write("\t\tSaveNZCV(IROps::OP_{});".format(op.Name.upper()))
 
             output_file.write("\t\tauto Op = AllocateOp<IROp_{}, IROps::OP_{}>();\n".format(op.Name, op.Name.upper()))
 

FEXCore/Source/CMakeLists.txt

@@ -90,7 +90,6 @@ set (SRCS
   Interface/Core/CPUBackend.cpp
   Interface/Core/CPUID.cpp
   Interface/Core/Frontend.cpp
-  Interface/Core/GdbServer.cpp
   Interface/Core/HostFeatures.cpp
   Interface/Core/ObjectCache/JobHandling.cpp
   Interface/Core/ObjectCache/NamedRegionObjectHandler.cpp
@@ -101,9 +100,7 @@ set (SRCS
   Interface/Core/OpcodeDispatcher/X87.cpp
   Interface/Core/OpcodeDispatcher/X87F64.cpp
   Interface/Core/OpcodeDispatcher.cpp
-  Interface/Core/SignalDelegator.cpp
   Interface/Core/X86Tables.cpp
-  Interface/Core/X86DebugInfo.cpp
   Interface/Core/X86HelperGen.cpp
   Interface/Core/ArchHelpers/Arm64Emitter.cpp
   Interface/Core/Dispatcher/Dispatcher.cpp
@@ -362,6 +359,7 @@ function(AddObject Name Type)
   add_library(${Name} ${Type} ${SRCS})
 
   target_link_libraries(${Name} FEXCore_Base)
+  target_compile_options(${Name} PRIVATE ${FEX_TUNE_COMPILE_FLAGS})
   AddDefaultOptionsToTarget(${Name})
 
   set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
@@ -370,6 +368,7 @@ endfunction()
 function(AddLibrary Name Type)
   add_library(${Name} ${Type} $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
   target_link_libraries(${Name} FEXCore_Base)
+  target_compile_options(${Name} PRIVATE ${FEX_TUNE_COMPILE_FLAGS})
   set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
   if (MINGW_BUILD)
     # Mingw build isn't building a linux shared library, so it can't have a SONAME.

FEXCore/Source/Interface/Config/Config.cpp

@@ -321,16 +321,6 @@ namespace DefaultValues {
     Meta->Load();
 
     // Do configuration option fix ups after everything is reloaded
-    {
-      // Always fix up the number of threads and create the configuration
-      // Otherwise the application could receive zero as the number of threads
-      FEX_CONFIG_OPT(Cores, THREADS);
-      if (Cores == 0) {
-        // When the number of emulated CPU cores is zero then auto detect
-        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_THREADS, fextl::fmt::format("{}", FEXCore::CPUInfo::CalculateNumberOfCPUs()));
-      }
-    }
-
     if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_CORE)) {
       // Sanitize Core option
       FEX_CONFIG_OPT(Core, CORE);

FEXCore/Source/Interface/Config/Config.json.in

@@ -31,15 +31,6 @@
           "Maximum number of instruction to store in a block"
         ]
       },
-      "Threads": {
-        "Type": "uint32",
-        "Default": "0",
-        "ShortArg": "T",
-        "Desc": [
-          "Number of physical hardware threads to tell the process we have.",
-          "0 will auto detect."
-        ]
-      },
       "CacheObjectCodeCompilation": {
         "Type": "uint32",
         "Default": "FEXCore::Config::ConfigObjectCodeHandler::CONFIG_NONE",

FEXCore/Source/Interface/Context/Context.cpp

@@ -26,12 +26,6 @@ namespace FEXCore::Context {
     return fextl::make_unique<FEXCore::Context::ContextImpl>();
   }
 
-  bool FEXCore::Context::ContextImpl::InitializeContext() {
-    // This should be used for generating things that are shared between threads
-    CPUID.Init(this);
-    return true;
-  }
-
   void FEXCore::Context::ContextImpl::SetExitHandler(ExitHandler handler) {
     CustomExitHandler = std::move(handler);
   }
@@ -52,22 +46,10 @@ namespace FEXCore::Context {
     CompileBlock(Thread->CurrentFrame, GuestRIP, MaxInst);
   }
 
-  FEXCore::Context::ExitReason FEXCore::Context::ContextImpl::GetExitReason() {
-    return ParentThread->ExitReason;
-  }
-
   bool FEXCore::Context::ContextImpl::IsDone() const {
     return IsPaused();
   }
 
-  void FEXCore::Context::ContextImpl::GetCPUState(FEXCore::Core::CPUState *State) const {
-    memcpy(State, ParentThread->CurrentFrame, sizeof(FEXCore::Core::CPUState));
-  }
-
-  void FEXCore::Context::ContextImpl::SetCPUState(const FEXCore::Core::CPUState *State) {
-    memcpy(ParentThread->CurrentFrame, State, sizeof(FEXCore::Core::CPUState));
-  }
-
   void FEXCore::Context::ContextImpl::SetCustomCPUBackendFactory(CustomCPUFactoryType Factory) {
     CustomCPUFactory = std::move(Factory);
   }

FEXCore/Source/Interface/Context/Context.h

@@ -14,8 +14,8 @@
 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/CompilerDefs.h>
-#include <FEXCore/Utils/DeferredSignalMutex.h>
 #include <FEXCore/Utils/Event.h>
+#include <FEXCore/Utils/SignalScopeGuards.h>
 #include <FEXCore/fextl/memory.h>
 #include <FEXCore/fextl/set.h>
 #include <FEXCore/fextl/string.h>
@@ -37,7 +37,6 @@
 namespace FEXCore {
 class CodeLoader;
 class ThunkHandler;
-class GdbServer;
 
 namespace CodeSerialize {
   class CodeObjectSerializeService;
@@ -73,8 +72,6 @@ namespace FEXCore::Context {
   class ContextImpl final : public FEXCore::Context::Context {
     public:
       // Context base class implementation.
-      bool InitializeContext() override;
-
       FEXCore::Core::InternalThreadState* InitCore(uint64_t InitialRIP, uint64_t StackPointer) override;
 
       void SetExitHandler(ExitHandler handler) override;
@@ -92,15 +89,8 @@ namespace FEXCore::Context {
       void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) override;
       void CompileRIPCount(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, uint64_t MaxInst) override;
 
-      int GetProgramStatus() const override;
-
-      ExitReason GetExitReason() override;
-
       bool IsDone() const override;
 
-      void GetCPUState(FEXCore::Core::CPUState *State) const override;
-      void SetCPUState(const FEXCore::Core::CPUState *State) override;
-
       void SetCustomCPUBackendFactory(CustomCPUFactoryType Factory) override;
 
       HostFeatures GetHostFeatures() const override;
@@ -108,31 +98,37 @@ namespace FEXCore::Context {
       void HandleCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) override;
 
       uint64_t RestoreRIPFromHostPC(FEXCore::Core::InternalThreadState *Thread, uint64_t HostPC) override;
-      uint32_t ReconstructCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread) override;
+      uint32_t ReconstructCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread, bool WasInJIT, uint64_t *HostGPRs, uint64_t PSTATE) override;
       void SetFlagsFromCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread, uint32_t EFLAGS) override;
 
       /**
        * @brief Used to create FEX thread objects in preparation for creating a true OS thread. Does set a TID or PID.
        *
-       * @param NewThreadState The initial thread state to setup for our state
+       * @param InitialRIP The starting RIP of this thread
+       * @param StackPointer The starting RSP of this thread
+       * @param NewThreadState The initial thread state to setup for our state, if inheriting.
        * @param ParentTID The PID that was the parent thread that created this
        *
        * @return The InternalThreadState object that tracks all of the emulated thread's state
        *
        * Usecases:
+       *  Parent thread Creation:
+       *    - Thread = CreateThread(InitialRIP, InitialStack, nullptr, 0);
+       *    - CTX->RunUntilExit(Thread);
        *  OS thread Creation:
-       *    - Thread = CreateThread(NewState, PPID);
+       *    - Thread = CreateThread(0, 0, NewState, PPID);
        *    - InitializeThread(Thread);
        *  OS fork (New thread created with a clone of thread state):
        *    - clone{2, 3}
-       *    - Thread = CreateThread(CopyOfThreadState, PPID);
+       *    - Thread = CreateThread(0, 0, CopyOfThreadState, PPID);
        *    - ExecutionThread(Thread); // Starts executing without creating another host thread
        *  Thunk callback executing guest code from native host thread
-       *    - Thread = CreateThread(NewState, PPID);
+       *    - Thread = CreateThread(0, 0, NewState, PPID);
        *    - InitializeThreadTLSData(Thread);
        *    - HandleCallback(Thread, RIP);
        */
-      FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) override;
+
+      FEXCore::Core::InternalThreadState* CreateThread(uint64_t InitialRIP, uint64_t StackPointer, FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) override;
 
       // Public for threading
       void ExecutionThread(FEXCore::Core::InternalThreadState *Thread) override;
@@ -290,17 +286,13 @@ namespace FEXCore::Context {
     void WaitForIdle() override;
     void SignalThread(FEXCore::Core::InternalThreadState *Thread, FEXCore::Core::SignalEvent Event);
 
-    bool GetGdbServerStatus() const { return DebugServer != nullptr; }
-    void StartGdbServer();
-    void StopGdbServer();
-
     static void ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
     static void ThreadAddBlockLink(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker);
 
     template<auto Fn>
     static uint64_t ThreadExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
       auto Thread = Frame->Thread;
-      ScopedDeferredSignalWithForkableSharedLock lk(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
+      auto lk = GuardSignalDeferringSection<std::shared_lock>(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
 
       return Fn(Frame, record);
     }
@@ -311,7 +303,7 @@ namespace FEXCore::Context {
       auto Thread = Frame->Thread;
 
       LogMan::Throw::AFmt(Thread->ThreadManager.GetTID() == FHU::Syscalls::gettid(), "Must be called from owning thread {}, not {}", Thread->ThreadManager.GetTID(), FHU::Syscalls::gettid());
-      ScopedDeferredSignalWithForkableUniqueLock lk(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
+      auto lk = GuardSignalDeferringSection(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
 
       ThreadRemoveCodeEntry(Thread, GuestRIP);
     }
@@ -424,15 +416,6 @@ namespace FEXCore::Context {
     }
 
   private:
-    /**
-     * @brief Does some final thread initialization
-     *
-     * @param Thread The internal FEX thread state object
-     *
-     * InitCore and CreateThread both call this to finish up thread object initialization
-     */
-    void InitializeThreadData(FEXCore::Core::InternalThreadState *Thread);
-
     /**
      * @brief Initializes the JIT compilers for the thread
      *
@@ -450,7 +433,6 @@ namespace FEXCore::Context {
 
     // Entry Cache
     std::mutex ExitMutex;
-    fextl::unique_ptr<GdbServer> DebugServer;
 
     IR::AOTIRCaptureCache IRCaptureCache;
     fextl::unique_ptr<FEXCore::CodeSerialize::CodeObjectSerializeService> CodeObjectCacheService;

FEXCore/Source/Interface/Core/ArchHelpers/Arm64Emitter.cpp

@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: MIT
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
+#include "FEXCore/Core/X86Enums.h"
 #include "FEXCore/Utils/AllocatorHooks.h"
 #include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/ArchHelpers/CodeEmitter/Registers.h"
@@ -28,7 +29,7 @@ namespace FEXCore::CPU {
 
 namespace x64 {
   // All but x19 and x29 are caller saved
-  constexpr std::array<FEXCore::ARMEmitter::Register, 16> SRA = {
+  constexpr std::array<FEXCore::ARMEmitter::Register, 18> SRA = {
     FEXCore::ARMEmitter::Reg::r4, FEXCore::ARMEmitter::Reg::r5,
     FEXCore::ARMEmitter::Reg::r6, FEXCore::ARMEmitter::Reg::r7,
     FEXCore::ARMEmitter::Reg::r8, FEXCore::ARMEmitter::Reg::r9,
@@ -36,23 +37,23 @@ namespace x64 {
     FEXCore::ARMEmitter::Reg::r12, FEXCore::ARMEmitter::Reg::r13,
     FEXCore::ARMEmitter::Reg::r14, FEXCore::ARMEmitter::Reg::r15,
     FEXCore::ARMEmitter::Reg::r16, FEXCore::ARMEmitter::Reg::r17,
-    FEXCore::ARMEmitter::Reg::r19, FEXCore::ARMEmitter::Reg::r29
+    FEXCore::ARMEmitter::Reg::r19, FEXCore::ARMEmitter::Reg::r29,
+    // PF/AF must be last.
+    REG_PF, REG_AF,
   };
 
-  constexpr std::array<FEXCore::ARMEmitter::Register, 9> RA = {
+  constexpr std::array<FEXCore::ARMEmitter::Register, 7> RA = {
     // All these callee saved
     FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21,
     FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23,
     FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25,
-    FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27,
     FEXCore::ARMEmitter::Reg::r30,
   };
 
-  constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 4> RAPair = {{
+  constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 3> RAPair = {{
     {FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21},
     {FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23},
     {FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25},
-    {FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27},
   }};
 
   // All are caller saved
@@ -175,19 +176,20 @@ namespace x64 {
 
 namespace x32 {
   // All but x19 and x29 are caller saved
-  constexpr std::array<FEXCore::ARMEmitter::Register, 8> SRA = {
+  constexpr std::array<FEXCore::ARMEmitter::Register, 10> SRA = {
     FEXCore::ARMEmitter::Reg::r4, FEXCore::ARMEmitter::Reg::r5,
     FEXCore::ARMEmitter::Reg::r6, FEXCore::ARMEmitter::Reg::r7,
     FEXCore::ARMEmitter::Reg::r8, FEXCore::ARMEmitter::Reg::r9,
     FEXCore::ARMEmitter::Reg::r10, FEXCore::ARMEmitter::Reg::r11,
+    // PF/AF must be last.
+    REG_PF, REG_AF,
   };
 
-  constexpr std::array<FEXCore::ARMEmitter::Register, 17> RA = {
+  constexpr std::array<FEXCore::ARMEmitter::Register, 15> RA = {
     // All these callee saved
     FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21,
     FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23,
     FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25,
-    FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27,
 
     // Registers only available on 32-bit
     // All these are caller saved (except for r19).
@@ -199,11 +201,10 @@ namespace x32 {
     FEXCore::ARMEmitter::Reg::r19,
   };
 
-  constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 8> RAPair = {{
+  constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 7> RAPair = {{
     {FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21},
     {FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23},
     {FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25},
-    {FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27},
 
     {FEXCore::ARMEmitter::Reg::r12, FEXCore::ARMEmitter::Reg::r13},
     {FEXCore::ARMEmitter::Reg::r14, FEXCore::ARMEmitter::Reg::r15},
@@ -368,7 +369,7 @@ Arm64Emitter::Arm64Emitter(FEXCore::Context::ContextImpl *ctx, void* EmissionPtr
     GeneralFPRegisters = x64::RAFPR;
   }
   else {
-    ConfiguredDynamicRegisterBase = std::span(x32::RA.begin() + 8, 8);
+    ConfiguredDynamicRegisterBase = std::span(x32::RA.begin() + 6, 8);
 
     StaticRegisters = x32::SRA;
     GeneralRegisters = x32::RA;
@@ -399,6 +400,15 @@ void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, ui
     Segments = 2;
   }
 
+  if (!Is64Bit && ((~Constant) & 0xFFFF0000) == 0) {
+    movn(s, Reg.W(), (~Constant) & 0xFFFF);
+
+    if (NOPPad) {
+      nop(); nop(); nop();
+    }
+    return;
+  }
+
   int RequiredMoveSegments{};
 
   // Count the number of move segments
@@ -581,6 +591,7 @@ void Arm64Emitter::SpillStaticRegs(FEXCore::ARMEmitter::Register TmpReg, bool FP
     // Disable FPCR.NEP and FPCR.AH
     // NEP(2): Changes ASIMD scalar instructions to insert in to the lower bits of the destination.
     // AH(1):  Changes NaN behaviour in some instructions. Specifically fmin, fmax.
+    //         Also interacts with RPRES to change reciprocal/rsqrt precision from 8-bit mantissa to 12-bit.
     //
     // Additional interesting AFP bits:
     // FIZ(0): Flush Inputs to Zero
@@ -592,10 +603,23 @@ void Arm64Emitter::SpillStaticRegs(FEXCore::ARMEmitter::Register TmpReg, bool FP
   }
 #endif
 
+  // Regardless of what GPRs/FPRs we're spilling, we need to spill NZCV since it
+  // is always static and almost certainly clobbered by the subsequent code.
+  //
+  // TODO: Can we prove that NZCV is not used across a call in some cases and
+  // omit this? Might help x87 perf? Future idea.
+  mrs(TmpReg, ARMEmitter::SystemRegister::NZCV);
+  str(TmpReg.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+
   if (!StaticRegisterAllocation()) {
     return;
   }
 
+  // PF/AF are special, remove them from the mask
+  uint32_t PFAFMask = ((1u << REG_PF.Idx()) | ((1u << REG_AF.Idx())));
+  unsigned PFAFSpillMask = GPRSpillMask & PFAFMask;
+  GPRSpillMask &= ~PFAFSpillMask;
+
   for (size_t i = 0; i < StaticRegisters.size(); i+=2) {
     auto Reg1 = StaticRegisters[i];
     auto Reg2 = StaticRegisters[i+1];
@@ -611,6 +635,14 @@ void Arm64Emitter::SpillStaticRegs(FEXCore::ARMEmitter::Register TmpReg, bool FP
     }
   }
 
+  // Now handle PF/AF
+  if (PFAFSpillMask) {
+    LOGMAN_THROW_A_FMT(PFAFSpillMask == PFAFMask, "PF/AF not spilled together");
+
+    str(REG_PF.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.pf_raw));
+    str(REG_AF.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.af_raw));
+  }
+
   if (FPRs) {
     if (EmitterCTX->HostFeatures.SupportsAVX) {
       for (size_t i = 0; i < StaticFPRegisters.size(); i++) {
@@ -658,7 +690,7 @@ void Arm64Emitter::SpillStaticRegs(FEXCore::ARMEmitter::Register TmpReg, bool FP
 void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRFillMask) {
   FEXCore::ARMEmitter::Register TmpReg = FEXCore::ARMEmitter::Reg::r0;
   LOGMAN_THROW_A_FMT(GPRFillMask != 0, "Must fill at least 1 GPR for a temp");
-  bool FoundRegister{};
+  [[maybe_unused]] bool FoundRegister{};
   for (auto Reg : StaticRegisters) {
     if (((1U << Reg.Idx()) & GPRFillMask)) {
       TmpReg = Reg;
@@ -688,6 +720,14 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF
   }
 #endif
 
+  // Regardless of what GPRs/FPRs we're filling, we need to fill NZCV since it
+  // is always static and was almost certainly clobbered.
+  //
+  // TODO: Can we prove that NZCV is not used across a call in some cases and
+  // omit this? Might help x87 perf? Future idea.
+  ldr(TmpReg.W(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.flags[24]));
+  msr(ARMEmitter::SystemRegister::NZCV, TmpReg);
+
   if (!StaticRegisterAllocation()) {
     return;
   }
@@ -745,6 +785,11 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF
     }
   }
 
+  // PF/AF are special, remove them from the mask
+  uint32_t PFAFMask = ((1u << REG_PF.Idx()) | ((1u << REG_AF.Idx())));
+  uint32_t PFAFFillMask = GPRFillMask & PFAFMask;
+  GPRFillMask &= ~PFAFMask;
+
   for (size_t i = 0; i < StaticRegisters.size(); i+=2) {
     auto Reg1 = StaticRegisters[i];
     auto Reg2 = StaticRegisters[i+1];
@@ -759,6 +804,14 @@ void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRF
       ldr(Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1]));
     }
   }
+
+  // Now handle PF/AF
+  if (PFAFFillMask) {
+    LOGMAN_THROW_A_FMT(PFAFFillMask == PFAFMask, "PF/AF not filled together");
+
+    ldr(REG_PF.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.pf_raw));
+    ldr(REG_AF.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.af_raw));
+  }
 }
 
 void Arm64Emitter::PushVectorRegisters(FEXCore::ARMEmitter::Register TmpReg, bool SVERegs, std::span<const FEXCore::ARMEmitter::VRegister> VRegs) {

FEXCore/Source/Interface/Core/ArchHelpers/Arm64Emitter.h

@@ -53,6 +53,10 @@ constexpr auto VTMP2 = FEXCore::ARMEmitter::VReg::v1;
 constexpr FEXCore::ARMEmitter::PRegister PRED_TMP_16B = FEXCore::ARMEmitter::PReg::p6;
 constexpr FEXCore::ARMEmitter::PRegister PRED_TMP_32B = FEXCore::ARMEmitter::PReg::p7;
 
+// We pin r26/r27 as PF/AF respectively, this is internal FEX ABI.
+constexpr auto REG_PF = FEXCore::ARMEmitter::Reg::r26;
+constexpr auto REG_AF = FEXCore::ARMEmitter::Reg::r27;
+
 // This class contains common emitter utility functions that can
 // be used by both Arm64 JIT and ARM64 Dispatcher
 class Arm64Emitter : public FEXCore::ARMEmitter::Emitter {
@@ -128,21 +132,21 @@ protected:
 
   void SpillForABICall(bool SupportsPreserveAllABI, FEXCore::ARMEmitter::Register TmpReg, bool FPRs = true) {
     if (SupportsPreserveAllABI) {
-      SpillForPreserveAllABICall(TMP1, true);
+      SpillForPreserveAllABICall(TmpReg, FPRs);
     }
     else {
-      SpillStaticRegs(TMP1);
-      PushDynamicRegsAndLR(TMP1);
+      SpillStaticRegs(TmpReg, FPRs);
+      PushDynamicRegsAndLR(TmpReg);
     }
   }
 
   void FillForABICall(bool SupportsPreserveAllABI, bool FPRs = true) {
     if (SupportsPreserveAllABI) {
-      FillForPreserveAllABICall(true);
+      FillForPreserveAllABICall(FPRs);
     }
     else {
       PopDynamicRegsAndLR();
-      FillStaticRegs();
+      FillStaticRegs(FPRs);
     }
   }
 

FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/ALUOps.inl

@@ -771,6 +771,16 @@ public:
     dc32(Op);
   }
 
+  void axflag() {
+    constexpr uint32_t Op = 0b1101'0101'0000'0000'0100'0000'0101'1111;
+    dc32(Op);
+  }
+
+  void xaflag() {
+    constexpr uint32_t Op = 0b1101'0101'0000'0000'0100'0000'0011'1111;
+    dc32(Op);
+  }
+
   // Conditional compare - register
   void ccmn(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::StatusFlags flags, FEXCore::ARMEmitter::Condition Cond) {
     constexpr uint32_t Op = 0b0011'1010'010 << 21;

FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/ASIMDOps.inl

@@ -60,7 +60,7 @@ public:
   }
   void sha256su1(FEXCore::ARMEmitter::VRegister rd, FEXCore::ARMEmitter::VRegister rn, FEXCore::ARMEmitter::VRegister rm) {
     constexpr uint32_t Op = 0b0101'1110'0000'0000'0000'00 << 10;
-    Crypto3RegSHA(Op, 0b100, rd, rn, rm);
+    Crypto3RegSHA(Op, 0b110, rd, rn, rm);
   }
 
   // Cryptographic two-register SHA

FEXCore/Source/Interface/Core/CPUID.cpp

@@ -106,11 +106,10 @@ static uint32_t GetCycleCounterFrequency() {
 }
 
 void CPUIDEmu::SetupHostHybridFlag() {
-  size_t CPUs = FEXCore::CPUInfo::CalculateNumberOfCPUs();
-  PerCPUData.resize(CPUs);
+  PerCPUData.resize(Cores);
 
   uint64_t MIDR{};
-  for (size_t i = 0; i < CPUs; ++i) {
+  for (size_t i = 0; i < Cores; ++i) {
     std::error_code ec{};
     fextl::string MIDRPath = fextl::fmt::format("/sys/devices/system/cpu/cpu{}/regs/identification/midr_el1", i);
 
@@ -218,7 +217,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
     fextl::vector<const CPUMIDR*> LittleCores;
 
     // Separate CPU cores out to big or little selected
-    for (size_t i = 0; i < CPUs; ++i) {
+    for (size_t i = 0; i < Cores; ++i) {
       uint32_t MIDR = PerCPUData[i].MIDR;
       auto MIDROption = FindDefinedMIDR(MIDR);
       if (MIDROption) {
@@ -334,7 +333,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
   }
   else {
     // If we aren't hybrid then just claim everything is big
-    for (size_t i = 0; i < CPUs; ++i) {
+    for (size_t i = 0; i < Cores; ++i) {
       uint32_t MIDR = PerCPUData[i].MIDR;
       auto MIDROption = FindDefinedMIDR(MIDR);
 
@@ -380,7 +379,6 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_0h(uint32_t Leaf) const {
 // Processor Info and Features bits
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {
   FEXCore::CPUID::FunctionResults Res{};
-  uint32_t CoreCount = Cores();
 
   // Hypervisor bit is normally set but some applications have issues with it.
   uint32_t Hypervisor = HideHypervisorBit() ? 0 : 1;
@@ -389,7 +387,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) const {
 
   Res.ebx = 0 | // Brand index
     (8 << 8) | // Cache line size in bytes
-    (CoreCount << 16) | // Number of addressable IDs for the logical cores in the physical CPU
+    (Cores << 16) | // Number of addressable IDs for the logical cores in the physical CPU
     (0 << 24); // Local APIC ID
 
   Res.ecx =
@@ -496,7 +494,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_04h(uint32_t Leaf) const {
 
   if (Leaf == 0) {
     // Report L1D
-    uint32_t CoreCount = Cores() - 1;
+    uint32_t CoreCount = Cores - 1;
 
     Res.eax = CacheType_Data | // Cache type
       (0b001 << 5) |      // Cache level
@@ -520,7 +518,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_04h(uint32_t Leaf) const {
   }
   else if (Leaf == 1) {
     // Report L1I
-    uint32_t CoreCount = Cores() - 1;
+    uint32_t CoreCount = Cores - 1;
 
     Res.eax = CacheType_Instruction | // Cache type
       (0b001 << 5) |      // Cache level
@@ -544,7 +542,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_04h(uint32_t Leaf) const {
   }
   else if (Leaf == 2) {
     // Report L2
-    uint32_t CoreCount = Cores() - 1;
+    uint32_t CoreCount = Cores - 1;
 
     Res.eax = CacheType_Unified | // Cache type
       (0b010 << 5) |      // Cache level
@@ -568,7 +566,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_04h(uint32_t Leaf) const {
   }
   else if (Leaf == 3) {
     // Report L3
-    uint32_t CoreCount = Cores() - 1;
+    uint32_t CoreCount = Cores - 1;
 
     Res.eax = CacheType_Unified | // Cache type
       (0b011 << 5) |      // Cache level
@@ -1070,7 +1068,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0008h(uint32_t Leaf) con
     (0 << 1) | // IRPerf: Instructions retired count support
     (CTX->HostFeatures.SupportsCLZERO << 0);  // CLZERO support
 
-  uint32_t CoreCount = Cores() - 1;
+  uint32_t CoreCount = Cores - 1;
   Res.ecx =
     (0 << 16) |       // PerfTscSize: Performance timestamp count size
     ((uint32_t)std::log2(CoreCount + 1) << 12) |       // ApicIdSize: Number of bits in ApicID
@@ -1168,7 +1166,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_001Dh(uint32_t Leaf) con
   }
   else if (Leaf == 3) {
     // Report L3
-    uint32_t CoreCount = Cores() - 1;
+    uint32_t CoreCount = Cores - 1;
 
     Res.eax = CacheType_Unified | // Cache type
       (0b011 << 5) |      // Cache level
@@ -1209,6 +1207,7 @@ FEXCore::CPUID::XCRResults CPUIDEmu::XCRFunction_0h() const {
 
 void CPUIDEmu::Init(FEXCore::Context::ContextImpl *ctx) {
   CTX = ctx;
+  Cores = FEXCore::CPUInfo::CalculateNumberOfCPUs();
 
   // Setup some state tracking
   SetupHostHybridFlag();

FEXCore/Source/Interface/Core/CPUID.h

@@ -113,7 +113,7 @@ public:
 private:
   FEXCore::Context::ContextImpl *CTX;
   bool Hybrid{};
-  FEX_CONFIG_OPT(Cores, THREADS);
+  uint32_t Cores{};
   FEX_CONFIG_OPT(HideHypervisorBit, HIDEHYPERVISORBIT);
 
   // XFEATURE_ENABLED_MASK

FEXCore/Source/Interface/Core/Core.cpp

@@ -9,12 +9,11 @@ $end_info$
 */
 
 #include <cstdint>
-#include "FEXCore/Utils/DeferredSignalMutex.h"
 #include "Interface/Context/Context.h"
+#include "Interface/Core/ArchHelpers//Arm64Emitter.h"
 #include "Interface/Core/LookupCache.h"
 #include "Interface/Core/CPUID.h"
 #include "Interface/Core/Frontend.h"
-#include "Interface/Core/GdbServer.h"
 #include "Interface/Core/ObjectCache/ObjectCacheService.h"
 #include "Interface/Core/OpcodeDispatcher.h"
 #include "Interface/Core/JIT/JITCore.h"
@@ -46,6 +45,7 @@ $end_info$
 #include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/File.h>
 #include <FEXCore/Utils/LogManager.h>
+#include "FEXCore/Utils/SignalScopeGuards.h"
 #include <FEXCore/Utils/Threads.h>
 #include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/fextl/fmt.h>
@@ -76,64 +76,6 @@ $end_info$
 #include <utility>
 #include <xxhash.h>
 
-namespace FEXCore::Core {
-struct ThreadLocalData {
-  FEXCore::Core::InternalThreadState* Thread;
-};
-
-constexpr std::array<std::string_view const, 22> FlagNames = {
-  "CF",
-  "",
-  "PF",
-  "",
-  "AF",
-  "",
-  "ZF",
-  "SF",
-  "TF",
-  "IF",
-  "DF",
-  "OF",
-  "IOPL",
-  "",
-  "NT",
-  "",
-  "RF",
-  "VM",
-  "AC",
-  "VIF",
-  "VIP",
-  "ID",
-};
-
-std::string_view const& GetFlagName(unsigned Flag) {
-  return FlagNames[Flag];
-}
-
-constexpr std::array<std::string_view const, 16> RegNames = {
-  "rax",
-  "rbx",
-  "rcx",
-  "rdx",
-  "rsi",
-  "rdi",
-  "rbp",
-  "rsp",
-  "r8",
-  "r9",
-  "r10",
-  "r11",
-  "r12",
-  "r13",
-  "r14",
-  "r15",
-};
-
-std::string_view const& GetGRegName(unsigned Reg) {
-  return RegNames[Reg];
-}
-} // namespace FEXCore::Core
-
 namespace FEXCore::Context {
   ContextImpl::ContextImpl()
   : IRCaptureCache {this} {
@@ -157,6 +99,8 @@ namespace FEXCore::Context {
 
     // Track atomic TSO emulation configuration.
     UpdateAtomicTSOEmulationConfig();
+
+    CPUID.Init(this);
   }
 
   ContextImpl::~ContextImpl() {
@@ -221,7 +165,7 @@ namespace FEXCore::Context {
     return Frame->State.rip;
   }
 
-  uint32_t ContextImpl::ReconstructCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread) {
+  uint32_t ContextImpl::ReconstructCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread, bool WasInJIT, uint64_t *HostGPRs, uint64_t PSTATE) {
     const auto Frame = Thread->CurrentFrame;
     uint32_t EFLAGS{};
 
@@ -243,9 +187,23 @@ namespace FEXCore::Context {
       }
     }
 
-    // SF/ZF/CF/OF are packed in a 32-bit value in RFLAG_NZCV_LOC.
     uint32_t Packed_NZCV{};
-    memcpy(&Packed_NZCV, &Frame->State.flags[X86State::RFLAG_NZCV_LOC], sizeof(Packed_NZCV));
+    if (WasInJIT) {
+      // If we were in the JIT then NZCV is in the CPU's PSTATE object.
+      // Packed in to the same bit locations as RFLAG_NZCV_LOC.
+      Packed_NZCV = PSTATE;
+
+      // If we were in the JIT then PF and AF are in registers.
+      // Move them to the CPUState frame now.
+      Frame->State.pf_raw = HostGPRs[CPU::REG_PF.Idx()];
+      Frame->State.af_raw = HostGPRs[CPU::REG_AF.Idx()];
+    }
+    else {
+      // If we were not in the JIT then the NZCV state is stored in the CPUState RFLAG_NZCV_LOC.
+      // SF/ZF/CF/OF are packed in a 32-bit value in RFLAG_NZCV_LOC.
+      memcpy(&Packed_NZCV, &Frame->State.flags[X86State::RFLAG_NZCV_LOC], sizeof(Packed_NZCV));
+    }
+
     uint32_t OF = (Packed_NZCV >> IR::OpDispatchBuilder::IndexNZCV(X86State::RFLAG_OF_RAW_LOC)) & 1;
     uint32_t CF = (Packed_NZCV >> IR::OpDispatchBuilder::IndexNZCV(X86State::RFLAG_CF_RAW_LOC)) & 1;
     uint32_t ZF = (Packed_NZCV >> IR::OpDispatchBuilder::IndexNZCV(X86State::RFLAG_ZF_RAW_LOC)) & 1;
@@ -259,13 +217,13 @@ namespace FEXCore::Context {
 
     // PF calculation is deferred, calculate it now.
     // Popcount the 8-bit flag and then extract the lower bit.
-    uint32_t PFByte = Frame->State.flags[X86State::RFLAG_PF_RAW_LOC];
+    uint32_t PFByte = Frame->State.pf_raw & 0xff;
     uint32_t PF = std::popcount(PFByte ^ 1) & 1;
     EFLAGS |= PF << X86State::RFLAG_PF_RAW_LOC;
 
     // AF calculation is deferred, calculate it now.
     // XOR with PF byte and extract bit 4.
-    uint32_t AF = ((Frame->State.flags[X86State::RFLAG_AF_RAW_LOC] ^ PFByte) & (1 << 4)) ? 1 : 0;
+    uint32_t AF = ((Frame->State.af_raw ^ PFByte) & (1 << 4)) ? 1 : 0;
     EFLAGS |= AF << X86State::RFLAG_AF_RAW_LOC;
 
     return EFLAGS;
@@ -285,11 +243,11 @@ namespace FEXCore::Context {
           // AF stored in bit 4 in our internal representation. It is also
           // XORed with byte 4 of the PF byte, but we write that as zero here so
           // we don't need any special handling for that.
-          Frame->State.flags[i] = (EFLAGS & (1U << i)) ? (1 << 4) : 0;
+          Frame->State.af_raw = (EFLAGS & (1U << i)) ? (1 << 4) : 0;
           break;
         case X86State::RFLAG_PF_RAW_LOC:
           // PF is inverted in our internal representation.
-          Frame->State.flags[i] = (EFLAGS & (1U << i)) ? 0 : 1;
+          Frame->State.pf_raw = (EFLAGS & (1U << i)) ? 0 : 1;
           break;
         default:
           Frame->State.flags[i] = (EFLAGS & (1U << i)) ? 1 : 0;
@@ -359,13 +317,6 @@ namespace FEXCore::Context {
     // Give this configuration to the SignalDelegator.
     SignalDelegation->SetConfig(SignalConfig);
 
-    if (Config.GdbServer) {
-      StartGdbServer();
-    }
-    else {
-      StopGdbServer();
-    }
-
 #ifndef _WIN32
     ThunkHandler = FEXCore::ThunkHandler::Create();
 #else
@@ -373,36 +324,21 @@ namespace FEXCore::Context {
     Config.NeedsPendingInterruptFaultCheck = true;
 #endif
 
-    using namespace FEXCore::Core;
+    if (Config.GdbServer) {
+      // If gdbserver is enabled then this needs to be enabled.
+      Config.NeedsPendingInterruptFaultCheck = true;
+      // FEX needs to start paused when gdb is enabled.
+      StartPaused = true;
+    }
 
-    FEXCore::Core::InternalThreadState *Thread = CreateThread(nullptr, 0);
+    FEXCore::Core::InternalThreadState *Thread = CreateThread(InitialRIP, StackPointer, nullptr, 0);
 
     // We are the parent thread
     ParentThread = Thread;
 
-    Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
-
-    Thread->CurrentFrame->State.rip = InitialRIP;
-
-    InitializeThreadData(Thread);
     return Thread;
   }
 
-  void ContextImpl::StartGdbServer() {
-#ifndef _WIN32
-    if (!DebugServer) {
-      DebugServer = fextl::make_unique<GdbServer>(this, SignalDelegation, SyscallHandler);
-      StartPaused = true;
-    }
-#endif
-  }
-
-  void ContextImpl::StopGdbServer() {
-#ifndef _WIN32
-    DebugServer.reset();
-#endif
-  }
-
   void ContextImpl::HandleCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) {
     static_cast<ContextImpl*>(Thread->CTX)->Dispatcher->ExecuteJITCallback(Thread->CurrentFrame, RIP);
   }
@@ -575,14 +511,6 @@ namespace FEXCore::Context {
     Dispatcher->ExecuteDispatch(Thread->CurrentFrame);
   }
 
-  int ContextImpl::GetProgramStatus() const {
-    return ParentThread->StatusCode;
-  }
-
-  void ContextImpl::InitializeThreadData(FEXCore::Core::InternalThreadState *Thread) {
-    Thread->CPUBackend->Initialize();
-  }
-
   struct ExecutionThreadHandler {
     ContextImpl *This;
     FEXCore::Core::InternalThreadState *Thread;
@@ -681,20 +609,22 @@ namespace FEXCore::Context {
     Thread->PassManager->Finalize();
   }
 
-  FEXCore::Core::InternalThreadState* ContextImpl::CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) {
+  FEXCore::Core::InternalThreadState* ContextImpl::CreateThread(uint64_t InitialRIP, uint64_t StackPointer, FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) {
     FEXCore::Core::InternalThreadState *Thread = new FEXCore::Core::InternalThreadState{};
 
+    Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
+    Thread->CurrentFrame->State.rip = InitialRIP;
+
     // Copy over the new thread state to the new object
     if (NewThreadState) {
-      memcpy(Thread->CurrentFrame, NewThreadState, sizeof(FEXCore::Core::CPUState));
+      memcpy(&Thread->CurrentFrame->State, NewThreadState, sizeof(FEXCore::Core::CPUState));
     }
-    Thread->CurrentFrame->Thread = Thread;
 
     // Set up the thread manager state
     Thread->ThreadManager.parent_tid = ParentTID;
+    Thread->CurrentFrame->Thread = Thread;
 
     InitializeCompiler(Thread);
-    InitializeThreadData(Thread);
 
     Thread->CurrentFrame->State.DeferredSignalRefCount.Store(0);
     Thread->CurrentFrame->State.DeferredSignalFaultAddress = reinterpret_cast<Core::NonAtomicRefCounter<uint64_t>*>(FEXCore::Allocator::VirtualAlloc(4096));
@@ -1113,7 +1043,7 @@ namespace FEXCore::Context {
     auto Thread = Frame->Thread;
 
     // Invalidate might take a unique lock on this, to guarantee that during invalidation no code gets compiled
-    ScopedDeferredSignalWithForkableSharedLock lk(CodeInvalidationMutex, Thread);
+    auto lk = GuardSignalDeferringSection<std::shared_lock>(CodeInvalidationMutex, Thread);
 
     // Is the code in the cache?
     // The backends only check L1 and L2, not L3
@@ -1296,7 +1226,7 @@ namespace FEXCore::Context {
     // Potential deferred since Thread might not be valid.
     // Thread object isn't valid very early in frontend's initialization.
     // To be more optimal the frontend should provide this code with a valid Thread object earlier.
-    ScopedPotentialDeferredSignalWithForkableUniqueLock lk(CodeInvalidationMutex, Thread);
+    auto lk = GuardSignalDeferringSectionWithFallback(CodeInvalidationMutex, Thread);
 
     InvalidateGuestCodeRangeInternal(this, Start, Length);
   }
@@ -1305,7 +1235,7 @@ namespace FEXCore::Context {
     // Potential deferred since Thread might not be valid.
     // Thread object isn't valid very early in frontend's initialization.
     // To be more optimal the frontend should provide this code with a valid Thread object earlier.
-    ScopedPotentialDeferredSignalWithForkableUniqueLock lk(CodeInvalidationMutex, Thread);
+    auto lk = GuardSignalDeferringSectionWithFallback(CodeInvalidationMutex, Thread);
 
     InvalidateGuestCodeRangeInternal(this, Start, Length);
     CallAfter(Start, Length);
@@ -1333,7 +1263,7 @@ namespace FEXCore::Context {
   }
 
   void ContextImpl::ThreadAddBlockLink(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker) {
-    ScopedDeferredSignalWithForkableSharedLock lk(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
+    auto lk = GuardSignalDeferringSection<std::shared_lock>(static_cast<ContextImpl*>(Thread->CTX)->CodeInvalidationMutex, Thread);
 
     Thread->LookupCache->AddBlockLink(GuestDestination, HostLink, delinker);
   }
@@ -1381,17 +1311,11 @@ namespace FEXCore::Context {
 
   IR::AOTIRCacheEntry *ContextImpl::LoadAOTIRCacheEntry(const fextl::string &filename) {
     auto rv = IRCaptureCache.LoadAOTIRCacheEntry(filename);
-    if (DebugServer) {
-      DebugServer->AlertLibrariesChanged();
-    }
     return rv;
   }
 
   void ContextImpl::UnloadAOTIRCacheEntry(IR::AOTIRCacheEntry *Entry) {
     IRCaptureCache.UnloadAOTIRCacheEntry(Entry);
-    if (DebugServer) {
-      DebugServer->AlertLibrariesChanged();
-    }
   }
 
   void ContextImpl::AppendThunkDefinitions(fextl::vector<FEXCore::IR::ThunkDefinition> const& Definitions) {

FEXCore/Source/Interface/Core/Dispatcher/Dispatcher.h

@@ -85,7 +85,9 @@ public:
 #endif
 
   uint16_t GetSRAGPRCount() const {
-    return StaticRegisters.size();
+    // PF/AF are the final two SRA registers.
+    // Only return the SRA for GPRs.
+    return StaticRegisters.size() - 2;
   }
 
   uint16_t GetSRAFPRCount() const {
@@ -93,7 +95,7 @@ public:
   }
 
   void GetSRAGPRMapping(uint8_t Mapping[16]) const {
-    for (size_t i = 0; i < StaticRegisters.size(); ++i) {
+    for (size_t i = 0; i < StaticRegisters.size() - 2; ++i) {
       Mapping[i] = StaticRegisters[i].Idx();
     }
   }

FEXCore/Source/Interface/Core/HostFeatures.cpp

@@ -180,11 +180,13 @@ static void OverrideFeatures(HostFeatures *Features) {
   if (EnableCrypto) {
     Features->SupportsAES = true;
     Features->SupportsCRC = true;
+    Features->SupportsSHA = true;
     Features->SupportsPMULL_128Bit = true;
   }
   else if (DisableCrypto) {
     Features->SupportsAES = false;
     Features->SupportsCRC = false;
+    Features->SupportsSHA = false;
     Features->SupportsPMULL_128Bit = false;
   }
   if (EnableRPRES) {
@@ -211,6 +213,8 @@ HostFeatures::HostFeatures() {
 
   SupportsAES = Features.Has(vixl::CPUFeatures::Feature::kAES);
   SupportsCRC = Features.Has(vixl::CPUFeatures::Feature::kCRC32);
+  SupportsSHA = Features.Has(vixl::CPUFeatures::Feature::kSHA1) &&
+                Features.Has(vixl::CPUFeatures::Feature::kSHA2);
   SupportsAtomics = Features.Has(vixl::CPUFeatures::Feature::kAtomics);
   SupportsRAND = Features.Has(vixl::CPUFeatures::Feature::kRNG);
 
@@ -238,13 +242,14 @@ HostFeatures::HostFeatures() {
 #endif
   // TODO: AVX2 is currently unsupported. Disable until the remaining features are implemented.
   SupportsAVX2 = false;
-  SupportsSHA = true;
   SupportsBMI1 = true;
   SupportsBMI2 = true;
   SupportsCLWB = true;
 
   // TODO: AFP is disabled until the scalar usage in the codebase can be audited to be working as expected.
   SupportsAFP = false;
+  // RPRES has a dependency on AFP. Disable it until AFP is enabled.
+  SupportsRPRES = false;
 
   if (!SupportsAtomics) {
     WARN_ONCE_FMT("Host CPU doesn't support atomics. Expect bad performance");
@@ -283,6 +288,8 @@ HostFeatures::HostFeatures() {
 #ifdef VIXL_SIMULATOR
   // simulator doesn't support dc(ZVA)
   SupportsCLZERO = false;
+  // Simulator doesn't support SHA
+  SupportsSHA = false;
 #else
   // Check if we can support cacheline clears
   uint32_t DCZID = GetDCZID();

FEXCore/Source/Interface/Core/JIT/Arm64/ALUOps.cpp

@@ -87,7 +87,7 @@ DEF_OP(Add) {
 
 DEF_OP(AddNZCV) {
   auto Op = IROp->C<IR::IROp_AddNZCV>();
-  const IR::OpSize OpSize = Op->Size;
+  const auto OpSize = IROp->Size;
 
   LOGMAN_THROW_AA_FMT(OpSize == IR::i32Bit || OpSize == IR::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
   const auto EmitSize = OpSize == IR::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
@@ -98,78 +98,63 @@ DEF_OP(AddNZCV) {
   } else {
     cmn(EmitSize, GetReg(Op->Src1.ID()), GetReg(Op->Src2.ID()));
   }
-
-  // TODO: Optimize this out
-  mrs(GetReg(Node), ARMEmitter::SystemRegister::NZCV);
 }
 
 DEF_OP(AdcNZCV) {
   auto Op = IROp->C<IR::IROp_AdcNZCV>();
-  const IR::OpSize OpSize = Op->Size;
+  const auto OpSize = IROp->Size;
 
   LOGMAN_THROW_AA_FMT(OpSize == IR::i32Bit || OpSize == IR::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
   const auto EmitSize = OpSize == IR::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
-  const auto Dst = GetReg(Node);
-
-  // TODO: Optimize this out
-  msr(ARMEmitter::SystemRegister::NZCV, GetReg(Op->NZCV.ID()));
 
   adcs(EmitSize, ARMEmitter::Reg::zr, GetReg(Op->Src1.ID()), GetReg(Op->Src2.ID()));
-
-  // TODO: Optimize this out
-  mrs(Dst, ARMEmitter::SystemRegister::NZCV);
 }
 
 DEF_OP(SbbNZCV) {
   auto Op = IROp->C<IR::IROp_SbbNZCV>();
-  const IR::OpSize OpSize = Op->Size;
+  const auto OpSize = IROp->Size;
 
   LOGMAN_THROW_AA_FMT(OpSize == IR::i32Bit || OpSize == IR::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
   const auto EmitSize = OpSize == IR::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
 
-  const auto Dst = GetReg(Node);
-
-  // Carry-in needs to be inverted for subtractions due to carry versus borrow
-  // distinction between x86 and arm.
-  // See below remarks on cfinv
-  eor(ARMEmitter::Size::i32Bit, TMP1, GetReg(Op->NZCV.ID()), 1u << 29);
-
-  // TODO: Optimize this out
-  msr(ARMEmitter::SystemRegister::NZCV, TMP1);
-
   sbcs(EmitSize, ARMEmitter::Reg::zr, GetReg(Op->Src1.ID()), GetReg(Op->Src2.ID()));
-
-  // TODO: Optimize this out
-  mrs(Dst, ARMEmitter::SystemRegister::NZCV);
-
-  // The carry flag produced by arm64 sbcs is inverted compared to the x86 carry
-  // flag. Invert it now.
-  //
-  // TODO: Once we optimize out the mrs, this will become a cfinv operation, but
-  // that's only available with Feat_FlagM. For now the portable way is to flip
-  // bit 29 (carry) manually.
-  eor(ARMEmitter::Size::i32Bit, Dst, Dst, 1u << 29);
 }
 
 DEF_OP(TestNZ) {
   auto Op = IROp->C<IR::IROp_TestNZ>();
-  const uint8_t OpSize = Op->Size;
+  const uint8_t OpSize = IROp->Size;
   const auto EmitSize = OpSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
 
-  const auto Dst = GetReg(Node);
-  auto Src = GetReg(Op->Src1.ID());
+  uint64_t Const;
+  auto Src1 = GetReg(Op->Src1.ID());
 
   // Shift the sign bit into place, clearing out the garbage in upper bits.
-  // setf+rmif would avoid the scratch register, but higher latency on M1.
+  // Adding zero does an effective test, setting NZ according to the result and
+  // zeroing CV.
   if (OpSize < 4) {
-    lsl(EmitSize, Dst, Src, 32 - (OpSize * 8));
-    Src = Dst;
+    // Cheaper to and+cmn than to lsl+lsl+tst, so do the and ourselves if
+    // needed.
+    if (Op->Src1 != Op->Src2) {
+      if (IsInlineConstant(Op->Src2, &Const)) {
+        and_(EmitSize, TMP1, Src1, Const);
+      } else {
+        auto Src2 = GetReg(Op->Src2.ID());
+        and_(EmitSize, TMP1, Src1, Src2);
+      }
+
+      Src1 = TMP1;
+    }
+
+    unsigned Shift = 32 - (OpSize * 8);
+    cmn(EmitSize, ARMEmitter::Reg::zr, Src1, ARMEmitter::ShiftType::LSL, Shift);
+  } else {
+    if (IsInlineConstant(Op->Src2, &Const)) {
+      tst(EmitSize, Src1, Const);
+    } else {
+      const auto Src2 = GetReg(Op->Src2.ID());
+      tst(EmitSize, Src1, Src2);
+    }
   }
-
-  tst(EmitSize, Src, Src);
-
-  // TODO: Optimize this out
-  mrs(Dst, ARMEmitter::SystemRegister::NZCV);
 }
 
 DEF_OP(Sub) {
@@ -199,7 +184,7 @@ DEF_OP(SubShift) {
 
 DEF_OP(SubNZCV) {
   auto Op = IROp->C<IR::IROp_SubNZCV>();
-  const IR::OpSize OpSize = Op->Size;
+  const auto OpSize = IROp->Size;
 
   LOGMAN_THROW_AA_FMT(OpSize == IR::i32Bit || OpSize == IR::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
   const auto EmitSize = OpSize == IR::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
@@ -213,20 +198,70 @@ DEF_OP(SubNZCV) {
   } else {
     cmp(EmitSize, GetReg(Op->Src1.ID()), GetReg(Op->Src2.ID()));
   }
+}
 
-  const auto Dst = GetReg(Node);
+DEF_OP(CarryInvert) {
+  LOGMAN_THROW_A_FMT(CTX->HostFeatures.SupportsFlagM, "Unsupported flagm op");
+  cfinv();
+}
 
-  // TODO: Optimize this out
-  mrs(Dst, ARMEmitter::SystemRegister::NZCV);
+DEF_OP(RmifNZCV) {
+  auto Op = IROp->C<IR::IROp_RmifNZCV>();
+  LOGMAN_THROW_A_FMT(CTX->HostFeatures.SupportsFlagM, "Unsupported flagm op");
 
-  if (Op->InvertCarry) {
-    // The carry flag produced by arm64 subs is inverted compared to the x86 carry
-    // flag. Invert it now.
-    //
-    // TODO: Once we optimize out the mrs, this will become a cfinv operation, but
-    // that's only available with Feat_FlagM. For now the portable way is to flip
-    // bit 29 (carry) manually.
-    eor(ARMEmitter::Size::i32Bit, Dst, Dst, 1u << 29);
+  rmif(GetReg(Op->Src.ID()).X(), Op->Rotate, Op->Mask);
+}
+
+DEF_OP(AXFlag) {
+  LOGMAN_THROW_A_FMT(CTX->HostFeatures.SupportsFlagM2, "Unsupported flagm2 op");
+  axflag();
+}
+
+ARMEmitter::Condition MapSelectCC(IR::CondClassType Cond) {
+  switch (Cond.Val) {
+  case FEXCore::IR::COND_EQ:  return ARMEmitter::Condition::CC_EQ;
+  case FEXCore::IR::COND_NEQ: return ARMEmitter::Condition::CC_NE;
+  case FEXCore::IR::COND_SGE: return ARMEmitter::Condition::CC_GE;
+  case FEXCore::IR::COND_SLT: return ARMEmitter::Condition::CC_LT;
+  case FEXCore::IR::COND_SGT: return ARMEmitter::Condition::CC_GT;
+  case FEXCore::IR::COND_SLE: return ARMEmitter::Condition::CC_LE;
+  case FEXCore::IR::COND_UGE: return ARMEmitter::Condition::CC_CS;
+  case FEXCore::IR::COND_ULT: return ARMEmitter::Condition::CC_CC;
+  case FEXCore::IR::COND_UGT: return ARMEmitter::Condition::CC_HI;
+  case FEXCore::IR::COND_ULE: return ARMEmitter::Condition::CC_LS;
+  case FEXCore::IR::COND_FLU: return ARMEmitter::Condition::CC_LT;
+  case FEXCore::IR::COND_FGE: return ARMEmitter::Condition::CC_GE;
+  case FEXCore::IR::COND_FLEU:return ARMEmitter::Condition::CC_LE;
+  case FEXCore::IR::COND_FGT: return ARMEmitter::Condition::CC_GT;
+  case FEXCore::IR::COND_FU:  return ARMEmitter::Condition::CC_VS;
+  case FEXCore::IR::COND_FNU: return ARMEmitter::Condition::CC_VC;
+  case FEXCore::IR::COND_VS:
+  case FEXCore::IR::COND_VC:
+  case FEXCore::IR::COND_MI: return ARMEmitter::Condition::CC_MI;
+  case FEXCore::IR::COND_PL: return ARMEmitter::Condition::CC_PL;
+  default:
+  LOGMAN_MSG_A_FMT("Unsupported compare type");
+  return ARMEmitter::Condition::CC_NV;
+  }
+}
+
+DEF_OP(CondAddNZCV) {
+  auto Op = IROp->C<IR::IROp_CondAddNZCV>();
+  const auto OpSize = IROp->Size;
+
+  LOGMAN_THROW_AA_FMT(OpSize == IR::i32Bit || OpSize == IR::i64Bit, "Unsupported {} size: {}", __func__, OpSize);
+  const auto EmitSize = OpSize == IR::i64Bit ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+
+  ARMEmitter::StatusFlags Flags = (ARMEmitter::StatusFlags)Op->FalseNZCV;
+  uint64_t Const = 0;
+  auto Src1 = IsInlineConstant(Op->Src1, &Const) ? ARMEmitter::Reg::zr :
+                                                   GetReg(Op->Src1.ID());
+  LOGMAN_THROW_A_FMT(Const == 0, "Unsupported inline constant");
+
+  if (IsInlineConstant(Op->Src2, &Const)) {
+    ccmn(EmitSize, Src1, Const, Flags, MapSelectCC(Op->Cond));
+  } else {
+    ccmn(EmitSize, Src1, GetReg(Op->Src2.ID()), Flags, MapSelectCC(Op->Cond));
   }
 }
 
@@ -237,27 +272,10 @@ DEF_OP(Neg) {
   LOGMAN_THROW_AA_FMT(OpSize == 4 || OpSize == 8, "Unsupported {} size: {}", __func__, OpSize);
   const auto EmitSize = OpSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
 
-  neg(EmitSize, GetReg(Node), GetReg(Op->Src.ID()));
-}
-
-DEF_OP(Abs) {
-  auto Op = IROp->C<IR::IROp_Abs>();
-  const uint8_t OpSize = IROp->Size;
-
-  LOGMAN_THROW_AA_FMT(OpSize == 4 || OpSize == 8, "Unsupported {} size: {}", __func__, OpSize);
-  const auto EmitSize = OpSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
-
-  const auto Dst = GetReg(Node);
-  auto Src = GetReg(Op->Src.ID());
-
-  if (CTX->HostFeatures.SupportsCSSC) {
-    // On CSSC supporting processors, this turns in to one instruction and doesn't modify flags.
-    abs(EmitSize, Dst, Src);
-  }
-  else {
-    cmp(EmitSize, Src, 0);
-    cneg(EmitSize, Dst, Src, ARMEmitter::Condition::CC_MI);
-  }
+  if (Op->Cond == FEXCore::IR::COND_AL)
+    neg(EmitSize, GetReg(Node), GetReg(Op->Src.ID()));
+  else
+    cneg(EmitSize, GetReg(Node), GetReg(Op->Src.ID()), MapSelectCC(Op->Cond));
 }
 
 DEF_OP(Mul) {
@@ -559,6 +577,16 @@ DEF_OP(Xor) {
   }
 }
 
+DEF_OP(XorShift) {
+  auto Op = IROp->C<IR::IROp_XorShift>();
+  const uint8_t OpSize = IROp->Size;
+
+  LOGMAN_THROW_AA_FMT(OpSize == 4 || OpSize == 8, "Unsupported {} size: {}", __func__, OpSize);
+  const auto EmitSize = OpSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+
+  eor(EmitSize, GetReg(Node), GetReg(Op->Src1.ID()), GetReg(Op->Src2.ID()), ConvertIRShiftType(Op->Shift), Op->ShiftAmount);
+}
+
 DEF_OP(Lshl) {
   auto Op = IROp->C<IR::IROp_Lshl>();
   const uint8_t OpSize = IROp->Size;
@@ -1307,36 +1335,6 @@ DEF_OP(Sbfe) {
   sbfx(EmitSize, Dst, Src, Op->lsb, Op->Width);
 }
 
-ARMEmitter::Condition MapSelectCC(IR::CondClassType Cond) {
-  switch (Cond.Val) {
-  case FEXCore::IR::COND_ANDZ:
-  case FEXCore::IR::COND_EQ:  return ARMEmitter::Condition::CC_EQ;
-  case FEXCore::IR::COND_ANDNZ:
-  case FEXCore::IR::COND_NEQ: return ARMEmitter::Condition::CC_NE;
-  case FEXCore::IR::COND_SGE: return ARMEmitter::Condition::CC_GE;
-  case FEXCore::IR::COND_SLT: return ARMEmitter::Condition::CC_LT;
-  case FEXCore::IR::COND_SGT: return ARMEmitter::Condition::CC_GT;
-  case FEXCore::IR::COND_SLE: return ARMEmitter::Condition::CC_LE;
-  case FEXCore::IR::COND_UGE: return ARMEmitter::Condition::CC_CS;
-  case FEXCore::IR::COND_ULT: return ARMEmitter::Condition::CC_CC;
-  case FEXCore::IR::COND_UGT: return ARMEmitter::Condition::CC_HI;
-  case FEXCore::IR::COND_ULE: return ARMEmitter::Condition::CC_LS;
-  case FEXCore::IR::COND_FLU: return ARMEmitter::Condition::CC_LT;
-  case FEXCore::IR::COND_FGE: return ARMEmitter::Condition::CC_GE;
-  case FEXCore::IR::COND_FLEU:return ARMEmitter::Condition::CC_LE;
-  case FEXCore::IR::COND_FGT: return ARMEmitter::Condition::CC_GT;
-  case FEXCore::IR::COND_FU:  return ARMEmitter::Condition::CC_VS;
-  case FEXCore::IR::COND_FNU: return ARMEmitter::Condition::CC_VC;
-  case FEXCore::IR::COND_VS:
-  case FEXCore::IR::COND_VC:
-  case FEXCore::IR::COND_MI:
-  case FEXCore::IR::COND_PL:
-  default:
-  LOGMAN_MSG_A_FMT("Unsupported compare type");
-  return ARMEmitter::Condition::CC_NV;
-  }
-}
-
 DEF_OP(Select) {
   auto Op = IROp->C<IR::IROp_Select>();
   const uint8_t OpSize = IROp->Size;
@@ -1345,28 +1343,15 @@ DEF_OP(Select) {
 
   uint64_t Const;
   auto cc = MapSelectCC(Op->Cond);
-  bool tests = Op->Cond == FEXCore::IR::COND_ANDZ ||
-               Op->Cond == FEXCore::IR::COND_ANDNZ;
-
-  LOGMAN_THROW_A_FMT(!tests || IsGPR(Op->Cmp1.ID()), "Only GPRs can be tested");
 
   if (IsGPR(Op->Cmp1.ID())) {
     const auto Src1 = GetReg(Op->Cmp1.ID());
 
-    if (tests) {
-      if (IsInlineConstant(Op->Cmp2, &Const))
-        tst(CompareEmitSize, Src1, Const);
-      else {
-        const auto Src2 = GetReg(Op->Cmp2.ID());
-        tst(CompareEmitSize, Src1, Src2);
-      }
-    } else {
-      if (IsInlineConstant(Op->Cmp2, &Const))
-        cmp(CompareEmitSize, Src1, Const);
-      else {
-        const auto Src2 = GetReg(Op->Cmp2.ID());
-        cmp(CompareEmitSize, Src1, Src2);
-      }
+    if (IsInlineConstant(Op->Cmp2, &Const))
+      cmp(CompareEmitSize, Src1, Const);
+    else {
+      const auto Src2 = GetReg(Op->Cmp2.ID());
+      cmp(CompareEmitSize, Src1, Src2);
     }
   }
   else if (IsGPRPair(Op->Cmp1.ID())) {
@@ -1405,6 +1390,38 @@ DEF_OP(Select) {
   }
 }
 
+DEF_OP(NZCVSelect) {
+  auto Op = IROp->C<IR::IROp_NZCVSelect>();
+  const uint8_t OpSize = IROp->Size;
+  const auto EmitSize = OpSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+
+  auto cc = MapSelectCC(Op->Cond);
+
+  uint64_t const_true, const_false;
+  bool is_const_true = IsInlineConstant(Op->TrueVal, &const_true);
+  bool is_const_false = IsInlineConstant(Op->FalseVal, &const_false);
+
+  uint64_t all_ones = OpSize == 8 ? 0xffff'ffff'ffff'ffffull : 0xffff'ffffull;
+
+  ARMEmitter::Register Dst = GetReg(Node);
+
+  if (is_const_true) {
+    if (is_const_false != true || !(const_true == 1 || const_true == all_ones) || const_false != 0) {
+      LOGMAN_MSG_A_FMT("NZCVSelect: Unsupported constant");
+    }
+
+    if (const_true == all_ones)
+      csetm(EmitSize, Dst, cc);
+    else
+      cset(EmitSize, Dst, cc);
+  } else if (is_const_false) {
+    LOGMAN_THROW_A_FMT(const_false == 0, "NZCVSelect: unsupported constant");
+    csel(EmitSize, Dst, GetReg(Op->TrueVal.ID()), ARMEmitter::Reg::zr, cc);
+  } else {
+    csel(EmitSize, Dst, GetReg(Op->TrueVal.ID()), GetReg(Op->FalseVal.ID()), cc);
+  }
+}
+
 DEF_OP(VExtractToGPR) {
   const auto Op = IROp->C<IR::IROp_VExtractToGPR>();
   const auto OpSize = IROp->Size;
@@ -1519,41 +1536,10 @@ DEF_OP(FCmp) {
   auto Op = IROp->C<IR::IROp_FCmp>();
   const auto EmitSubSize = Op->ElementSize == 8 ? ARMEmitter::ScalarRegSize::i64Bit : ARMEmitter::ScalarRegSize::i32Bit;
 
-  ARMEmitter::Register Dst = GetReg(Node);
   ARMEmitter::VRegister Scalar1 = GetVReg(Op->Scalar1.ID());
   ARMEmitter::VRegister Scalar2 = GetVReg(Op->Scalar2.ID());
 
   fcmp(EmitSubSize, Scalar1, Scalar2);
-  bool set = false;
-
-  if (Op->Flags & (1 << IR::FCMP_FLAG_EQ)) {
-    LOGMAN_THROW_AA_FMT(IR::FCMP_FLAG_EQ == 0, "IR::FCMP_FLAG_EQ must equal 0");
-    // EQ or unordered
-    cset(ARMEmitter::Size::i64Bit, Dst, ARMEmitter::Condition::CC_EQ); // Z = 1
-    csinc(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::zr, ARMEmitter::Condition::CC_VC); // IF !V ? Z : 1
-    set = true;
-  }
-
-  if (Op->Flags & (1 << IR::FCMP_FLAG_LT)) {
-    // LT or unordered
-    cset(ARMEmitter::Size::i64Bit, TMP2, ARMEmitter::Condition::CC_LT);
-    if (!set) {
-      lsl(ARMEmitter::Size::i64Bit, Dst, TMP2, IR::FCMP_FLAG_LT);
-      set = true;
-    } else {
-      bfi(ARMEmitter::Size::i64Bit, Dst, TMP2, IR::FCMP_FLAG_LT, 1);
-    }
-  }
-
-  if (Op->Flags & (1 << IR::FCMP_FLAG_UNORDERED)) {
-    cset(ARMEmitter::Size::i64Bit, TMP2, ARMEmitter::Condition::CC_VS);
-    if (!set) {
-      lsl(ARMEmitter::Size::i64Bit, Dst, TMP2, IR::FCMP_FLAG_UNORDERED);
-      set = true;
-    } else {
-      bfi(ARMEmitter::Size::i64Bit, Dst, TMP2, IR::FCMP_FLAG_UNORDERED, 1);
-    }
-  }
 }
 
 #undef DEF_OP

FEXCore/Source/Interface/Core/JIT/Arm64/BranchOps.cpp

@@ -97,9 +97,7 @@ DEF_OP(Jump) {
 
 static ARMEmitter::Condition MapBranchCC(IR::CondClassType Cond) {
   switch (Cond.Val) {
-  case FEXCore::IR::COND_ANDZ:
   case FEXCore::IR::COND_EQ:  return ARMEmitter::Condition::CC_EQ;
-  case FEXCore::IR::COND_ANDNZ:
   case FEXCore::IR::COND_NEQ: return ARMEmitter::Condition::CC_NE;
   case FEXCore::IR::COND_SGE: return ARMEmitter::Condition::CC_GE;
   case FEXCore::IR::COND_SLT: return ARMEmitter::Condition::CC_LT;
@@ -117,8 +115,8 @@ static ARMEmitter::Condition MapBranchCC(IR::CondClassType Cond) {
   case FEXCore::IR::COND_FNU: return ARMEmitter::Condition::CC_VC;
   case FEXCore::IR::COND_VS:
   case FEXCore::IR::COND_VC:
-  case FEXCore::IR::COND_MI:
-  case FEXCore::IR::COND_PL:
+  case FEXCore::IR::COND_MI:  return ARMEmitter::Condition::CC_MI;
+  case FEXCore::IR::COND_PL:  return ARMEmitter::Condition::CC_PL;
   default:
   LOGMAN_MSG_A_FMT("Unsupported compare type");
   return ARMEmitter::Condition::CC_NV;
@@ -130,42 +128,27 @@ DEF_OP(CondJump) {
 
   auto TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;
 
-  uint64_t Const;
-  const bool isConst = IsInlineConstant(Op->Cmp2, &Const);
-  bool tests = Op->Cond == FEXCore::IR::COND_ANDZ ||
-               Op->Cond == FEXCore::IR::COND_ANDNZ;
-
-  const auto Size = Op->CompareSize == 4 ? ARMEmitter::Size::i32Bit : ARMEmitter::Size::i64Bit;
-  const auto SubSize = ARMEmitter::ToVectorSizePair(Op->CompareSize == 4 ? ARMEmitter::SubRegSize::i32Bit : ARMEmitter::SubRegSize::i64Bit);
-
-  if (isConst && Const == 0 && Op->Cond.Val == FEXCore::IR::COND_EQ) {
-    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
-    cbz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
-  } else if (isConst && Const == 0 && Op->Cond.Val == FEXCore::IR::COND_NEQ) {
-    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
-    cbnz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
+  if (Op->FromNZCV) {
+    b(MapBranchCC(Op->Cond), TrueTargetLabel);
   } else {
-    if (IsGPR(Op->Cmp1.ID())) {
-      if (tests) {
-        if (isConst) {
-          tst(Size, GetReg(Op->Cmp1.ID()), Const);
-        } else {
-          tst(Size, GetReg(Op->Cmp1.ID()), GetReg(Op->Cmp2.ID()));
-        }
-      } else {
-        if (isConst) {
-          cmp(Size, GetReg(Op->Cmp1.ID()), Const);
-        } else {
-          cmp(Size, GetReg(Op->Cmp1.ID()), GetReg(Op->Cmp2.ID()));
-        }
-      }
-    } else if (IsFPR(Op->Cmp1.ID())) {
-      fcmp(SubSize.Scalar, GetVReg(Op->Cmp1.ID()), GetVReg(Op->Cmp2.ID()));
-    } else {
-      LOGMAN_MSG_A_FMT("CondJump: Expected GPR or FPR");
+    uint64_t Const;
+    const bool isConst = IsInlineConstant(Op->Cmp2, &Const);
+
+    const auto Size = Op->CompareSize == 4 ? ARMEmitter::Size::i32Bit : ARMEmitter::Size::i64Bit;
+
+    LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
+    LOGMAN_THROW_A_FMT(isConst && Const == 0, "CondJump: Expected 0 source");
+    LOGMAN_THROW_A_FMT(Op->Cond.Val == FEXCore::IR::COND_EQ ||
+                       Op->Cond.Val == FEXCore::IR::COND_NEQ,
+                       "CondJump: Expected simple condition");
+
+    if (Op->Cond.Val == FEXCore::IR::COND_EQ) {
+      cbz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
+    } else  {
+      cbnz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
     }
 
-    b(MapBranchCC(Op->Cond), TrueTargetLabel);
+    // TODO: Wire up tbz/tbnz
   }
 
   PendingTargetLabel = &JumpTargets.try_emplace(Op->FalseBlock.ID()).first->second;

FEXCore/Source/Interface/Core/JIT/Arm64/EncryptionOps.cpp

@@ -179,6 +179,32 @@ DEF_OP(CRC32) {
   }
 }
 
+DEF_OP(VSha1H) {
+  auto Op = IROp->C<IR::IROp_VSha1H>();
+
+  const auto Dst = GetVReg(Node);
+  const auto Src = GetVReg(Op->Src.ID());
+
+  sha1h(Dst.S(), Src.S());
+}
+
+DEF_OP(VSha256U0) {
+  auto Op = IROp->C<IR::IROp_VSha256U0>();
+
+  const auto Dst = GetVReg(Node);
+  const auto Src1 = GetVReg(Op->Src1.ID());
+  const auto Src2 = GetVReg(Op->Src2.ID());
+
+  if (Dst == Src1) {
+    sha256su0(Dst, Src2);
+  }
+  else {
+    mov(VTMP1.Q(), Src1.Q());
+    sha256su0(VTMP1, Src2);
+    mov(Dst.Q(), Src1.Q());
+  }
+}
+
 DEF_OP(PCLMUL) {
   const auto Op = IROp->C<IR::IROp_PCLMUL>();
   const auto OpSize = IROp->Size;

FEXCore/Source/Interface/Core/JIT/Arm64/MemoryOps.cpp

@@ -5,6 +5,7 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "FEXCore/Core/X86Enums.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/ArchHelpers/CodeEmitter/Registers.h"
@@ -295,7 +296,11 @@ DEF_OP(LoadRegisterSRA) {
   const auto OpSize = IROp->Size;
 
   if (Op->Class == IR::GPRClass) {
-    const auto regId = (Op->Offset - offsetof(Core::CpuStateFrame, State.gregs[0])) / Core::CPUState::GPR_REG_SIZE;
+    const auto regId =
+      Op->Offset == offsetof(Core::CpuStateFrame, State.pf_raw) ? (StaticRegisters.size() - 2) :
+      Op->Offset == offsetof(Core::CpuStateFrame, State.af_raw) ? (StaticRegisters.size() - 1) :
+      (Op->Offset - offsetof(Core::CpuStateFrame, State.gregs[0])) / Core::CPUState::GPR_REG_SIZE;
+
     const auto regOffs = Op->Offset & 7;
 
     LOGMAN_THROW_A_FMT(regId < StaticRegisters.size(), "out of range regId");
@@ -473,10 +478,14 @@ DEF_OP(StoreRegisterSRA) {
   const auto OpSize = IROp->Size;
 
   if (Op->Class == IR::GPRClass) {
-    const auto regId = (Op->Offset / Core::CPUState::GPR_REG_SIZE) - 1;
     const auto regOffs = Op->Offset & 7;
 
-    LOGMAN_THROW_A_FMT(regId < StaticFPRegisters.size(), "out of range regId");
+    const auto regId =
+      Op->Offset == offsetof(Core::CpuStateFrame, State.pf_raw) ? (StaticRegisters.size() - 2) :
+      Op->Offset == offsetof(Core::CpuStateFrame, State.af_raw) ? (StaticRegisters.size() - 1) :
+      (Op->Offset - offsetof(Core::CpuStateFrame, State.gregs[0])) / Core::CPUState::GPR_REG_SIZE;
+
+    LOGMAN_THROW_A_FMT(regId < StaticRegisters.size(), "out of range regId");
 
     const auto reg = StaticRegisters[regId];
     const auto Src = GetReg(Op->Value.ID());
@@ -1031,23 +1040,37 @@ DEF_OP(FillRegister) {
   }
 }
 
+DEF_OP(LoadNZCV) {
+  auto Dst = GetReg(Node);
+
+  mrs(Dst, ARMEmitter::SystemRegister::NZCV);
+}
+
+DEF_OP(StoreNZCV) {
+  auto Op = IROp->C<IR::IROp_StoreNZCV>();
+
+  msr(ARMEmitter::SystemRegister::NZCV, GetReg(Op->Value.ID()));
+}
+
 DEF_OP(LoadFlag) {
   auto Op = IROp->C<IR::IROp_LoadFlag>();
   auto Dst = GetReg(Node);
 
-  if (Op->Flag == 24 /* NZCV */)
-    ldr(Dst.W(), STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag);
-  else
-    ldrb(Dst, STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag);
+  LOGMAN_THROW_A_FMT(Op->Flag != X86State::RFLAG_PF_RAW_LOC &&
+                     Op->Flag != X86State::RFLAG_AF_RAW_LOC,
+                     "PF/AF must be accessed as registers");
+
+  ldrb(Dst, STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag);
 }
 
 DEF_OP(StoreFlag) {
   auto Op = IROp->C<IR::IROp_StoreFlag>();
 
-  if (Op->Flag == 24 /* NZCV */)
-    str(GetReg(Op->Value.ID()).W(), STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag);
-  else
-    strb(GetReg(Op->Value.ID()), STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag);
+  LOGMAN_THROW_A_FMT(Op->Flag != X86State::RFLAG_PF_RAW_LOC &&
+                     Op->Flag != X86State::RFLAG_AF_RAW_LOC,
+                     "PF/AF must be accessed as registers");
+
+  strb(GetReg(Op->Value.ID()), STATE, offsetof(FEXCore::Core::CPUState, flags[0]) + Op->Flag);
 }
 
 FEXCore::ARMEmitter::ExtendedMemOperand Arm64JITCore::GenerateMemOperand(uint8_t AccessSize,
@@ -1840,9 +1863,15 @@ DEF_OP(MemSet) {
   const auto MemReg = GetReg(Op->Addr.ID());
   const auto Value = GetReg(Op->Value.ID());
   const auto Length = GetReg(Op->Length.ID());
-  const auto Direction = GetReg(Op->Direction.ID());
   const auto Dst = GetReg(Node);
 
+  uint64_t DirectionConstant;
+  bool DirectionIsInline = IsInlineConstant(Op->Direction, &DirectionConstant);
+  FEXCore::ARMEmitter::Register DirectionReg = ARMEmitter::Reg::r0;
+  if (!DirectionIsInline) {
+    DirectionReg = GetReg(Op->Direction.ID());
+  }
+
   // If Direction == 0 then:
   //   MemReg is incremented (by size)
   // else:
@@ -1862,8 +1891,10 @@ DEF_OP(MemSet) {
     add(TMP2, Prefix.X(), MemReg.X());
   }
 
-  // Backward or forwards implementation depends on flag
-  cbnz(ARMEmitter::Size::i64Bit, Direction, &BackwardImpl);
+  if (!DirectionIsInline) {
+    // Backward or forwards implementation depends on flag
+    cbnz(ARMEmitter::Size::i64Bit, DirectionReg, &BackwardImpl);
+  }
 
   auto MemStore = [this](auto Value, uint32_t OpSize, int32_t Size) {
     switch (OpSize) {
@@ -1917,8 +1948,7 @@ DEF_OP(MemSet) {
     }
   };
 
-  // Emit forward direction memset then backward direction memset.
-  for (int32_t Direction :  { 1, -1 }) {
+  auto EmitMemset = [&](int32_t Direction) {
     const int32_t OpSize = Size;
     const int32_t SizeDirection = Size * Direction;
 
@@ -1978,15 +2008,26 @@ DEF_OP(MemSet) {
           break;
       }
     }
+  };
 
-    if (Direction == 1) {
-      b(&Done);
-      Bind(&BackwardImpl);
-    }
+  if (DirectionIsInline) {
+    // If the direction constant is set then the direction is negative.
+    EmitMemset(DirectionConstant ? -1 : 1);
   }
+  else {
+    // Emit forward direction memset then backward direction memset.
+    for (int32_t Direction :  { 1, -1 }) {
+      EmitMemset(Direction);
 
-  Bind(&Done);
-  // Destination already set to the final pointer.
+      if (Direction == 1) {
+        b(&Done);
+        Bind(&BackwardImpl);
+      }
+    }
+
+    Bind(&Done);
+    // Destination already set to the final pointer.
+  }
 }
 
 DEF_OP(MemCpy) {
@@ -2001,7 +2042,12 @@ DEF_OP(MemCpy) {
   const auto MemRegSrc = GetReg(Op->AddrSrc.ID());
 
   const auto Length = GetReg(Op->Length.ID());
-  const auto Direction = GetReg(Op->Direction.ID());
+  uint64_t DirectionConstant;
+  bool DirectionIsInline = IsInlineConstant(Op->Direction, &DirectionConstant);
+  FEXCore::ARMEmitter::Register DirectionReg = ARMEmitter::Reg::r0;
+  if (!DirectionIsInline) {
+    DirectionReg = GetReg(Op->Direction.ID());
+  }
 
   auto Dst = GetRegPair(Node);
   // If Direction == 0 then:
@@ -2038,8 +2084,10 @@ DEF_OP(MemCpy) {
   // TMP3 = Src
   // TMP4 = load+store temp value
 
-  // Backward or forwards implementation depends on flag
-  cbnz(ARMEmitter::Size::i64Bit, Direction, &BackwardImpl);
+  if (!DirectionIsInline) {
+    // Backward or forwards implementation depends on flag
+    cbnz(ARMEmitter::Size::i64Bit, DirectionReg, &BackwardImpl);
+  }
 
   auto MemCpy = [this](uint32_t OpSize, int32_t Size) {
     switch (OpSize) {
@@ -2161,8 +2209,7 @@ DEF_OP(MemCpy) {
     }
   };
 
-  // Emit forward direction memset then backward direction memset.
-  for (int32_t Direction :  { 1, -1 }) {
+  auto EmitMemcpy = [&](int32_t Direction) {
     const int32_t OpSize = Size;
     const int32_t SizeDirection = Size * Direction;
 
@@ -2235,15 +2282,24 @@ DEF_OP(MemCpy) {
           break;
       }
     }
+  };
 
-    if (Direction == 1) {
-      b(&Done);
-      Bind(&BackwardImpl);
+  if (DirectionIsInline) {
+    // If the direction constant is set then the direction is negative.
+    EmitMemcpy(DirectionConstant ? -1 : 1);
+  }
+  else {
+    // Emit forward direction memset then backward direction memset.
+    for (int32_t Direction :  { 1, -1 }) {
+      EmitMemcpy(Direction);
+      if (Direction == 1) {
+        b(&Done);
+        Bind(&BackwardImpl);
+      }
     }
+    Bind(&Done);
+    // Destination already set to the final pointer.
   }
-
-  Bind(&Done);
-  // Destination already set to the final pointer.
 }
 
 DEF_OP(ParanoidLoadMemTSO) {

FEXCore/Source/Interface/Core/JIT/Arm64/VectorOps.cpp

@@ -1799,6 +1799,9 @@ DEF_OP(VFMin) {
     }
   } else {
     if (IsScalar) {
+      // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+      mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
       switch (ElementSize) {
         case 2: {
           fcmp(Vector1.H(), Vector2.H());
@@ -1818,6 +1821,9 @@ DEF_OP(VFMin) {
         default:
           break;
       }
+
+      // Restore NZCV
+      msr(ARMEmitter::SystemRegister::NZCV, TMP1);
     } else {
       if (Dst == Vector1) {
         // Destination is already Vector1, need to insert Vector2 on false.
@@ -1878,6 +1884,9 @@ DEF_OP(VFMax) {
     }
   } else {
     if (IsScalar) {
+      // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+      mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
       switch (ElementSize) {
         case 2: {
           fcmp(Vector1.H(), Vector2.H());
@@ -1897,6 +1906,9 @@ DEF_OP(VFMax) {
         default:
           break;
       }
+
+      // Restore NZCV
+      msr(ARMEmitter::SystemRegister::NZCV, TMP1);
     } else {
       if (Dst == Vector1) {
         // Destination is already Vector1, need to insert Vector2 on true.
@@ -2654,6 +2666,9 @@ DEF_OP(VCMPEQ) {
     ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i128Bit);
 
   if (HostSupportsSVE256 && Is256Bit) {
+    // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
     const auto Mask = PRED_TMP_32B.Zeroing();
     const auto ComparePred = ARMEmitter::PReg::p0;
 
@@ -2664,6 +2679,9 @@ DEF_OP(VCMPEQ) {
     not_(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), Vector1.Z());
     movprfx(SubRegSize.Vector, Dst.Z(), ComparePred.Zeroing(), Vector1.Z());
     orr(SubRegSize.Vector, Dst.Z(), ComparePred.Merging(), Dst.Z(), VTMP1.Z());
+
+    // Restore NZCV
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
   } else {
     if (IsScalar) {
       cmeq(SubRegSize.Scalar, Dst, Vector1, Vector2);
@@ -2695,6 +2713,9 @@ DEF_OP(VCMPEQZ) {
     const auto Mask = PRED_TMP_32B.Zeroing();
     const auto ComparePred = ARMEmitter::PReg::p0;
 
+    // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
     // Ensure no junk is in the temp (important for ensuring
     // non-equal entries remain as zero).
     mov_imm(ARMEmitter::SubRegSize::i64Bit, VTMP1.Z(), 0);
@@ -2705,6 +2726,9 @@ DEF_OP(VCMPEQZ) {
     cmpeq(SubRegSize.Vector, ComparePred, Mask, Vector.Z(), 0);
     not_(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), Vector.Z());
     mov(Dst.Z(), VTMP1.Z());
+
+    // Restore NZCV
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
   } else {
     if (IsScalar) {
       cmeq(SubRegSize.Scalar, Dst, Vector);
@@ -2737,6 +2761,9 @@ DEF_OP(VCMPGT) {
     const auto Mask = PRED_TMP_32B.Zeroing();
     const auto ComparePred = ARMEmitter::PReg::p0;
 
+    // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
     // General idea is to compare for greater-than, bitwise NOT
     // the valid values, then ORR the NOTed values with the original
     // values to form entries that are all 1s.
@@ -2744,6 +2771,9 @@ DEF_OP(VCMPGT) {
     not_(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), Vector1.Z());
     movprfx(SubRegSize.Vector, Dst.Z(), ComparePred.Zeroing(), Vector1.Z());
     orr(SubRegSize.Vector, Dst.Z(), ComparePred.Merging(), Dst.Z(), VTMP1.Z());
+
+    // Restore NZCV
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
   } else {
     if (IsScalar) {
       cmgt(SubRegSize.Scalar, Dst, Vector1, Vector2);
@@ -2775,6 +2805,9 @@ DEF_OP(VCMPGTZ) {
     const auto Mask = PRED_TMP_32B.Zeroing();
     const auto ComparePred = ARMEmitter::PReg::p0;
 
+    // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
     // Ensure no junk is in the temp (important for ensuring
     // non greater-than values remain as zero).
     mov_imm(ARMEmitter::SubRegSize::i64Bit, VTMP1.Z(), 0);
@@ -2782,6 +2815,9 @@ DEF_OP(VCMPGTZ) {
     not_(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), Vector.Z());
     orr(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), VTMP1.Z(), Vector.Z());
     mov(Dst.Z(), VTMP1.Z());
+
+    // Restore NZCV
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
   } else {
     if (IsScalar) {
       cmgt(SubRegSize.Scalar, Dst, Vector);
@@ -2813,6 +2849,9 @@ DEF_OP(VCMPLTZ) {
     const auto Mask = PRED_TMP_32B.Zeroing();
     const auto ComparePred = ARMEmitter::PReg::p0;
 
+    // FIXME: We should rework this op to avoid the NZCV spill/fill dance.
+    mrs(TMP1, ARMEmitter::SystemRegister::NZCV);
+
     // Ensure no junk is in the temp (important for ensuring
     // non less-than values remain as zero).
     mov_imm(ARMEmitter::SubRegSize::i64Bit, VTMP1.Z(), 0);
@@ -2820,6 +2859,9 @@ DEF_OP(VCMPLTZ) {
     not_(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), Vector.Z());
     orr(SubRegSize.Vector, VTMP1.Z(), ComparePred.Merging(), VTMP1.Z(), Vector.Z());
     mov(Dst.Z(), VTMP1.Z());
+
+    // Restore NZCV
+    msr(ARMEmitter::SystemRegister::NZCV, TMP1);
   } else {
     if (IsScalar) {
       cmlt(SubRegSize.Scalar, Dst, Vector);
@@ -3904,6 +3946,58 @@ DEF_OP(VUShrI) {
   }
 }
 
+DEF_OP(VUShraI) {
+  const auto Op = IROp->C<IR::IROp_VUShraI>();
+  const auto OpSize = IROp->Size;
+
+  const auto BitShift = Op->BitShift;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  const auto Dst = GetVReg(Node);
+  const auto DestVector = GetVReg(Op->DestVector.ID());
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 1 || ElementSize == 2 || ElementSize == 4 || ElementSize == 8, "Invalid size");
+  const auto SubRegSize =
+    ElementSize == 1 ? ARMEmitter::SubRegSize::i8Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i8Bit;
+
+  if (HostSupportsSVE256 && Is256Bit) {
+    if (Dst == DestVector) {
+      usra(SubRegSize, Dst.Z(), Vector.Z(), BitShift);
+    }
+    else {
+      if (Dst != Vector) {
+        mov(Dst.Z(), DestVector.Z());
+        usra(SubRegSize, Dst.Z(), Vector.Z(), BitShift);
+      }
+      else {
+        mov(VTMP1.Z(), DestVector.Z());
+        usra(SubRegSize, Dst.Z(), Vector.Z(), BitShift);
+        mov(Dst.Z(), VTMP1.Z());
+      }
+    }
+  } else {
+    if (Dst == DestVector) {
+      usra(SubRegSize, Dst.Q(), Vector.Q(), BitShift);
+    }
+    else {
+      if (Dst != Vector) {
+        mov(Dst.Q(), DestVector.Q());
+        usra(SubRegSize, Dst.Q(), Vector.Q(), BitShift);
+      }
+      else {
+        mov(VTMP1.Q(), DestVector.Q());
+        usra(SubRegSize, VTMP1.Q(), Vector.Q(), BitShift);
+        mov(Dst.Q(), VTMP1.Q());
+      }
+    }
+  }
+}
+
 DEF_OP(VSShrI) {
   const auto Op = IROp->C<IR::IROp_VSShrI>();
   const auto OpSize = IROp->Size;

FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -38,6 +38,13 @@ enum class MemoryAccessType {
   STREAM,
 };
 
+enum class BTAction {
+  BTNone,
+  BTClear,
+  BTSet,
+  BTComplement,
+};
+
 struct LoadSourceOptions {
   // Alignment of the load in bytes. -1 signifies unaligned
   int8_t Align = -1;
@@ -89,7 +96,6 @@ public:
     TYPE_RORI,
     TYPE_ROL,
     TYPE_ROLI,
-    TYPE_FCMP,
     TYPE_BEXTR,
     TYPE_BLSI,
     TYPE_BLSMSK,
@@ -98,7 +104,6 @@ public:
     TYPE_BZHI,
     TYPE_TZCNT,
     TYPE_LZCNT,
-    TYPE_BITSELECT,
     TYPE_RDRAND,
   };
 
@@ -122,11 +127,10 @@ public:
   }
 
   void StartNewBlock() {
-    flagsOp = SelectionFlag::Nothing;
-
     // If we loaded flags but didn't change them, invalidate the cached copy and move on.
     // Changes get stored out by CalculateDeferredFlags.
     CachedNZCV = nullptr;
+    PossiblySetNZCVBits = ~0U;
 
     // New block needs to reset segment telemetry.
     SegmentsNeedReadCheck = ~0U;
@@ -155,6 +159,14 @@ public:
     CalculateDeferredFlags();
     return _CondJump(ssa0, ssa1, ssa2, cond);
   }
+  IRPair<IROp_CondJump> CondJumpNZCV(CondClassType Cond) {
+    CalculateDeferredFlags();
+
+    // The jump will ignore the sources, so it doesn't matter what we put here.
+    // Put an inline constant so RA+codegen will ignore altogether.
+    auto Placeholder = _InlineConstant(0);
+    return _CondJump(Placeholder, Placeholder, InvalidNode, InvalidNode, Cond, 0, true);
+  }
 
   bool FinishOp(uint64_t NextRIP, bool LastOp) {
     // If we are switching to a new block and this current block has yet to set a RIP
@@ -324,14 +336,10 @@ public:
   void RCLOp1Bit(OpcodeArgs);
   void RCLOp(OpcodeArgs);
   void RCLSmallerOp(OpcodeArgs);
-  template<uint32_t SrcIndex>
+
+  template<uint32_t SrcIndex, enum BTAction Action>
   void BTOp(OpcodeArgs);
-  template<uint32_t SrcIndex>
-  void BTROp(OpcodeArgs);
-  template<uint32_t SrcIndex>
-  void BTSOp(OpcodeArgs);
-  template<uint32_t SrcIndex>
-  void BTCOp(OpcodeArgs);
+
   void IMUL1SrcOp(OpcodeArgs);
   void IMUL2SrcOp(OpcodeArgs);
   void IMULOp(OpcodeArgs);
@@ -916,17 +924,36 @@ public:
   }
 
 protected:
-  void SaveNZCV() override {
+  void SaveNZCV(IROps Op = OP_DUMMY) override {
+    /* Some opcodes are conservatively marked as clobbering flags, but in fact
+     * do not clobber flags in certain conditions. Check for that here as an
+     * optimization.
+     */
+    switch (Op) {
+    case OP_VFMINSCALARINSERT:
+    case OP_VFMAXSCALARINSERT:
+      /* On AFP platforms, becomes fmin/fmax and preserves NZCV. Otherwise
+       * becomes fcmp and clobbers.
+       */
+      if (CTX->HostFeatures.SupportsAFP)
+        return;
+
+      break;
+    default:
+      break;
+    }
+
+    // Invariant: When executing instructions that clobber NZCV, the flags must
+    // be resident in a GPR, which is equivalent to CachedNZCV != nullptr. Get
+    // the NZCV which fills the cache if necessary.
+    if (CachedNZCV == nullptr)
+      GetNZCV();
+
+    // Assume we'll need a reload.
+    NZCVDirty = true;
   }
 
 private:
-  enum class SelectionFlag {
-    Nothing,  // must rely on x86 flags
-    CMP,      // flags were set by a CMP between flagsOpDest/flagsOpDestSigned and flagsOpSrc/flagsOpSrcSigned with flagsOpSize size
-    AND,      // flags were set by an AND/TEST, flagsOpDest contains the resulting value of flagsOpSize size
-    FCMP,     // flags were set by a ucomis* / comis*
-  };
-
   struct JumpTargetInfo {
     OrderedNode* BlockEntry;
     bool HaveEmitted;
@@ -934,13 +961,6 @@ private:
 
   FEXCore::Context::ContextImpl *CTX{};
 
-  SelectionFlag flagsOp{};
-  uint8_t flagsOpSize{};
-  OrderedNode* flagsOpDest{};
-  OrderedNode* flagsOpSrc{};
-  OrderedNode* flagsOpDestSigned{};
-  OrderedNode* flagsOpSrcSigned{};
-
   constexpr static unsigned FullNZCVMask =
     (1U << FEXCore::X86State::RFLAG_CF_RAW_LOC) |
     (1U << FEXCore::X86State::RFLAG_ZF_RAW_LOC) |
@@ -1243,7 +1263,7 @@ private:
 
   OrderedNode *GetNZCV() {
     if (!CachedNZCV) {
-      CachedNZCV = _LoadFlag(FEXCore::X86State::RFLAG_NZCV_LOC);
+      CachedNZCV = _LoadNZCV();
 
       // We don't know what's set
       PossiblySetNZCVBits = ~0;
@@ -1275,37 +1295,85 @@ private:
   }
 
   void SetNZ_ZeroCV(unsigned SrcSize, OrderedNode *Res) {
-    CachedNZCV = _TestNZ(SrcSize, Res);
+    _TestNZ(IR::SizeToOpSize(SrcSize), Res, Res);
+    CachedNZCV = _LoadNZCV();
     PossiblySetNZCVBits = (1u << 31) | (1u << 30);
-    NZCVDirty = true;
+    NZCVDirty = false;
   }
 
-  OrderedNode *InsertNZCV(OrderedNode *NZCV, unsigned BitOffset, OrderedNode *Value) {
-    unsigned Bit = IndexNZCV(BitOffset);
+  void InsertNZCV(unsigned BitOffset, OrderedNode *Value, signed FlagOffset, bool MustMask) {
+    signed Bit = IndexNZCV(BitOffset);
+
+    // If NZCV is not dirty, we always want to use rmif, it's 1 instruction to
+    // implement this. But if NZCV is dirty, it might still be cheaper to copy
+    // the GPR flags to NZCV and rmif. This is a heuristic for cases where we
+    // expect that 2 instruction sequence to be a win (versus something like
+    // bfe+mov+bfi+mov which can happen with our RA..). It's not totally
+    // conservative but it's pretty good in practice.
+    bool PreferRmif = !NZCVDirty || FlagOffset || MustMask ||
+                      (PossiblySetNZCVBits & (1u << Bit));
+
+    if (CTX->HostFeatures.SupportsFlagM && PreferRmif) {
+      // Update NZCV
+      if (NZCVDirty && CachedNZCV)
+        _StoreNZCV(CachedNZCV);
+
+      CachedNZCV = nullptr;
+      NZCVDirty = false;
+
+      // Insert as NZCV.
+      signed RmifBit = Bit - 28;
+      _RmifNZCV(Value, (64 + FlagOffset - RmifBit) % 64, 1u << RmifBit);
+      CachedNZCV = nullptr;
+    } else {
+      // Insert as GPR
+      if (FlagOffset || MustMask)
+        Value = _Bfe(OpSize::i64Bit, 1, FlagOffset, Value);
+
+      if (PossiblySetNZCVBits == 0)
+        SetNZCV(_Lshl(OpSize::i64Bit, Value, _Constant(Bit)));
+      else if ((PossiblySetNZCVBits & (1u << Bit)) == 0)
+        SetNZCV(_Orlshl(OpSize::i32Bit, GetNZCV(), Value, Bit));
+      else
+        SetNZCV(_Bfi(OpSize::i32Bit, 1, Bit, GetNZCV(), Value));
+    }
 
-    uint32_t SetBits = PossiblySetNZCVBits;
     PossiblySetNZCVBits |= (1u << Bit);
+  }
 
-    if (SetBits == 0)
-      return _Lshl(OpSize::i64Bit, Value, _Constant(Bit));
-    else if ((SetBits & (1u << Bit)) == 0)
-      return _Orlshl(OpSize::i32Bit, NZCV, Value, Bit);
-    else
-      return _Bfi(OpSize::i32Bit, 1, Bit, NZCV, Value);
+  void CarryInvert() {
+    unsigned Bit = IndexNZCV(FEXCore::X86State::RFLAG_CF_RAW_LOC);
+
+    if (CTX->HostFeatures.SupportsFlagM && !NZCVDirty) {
+      // Invert as NZCV.
+      _CarryInvert();
+      CachedNZCV = nullptr;
+    } else {
+      // Invert as a GPR
+      SetNZCV(_Xor(OpSize::i32Bit, GetNZCV(), _Constant(1u << Bit)));
+    }
+
+    PossiblySetNZCVBits |= 1u << Bit;
   }
 
   template<unsigned BitOffset>
-  void SetRFLAG(OrderedNode *Value) {
-    SetRFLAG(Value, BitOffset);
+  void SetRFLAG(OrderedNode *Value, unsigned ValueOffset = 0, bool MustMask = false) {
+    SetRFLAG(Value, BitOffset, ValueOffset, MustMask);
   }
 
-  void SetRFLAG(OrderedNode *Value, unsigned BitOffset) {
-    flagsOp = SelectionFlag::Nothing;
+  void SetRFLAG(OrderedNode *Value, unsigned BitOffset, unsigned ValueOffset = 0, bool MustMask = false) {
+    if (IsNZCV(BitOffset)) {
+      InsertNZCV(BitOffset, Value, ValueOffset, MustMask);
+    } else if (BitOffset == FEXCore::X86State::RFLAG_PF_RAW_LOC) {
+      _StoreRegister(Value, false, offsetof(FEXCore::Core::CPUState, pf_raw), GPRClass, GPRFixedClass, CTX->GetGPRSize());
+    } else if (BitOffset == FEXCore::X86State::RFLAG_AF_RAW_LOC) {
+      _StoreRegister(Value, false, offsetof(FEXCore::Core::CPUState, af_raw), GPRClass, GPRFixedClass, CTX->GetGPRSize());
+    } else {
+      if (ValueOffset || MustMask)
+        Value = _Bfe(OpSize::i32Bit, 1, ValueOffset, Value);
 
-    if (IsNZCV(BitOffset))
-      SetNZCV(InsertNZCV(PossiblySetNZCVBits ? GetNZCV() : nullptr, BitOffset, Value));
-    else
       _StoreFlag(Value, BitOffset);
+    }
   }
 
   void SetAF(unsigned Constant) {
@@ -1318,17 +1386,134 @@ private:
 
   void ZeroMultipleFlags(uint32_t BitMask);
 
-  OrderedNode *GetRFLAG(unsigned BitOffset) {
+  CondClassType CondForNZCVBit(unsigned BitOffset, bool Invert) {
+    switch (BitOffset) {
+    case FEXCore::X86State::RFLAG_SF_RAW_LOC:
+      return Invert ? CondClassType{COND_PL} : CondClassType{COND_MI};
+
+    case FEXCore::X86State::RFLAG_ZF_RAW_LOC:
+      return Invert ? CondClassType{COND_NEQ} : CondClassType{COND_EQ};
+
+    case FEXCore::X86State::RFLAG_CF_RAW_LOC:
+      return Invert ? CondClassType{COND_ULT} : CondClassType{COND_UGE};
+
+    case FEXCore::X86State::RFLAG_OF_RAW_LOC:
+      return Invert ? CondClassType{COND_FNU} : CondClassType{COND_FU};
+
+    default:
+      FEX_UNREACHABLE;
+    }
+  }
+
+  OrderedNode *GetRFLAG(unsigned BitOffset, bool Invert = false) {
     if (IsNZCV(BitOffset)) {
-      if (!CachedNZCV || (PossiblySetNZCVBits & (1u << IndexNZCV(BitOffset))))
-        return _Bfe(OpSize::i32Bit, 1, IndexNZCV(BitOffset), GetNZCV());
-      else
-        return _Constant(0);
+      if (!(PossiblySetNZCVBits & (1u << IndexNZCV(BitOffset)))) {
+        return _Constant(Invert ? 1 : 0);
+      } else if (NZCVDirty) {
+        auto Value = _Bfe(OpSize::i32Bit, 1, IndexNZCV(BitOffset), GetNZCV());
+
+        if (Invert)
+          return _Xor(OpSize::i32Bit, Value, _Constant(1));
+        else
+          return Value;
+      } else {
+        return _NZCVSelect(OpSize::i32Bit, CondForNZCVBit(BitOffset, Invert),
+                           _Constant(1), _Constant(0));
+      }
+    } else if (BitOffset == FEXCore::X86State::RFLAG_PF_RAW_LOC) {
+      return _LoadRegister(false, offsetof(FEXCore::Core::CPUState, pf_raw), GPRClass, GPRFixedClass, CTX->GetGPRSize());
+    } else if (BitOffset == FEXCore::X86State::RFLAG_AF_RAW_LOC) {
+      return _LoadRegister(false, offsetof(FEXCore::Core::CPUState, af_raw), GPRClass, GPRFixedClass, CTX->GetGPRSize());
     } else {
       return _LoadFlag(BitOffset);
     }
   }
 
+  // Set SSE comparison flags based on the result set by Arm FCMP. This converts
+  // NZCV from the Arm representation to an eXternal representation that's
+  // totally not a euphemism for x86 or anything, nuh-uh.
+  void ConvertNZCVToSSE() {
+    if (CTX->HostFeatures.SupportsFlagM2) {
+      LOGMAN_THROW_A_FMT(!NZCVDirty, "only expected after fcmp");
+
+      // We need to set PF according to the unordered flag. We'd rather do this
+      // after axflag, since some impls fuse fcmp+axflag, so we want to do this
+      // after. We can recover "unordered" after axflag as (Z && !C), but
+      // there's no condition code for this so it would take 2 instructions
+      // instead of one, which seems worse than doing 1 op before and breaking
+      // the fusion.
+      //
+      // We set PF to unordered (V), but our PF representation is inverted so we
+      // actually set to !V. This is one instruction with the VC cond code.
+      OrderedNode *PFInvert =
+        _NZCVSelect(OpSize::i32Bit, CondClassType{COND_FNU}, _Constant(1), _Constant(0));
+
+      SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(PFInvert);
+
+      // For the rest, this one weird a64 instruction maps exactly to what x86
+      // needs. What a coincidence!
+      _AXFlag();
+      PossiblySetNZCVBits = ~0;
+
+      // It does assume we invert CF internally, which is still TODO for us. For
+      // now, add a cfinv to deal. Hopefully we delete this later.
+      CarryInvert();
+    } else {
+      OrderedNode *Z = GetRFLAG(FEXCore::X86State::RFLAG_ZF_RAW_LOC);
+      OrderedNode *C_inv = GetRFLAG(FEXCore::X86State::RFLAG_CF_RAW_LOC, true);
+      OrderedNode *V = GetRFLAG(FEXCore::X86State::RFLAG_OF_RAW_LOC);
+
+      // We want to zero SF/OF, and then set CF/ZF. Zeroing up front lets us do
+      // this all with shifted-or's on non-flagm platforms.
+      ZeroNZCV();
+
+      SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(_Or(OpSize::i32Bit, C_inv, V));
+      SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(_Or(OpSize::i32Bit, Z, V));
+
+      // Note that we store PF inverted.
+      // TODO: We could maybe optimize this xor out for non-flagm platforms with
+      // bfi/bfxil?
+      SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(_Xor(OpSize::i32Bit, V, _Constant(1)));
+    }
+  }
+
+  // Set x87 comparison flags based on the result set by Arm FCMP. Clobbers
+  // NZCV on flagm2 platforms.
+  void ConvertNZCVToX87() {
+    OrderedNode *V = GetRFLAG(FEXCore::X86State::RFLAG_OF_RAW_LOC);
+
+    if (CTX->HostFeatures.SupportsFlagM2) {
+      LOGMAN_THROW_A_FMT(!NZCVDirty, "only expected after fcmp");
+
+      // Convert to x86 flags, saves us from or'ing after.
+      _AXFlag();
+      PossiblySetNZCVBits = ~0;
+
+      // Copy the values. CF is inverted from the axflag result, ZF is as-is.
+      SetRFLAG<FEXCore::X86State::X87FLAG_C0_LOC>(GetRFLAG(FEXCore::X86State::RFLAG_CF_RAW_LOC, true));
+      SetRFLAG<FEXCore::X86State::X87FLAG_C3_LOC>(GetRFLAG(FEXCore::X86State::RFLAG_ZF_RAW_LOC));
+    } else {
+      OrderedNode *Z = GetRFLAG(FEXCore::X86State::RFLAG_ZF_RAW_LOC);
+      OrderedNode *N = GetRFLAG(FEXCore::X86State::RFLAG_SF_RAW_LOC);
+
+      SetRFLAG<FEXCore::X86State::X87FLAG_C0_LOC>(_Or(OpSize::i32Bit, N, V));
+      SetRFLAG<FEXCore::X86State::X87FLAG_C3_LOC>(_Or(OpSize::i32Bit, Z, V));
+    }
+
+    SetRFLAG<FEXCore::X86State::X87FLAG_C1_LOC>(_Constant(0));
+    SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(V);
+  }
+
+  // Helper to derive Dest by a given builder-using Expression with the opcode
+  // replaced with NewOp. Useful for generic building code. Not safe in general.
+  // but does the right handling of ImplicitFlagClobber at least and must be
+  // used instead of raw Op mutation.
+#define DeriveOp(Dest, NewOp, Expr) \
+  if (ImplicitFlagClobber(NewOp))   \
+    SaveNZCV(NewOp);                \
+  auto Dest = (Expr);               \
+  Dest.first->Header.Op = (NewOp)
+
   // Named constant cache for the current block.
   // Different arrays for sizes 1,2,4,8,16,32.
   OrderedNode *CachedNamedVectorConstants[FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_MAX][6]{};
@@ -1382,8 +1567,8 @@ private:
     CachedIndexedNamedVectorConstants.clear();
   }
 
-  OrderedNode *SelectMask(OrderedNode *Cmp, uint64_t Mask, bool Invert, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
-  OrderedNode *SelectNZCV(unsigned BitOffset, bool Invert, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
+  std::pair<bool, CondClassType> DecodeNZCVCondition(uint8_t OP) const;
+  OrderedNode *SelectBit(OrderedNode *Cmp, bool Invert, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
   OrderedNode *SelectCC(uint8_t OP, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
 
   /**
@@ -1416,7 +1601,7 @@ private:
     OrderedNode *Res{};
 
     union {
-      // UMUL, BEXTR, BLSI, BLSMSK, POPCOUNT, TZCNT, LZCNT, BITSELECT, RDRAND
+      // UMUL, BEXTR, BLSI, BLSMSK, POPCOUNT, TZCNT, LZCNT, RDRAND
       struct {
       } NoSource;
 
@@ -1486,13 +1671,48 @@ private:
     return CurrentDeferredFlags.Type == FlagsGenerationType::TYPE_NONE;
   }
 
+  template <typename F>
+  void CalculateFlags_ShiftVariable(OrderedNode *Shift, F&& CalculateFlags) {
+    // We are the ones calculating the deferred flags. Don't recurse!
+    InvalidateDeferredFlags();
+
+    // RCR can call this with constants, so handle that without branching.
+    uint64_t Const;
+    if (IsValueConstant(WrapNode(Shift), &Const)) {
+      if (Const)
+        CalculateFlags();
+
+      return;
+    }
+
+    // Otherwise, prepare to branch.
+    uint32_t OldSetNZCVBits = PossiblySetNZCVBits;
+    auto Zero = _Constant(0);
+
+    // If the shift is zero, do not touch the flags.
+    auto SetBlock = CreateNewCodeBlockAfter(GetCurrentBlock());
+    auto EndBlock = CreateNewCodeBlockAfter(SetBlock);
+    CondJump(Shift, Zero, EndBlock, SetBlock, {COND_EQ});
+
+    SetCurrentCodeBlock(SetBlock);
+    StartNewBlock();
+    {
+      CalculateFlags();
+      Jump(EndBlock);
+    }
+
+    SetCurrentCodeBlock(EndBlock);
+    StartNewBlock();
+    PossiblySetNZCVBits |= OldSetNZCVBits;
+  }
+
   /**
    * @name These functions are used by the deferred flag handling while it is calculating and storing flags in to RFLAGs.
    * @{ */
   OrderedNode *LoadPFRaw();
   OrderedNode *LoadAF();
   void FixupAF();
-  void CalculatePF(OrderedNode *Res, OrderedNode *condition = nullptr);
+  void CalculatePF(OrderedNode *Res);
   void CalculateAF(OpSize OpSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
 
   void CalculateOF(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2, bool Sub);
@@ -1515,7 +1735,6 @@ private:
   void CalculateFlags_RotateLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
   void CalculateFlags_RotateRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
   void CalculateFlags_RotateLeftImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift);
-  void CalculateFlags_FCMP(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2);
   void CalculateFlags_BEXTR(OrderedNode *Src);
   void CalculateFlags_BLSI(uint8_t SrcSize, OrderedNode *Src);
   void CalculateFlags_BLSMSK(OrderedNode *Src);
@@ -1524,7 +1743,6 @@ private:
   void CalculateFlags_BZHI(uint8_t SrcSize, OrderedNode *Result, OrderedNode *Src);
   void CalculateFlags_TZCNT(OrderedNode *Src);
   void CalculateFlags_LZCNT(uint8_t SrcSize, OrderedNode *Src);
-  void CalculateFlags_BITSELECT(OrderedNode *Src);
   void CalculateFlags_RDRAND(OrderedNode *Src);
   /**  @} */
 
@@ -1827,20 +2045,6 @@ private:
     };
   }
 
-  void GenerateFlags_FCMP(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-    CurrentDeferredFlags = DeferredFlagData {
-      .Type = FlagsGenerationType::TYPE_FCMP,
-      .SrcSize = GetSrcSize(Op),
-      .Res = Res,
-      .Sources = {
-        .TwoSource = {
-          .Src1 = Src1,
-          .Src2 = Src2,
-        },
-      }
-    };
-  }
-
   void GenerateFlags_BEXTR(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
     CurrentDeferredFlags = DeferredFlagData {
       .Type = FlagsGenerationType::TYPE_BEXTR,
@@ -1915,14 +2119,6 @@ private:
     };
   }
 
-  void GenerateFlags_BITSELECT(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
-    CurrentDeferredFlags = DeferredFlagData {
-      .Type = FlagsGenerationType::TYPE_BITSELECT,
-      .SrcSize = GetSrcSize(Op),
-      .Res = Src,
-    };
-  }
-
   void GenerateFlags_RDRAND(FEXCore::X86Tables::DecodedOp Op, OrderedNode *Src) {
     CurrentDeferredFlags = DeferredFlagData {
       .Type = FlagsGenerationType::TYPE_RDRAND,

FEXCore/Source/Interface/Core/OpcodeDispatcher/Crypto.cpp

@@ -26,9 +26,24 @@ void OpDispatchBuilder::SHA1NEXTEOp(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
 
-  auto Tmp = _Ror(OpSize::i32Bit, _VExtractToGPR(16, 4, Dest, 3), _Constant(32, 2));
-  auto Top = _Add(OpSize::i32Bit, _VExtractToGPR(16, 4, Src, 3), Tmp);
-  auto Result = _VInsGPR(16, 4, 3, Src, Top);
+  OrderedNode *RotatedNode{};
+  if (CTX->HostFeatures.SupportsSHA) {
+    // ARMv8 SHA1 extension provides a `SHA1H` instruction which does a fixed rotate by 30.
+    // This only operates on element 0 rather than element 3. We don't have the luxury of rewriting the x86 SHA algorithm to take advantage of this.
+    // Move the element to zero, rotate, and then move back (Using duplicates).
+    // Saves one instruction versus that path that doesn't support SHA extension.
+    auto Duplicated = _VDupElement(OpSize::i128Bit, OpSize::i32Bit, Dest, 3);
+    auto Sha1HRotated = _VSha1H(Duplicated);
+    RotatedNode = _VDupElement(OpSize::i128Bit, OpSize::i32Bit, Sha1HRotated, 0);
+  }
+  else {
+    // SHA1 extension missing, manually rotate.
+    // Emulate rotate.
+    auto ShiftLeft = _VShlI(OpSize::i128Bit, OpSize::i32Bit, Dest, 30);
+    RotatedNode = _VUShraI(OpSize::i128Bit, OpSize::i32Bit, ShiftLeft, Dest, 2);
+  }
+  auto Tmp = _VAdd(OpSize::i128Bit, OpSize::i32Bit, Src, RotatedNode);
+  auto Result = _VInsElement(OpSize::i128Bit, OpSize::i32Bit, 3, 3, Src, Tmp);
 
   StoreResult(FPRClass, Op, Result, -1);
 }
@@ -49,23 +64,31 @@ void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
 
-  // ROR by 31 is equivalent to a ROL by 1
-  auto ThirtyOne = _Constant(32, 31);
+  // This instruction mostly matches ARMv8's SHA1SU1 instruction but one of the elements are flipped in an unexpected way.
+  // Do all the work without it.
 
-  auto W13 = _VExtractToGPR(16, 4, Src, 2);
-  auto W14 = _VExtractToGPR(16, 4, Src, 1);
-  auto W15 = _VExtractToGPR(16, 4, Src, 0);
-  auto W16 = _Ror(OpSize::i32Bit, _Xor(OpSize::i32Bit, _VExtractToGPR(16, 4, Dest, 3), W13), ThirtyOne);
-  auto W17 = _Ror(OpSize::i32Bit, _Xor(OpSize::i32Bit, _VExtractToGPR(16, 4, Dest, 2), W14), ThirtyOne);
-  auto W18 = _Ror(OpSize::i32Bit, _Xor(OpSize::i32Bit, _VExtractToGPR(16, 4, Dest, 1), W15), ThirtyOne);
-  auto W19 = _Ror(OpSize::i32Bit, _Xor(OpSize::i32Bit, _VExtractToGPR(16, 4, Dest, 0), W16), ThirtyOne);
+  const auto ZeroRegister = LoadAndCacheNamedVectorConstant(OpSize::i32Bit, FEXCore::IR::NamedVectorConstant::NAMED_VECTOR_ZERO);
 
-  auto D3 = _VInsGPR(16, 4, 3, Dest, W16);
-  auto D2 = _VInsGPR(16, 4, 2, D3, W17);
-  auto D1 = _VInsGPR(16, 4, 1, D2, W18);
-  auto D0 = _VInsGPR(16, 4, 0, D1, W19);
+  // Shift the incoming source left by a 32-bit element, inserting Zeros.
+  // This could be slightly improved to use a VInsGPR with the zero register.
+  auto Src2Shift = _VExtr(OpSize::i128Bit, OpSize::i8Bit, Src, ZeroRegister, 12);
+  auto Xor1 = _VXor(OpSize::i128Bit, OpSize::i8Bit, Dest, Src2Shift);
 
-  StoreResult(FPRClass, Op, D0, -1);
+  // Emulate rotate.
+  auto ShiftLeftXor1 = _VShlI(OpSize::i128Bit, OpSize::i32Bit, Xor1, 1);
+  auto RotatedXor1 = _VUShraI(OpSize::i128Bit, OpSize::i32Bit, ShiftLeftXor1, Xor1, 31);
+
+  // Element0 didn't get XOR'd with anything, so do it now.
+  auto ExtractUpper = _VDupElement(OpSize::i128Bit, OpSize::i32Bit, RotatedXor1, 3);
+  auto XorLower = _VXor(OpSize::i128Bit, OpSize::i8Bit, Dest, ExtractUpper);
+
+  // Emulate rotate.
+  auto ShiftLeftXorLower = _VShlI(OpSize::i128Bit, OpSize::i32Bit, XorLower, 1);
+  auto RotatedXorLower = _VUShraI(OpSize::i128Bit, OpSize::i32Bit, ShiftLeftXorLower, XorLower, 31);
+
+  auto Result = _VInsElement(OpSize::i128Bit, OpSize::i32Bit, 0, 0, RotatedXor1, RotatedXorLower);
+
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
@@ -151,30 +174,37 @@ void OpDispatchBuilder::SHA1RNDS4Op(OpcodeArgs) {
 }
 
 void OpDispatchBuilder::SHA256MSG1Op(OpcodeArgs) {
-  const auto Sigma0 = [this](OrderedNode* W) -> OrderedNode* {
-    return _Xor(OpSize::i32Bit, _Xor(OpSize::i32Bit, _Ror(OpSize::i32Bit, W, _Constant(32, 7)), _Ror(OpSize::i32Bit, W, _Constant(32, 18))), _Lshr(OpSize::i32Bit, W, _Constant(32, 3)));
-  };
-
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
 
-  auto W4 = _VExtractToGPR(16, 4, Src, 0);
-  auto W3 = _VExtractToGPR(16, 4, Dest, 3);
-  auto W2 = _VExtractToGPR(16, 4, Dest, 2);
-  auto W1 = _VExtractToGPR(16, 4, Dest, 1);
-  auto W0 = _VExtractToGPR(16, 4, Dest, 0);
+  OrderedNode *Result{};
 
-  auto Sig3 = _Add(OpSize::i32Bit, W3, Sigma0(W4));
-  auto Sig2 = _Add(OpSize::i32Bit, W2, Sigma0(W3));
-  auto Sig1 = _Add(OpSize::i32Bit, W1, Sigma0(W2));
-  auto Sig0 = _Add(OpSize::i32Bit, W0, Sigma0(W1));
+  if (CTX->HostFeatures.SupportsSHA) {
+    Result = _VSha256U0(Dest, Src);
+  }
+  else {
+    const auto Sigma0 = [this](OrderedNode* W) -> OrderedNode* {
+      return _Xor(OpSize::i32Bit, _Xor(OpSize::i32Bit, _Ror(OpSize::i32Bit, W, _Constant(32, 7)), _Ror(OpSize::i32Bit, W, _Constant(32, 18))), _Lshr(OpSize::i32Bit, W, _Constant(32, 3)));
+    };
 
-  auto D3 = _VInsGPR(16, 4, 3, Dest, Sig3);
-  auto D2 = _VInsGPR(16, 4, 2, D3, Sig2);
-  auto D1 = _VInsGPR(16, 4, 1, D2, Sig1);
-  auto D0 = _VInsGPR(16, 4, 0, D1, Sig0);
+    auto W4 = _VExtractToGPR(16, 4, Src, 0);
+    auto W3 = _VExtractToGPR(16, 4, Dest, 3);
+    auto W2 = _VExtractToGPR(16, 4, Dest, 2);
+    auto W1 = _VExtractToGPR(16, 4, Dest, 1);
+    auto W0 = _VExtractToGPR(16, 4, Dest, 0);
 
-  StoreResult(FPRClass, Op, D0, -1);
+    auto Sig3 = _Add(OpSize::i32Bit, W3, Sigma0(W4));
+    auto Sig2 = _Add(OpSize::i32Bit, W2, Sigma0(W3));
+    auto Sig1 = _Add(OpSize::i32Bit, W1, Sigma0(W2));
+    auto Sig0 = _Add(OpSize::i32Bit, W0, Sigma0(W1));
+
+    auto D3 = _VInsGPR(16, 4, 3, Dest, Sig3);
+    auto D2 = _VInsGPR(16, 4, 2, D3, Sig2);
+    auto D1 = _VInsGPR(16, 4, 1, D2, Sig1);
+    Result = _VInsGPR(16, 4, 0, D1, Sig0);
+  }
+
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::SHA256MSG2Op(OpcodeArgs) {

FEXCore/Source/Interface/Core/OpcodeDispatcher/Flags.cpp

@@ -40,7 +40,6 @@ constexpr std::array<uint32_t, 17> FlagOffsets = {
 };
 
 void OpDispatchBuilder::ZeroMultipleFlags(uint32_t FlagsMask) {
-  flagsOp = SelectionFlag::Nothing;
   auto ZeroConst = _Constant(0);
 
   if (ContainsNZCV(FlagsMask)) {
@@ -130,8 +129,7 @@ void OpDispatchBuilder::SetPackedRFLAG(bool Lower8, OrderedNode *Src) {
       Tmp = _Xor(OpSize::i32Bit, Tmp, _Constant(1));
       SetRFLAG(Tmp, FlagOffset);
     } else {
-      auto Tmp = _Bfe(OpSize::i32Bit, 1, FlagOffset, Src);
-      SetRFLAG(Tmp, FlagOffset);
+      SetRFLAG(Src, FlagOffset, FlagOffset, true);
     }
   }
 }
@@ -237,17 +235,17 @@ void OpDispatchBuilder::CalculateOF(uint8_t SrcSize, OrderedNode *Res, OrderedNo
       Anded = _Andn(OpSize, XorOp2, XorOp1);
   }
 
-  auto OF = _Bfe(OpSize, 1, SrcSize * 8 - 1, Anded);
-  SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(OF);
+  SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(Anded, SrcSize * 8 - 1, true);
 }
 
 OrderedNode *OpDispatchBuilder::LoadPFRaw() {
-  // Read the stored byte. This is the original 8-bit result, it needs parity calculated.
-  auto PFByte = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
+  // Read the stored byte. This is the original result (up to 64-bits), it needs
+  // parity calculated.
+  auto Result = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
 
   // Cast the input to a 32-bit FPR. Logically we only need 8-bit, but that would
   // generate unwanted an ubfx instruction. VPopcount will ignore the upper bits anyway.
-  auto InputFPR = _VCastFromGPR(4, 4, PFByte);
+  auto InputFPR = _VCastFromGPR(4, 4, Result);
 
   // Calculate the popcount.
   auto Count = _VPopcount(1, 1, InputFPR);
@@ -255,14 +253,14 @@ OrderedNode *OpDispatchBuilder::LoadPFRaw() {
 }
 
 OrderedNode *OpDispatchBuilder::LoadAF() {
-  // Read the stored byte. This is the XOR of the arguments.
-  auto AFByte = GetRFLAG(FEXCore::X86State::RFLAG_AF_RAW_LOC);
+  // Read the stored value. This is the XOR of the arguments.
+  auto AFWord = GetRFLAG(FEXCore::X86State::RFLAG_AF_RAW_LOC);
 
-  // Read the result, stored as the PF byte for deferred PF calculation.
-  auto PFByte = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
+  // Read the result, stored for PF.
+  auto Result = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
 
   // What's left is to XOR and extract. This is the deferred part.
-  return _Bfe(OpSize::i32Bit, 1, 4, _Xor(OpSize::i32Bit, AFByte, PFByte));
+  return _Bfe(OpSize::i32Bit, 1, 4, _Xor(OpSize::i32Bit, AFWord, Result));
 }
 
 void OpDispatchBuilder::FixupAF() {
@@ -272,23 +270,14 @@ void OpDispatchBuilder::FixupAF() {
   //
   //  (AF[4] ^ PF[4]) ^ PF[4] = AF[4]
 
-  auto PFByte = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
-  auto AFByte = GetRFLAG(FEXCore::X86State::RFLAG_AF_RAW_LOC);
+  auto PFRaw = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
+  auto AFRaw = GetRFLAG(FEXCore::X86State::RFLAG_AF_RAW_LOC);
 
-  OrderedNode *XorRes = _Xor(OpSize::i32Bit, AFByte, PFByte);
+  OrderedNode *XorRes = _Xor(OpSize::i32Bit, AFRaw, PFRaw);
   SetRFLAG<FEXCore::X86State::RFLAG_AF_RAW_LOC>(XorRes);
 }
 
-void OpDispatchBuilder::CalculatePF(OrderedNode *Res, OrderedNode *condition) {
-  // For shifts, we can only update for nonzero shift. If zero, we nop out the flag write by
-  // writing the existing value. Note we call GetRFLAG directly, rather than LoadPFRaw, because
-  // we need the existing /encoded/ value rather than the decoded PF value. In particular,
-  // this does not calculate a popcount.
-  if (condition) {
-    auto OldFlag = GetRFLAG(FEXCore::X86State::RFLAG_PF_RAW_LOC);
-    Res = _Select(FEXCore::IR::COND_EQ, condition, _Constant(0), OldFlag, Res);
-  }
-
+void OpDispatchBuilder::CalculatePF(OrderedNode *Res) {
   // Calculation is entirely deferred until load, just store the 8-bit result.
   SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(Res);
 }
@@ -313,7 +302,7 @@ void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
   if (CurrentDeferredFlags.Type == FlagsGenerationType::TYPE_NONE) {
     // Nothing to do
     if (NZCVDirty && CachedNZCV)
-      _StoreFlag(CachedNZCV, FEXCore::X86State::RFLAG_NZCV_LOC);
+      _StoreNZCV(CachedNZCV);
 
     CachedNZCV = nullptr;
     NZCVDirty = false;
@@ -446,13 +435,6 @@ void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
         CurrentDeferredFlags.Sources.OneSrcImmediate.Src1,
         CurrentDeferredFlags.Sources.OneSrcImmediate.Imm);
       break;
-    case FlagsGenerationType::TYPE_FCMP:
-      CalculateFlags_FCMP(
-        CurrentDeferredFlags.SrcSize,
-        CurrentDeferredFlags.Res,
-        CurrentDeferredFlags.Sources.TwoSource.Src1,
-        CurrentDeferredFlags.Sources.TwoSource.Src2);
-      break;
     case FlagsGenerationType::TYPE_BEXTR:
       CalculateFlags_BEXTR(CurrentDeferredFlags.Res);
       break;
@@ -487,9 +469,6 @@ void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
         CurrentDeferredFlags.SrcSize,
         CurrentDeferredFlags.Res);
       break;
-    case FlagsGenerationType::TYPE_BITSELECT:
-      CalculateFlags_BITSELECT(CurrentDeferredFlags.Res);
-      break;
     case FlagsGenerationType::TYPE_RDRAND:
       CalculateFlags_RDRAND(CurrentDeferredFlags.Res);
       break;
@@ -501,7 +480,7 @@ void OpDispatchBuilder::CalculateDeferredFlags(uint32_t FlagsToCalculateMask) {
   CurrentDeferredFlags.Type = FlagsGenerationType::TYPE_NONE;
 
   if (NZCVDirty && CachedNZCV)
-    _StoreFlag(CachedNZCV, FEXCore::X86State::RFLAG_NZCV_LOC);
+    _StoreNZCV(CachedNZCV);
 
   CachedNZCV = nullptr;
   NZCVDirty = false;
@@ -516,7 +495,12 @@ void OpDispatchBuilder::CalculateFlags_ADC(uint8_t SrcSize, OrderedNode *Res, Or
   CalculatePF(Res);
 
   if (SrcSize >= 4) {
-    SetNZCV(_AdcNZCV(OpSize, Src1, Src2, GetNZCV()));
+    if (NZCVDirty && CachedNZCV)
+      _StoreNZCV(CachedNZCV);
+    CachedNZCV = nullptr;
+
+    _AdcNZCV(OpSize, Src1, Src2);
+    PossiblySetNZCVBits = ~0;
   } else {
     // SF/ZF
     SetNZ_ZeroCV(SrcSize, Res);
@@ -544,7 +528,19 @@ void OpDispatchBuilder::CalculateFlags_SBB(uint8_t SrcSize, OrderedNode *Res, Or
   CalculatePF(Res);
 
   if (SrcSize >= 4) {
-    SetNZCV(_SbbNZCV(OpSize, Src1, Src2, GetNZCV()));
+    // Rectify input carry
+    CarryInvert();
+
+    if (NZCVDirty && CachedNZCV)
+      _StoreNZCV(CachedNZCV);
+    CachedNZCV = nullptr;
+    NZCVDirty = false;
+
+    _SbbNZCV(OpSize, Src1, Src2);
+    PossiblySetNZCVBits = ~0;
+
+    // Rectify output carry
+    CarryInvert();
   } else {
     // SF/ZF
     SetNZ_ZeroCV(SrcSize, Res);
@@ -574,8 +570,14 @@ void OpDispatchBuilder::CalculateFlags_SUB(uint8_t SrcSize, OrderedNode *Res, Or
 
   // TODO: Could do this path for small sources if we have FEAT_FlagM
   if (SrcSize >= 4) {
+    _SubNZCV(OpSize, Src1, Src2);
+    CachedNZCV = nullptr;
+    NZCVDirty = false;
+    PossiblySetNZCVBits = ~0;
+
     // We only bother inverting CF if we're actually going to update CF.
-    SetNZCV(_SubNZCV(OpSize, Src1, Src2, UpdateCF));
+    if (UpdateCF)
+      CarryInvert();
   } else {
     // SF/ZF
     SetNZ_ZeroCV(SrcSize, Res);
@@ -583,8 +585,7 @@ void OpDispatchBuilder::CalculateFlags_SUB(uint8_t SrcSize, OrderedNode *Res, Or
     // CF
     if (UpdateCF) {
       // Grab carry bit from unmasked output.
-      auto Bfe = _Bfe(OpSize::i32Bit, 1, SrcSize * 8, Res);
-      SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Bfe);
+      SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Res, SrcSize * 8, true);
     }
 
     CalculateOF(SrcSize, Res, Src1, Src2, true);
@@ -606,7 +607,10 @@ void OpDispatchBuilder::CalculateFlags_ADD(uint8_t SrcSize, OrderedNode *Res, Or
 
   // TODO: Could do this path for small sources if we have FEAT_FlagM
   if (SrcSize >= 4) {
-    SetNZCV(_AddNZCV(OpSize, Src1, Src2));
+    _AddNZCV(OpSize, Src1, Src2);
+    CachedNZCV = nullptr;
+    NZCVDirty = false;
+    PossiblySetNZCVBits = ~0;
   } else {
     // SF/ZF
     SetNZ_ZeroCV(SrcSize, Res);
@@ -614,8 +618,7 @@ void OpDispatchBuilder::CalculateFlags_ADD(uint8_t SrcSize, OrderedNode *Res, Or
     // CF
     if (UpdateCF) {
       // Grab carry bit from unmasked output
-      auto Bfe = _Bfe(OpSize::i32Bit, 1, SrcSize * 8, Res);
-      SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Bfe);
+      SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Res, SrcSize * 8, true);
     }
 
     CalculateOF(SrcSize, Res, Src1, Src2, false);
@@ -627,8 +630,6 @@ void OpDispatchBuilder::CalculateFlags_ADD(uint8_t SrcSize, OrderedNode *Res, Or
 }
 
 void OpDispatchBuilder::CalculateFlags_MUL(uint8_t SrcSize, OrderedNode *Res, OrderedNode *High) {
-  auto Zero = _Constant(0);
-
   // PF/AF/ZF/SF
   // Undefined
   {
@@ -640,19 +641,22 @@ void OpDispatchBuilder::CalculateFlags_MUL(uint8_t SrcSize, OrderedNode *Res, Or
   {
     // CF and OF are set if the result of the operation can't be fit in to the destination register
     // If the value can fit then the top bits will be zero
-
     auto SignBit = _Sbfe(OpSize::i64Bit, 1, SrcSize * 8 - 1, Res);
+    _SubNZCV(OpSize::i64Bit, High, SignBit);
 
-    auto CV = _Constant((1u << IndexNZCV(FEXCore::X86State::RFLAG_CF_RAW_LOC)) |
-                        (1u << IndexNZCV(FEXCore::X86State::RFLAG_OF_RAW_LOC)));
-
-    // Set CV accordingly and zero NZ regardless
-    SetNZCV(_Select(FEXCore::IR::COND_EQ, High, SignBit, Zero, CV));
+    // If High = SignBit, then sets to nZcv. Else sets to nzCV. Since SF/ZF
+    // undefined, this does what we need.
+    auto Zero = _Constant(0);
+    _CondAddNZCV(OpSize::i64Bit, Zero, Zero, CondClassType{COND_EQ}, 0x3 /* nzCV */);
+    CachedNZCV = nullptr;
+    NZCVDirty = false;
+    PossiblySetNZCVBits = ~0;
   }
 }
 
 void OpDispatchBuilder::CalculateFlags_UMUL(OrderedNode *High) {
   auto Zero = _Constant(0);
+  OpSize Size = IR::SizeToOpSize(GetOpSize(High));
 
   // AF/SF/PF/ZF
   // Undefined
@@ -665,11 +669,14 @@ void OpDispatchBuilder::CalculateFlags_UMUL(OrderedNode *High) {
   {
     // CF and OF are set if the result of the operation can't be fit in to the destination register
     // The result register will be all zero if it can't fit due to how multiplication behaves
+    _SubNZCV(Size, High, Zero);
 
-    auto CV = _Constant((1u << IndexNZCV(FEXCore::X86State::RFLAG_CF_RAW_LOC)) |
-                        (1u << IndexNZCV(FEXCore::X86State::RFLAG_OF_RAW_LOC)));
-
-    SetNZCV(_Select(FEXCore::IR::COND_EQ, High, Zero, Zero, CV));
+    // If High = 0, then sets to nZcv. Else sets to nzCV. Since SF/ZF undefined,
+    // this does what we need.
+    _CondAddNZCV(Size, Zero, Zero, CondClassType{COND_EQ}, 0x3 /* nzCV */);
+    CachedNZCV = nullptr;
+    NZCVDirty = false;
+    PossiblySetNZCVBits = ~0;
   }
 }
 
@@ -685,118 +692,79 @@ void OpDispatchBuilder::CalculateFlags_Logical(uint8_t SrcSize, OrderedNode *Res
 }
 
 void OpDispatchBuilder::CalculateFlags_ShiftLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
+  CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res, Src1, Src2](){
+    const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
+    SetNZ_ZeroCV(SrcSize, Res);
 
-  auto Zero = _Constant(0);
-
-  auto OldNZCV = GetNZCV();
-  uint32_t OldSetNZCVBits = PossiblySetNZCVBits;
-
-  SetNZ_ZeroCV(SrcSize, Res);
-
-  // CF
-  {
     // Extract the last bit shifted in to CF
     auto Size = _Constant(SrcSize * 8);
     auto ShiftAmt = _Sub(OpSize, Size, Src2);
-    auto LastBit = _Bfe(OpSize, 1, 0, _Lshr(OpSize, Src1, ShiftAmt));
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit);
-  }
+    auto LastBit = _Lshr(OpSize, Src1, ShiftAmt);
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit, 0, true);
 
-  CalculatePF(Res, Src2);
+    CalculatePF(Res);
 
-  // AF
-  // Undefined
-  _InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
+    // AF
+    // Undefined
+    _InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
 
-  // OF
-  {
     // In the case of left shift. OF is only set from the result of <Top Source Bit> XOR <Top Result Bit>
     // When Shift > 1 then OF is undefined
-    auto val = _Bfe(OpSize, 1, SrcSize * 8 - 1, _Xor(OpSize, Src1, Res));
-    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(val);
-  }
-
-  // Now select between the two
-  SetNZCV(_Select(FEXCore::IR::COND_EQ, Src2, Zero, OldNZCV, GetNZCV()));
-  PossiblySetNZCVBits |= OldSetNZCVBits;
+    auto OFXor = _Xor(OpSize, Src1, Res);
+    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(OFXor, SrcSize * 8 - 1, true);
+  });
 }
 
 void OpDispatchBuilder::CalculateFlags_ShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
-  auto Zero = _Constant(0);
-  auto One = _Constant(1);
+  CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res, Src1, Src2](){
+    const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
+    SetNZ_ZeroCV(SrcSize, Res);
 
-  auto OldNZCV = GetNZCV();
-  uint32_t OldSetNZCVBits = PossiblySetNZCVBits;
-
-  SetNZ_ZeroCV(SrcSize, Res);
-
-  // CF
-  {
     // Extract the last bit shifted in to CF
-    auto ShiftAmt = _Sub(OpSize::i64Bit, Src2, One);
+    auto ShiftAmt = _Sub(OpSize::i64Bit, Src2, _Constant(1));
     const auto CFSize = IR::SizeToOpSize(std::max<uint8_t>(4u, SrcSize));
-    auto LastBit = _Bfe(CFSize, 1, 0, _Lshr(CFSize, Src1, ShiftAmt));
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit);
-  }
+    auto LastBit = _Lshr(CFSize, Src1, ShiftAmt);
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit, 0, true);
 
-  CalculatePF(Res, Src2);
+    CalculatePF(Res);
 
-  // AF
-  // Undefined
-  _InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
+    // AF
+    // Undefined
+    _InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
 
-  // OF
-  {
     // Only defined when Shift is 1 else undefined
     // OF flag is set if a sign change occurred
-    auto val = _Bfe(OpSize, 1, SrcSize * 8 - 1, _Xor(OpSize, Src1, Res));
-    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(val);
-  }
-
-  // Now select between the two
-  SetNZCV(_Select(FEXCore::IR::COND_EQ, Src2, Zero, OldNZCV, GetNZCV()));
-  PossiblySetNZCVBits |= OldSetNZCVBits;
+    auto val = _Xor(OpSize, Src1, Res);
+    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(val, SrcSize * 8 - 1, true);
+  });
 }
 
 void OpDispatchBuilder::CalculateFlags_SignShiftRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  auto Zero = _Constant(0);
-  auto One = _Constant(1);
+  CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res, Src1, Src2](){
+    // SF/ZF/OF
+    SetNZ_ZeroCV(SrcSize, Res);
 
-  auto OldNZCV = GetNZCV();
-  uint32_t OldSetNZCVBits = PossiblySetNZCVBits;
-
-  // SF/ZF/OF
-  SetNZ_ZeroCV(SrcSize, Res);
-
-  // CF
-  {
     // Extract the last bit shifted in to CF
     const auto CFSize = IR::SizeToOpSize(std::max<uint32_t>(4u, GetOpSize(Src1)));
-    auto ShiftAmt = _Sub(OpSize::i64Bit, Src2, One);
-    auto LastBit = _Bfe(CFSize, 1, 0, _Lshr(CFSize, Src1, ShiftAmt));
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit);
-  }
+    auto ShiftAmt = _Sub(OpSize::i64Bit, Src2, _Constant(1));
+    auto LastBit = _Lshr(CFSize, Src1, ShiftAmt);
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(LastBit, 0, true);
 
-  CalculatePF(Res, Src2);
+    CalculatePF(Res);
 
-  // AF
-  // Undefined
-  _InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
-
-  // Now select between the two
-  SetNZCV(_Select(FEXCore::IR::COND_EQ, Src2, Zero, OldNZCV, GetNZCV()));
-  PossiblySetNZCVBits |= OldSetNZCVBits;
+    // AF
+    // Undefined
+    _InvalidateFlags(1 << X86State::RFLAG_AF_RAW_LOC);
+  });
 }
 
-void OpDispatchBuilder::CalculateFlags_ShiftLeftImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
+void OpDispatchBuilder::CalculateFlags_ShiftLeftImmediate(uint8_t SrcSize, OrderedNode *UnmaskedRes, OrderedNode *Src1, uint64_t Shift) {
   // No flags changed if shift is zero
   if (Shift == 0) return;
 
   auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
 
-  SetNZ_ZeroCV(SrcSize, Res);
+  SetNZ_ZeroCV(SrcSize, UnmaskedRes);
 
   // CF
   {
@@ -805,10 +773,10 @@ void OpDispatchBuilder::CalculateFlags_ShiftLeftImmediate(uint8_t SrcSize, Order
     if (SrcSizeBits < Shift) {
       Shift &= (SrcSizeBits - 1);
     }
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(_Bfe(OpSize, 1, SrcSizeBits - Shift, Src1));
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Src1, SrcSizeBits - Shift, true);
   }
 
-  CalculatePF(Res);
+  CalculatePF(UnmaskedRes);
 
   // AF
   // Undefined
@@ -817,9 +785,8 @@ void OpDispatchBuilder::CalculateFlags_ShiftLeftImmediate(uint8_t SrcSize, Order
   // OF
   // In the case of left shift. OF is only set from the result of <Top Source Bit> XOR <Top Result Bit>
   if (Shift == 1) {
-    auto Xor = _Xor(OpSize, Res, Src1);
-    auto OF = _Bfe(OpSize, 1, SrcSize * 8 - 1, Xor);
-    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(OF);
+    auto Xor = _Xor(OpSize, UnmaskedRes, Src1);
+    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(Xor, SrcSize * 8 - 1, true);
   } else {
     // Undefined, we choose to zero as part of SetNZ_ZeroCV
   }
@@ -834,7 +801,7 @@ void OpDispatchBuilder::CalculateFlags_SignShiftRightImmediate(uint8_t SrcSize,
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(_Bfe(IR::SizeToOpSize(std::max<uint32_t>(4u, GetOpSize(Src1))), 1, Shift-1, Src1));
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Src1, Shift-1, true);
   }
 
   CalculatePF(Res);
@@ -850,8 +817,6 @@ void OpDispatchBuilder::CalculateFlags_SignShiftRightImmediate(uint8_t SrcSize,
 }
 
 void OpDispatchBuilder::CalculateFlags_ShiftRightImmediateCommon(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
-
   // Stash OF before overwriting it
   auto OldOF = Shift != 1 ? GetRFLAG(FEXCore::X86State::RFLAG_OF_RAW_LOC) : NULL;
   SetNZ_ZeroCV(SrcSize, Res);
@@ -859,7 +824,7 @@ void OpDispatchBuilder::CalculateFlags_ShiftRightImmediateCommon(uint8_t SrcSize
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(_Bfe(OpSize, 1, Shift-1, Src1));
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Src1, Shift-1, true);
   }
 
   CalculatePF(Res);
@@ -878,7 +843,6 @@ void OpDispatchBuilder::CalculateFlags_ShiftRightImmediate(uint8_t SrcSize, Orde
   // No flags changed if shift is zero
   if (Shift == 0) return;
 
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
   CalculateFlags_ShiftRightImmediateCommon(SrcSize, Res, Src1, Shift);
 
   // OF
@@ -886,7 +850,7 @@ void OpDispatchBuilder::CalculateFlags_ShiftRightImmediate(uint8_t SrcSize, Orde
     // Only defined when Shift is 1 else undefined
     // Is set to the MSB of the original value
     if (Shift == 1) {
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(_Bfe(OpSize, 1, SrcSize * 8 - 1, Src1));
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(Src1, SrcSize * 8 - 1, true);
     }
   }
 }
@@ -904,62 +868,53 @@ void OpDispatchBuilder::CalculateFlags_ShiftRightDoubleImmediate(uint8_t SrcSize
     // Is set if the MSB bit changes.
     // XOR of Result and Src1
     if (Shift == 1) {
-      auto val = _Bfe(OpSize, 1, SrcSize * 8 - 1, _Xor(OpSize, Src1, Res));
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(val);
+      auto val = _Xor(OpSize, Src1, Res);
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(val, SrcSize * 8 - 1, true);
     }
   }
 }
 
 void OpDispatchBuilder::CalculateFlags_RotateRight(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  auto Zero = _Constant(0);
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
-  auto SizeBits = SrcSize * 8;
+  CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res](){
+    auto SizeBits = SrcSize * 8;
+    const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
 
-  auto OldNZCV = GetNZCV();
-  auto OldSetNZCVBits = PossiblySetNZCVBits;
-  ZeroCV();
+    // Ends up faster overall if we don't have FlagM, slower if we do...
+    // If Shift != 1, OF is undefined so we choose to zero here.
+    if (!CTX->HostFeatures.SupportsFlagM)
+      ZeroCV();
 
-  // Extract the last bit shifted in to CF
-  auto NewCF = _Bfe(OpSize, 1, SizeBits - 1, Res);
-  SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(NewCF);
+    // Extract the last bit shifted in to CF
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Res, SizeBits - 1, true);
 
-  // OF is set to the XOR of the new CF bit and the most significant bit of the result
-  // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
-  auto NewOF = _Xor(OpSize, _Bfe(OpSize, 1, SizeBits - 2, Res), NewCF);
-  SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF);
-
-  // Now select: if shift == 0, don't update flags
-  SetNZCV(_Select(FEXCore::IR::COND_EQ, Src2, Zero, OldNZCV, GetNZCV()));
-  PossiblySetNZCVBits |= OldSetNZCVBits;
+    // OF is set to the XOR of the new CF bit and the most significant bit of the result
+    // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
+    auto NewOF = _XorShift(OpSize, Res, Res, ShiftType::LSR, 1);
+    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF, SizeBits - 2, true);
+  });
 }
 
 void OpDispatchBuilder::CalculateFlags_RotateLeft(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  auto Zero = _Constant(0);
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
-  auto SizeBits = SrcSize * 8;
+  CalculateFlags_ShiftVariable(Src2, [this, SrcSize, Res](){
+    const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
+    auto SizeBits = SrcSize * 8;
 
-  auto OldNZCV = GetNZCV();
-  auto OldSetNZCVBits = PossiblySetNZCVBits;
+    // Ends up faster overall if we don't have FlagM, slower if we do...
+    // If Shift != 1, OF is undefined so we choose to zero here.
+    if (!CTX->HostFeatures.SupportsFlagM)
+      ZeroCV();
 
-  // Ends up faster overall.
-  // XXX: can do much better if we have FlagM (with RMIF).
-  ZeroCV();
+    // Extract the last bit shifted in to CF
+    //auto Size = _Constant(GetSrcSize(Res) * 8);
+    //auto ShiftAmt = _Sub(OpSize::i64Bit, Size, Src2);
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Res, 0, true);
 
-  // Extract the last bit shifted in to CF
-  //auto Size = _Constant(GetSrcSize(Res) * 8);
-  //auto ShiftAmt = _Sub(OpSize::i64Bit, Size, Src2);
-  auto NewCF = _Bfe(OpSize, 1, 0, Res);
-  SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(NewCF);
-
-  // OF is the LSB and MSB XOR'd together.
-  // OF is set to the XOR of the new CF bit and the most significant bit of the result.
-  // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
-  auto NewOF = _Xor(OpSize, _Bfe(OpSize, 1, SizeBits - 1, Res), NewCF);
-  SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF);
-
-  // Now select: if shift == 0, don't update flags
-  SetNZCV(_Select(FEXCore::IR::COND_EQ, Src2, Zero, OldNZCV, GetNZCV()));
-  PossiblySetNZCVBits |= OldSetNZCVBits;
+    // OF is the LSB and MSB XOR'd together.
+    // OF is set to the XOR of the new CF bit and the most significant bit of the result.
+    // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
+    auto NewOF = _XorShift(OpSize, Res, Res, ShiftType::LSR, SizeBits - 1);
+    SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF, 0, true);
+  });
 }
 
 void OpDispatchBuilder::CalculateFlags_RotateRightImmediate(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, uint64_t Shift) {
@@ -967,16 +922,16 @@ void OpDispatchBuilder::CalculateFlags_RotateRightImmediate(uint8_t SrcSize, Ord
 
   const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
   auto SizeBits = SrcSize * 8;
-  auto NewCF = _Bfe(OpSize, 1, SizeBits - 1, Res);
 
-  // Ends up faster overall. If Shift != 1, OF is undefined so we choose to zero here.
-  // XXX: can do much better if we have FlagM (with RMIF).
-  ZeroCV();
+  // Ends up faster overall if we don't have FlagM, slower if we do...
+  // If Shift != 1, OF is undefined so we choose to zero here.
+  if (!CTX->HostFeatures.SupportsFlagM)
+    ZeroCV();
 
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(NewCF);
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Res, SizeBits - 1, true);
   }
 
   // OF
@@ -984,8 +939,8 @@ void OpDispatchBuilder::CalculateFlags_RotateRightImmediate(uint8_t SrcSize, Ord
     if (Shift == 1) {
       // OF is the top two MSBs XOR'd together
       // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
-      auto NewOF = _Xor(OpSize, _Bfe(OpSize, 1, SizeBits - 2, Res), NewCF);
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF);
+      auto NewOF = _XorShift(OpSize, Res, Res, ShiftType::LSR, 1);
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF, SizeBits - 2, 1);
     }
   }
 }
@@ -996,16 +951,15 @@ void OpDispatchBuilder::CalculateFlags_RotateLeftImmediate(uint8_t SrcSize, Orde
   const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
   auto SizeBits = SrcSize * 8;
 
-  auto NewCF = _Bfe(OpSize, 1, 0, Res);
-
-  // Ends up faster overall. If Shift != 1, OF is undefined so we choose to zero here.
-  // XXX: can do much better if we have FlagM (with RMIF).
-  ZeroCV();
+  // Ends up faster overall if we don't have FlagM, slower if we do...
+  // If Shift != 1, OF is undefined so we choose to zero here.
+  if (!CTX->HostFeatures.SupportsFlagM)
+    ZeroCV();
 
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(NewCF);
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(Res, 0, true);
   }
 
   // OF
@@ -1014,37 +968,13 @@ void OpDispatchBuilder::CalculateFlags_RotateLeftImmediate(uint8_t SrcSize, Orde
       // OF is the LSB and MSB XOR'd together.
       // OF is set to the XOR of the new CF bit and the most significant bit of the result.
       // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
-      auto NewOF = _Xor(OpSize, _Bfe(OpSize, 1, SizeBits - 1, Res), NewCF);
+      auto NewOF = _XorShift(OpSize, Res, Res, ShiftType::LSR, SizeBits - 1);
 
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF);
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(NewOF, 0, true);
     }
   }
 }
 
-void OpDispatchBuilder::CalculateFlags_FCMP(uint8_t SrcSize, OrderedNode *Res, OrderedNode *Src1, OrderedNode *Src2) {
-  OrderedNode *HostFlag_CF = _GetHostFlag(Res, FCMP_FLAG_LT);
-  OrderedNode *HostFlag_ZF = _GetHostFlag(Res, FCMP_FLAG_EQ);
-  OrderedNode *HostFlag_Unordered  = _GetHostFlag(Res, FCMP_FLAG_UNORDERED);
-
-  SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(HostFlag_CF);
-  SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(HostFlag_ZF);
-
-  // PF is stored inverted, so invert from the host flag.
-  // TODO: This could perhaps be optimized?
-  auto PF = _Xor(OpSize::i32Bit, HostFlag_Unordered, _Constant(1));
-  SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(PF);
-
-  // Zero AF. Note that we set the PF byte to 0/1 above, so PF[4] is 0 so the
-  // XOR with PF will have no effect, so setting the AF byte to zero will indeed
-  // zero AF as intended.
-  uint32_t FlagsMaskToZero =
-    (1U << X86State::RFLAG_AF_RAW_LOC) |
-    (1U << X86State::RFLAG_SF_RAW_LOC) |
-    (1U << X86State::RFLAG_OF_RAW_LOC);
-
-  ZeroMultipleFlags(FlagsMaskToZero);
-}
-
 void OpDispatchBuilder::CalculateFlags_BEXTR(OrderedNode *Src) {
   auto Zero = _Constant(0);
   auto One = _Constant(1);
@@ -1155,34 +1085,12 @@ void OpDispatchBuilder::CalculateFlags_POPCOUNT(OrderedNode *Src) {
 }
 
 void OpDispatchBuilder::CalculateFlags_BZHI(uint8_t SrcSize, OrderedNode *Result, OrderedNode *Src) {
-  // Now for the flags
-
-  auto Bounds = _Constant(SrcSize * 8- 1);
-  auto Zero = _Constant(0);
-  auto One = _Constant(1);
-
-  // OF cleared
-  SetRFLAG<X86State::RFLAG_OF_RAW_LOC>(Zero);
-
   // PF/AF undefined
   _InvalidateFlags((1UL << X86State::RFLAG_PF_RAW_LOC) |
                    (1UL << X86State::RFLAG_AF_RAW_LOC));
 
-  // ZF
-  {
-    auto ZFOp = _Select(IR::COND_EQ,
-                        Result, Zero,
-                        One, Zero);
-    SetRFLAG<X86State::RFLAG_ZF_RAW_LOC>(ZFOp);
-  }
-
-  // CF
-  {
-    auto CFOp = _Select(IR::COND_UGT,
-                        Src, Bounds,
-                        One, Zero);
-    SetRFLAG<X86State::RFLAG_CF_RAW_LOC>(CFOp);
-  }
+  SetNZ_ZeroCV(SrcSize, Result);
+  SetRFLAG<X86State::RFLAG_CF_RAW_LOC>(Src);
 }
 
 void OpDispatchBuilder::CalculateFlags_TZCNT(OrderedNode *Src) {
@@ -1196,12 +1104,10 @@ void OpDispatchBuilder::CalculateFlags_TZCNT(OrderedNode *Src) {
 
   // Set flags
   SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(ZFResult);
-  SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(_Bfe(OpSize::i32Bit, 1, 0, Src));
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(Src, 0, true);
 }
 
 void OpDispatchBuilder::CalculateFlags_LZCNT(uint8_t SrcSize, OrderedNode *Src) {
-  const auto OpSize = SrcSize == 8 ? OpSize::i64Bit : OpSize::i32Bit;
-
   // OF, SF, AF, PF all undefined
   ZeroNZCV();
 
@@ -1212,22 +1118,7 @@ void OpDispatchBuilder::CalculateFlags_LZCNT(uint8_t SrcSize, OrderedNode *Src)
 
   // Set flags
   SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(ZFResult);
-  SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(_Bfe(OpSize, 1, SrcSize * 8 - 1, Src));
-}
-
-void OpDispatchBuilder::CalculateFlags_BITSELECT(OrderedNode *Src) {
-  // OF, SF, AF, PF, CF all undefined
-  ZeroNZCV();
-
-  auto ZeroConst = _Constant(0);
-  auto OneConst = _Constant(1);
-
-  // ZF is set to 1 if the source was zero
-  auto ZFSelectOp = _Select(FEXCore::IR::COND_EQ,
-      Src, ZeroConst,
-      OneConst, ZeroConst);
-
-  SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(ZFSelectOp);
+  SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(Src, SrcSize * 8 - 1, true);
 }
 
 void OpDispatchBuilder::CalculateFlags_RDRAND(OrderedNode *Src) {

FEXCore/Source/Interface/Core/OpcodeDispatcher/Vector.cpp

@@ -225,9 +225,7 @@ void OpDispatchBuilder::VectorALUOpImpl(OpcodeArgs, IROps IROp, size_t ElementSi
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
 
-  auto ALUOp = _VAdd(Size, ElementSize, Dest, Src);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
+  DeriveOp(ALUOp, IROp, _VAdd(Size, ElementSize, Dest, Src));
 
   StoreResult(FPRClass, Op, ALUOp, -1);
 }
@@ -371,9 +369,7 @@ void OpDispatchBuilder::AVXVectorALUOpImpl(OpcodeArgs, IROps IROp, size_t Elemen
   OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
   OrderedNode *Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
 
-  auto ALUOp = _VAdd(Size, ElementSize, Src1, Src2);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
+  DeriveOp(ALUOp, IROp, _VAdd(Size, ElementSize, Src1, Src2));
 
   StoreResult(FPRClass, Op, ALUOp, -1);
 }
@@ -506,9 +502,7 @@ void OpDispatchBuilder::VectorALUROpImpl(OpcodeArgs, IROps IROp, size_t ElementS
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags);
 
-  auto ALUOp = _VAdd(Size, ElementSize, Src, Dest);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
+  DeriveOp(ALUOp, IROp, _VAdd(Size, ElementSize, Src, Dest));
 
   StoreResult(FPRClass, Op, ALUOp, -1);
 }
@@ -540,10 +534,8 @@ OrderedNode* OpDispatchBuilder::VectorScalarInsertALUOpImpl(OpcodeArgs, IROps IR
                                             {.AllowUpperGarbage = true});
 
   // If OpSize == ElementSize then it only does the lower scalar op
-  auto ALUOp = _VFAddScalarInsert(IR::SizeToOpSize(DstSize), ElementSize, Src1, Src2, ZeroUpperBits);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
-
+  DeriveOp(ALUOp, IROp,
+           _VFAddScalarInsert(IR::SizeToOpSize(DstSize), ElementSize, Src1, Src2, ZeroUpperBits));
   return ALUOp;
 }
 
@@ -626,10 +618,7 @@ OrderedNode* OpDispatchBuilder::VectorScalarUnaryInsertALUOpImpl(OpcodeArgs, IRO
                                             {.AllowUpperGarbage = true});
 
   // If OpSize == ElementSize then it only does the lower scalar op
-  auto ALUOp = _VFSqrtScalarInsert(IR::SizeToOpSize(DstSize), ElementSize, Src1, Src2, ZeroUpperBits);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
-
+  DeriveOp(ALUOp, IROp, _VFSqrtScalarInsert(IR::SizeToOpSize(DstSize), ElementSize, Src1, Src2, ZeroUpperBits));
   return ALUOp;
 }
 
@@ -940,9 +929,7 @@ void OpDispatchBuilder::VectorUnaryOpImpl(OpcodeArgs, IROps IROp, size_t Element
 
   OrderedNode *Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
 
-  auto ALUOp = _VFSqrt(OpSize, ElementSize, Src);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
+  DeriveOp(ALUOp, IROp, _VFSqrt(OpSize, ElementSize, Src));
 
   StoreResult(FPRClass, Op, ALUOp, -1);
 }
@@ -979,9 +966,7 @@ void OpDispatchBuilder::AVXVectorUnaryOpImpl(OpcodeArgs, IROps IROp, size_t Elem
 
   OrderedNode *Src = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
 
-  auto ALUOp = _VFSqrt(OpSize, ElementSize, Src);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
+  DeriveOp(ALUOp, IROp, _VFSqrt(OpSize, ElementSize, Src));
 
   // NOTE: We don't need to clear the upper lanes here, since the
   //       IR ops make use of 128-bit AdvSimd for 128-bit cases,
@@ -1017,9 +1002,7 @@ void OpDispatchBuilder::VectorUnaryDuplicateOpImpl(OpcodeArgs, IROps IROp, size_
 
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
 
-  auto ALUOp = _VFSqrt(ElementSize, ElementSize, Src);
-  // Overwrite our IR's op type
-  ALUOp.first->Header.Op = IROp;
+  DeriveOp(ALUOp, IROp, _VFSqrt(ElementSize, ElementSize, Src));
 
   // Duplicate the lower bits
   auto Result = _VDupElement(Size, ElementSize, ALUOp, 0);
@@ -1746,8 +1729,7 @@ void OpDispatchBuilder::VHADDPOp(OpcodeArgs) {
   OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags);
   OrderedNode *Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags);
 
-  auto Res = _VFAddP(SrcSize, ElementSize, Src1, Src2);
-  Res.first->Header.Op = IROp;
+  DeriveOp(Res, IROp, _VFAddP(SrcSize, ElementSize, Src1, Src2));
 
   OrderedNode *Dest = Res;
    if (Is256Bit) {
@@ -2439,8 +2421,7 @@ void OpDispatchBuilder::AVXVariableShiftImpl(OpcodeArgs, IROps IROp) {
   OrderedNode *Vector = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], DstSize, Op->Flags);
   OrderedNode *ShiftVector = LoadSource_WithOpSize(FPRClass, Op, Op->Src[1], DstSize, Op->Flags);
 
-  auto Shift = _VUShr(DstSize, SrcSize, Vector, ShiftVector, true);
-  Shift.first->Header.Op = IROp;
+  DeriveOp(Shift, IROp, _VUShr(DstSize, SrcSize, Vector, ShiftVector, true));
 
   StoreResult(FPRClass, Op, Shift, -1);
 }
@@ -3441,20 +3422,21 @@ void OpDispatchBuilder::VPALIGNROp(OpcodeArgs) {
 
 template<size_t ElementSize>
 void OpDispatchBuilder::UCOMISxOp(OpcodeArgs) {
+  InvalidateDeferredFlags();
+
   const auto SrcSize = Op->Src[0].IsGPR() ? GetGuestVectorLength() : GetSrcSize(Op);
   OrderedNode *Src1 = LoadSource_WithOpSize(FPRClass, Op, Op->Dest, GetGuestVectorLength(), Op->Flags);
   OrderedNode *Src2 = LoadSource_WithOpSize(FPRClass, Op, Op->Src[0], SrcSize, Op->Flags);
-  OrderedNode *Res = _FCmp(ElementSize, Src1, Src2,
-    (1 << FCMP_FLAG_EQ) |
-    (1 << FCMP_FLAG_LT) |
-    (1 << FCMP_FLAG_UNORDERED));
 
-  GenerateFlags_FCMP(Op, Res, Src1, Src2);
+  CachedNZCV = nullptr;
+  _FCmp(ElementSize, Src1, Src2);
+  PossiblySetNZCVBits = ~0;
+  ConvertNZCVToSSE();
 
-  flagsOp = SelectionFlag::FCMP;
-  flagsOpDest = Src1;
-  flagsOpSrc = Src2;
-  flagsOpSize = GetSrcSize(Op);
+  // Zero AF. Note that the comparison sets the raw PF to 0/1 above, so PF[4] is
+  // 0 so the XOR with PF will have no effect, so setting the AF byte to zero
+  // will indeed zero AF as intended.
+  SetRFLAG<FEXCore::X86State::RFLAG_AF_RAW_LOC>(_Constant(0));
 }
 
 template

FEXCore/Source/Interface/Core/OpcodeDispatcher/X87.cpp

@@ -247,10 +247,13 @@ void OpDispatchBuilder::FILD(OpcodeArgs) {
     data = _Sbfe(OpSize::i64Bit, read_width * 8, 0, data);
   }
 
-  // Extract sign and make interger absolute
-  auto sign = _Select(COND_SLT, data, zero, _Constant(0x8000), zero);
+  // We're about to clobber flags to grab the sign, so save NZCV.
+  SaveNZCV();
 
-  auto absolute = _Abs(OpSize::i64Bit, data);
+  // Extract sign and make interger absolute
+  _SubNZCV(OpSize::i64Bit, data, zero);
+  auto sign = _NZCVSelect(OpSize::i64Bit, CondClassType{COND_SLT}, _Constant(0x8000), zero);
+  auto absolute = _Neg(OpSize::i64Bit, data, CondClassType{COND_MI});
 
   // left justify the absolute interger
   auto shift = _Sub(OpSize::i64Bit, _Constant(63), _FindMSB(IR::OpSize::i64Bit, absolute));
@@ -856,9 +859,7 @@ void OpDispatchBuilder::X87UnaryOp(OpcodeArgs) {
   auto top = GetX87Top();
   auto a = _LoadContextIndexed(top, 16, MMBaseOffset(), 16, FPRClass);
 
-  auto result = _F80Round(a);
-  // Overwrite the op
-  result.first->Header.Op = IROp;
+  DeriveOp(result, IROp, _F80Round(a));
 
   if constexpr (IROp == IR::OP_F80SIN ||
                 IROp == IR::OP_F80COS) {
@@ -889,9 +890,7 @@ void OpDispatchBuilder::X87BinaryOp(OpcodeArgs) {
   auto a = _LoadContextIndexed(top, 16, MMBaseOffset(), 16, FPRClass);
   st1 = _LoadContextIndexed(st1, 16, MMBaseOffset(), 16, FPRClass);
 
-  auto result = _F80Add(a, st1);
-  // Overwrite the op
-  result.first->Header.Op = IROp;
+  DeriveOp(result, IROp, _F80Add(a, st1));
 
   if constexpr (IROp == IR::OP_F80FPREM ||
     IROp == IR::OP_F80FPREM1) {

FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp

@@ -601,21 +601,13 @@ void OpDispatchBuilder::FTSTF64(OpcodeArgs) {
   auto low = _Constant(0);
   OrderedNode *data = _VCastFromGPR(8, 8, low);
 
-  OrderedNode *Res = _FCmp(8, a, data,
-    (1 << FCMP_FLAG_EQ) |
-    (1 << FCMP_FLAG_LT) |
-    (1 << FCMP_FLAG_UNORDERED));
+  // We are going to clobber NZCV, make sure it's in a GPR first.
+  GetNZCV();
 
-  OrderedNode *HostFlag_CF = _GetHostFlag(Res, FCMP_FLAG_LT);
-  OrderedNode *HostFlag_ZF = _GetHostFlag(Res, FCMP_FLAG_EQ);
-  OrderedNode *HostFlag_Unordered  = _GetHostFlag(Res, FCMP_FLAG_UNORDERED);
-  HostFlag_CF = _Or(OpSize::i32Bit, HostFlag_CF, HostFlag_Unordered);
-  HostFlag_ZF = _Or(OpSize::i32Bit, HostFlag_ZF, HostFlag_Unordered);
-
-  SetRFLAG<FEXCore::X86State::X87FLAG_C0_LOC>(HostFlag_CF);
-  SetRFLAG<FEXCore::X86State::X87FLAG_C1_LOC>(_Constant(0));
-  SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(HostFlag_Unordered);
-  SetRFLAG<FEXCore::X86State::X87FLAG_C3_LOC>(HostFlag_ZF);
+  // Now we do our comparison.
+  _FCmp(8, a, data);
+  PossiblySetNZCVBits = ~0;
+  ConvertNZCVToX87();
 }
 
 //TODO: This should obey rounding mode
@@ -681,36 +673,22 @@ void OpDispatchBuilder::FCOMIF64(OpcodeArgs) {
 
   auto a = _LoadContextIndexed(top, 8, MMBaseOffset(), 16, FPRClass);
 
-  OrderedNode *Res = _FCmp(8, a, b,
-    (1 << FCMP_FLAG_EQ) |
-    (1 << FCMP_FLAG_LT) |
-    (1 << FCMP_FLAG_UNORDERED));
-
-  OrderedNode *HostFlag_CF = _GetHostFlag(Res, FCMP_FLAG_LT);
-  OrderedNode *HostFlag_ZF = _GetHostFlag(Res, FCMP_FLAG_EQ);
-  OrderedNode *HostFlag_Unordered  = _GetHostFlag(Res, FCMP_FLAG_UNORDERED);
-
-  HostFlag_CF = _Or(OpSize::i32Bit, HostFlag_CF, HostFlag_Unordered);
-  HostFlag_ZF = _Or(OpSize::i32Bit, HostFlag_ZF, HostFlag_Unordered);
-
   if constexpr (whichflags == FCOMIFlags::FLAGS_X87) {
-    SetRFLAG<FEXCore::X86State::X87FLAG_C0_LOC>(HostFlag_CF);
-    SetRFLAG<FEXCore::X86State::X87FLAG_C1_LOC>(_Constant(0));
-    SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(HostFlag_Unordered);
-    SetRFLAG<FEXCore::X86State::X87FLAG_C3_LOC>(HostFlag_ZF);
+    // We are going to clobber NZCV, make sure it's in a GPR first.
+    GetNZCV();
+
+    _FCmp(8, a, b);
+    PossiblySetNZCVBits = ~0;
+    ConvertNZCVToX87();
   }
   else {
     // Invalidate deferred flags early
     // OF, SF, AF, PF all undefined
     InvalidateDeferredFlags();
 
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_RAW_LOC>(HostFlag_CF);
-    SetRFLAG<FEXCore::X86State::RFLAG_ZF_RAW_LOC>(HostFlag_ZF);
-
-    // PF is stored inverted, so invert from the host flag.
-    // TODO: This could perhaps be optimized?
-    auto PF = _Xor(OpSize::i32Bit, HostFlag_Unordered, _Constant(1));
-    SetRFLAG<FEXCore::X86State::RFLAG_PF_RAW_LOC>(PF);
+    _FCmp(8, a, b);
+    PossiblySetNZCVBits = ~0;
+    ConvertNZCVToSSE();
   }
 
   if constexpr (poptwice) {
@@ -767,9 +745,7 @@ void OpDispatchBuilder::X87UnaryOpF64(OpcodeArgs) {
   auto top = GetX87Top();
   auto a = _LoadContextIndexed(top, 8, MMBaseOffset(), 16, FPRClass);
 
-  auto result = _F64SIN(a);
-  // Overwrite the op
-  result.first->Header.Op = IROp;
+  DeriveOp(result, IROp, _F64SIN(a));
 
   if constexpr (IROp == IR::OP_F64SIN ||
                 IROp == IR::OP_F64COS) {
@@ -799,9 +775,7 @@ void OpDispatchBuilder::X87BinaryOpF64(OpcodeArgs) {
   auto a = _LoadContextIndexed(top, 8, MMBaseOffset(), 16, FPRClass);
   st1 = _LoadContextIndexed(st1, 8, MMBaseOffset(), 16, FPRClass);
 
-  auto result = _F64ATAN(a, st1);
-  // Overwrite the op
-  result.first->Header.Op = IROp;
+  DeriveOp(result, IROp, _F64ATAN(a, st1));
 
   if constexpr (IROp == IR::OP_F64FPREM ||
     IROp == IR::OP_F64FPREM1) {

FEXCore/Source/Interface/Core/SignalDelegator.cpp

@@ -1,41 +0,0 @@
-// SPDX-License-Identifier: MIT
-#include <FEXCore/Core/SignalDelegator.h>
-#include <FEXCore/Utils/LogManager.h>
-#include <FEXHeaderUtils/Syscalls.h>
-
-#include <unistd.h>
-#include <signal.h>
-
-namespace FEXCore {
-  void SignalDelegator::RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
-    SetHostSignalHandler(Signal, Func, Required);
-    FrontendRegisterHostSignalHandler(Signal, Func, Required);
-  }
-
-  void SignalDelegator::HandleSignal(int Signal, void *Info, void *UContext) {
-    // Let the host take first stab at handling the signal
-    auto Thread = GetTLSThread();
-    HostSignalHandler &Handler = HostHandlers[Signal];
-
-    if (!Thread) {
-      LogMan::Msg::AFmt("[{}] Thread has received a signal and hasn't registered itself with the delegate! Programming error!", FHU::Syscalls::gettid());
-    }
-    else {
-      for (auto &Handler : Handler.Handlers) {
-        if (Handler(Thread, Signal, Info, UContext)) {
-          // If the host handler handled the fault then we can continue now
-          return;
-        }
-      }
-
-      if (Handler.FrontendHandler &&
-          Handler.FrontendHandler(Thread, Signal, Info, UContext)) {
-        return;
-      }
-
-      // Now let the frontend handle the signal
-      // It's clearly a guest signal and this ends up being an OS specific issue
-      HandleGuestSignal(Thread, Signal, Info, UContext);
-    }
-  }
-}

FEXCore/Source/Interface/Core/X86DebugInfo.cpp

@@ -1,84 +0,0 @@
-// SPDX-License-Identifier: MIT
-#ifndef NDEBUG
-
-#include "Interface/Core/X86Tables/X86Tables.h"
-
-#include <FEXCore/Utils/LogManager.h>
-#include <tuple>
-
-namespace FEXCore::X86Tables::X86InstDebugInfo {
-void InstallDebugInfo() {
-  const std::tuple<uint8_t, uint8_t, Flags> BaseOpTable[] = {
-    {0x50, 8, {FLAGS_MEM_ACCESS}},
-    {0x58, 8, {FLAGS_MEM_ACCESS}},
-
-    {0x68, 1, {FLAGS_MEM_ACCESS}},
-    {0x6A, 1, {FLAGS_MEM_ACCESS}},
-
-    {0xAA, 4, {FLAGS_MEM_ACCESS}},
-
-    {0xC8, 1, {FLAGS_MEM_ACCESS}},
-
-    {0xCC, 2, {FLAGS_DEBUG}},
-
-    {0xD7, 1, {FLAGS_MEM_ACCESS}},
-
-    {0xF1, 1, {FLAGS_DEBUG}},
-    {0xF4, 1, {FLAGS_DEBUG}},
-  };
-
-  const std::tuple<uint8_t, uint8_t, Flags> TwoByteOpTable[] = {
-    {0x0B, 1, {FLAGS_DEBUG}},
-    {0x19, 7, {FLAGS_DEBUG}},
-    {0x28, 2, {FLAGS_MEM_ALIGN_16}},
-
-    {0x31, 1, {FLAGS_DEBUG}},
-
-    {0xA2, 1, {FLAGS_DEBUG}},
-    {0xA3, 1, {FLAGS_MEM_ACCESS}},
-    {0xAB, 1, {FLAGS_MEM_ACCESS}},
-    {0xB3, 1, {FLAGS_MEM_ACCESS}},
-    {0xBB, 1, {FLAGS_MEM_ACCESS}},
-
-    {0xFF, 1, {FLAGS_DEBUG}},
-  };
-
-  const std::tuple<uint8_t, uint8_t, Flags> PrimaryGroupOpTable[] = {
-#define OPD(group, prefix, Reg) (((group - FEXCore::X86Tables::TYPE_GROUP_1) << 6) | (prefix) << 3 | (Reg))
-    {OPD(TYPE_GROUP_3, OpToIndex(0xF6), 6), 2, {FLAGS_DIVIDE}},
-    {OPD(TYPE_GROUP_3, OpToIndex(0xF7), 6), 2, {FLAGS_DIVIDE}},
-#undef OPD
-  };
-
-  const std::tuple<uint16_t, uint8_t, Flags> SecondaryExtensionOpTable[] = {
-#define PF_NONE 0
-#define PF_F3 1
-#define PF_66 2
-#define PF_F2 3
-#define OPD(group, prefix, Reg) (((group - FEXCore::X86Tables::TYPE_GROUP_6) << 5) | (prefix) << 3 | (Reg))
-    {OPD(TYPE_GROUP_15, PF_NONE, 2), 1, {FLAGS_DEBUG}},
-    {OPD(TYPE_GROUP_15, PF_NONE, 3), 1, {FLAGS_DEBUG}},
-#undef PF_F3
-#undef PF_66
-#undef PF_F2
-#undef OPD
-  };
-
-  auto GenerateDebugTable = [](auto& FinalTable, auto& LocalTable) {
-    for (auto Op : LocalTable) {
-      auto OpNum = std::get<0>(Op);
-      auto DebugInfo = std::get<2>(Op);
-      for (uint8_t i = 0; i < std::get<1>(Op); ++i) {
-        memcpy(&FinalTable[OpNum+i].DebugInfo, &DebugInfo, sizeof(X86InstDebugInfo::Flags));
-      }
-    }
-  };
-
-  GenerateDebugTable(BaseOps, BaseOpTable);
-  GenerateDebugTable(SecondBaseOps, TwoByteOpTable);
-  GenerateDebugTable(PrimaryInstGroupOps, PrimaryGroupOpTable);
-
-  GenerateDebugTable(SecondInstGroupOps, SecondaryExtensionOpTable);
-}
-}
-#endif

FEXCore/Source/Interface/Core/X86Tables.cpp

@@ -44,11 +44,6 @@ void InitializeVEXTables();
 void InitializeXOPTables();
 void InitializeEVEXTables();
 
-#ifndef NDEBUG
-uint64_t Total{};
-uint64_t NumInsts{};
-#endif
-
 void InitializeInfoTables(Context::OperatingMode Mode) {
   InitializeBaseTables(Mode);
   InitializeSecondaryTables(Mode);
@@ -62,10 +57,6 @@ void InitializeInfoTables(Context::OperatingMode Mode) {
   InitializeVEXTables();
   InitializeXOPTables();
   InitializeEVEXTables();
-
-#ifndef NDEBUG
-  X86InstDebugInfo::InstallDebugInfo();
-#endif
 }
 
 }

FEXCore/Source/Interface/Core/X86Tables/BaseTables.cpp

@@ -100,10 +100,10 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0x6B, 1, X86InstInfo{"IMUL",   TYPE_INST, FLAGS_MODRM | FLAGS_SRC_SEXT ,                                    1, nullptr}},
 
     // This should just throw a GP
-    {0x6C, 1, X86InstInfo{"INSB",   TYPE_INVALID, FLAGS_SUPPORTS_REP, 0, nullptr}},
-    {0x6D, 1, X86InstInfo{"INSW",   TYPE_INVALID, FLAGS_SUPPORTS_REP, 0, nullptr}},
-    {0x6E, 1, X86InstInfo{"OUTS",   TYPE_INVALID, FLAGS_SUPPORTS_REP, 0, nullptr}},
-    {0x6F, 1, X86InstInfo{"OUTS",   TYPE_INVALID, FLAGS_SUPPORTS_REP, 0, nullptr}},
+    {0x6C, 1, X86InstInfo{"INSB",   TYPE_INVALID, FLAGS_NONE, 0, nullptr}},
+    {0x6D, 1, X86InstInfo{"INSW",   TYPE_INVALID, FLAGS_NONE, 0, nullptr}},
+    {0x6E, 1, X86InstInfo{"OUTS",   TYPE_INVALID, FLAGS_NONE, 0, nullptr}},
+    {0x6F, 1, X86InstInfo{"OUTS",   TYPE_INVALID, FLAGS_NONE, 0, nullptr}},
 
     {0x70, 1, X86InstInfo{"JO",     TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_SETS_RIP | FLAGS_SRC_SEXT , 1, nullptr}},
     {0x71, 1, X86InstInfo{"JNO",    TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_SETS_RIP | FLAGS_SRC_SEXT , 1, nullptr}},
@@ -147,19 +147,19 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0x9E, 1, X86InstInfo{"SAHF",   TYPE_INST, FLAGS_NONE,                              0, nullptr}},
     {0x9F, 1, X86InstInfo{"LAHF",   TYPE_INST, FLAGS_NONE,                              0, nullptr}},
 
-    {0xA4, 1, X86InstInfo{"MOVSB",  TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS  | FLAGS_SUPPORTS_REP,                                            0, nullptr}},
-    {0xA5, 1, X86InstInfo{"MOVS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP,                                                            0, nullptr}},
-    {0xA6, 1, X86InstInfo{"CMPSB",  TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS  | FLAGS_SUPPORTS_REP,                                            0, nullptr}},
-    {0xA7, 1, X86InstInfo{"CMPS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP,                                                            0, nullptr}},
+    {0xA4, 1, X86InstInfo{"MOVSB",  TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS,                                            0, nullptr}},
+    {0xA5, 1, X86InstInfo{"MOVS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS,                                                            0, nullptr}},
+    {0xA6, 1, X86InstInfo{"CMPSB",  TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS,                                            0, nullptr}},
+    {0xA7, 1, X86InstInfo{"CMPS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS,                                                            0, nullptr}},
 
     {0xA8, 1, X86InstInfo{"TEST",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_SF_DST_RAX ,                                             1, nullptr}},
     {0xA9, 1, X86InstInfo{"TEST",   TYPE_INST, FLAGS_SF_DST_RAX | FLAGS_SRC_SEXT | FLAGS_DISPLACE_SIZE_DIV_2,                4, nullptr}},
-    {0xAA, 1, X86InstInfo{"STOS",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS  | FLAGS_SUPPORTS_REP | FLAGS_SF_SRC_RAX,                   0, nullptr}},
-    {0xAB, 1, X86InstInfo{"STOS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP | FLAGS_SF_SRC_RAX,                                   0, nullptr}},
-    {0xAC, 1, X86InstInfo{"LODS",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_SF_DST_RAX | FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP,                                                      0, nullptr}},
-    {0xAD, 1, X86InstInfo{"LODS",   TYPE_INST, FLAGS_SF_DST_RAX | FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP,                                                      0, nullptr}},
-    {0xAE, 1, X86InstInfo{"SCAS",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP | FLAGS_SF_SRC_RAX,                                   0, nullptr}},
-    {0xAF, 1, X86InstInfo{"SCAS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_SUPPORTS_REP | FLAGS_SF_SRC_RAX,                                   0, nullptr}},
+    {0xAA, 1, X86InstInfo{"STOS",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS | FLAGS_SF_SRC_RAX,                   0, nullptr}},
+    {0xAB, 1, X86InstInfo{"STOS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_SF_SRC_RAX,                                   0, nullptr}},
+    {0xAC, 1, X86InstInfo{"LODS",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_SF_DST_RAX | FLAGS_DEBUG_MEM_ACCESS,                                                      0, nullptr}},
+    {0xAD, 1, X86InstInfo{"LODS",   TYPE_INST, FLAGS_SF_DST_RAX | FLAGS_DEBUG_MEM_ACCESS,                                                      0, nullptr}},
+    {0xAE, 1, X86InstInfo{"SCAS",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_DEBUG_MEM_ACCESS | FLAGS_SF_SRC_RAX,                                   0, nullptr}},
+    {0xAF, 1, X86InstInfo{"SCAS",   TYPE_INST, FLAGS_DEBUG_MEM_ACCESS | FLAGS_SF_SRC_RAX,                                   0, nullptr}},
 
     {0xB0, 8, X86InstInfo{"MOV",    TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_SF_REX_IN_BYTE ,                                         1, nullptr}},
     {0xB8, 8, X86InstInfo{"MOV",    TYPE_INST, FLAGS_SF_REX_IN_BYTE | FLAGS_DISPLACE_SIZE_DIV_2 | FLAGS_DISPLACE_SIZE_MUL_2, 4, nullptr}},
@@ -169,7 +169,7 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0xC8, 1, X86InstInfo{"ENTER",  TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_DEBUG_MEM_ACCESS ,                                      3, nullptr}},
     {0xC9, 1, X86InstInfo{"LEAVE",  TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_DEBUG_MEM_ACCESS ,                                                0, nullptr}},
     {0xCA, 2, X86InstInfo{"RETF",   TYPE_PRIV, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_SETS_RIP | FLAGS_BLOCK_END,                                                              0, nullptr}},
-    {0xCC, 1, X86InstInfo{"INT3",   TYPE_INST, FLAGS_DEBUG,                                                                                      0, nullptr}},
+    {0xCC, 1, X86InstInfo{"INT3",   TYPE_INST, FLAGS_NONE,                                                                                      0, nullptr}},
     {0xCD, 1, X86InstInfo{"INT",    TYPE_INST, DEFAULT_SYSCALL_FLAGS,                                                                  1, nullptr}},
     {0xCF, 1, X86InstInfo{"IRET",   TYPE_INST, FLAGS_SETS_RIP | FLAGS_BLOCK_END,                                                                                    0, nullptr}},
 
@@ -192,8 +192,8 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0xEC, 2, X86InstInfo{"IN",     TYPE_INVALID, FLAGS_NONE,             0, nullptr}},
     {0xEE, 2, X86InstInfo{"OUT",    TYPE_INVALID, FLAGS_NONE,             0, nullptr}},
 
-    {0xF1, 1, X86InstInfo{"INT1",   TYPE_INST, FLAGS_DEBUG,               0, nullptr}},
-    {0xF4, 1, X86InstInfo{"HLT",    TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,               0, nullptr}},
+    {0xF1, 1, X86InstInfo{"INT1",   TYPE_INST, FLAGS_NONE,               0, nullptr}},
+    {0xF4, 1, X86InstInfo{"HLT",    TYPE_INST, FLAGS_BLOCK_END,               0, nullptr}},
     {0xF5, 1, X86InstInfo{"CMC",    TYPE_INST, FLAGS_NONE,                0, nullptr}},
     {0xF8, 1, X86InstInfo{"CLC",    TYPE_INST, FLAGS_NONE,                0, nullptr}},
     {0xF9, 1, X86InstInfo{"STC",    TYPE_INST, FLAGS_NONE,                0, nullptr}},

FEXCore/Source/Interface/Core/X86Tables/SecondaryGroupTables.cpp

@@ -183,41 +183,41 @@ void InitializeSecondaryGroupTables() {
     {OPD(TYPE_GROUP_9, PF_F2, 7), 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NONE,     0, nullptr}},
 
     // GROUP 10
-    {OPD(TYPE_GROUP_10, PF_NONE, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_NONE, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_NONE, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END, 0, nullptr}},
 
-    {OPD(TYPE_GROUP_10, PF_F3, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F3, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F3, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
 
-    {OPD(TYPE_GROUP_10, PF_66, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_66, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_66, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
 
-    {OPD(TYPE_GROUP_10, PF_F2, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
-    {OPD(TYPE_GROUP_10, PF_F2, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 0), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 1), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 2), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 3), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 4), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 5), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 6), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
+    {OPD(TYPE_GROUP_10, PF_F2, 7), 1, X86InstInfo{"UD1", TYPE_INST, FLAGS_BLOCK_END,   0, nullptr}},
 
     // GROUP 12
     {OPD(TYPE_GROUP_12, PF_NONE, 0), 1, X86InstInfo{"",      TYPE_INVALID, FLAGS_NONE,                                      0, nullptr}},

FEXCore/Source/Interface/Core/X86Tables/SecondaryTables.cpp

@@ -29,7 +29,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x08, 1, X86InstInfo{"INVD",       TYPE_PRIV, FLAGS_NO_OVERLAY,                                                                                    0, nullptr}},
     {0x09, 1, X86InstInfo{"WBINVD",     TYPE_PRIV, FLAGS_NO_OVERLAY,                                                                                    0, nullptr}},
     {0x0A, 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                                 0, nullptr}},
-    {0x0B, 1, X86InstInfo{"UD2",        TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END | FLAGS_NO_OVERLAY,                                                    0, nullptr}},
+    {0x0B, 1, X86InstInfo{"UD2",        TYPE_INST, FLAGS_BLOCK_END | FLAGS_NO_OVERLAY,                                                    0, nullptr}},
     {0x0C, 1, X86InstInfo{"",           TYPE_INVALID, FLAGS_NO_OVERLAY,                                                                                 0, nullptr}},
     {0x0D, 1, X86InstInfo{"",           TYPE_GROUP_P, FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                                                 0, nullptr}},
     {0x0E, 1, X86InstInfo{"FEMMS",      TYPE_INST, FLAGS_NO_OVERLAY,                                                            0, nullptr}},
@@ -44,7 +44,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x16, 1, X86InstInfo{"MOVLHPS",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                                                 0, nullptr}},
     {0x17, 1, X86InstInfo{"MOVHPS",     TYPE_INST, GenFlagsSizes(SIZE_64BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS, 0, nullptr}},
     {0x18, 1, X86InstInfo{"",           TYPE_GROUP_16, FLAGS_NO_OVERLAY,                                                                                      0, nullptr}},
-    {0x19, 7, X86InstInfo{"NOP",        TYPE_INST, FLAGS_DEBUG | FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                                                     0, nullptr}},
+    {0x19, 7, X86InstInfo{"NOP",        TYPE_INST, FLAGS_MODRM | FLAGS_NO_OVERLAY,                                                                                     0, nullptr}},
 
     {0x20, 2, X86InstInfo{"MOV",        TYPE_PRIV, GenFlagsSameSize(SIZE_64BIT) | FLAGS_NO_OVERLAY,                                                     0, nullptr}},
     {0x22, 2, X86InstInfo{"MOV",        TYPE_PRIV, GenFlagsSameSize(SIZE_64BIT) | FLAGS_NO_OVERLAY,                                                     0, nullptr}},
@@ -59,7 +59,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x2F, 1, X86InstInfo{"COMISS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                                                                   0, nullptr}},
 
     {0x30, 1, X86InstInfo{"WRMSR",      TYPE_PRIV, FLAGS_NO_OVERLAY,                                                                             0, nullptr}},
-    {0x31, 1, X86InstInfo{"RDTSC",      TYPE_INST, FLAGS_DEBUG | FLAGS_NO_OVERLAY,                                                               0, nullptr}},
+    {0x31, 1, X86InstInfo{"RDTSC",      TYPE_INST, FLAGS_NO_OVERLAY,                                                               0, nullptr}},
     {0x32, 1, X86InstInfo{"RDMSR",      TYPE_PRIV, FLAGS_NO_OVERLAY,                                                                             0, nullptr}},
     {0x33, 1, X86InstInfo{"RDPMC",      TYPE_PRIV, FLAGS_NO_OVERLAY,                                                                             0, nullptr}},
     {0x34, 1, X86InstInfo{"SYSENTER",   TYPE_PRIV, FLAGS_NO_OVERLAY,                                                                             0, nullptr}},
@@ -166,7 +166,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x9E, 1, X86InstInfo{"SETLE",   TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_NO_OVERLAY,                        0, nullptr}},
     {0x9F, 1, X86InstInfo{"SETNLE",  TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_NO_OVERLAY,                        0, nullptr}},
 
-    {0xA2, 1, X86InstInfo{"CPUID",   TYPE_INST,     FLAGS_DEBUG | FLAGS_SF_SRC_RAX | FLAGS_NO_OVERLAY,                                              0, nullptr}},
+    {0xA2, 1, X86InstInfo{"CPUID",   TYPE_INST,     FLAGS_SF_SRC_RAX | FLAGS_NO_OVERLAY,                                              0, nullptr}},
     {0xA3, 1, X86InstInfo{"BT",      TYPE_INST,     FLAGS_DEBUG_MEM_ACCESS | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_NO_OVERLAY,                     0, nullptr}},
     {0xA4, 1, X86InstInfo{"SHLD",    TYPE_INST,     FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_NO_OVERLAY,                                              1, nullptr}},
     {0xA5, 1, X86InstInfo{"SHLD",    TYPE_INST,     FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_SRC_RCX | FLAGS_NO_OVERLAY,                           0, nullptr}},
@@ -254,7 +254,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0xFC, 1, X86InstInfo{"PADDB",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                                      0, nullptr}},
     {0xFD, 1, X86InstInfo{"PADDW",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                                      0, nullptr}},
     {0xFE, 1, X86InstInfo{"PADDD",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX,                                      0, nullptr}},
-    {0xFF, 1, X86InstInfo{"UD0",      TYPE_INST, FLAGS_DEBUG | FLAGS_BLOCK_END,                                                                                           0, nullptr}},
+    {0xFF, 1, X86InstInfo{"UD0",      TYPE_INST, FLAGS_BLOCK_END,                                                                                           0, nullptr}},
 
     // FEX reserved instructions
     // Unused x86 encoding instruction.

FEXCore/Source/Interface/Core/X86Tables/X86Tables.h

@@ -279,9 +279,15 @@ namespace InstFlags {
 using InstFlagType = uint64_t;
 
 constexpr InstFlagType FLAGS_NONE                  = 0;
-constexpr InstFlagType FLAGS_DEBUG                 = (1ULL << 1);
+// The secondary Opcode Map uses prefix bytes to overlay more instruction
+// But some instructions need to ignore this overlay and consume these prefixes.
+constexpr InstFlagType FLAGS_NO_OVERLAY            = (1ULL << 0);
+// Some instructions partially ignore overlay
+// Ignore OpSize (0x66) in this case
+constexpr InstFlagType FLAGS_NO_OVERLAY66          = (1ULL << 1);
 constexpr InstFlagType FLAGS_DEBUG_MEM_ACCESS      = (1ULL << 2);
-constexpr InstFlagType FLAGS_SUPPORTS_REP          = (1ULL << 3);
+// Only SEXT if the instruction is operating in 64bit operand size
+constexpr InstFlagType FLAGS_SRC_SEXT64BIT         = (1ULL << 3);
 constexpr InstFlagType FLAGS_BLOCK_END             = (1ULL << 4);
 constexpr InstFlagType FLAGS_SETS_RIP              = (1ULL << 5);
 
@@ -331,27 +337,17 @@ constexpr InstFlagType FLAGS_MODRM                 = (1ULL << 16);
   constexpr InstFlagType FLAGS_SF_MOD_MEM_ONLY       = (1ULL << 18);
   constexpr InstFlagType FLAGS_SF_MOD_REG_ONLY       = (1ULL << 19);
 
-// The secondary Opcode Map uses prefix bytes to overlay more instruction
-// But some instructions need to ignore this overlay and consume these prefixes.
-constexpr InstFlagType FLAGS_NO_OVERLAY           = (1ULL << 20);
-// Some instructions partially ignore overlay
-// Ignore OpSize (0x66) in this case
-constexpr InstFlagType FLAGS_NO_OVERLAY66         = (1ULL << 21);
-
 // x87
-constexpr InstFlagType FLAGS_POP                  = (1ULL << 22);
-
-// Only SEXT if the instruction is operating in 64bit operand size
-constexpr InstFlagType FLAGS_SRC_SEXT64BIT        = (1ULL << 23);
+constexpr InstFlagType FLAGS_POP                  = (1ULL << 20);
 
 // Whether or not the instruction has a VEX prefix for the first source operand
-constexpr InstFlagType FLAGS_VEX_1ST_SRC          = (1ULL << 24);
+constexpr InstFlagType FLAGS_VEX_1ST_SRC          = (1ULL << 21);
 // Whether or not the instruction has a VEX prefix for the second source operand
-constexpr InstFlagType FLAGS_VEX_2ND_SRC          = (1ULL << 25);
+constexpr InstFlagType FLAGS_VEX_2ND_SRC          = (1ULL << 22);
 // Whether or not the instruction has a VEX prefix for the destination
-constexpr InstFlagType FLAGS_VEX_DST              = (1ULL << 26);
+constexpr InstFlagType FLAGS_VEX_DST              = (1ULL << 23);
 // Whether or not the instruction has a VSIB byte
-constexpr InstFlagType FLAGS_VEX_VSIB             = (1ULL << 27);
+constexpr InstFlagType FLAGS_VEX_VSIB             = (1ULL << 24);
 
 constexpr InstFlagType FLAGS_SIZE_DST_OFF = 58;
 constexpr InstFlagType FLAGS_SIZE_SRC_OFF = FLAGS_SIZE_DST_OFF + 3;
@@ -419,35 +415,12 @@ constexpr uint8_t OpToIndex(uint8_t Op) {
 using DecodedOp = DecodedInst const*;
 using OpDispatchPtr = void (IR::OpDispatchBuilder::*)(DecodedOp);
 
-#ifndef NDEBUG
-namespace X86InstDebugInfo {
-constexpr uint64_t FLAGS_MEM_ALIGN_4    = (1 << 0);
-constexpr uint64_t FLAGS_MEM_ALIGN_8    = (1 << 1);
-constexpr uint64_t FLAGS_MEM_ALIGN_16   = (1 << 2);
-constexpr uint64_t FLAGS_MEM_ALIGN_SIZE = (1 << 3); // If instruction size changes depending on prefixes
-constexpr uint64_t FLAGS_MEM_ACCESS     = (1 << 4);
-constexpr uint64_t FLAGS_DEBUG          = (1 << 5);
-constexpr uint64_t FLAGS_DIVIDE         = (1 << 6);
-
-
-struct Flags {
-  uint64_t DebugFlags;
-};
-void InstallDebugInfo();
-}
-
-#endif
-
 struct X86InstInfo {
   char const *Name;
   InstType Type;
   InstFlags::InstFlagType Flags; ///< Must be larger than InstFlags enum
   uint8_t MoreBytes;
   OpDispatchPtr OpcodeDispatcher;
-#ifndef NDEBUG
-  X86InstDebugInfo::Flags DebugInfo;
-  uint32_t NumUnitTestsGenerated;
-#endif
 
   bool operator==(const X86InstInfo &b) const {
     if (strcmp(Name, b.Name) != 0 ||
@@ -524,12 +497,6 @@ extern std::array<X86InstInfo, MAX_XOP_GROUP_TABLE_SIZE> XOPTableGroupOps;
 // EVEX
 extern std::array<X86InstInfo, MAX_EVEX_TABLE_SIZE> EVEXTableOps;
 
-
-#ifndef NDEBUG
-extern uint64_t Total;
-extern uint64_t NumInsts;
-#endif
-
 template <typename OpcodeType>
 struct X86TablesInfoStruct {
   OpcodeType first;
@@ -548,11 +515,6 @@ static inline void GenerateTable(X86InstInfo *FinalTable, X86TablesInfoStruct<Op
     for (uint32_t i = 0; i < Op.second; ++i) {
       LOGMAN_THROW_AA_FMT(FinalTable[OpNum + i].Type == TYPE_UNKNOWN, "Duplicate Entry {}->{}", FinalTable[OpNum + i].Name, Info.Name);
       FinalTable[OpNum + i] = Info;
-#ifndef NDEBUG
-      ++Total;
-      if (Info.Type == TYPE_INST)
-        NumInsts++;
-#endif
     }
   }
 };
@@ -570,11 +532,6 @@ static inline void GenerateTableWithCopy(X86InstInfo *FinalTable, X86TablesInfoS
       }
       else {
         FinalTable[OpNum + i] = Info;
-#ifndef NDEBUG
-        ++Total;
-        if (Info.Type == TYPE_INST)
-          NumInsts++;
-#endif
       }
     }
   }
@@ -602,11 +559,6 @@ static inline void GenerateX87Table(X86InstInfo *FinalTable, X86TablesInfoStruct
           }
         }
       }
-#ifndef NDEBUG
-        ++Total;
-        if (Info.Type == TYPE_INST)
-          NumInsts++;
-#endif
     }
   }
 };

FEXCore/Source/Interface/IR/AOTIR.cpp

@@ -363,18 +363,9 @@ namespace FEXCore::IR {
 
       // Insert to caches if we generated IR
       if (GeneratedIR) {
-        if (CTX->GetGdbServerStatus()) {
-          // Add to thread local ir cache
-          Core::LocalIREntry Entry = {StartAddr, Length, decltype(Entry.IR)(IRList), std::move(RAData), decltype(Entry.DebugData)(DebugData)};
-
-          std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
-          Thread->DebugStore.insert({GuestRIP, std::move(Entry)});
-        }
-        else {
-          // If the IR doesn't need to be retained then we can just delete it now
-          delete DebugData;
-          if (IRList->IsCopy()) delete IRList;
-        }
+        // If the IR doesn't need to be retained then we can just delete it now
+        delete DebugData;
+        if (IRList->IsCopy()) delete IRList;
       }
     }
 

FEXCore/Source/Interface/IR/IR.json

@@ -67,8 +67,6 @@
     "constexpr uint8_t COND_SLT  = 11",
     "constexpr uint8_t COND_SGT  = 12",
     "constexpr uint8_t COND_SLE  = 13",
-    "constexpr uint8_t COND_ANDZ  = 14 /* (a & b) == 0 */",
-    "constexpr uint8_t COND_ANDNZ = 15 /* (a & b) != 0 */",
 
     "constexpr uint8_t COND_FLU  = 16 /* float less or unordred */",
     "constexpr uint8_t COND_FGE  = 17 /* float greater or equal */",
@@ -77,6 +75,8 @@
     "constexpr uint8_t COND_FU   = 20 /* float unordred */",
     "constexpr uint8_t COND_FNU  = 21 /* float not unordred */",
 
+    "constexpr uint8_t COND_AL   = 32 /* always */",
+
     "constexpr FEXCore::IR::RegisterClassType GPRClass {0}",
     "constexpr FEXCore::IR::RegisterClassType GPRFixedClass {1}",
     "constexpr FEXCore::IR::RegisterClassType FPRClass {2}",
@@ -264,7 +264,7 @@
         "HasSideEffects": true,
         "RAOverride": "0"
       },
-      "CondJump SSA:$Cmp1, SSA:$Cmp2, SSA:$TrueBlock, SSA:$FalseBlock, CondClass:$Cond{{COND_NEQ}}, u8:$CompareSize{0}": {
+      "CondJump SSA:$Cmp1, SSA:$Cmp2, SSA:$TrueBlock, SSA:$FalseBlock, CondClass:$Cond{{COND_NEQ}}, u8:$CompareSize{0}, i1:$FromNZCV{false}": {
         "HasSideEffects": true,
         "RAOverride": "2",
         "EmitValidation": [
@@ -447,6 +447,17 @@
         ]
       },
 
+      "GPR = LoadNZCV": {
+        "Desc": ["Loads value of NZCV register"],
+        "DestSize": "4"
+      },
+
+      "StoreNZCV GPR:$Value": {
+        "HasSideEffects": true,
+        "Desc": ["Stores value to NZCV register"],
+        "DestSize": "4"
+      },
+
       "GPR = LoadFlag u32:$Flag": {
         "Desc": ["Loads an x86-64 flag from the context object",
                  "Specialized to allow flexible implementation of flag handling"
@@ -854,9 +865,9 @@
         "DestSize": "8"
       },
 
-      "GPR = Neg OpSize:#Size, GPR:$Src": {
-        "Desc": ["Integer negation",
-                 "Dest = -Src",
+      "GPR = Neg OpSize:#Size, GPR:$Src, CondClass:$Cond{{COND_AL}}": {
+        "Desc": ["Integer negation, with optional predication",
+                 "Dest = Cond ? -Src : Src",
                  "Will truncate to 64 or 32bits"
                 ],
         "DestSize": "Size",
@@ -864,17 +875,6 @@
           "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
-      "GPR = Abs OpSize:#Size, GPR:$Src": {
-        "Desc": ["Integer 2's complement absolute value",
-                 "Dest = std::abs(Src)",
-                 "Will truncate to 64 or 32bits"
-                ],
-        "DestSize": "Size",
-        "ImplicitFlagClobber": true,
-        "EmitValidation": [
-          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
-        ]
-      },
       "GPR = Not OpSize:#Size, GPR:$Src": {
         "Desc": ["Integer binary not",
                  "op:",
@@ -953,28 +953,48 @@
           "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
-      "GPR = AddNZCV OpSize:$Size, GPR:$Src1, GPR:$Src2": {
-        "Desc": ["Return NZCV for the sum of two GPRs"],
-        "DestSize": "4",
-        "ImplicitFlagClobber": true,
+      "AddNZCV OpSize:#Size, GPR:$Src1, GPR:$Src2": {
+        "Desc": ["Set NZCV for the sum of two GPRs"],
+        "HasSideEffects": true,
+        "DestSize": "Size",
         "EmitValidation": [
-          "_Size == FEXCore::IR::OpSize::i32Bit || _Size == FEXCore::IR::OpSize::i64Bit"
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
-      "GPR = AdcNZCV OpSize:$Size, GPR:$Src1, GPR:$Src2, GPR:$NZCV": {
-        "Desc": ["Return NZCV for the sum of two GPRs and carry-in given as NZCV"],
-        "DestSize": "4",
-        "ImplicitFlagClobber": true,
+      "CarryInvert": {
+        "Desc": ["Invert carry flag in NZCV"],
+        "HasSideEffects": true
+      },
+      "AXFlag": {
+        "Desc": ["After an FCmp, converts NZCV flags from the Arm format to a mysterious eXternal format"],
+        "HasSideEffects": true
+      },
+      "RmifNZCV GPR:$Src, u8:$Rotate, u8:$Mask": {
+        "Desc": ["Rotate, mask, and insert into NZCV on FlagM platforms"],
+        "HasSideEffects": true
+      },
+      "CondAddNZCV OpSize:#Size, GPR:$Src1, GPR:$Src2, CondClass:$Cond, u8:$FalseNZCV": {
+        "Desc": ["If condition is true, set NZCV per sum of GPRs, else force NZCV to a constant."],
+        "HasSideEffects": true,
+        "DestSize": "Size",
         "EmitValidation": [
-          "_Size == FEXCore::IR::OpSize::i32Bit || _Size == FEXCore::IR::OpSize::i64Bit"
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
-      "GPR = SbbNZCV OpSize:$Size, GPR:$Src1, GPR:$Src2, GPR:$NZCV": {
-        "Desc": ["Return NZCV for the sum of two GPRs and carry-in given as NZCV"],
-        "DestSize": "4",
-        "ImplicitFlagClobber": true,
+      "AdcNZCV OpSize:#Size, GPR:$Src1, GPR:$Src2": {
+        "Desc": ["Set NZCV for the sum of two GPRs and carry-in given as NZCV"],
+        "HasSideEffects": true,
+        "DestSize": "Size",
         "EmitValidation": [
-          "_Size == FEXCore::IR::OpSize::i32Bit || _Size == FEXCore::IR::OpSize::i64Bit"
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
+        ]
+      },
+      "SbbNZCV OpSize:#Size, GPR:$Src1, GPR:$Src2": {
+        "Desc": ["Set NZCV for the difference of two GPRs and carry-in given as NZCV"],
+        "HasSideEffects": true,
+        "DestSize": "Size",
+        "EmitValidation": [
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
       "GPR = Sub OpSize:#Size, GPR:$Src1, GPR:$Src2": {
@@ -996,14 +1016,14 @@
           "_Shift != ShiftType::ROR"
         ]
       },
-      "GPR = SubNZCV OpSize:$Size, GPR:$Src1, GPR:$Src2, u8:$InvertCarry": {
-        "Desc": ["Return NZCV for the difference of two GPRs. ",
-                 "If InvertCarry is nonzero, carry flag uses x86 definition, inverted from arm64.",
+      "SubNZCV OpSize:#Size, GPR:$Src1, GPR:$Src2": {
+        "Desc": ["Set NZCV for the difference of two GPRs. ",
+                 "Carry flag uses arm64 definition, inverted x86.",
                  ""],
-        "DestSize": "4",
-        "ImplicitFlagClobber": true,
+        "DestSize": "Size",
+        "HasSideEffects": true,
         "EmitValidation": [
-          "_Size == FEXCore::IR::OpSize::i32Bit || _Size == FEXCore::IR::OpSize::i64Bit"
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
       "GPR = Or OpSize:#Size, GPR:$Src1, GPR:$Src2": {
@@ -1046,6 +1066,13 @@
           "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
+      "GPR = XorShift OpSize:#Size, GPR:$Src1, GPR:$Src2, ShiftType:$Shift{ShiftType::LSL}, u8:$ShiftAmount{0}": {
+        "Desc": [ "Integer binary exclusive or with shifted register"],
+        "DestSize": "Size",
+        "EmitValidation": [
+          "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
+        ]
+      },
       "GPR = And OpSize:#Size, GPR:$Src1, GPR:$Src2": {
         "Desc": ["Integer binary and"
                 ],
@@ -1061,10 +1088,10 @@
           "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
-      "GPR = TestNZ u8:$Size, GPR:$Src1": {
-        "Desc": ["Return NZCV for a GPR, setting N and Z accordingly and zeroing C and V"],
-        "ImplicitFlagClobber": true,
-        "DestSize": "4"
+      "TestNZ OpSize:#Size, GPR:$Src1, GPR:$Src2": {
+        "Desc": ["Set NZCV for the binary AND of two GPRs, setting N and Z accordingly and zeroing C and V"],
+        "DestSize": "Size",
+        "HasSideEffects": true
       },
       "GPR = Lshl OpSize:#Size, GPR:$Src1, GPR:$Src2": {
         "Desc": ["Integer logical shift left"
@@ -1209,6 +1236,16 @@
           "Size == FEXCore::IR::OpSize::i32Bit || Size == FEXCore::IR::OpSize::i64Bit"
         ]
       },
+      "GPR = NZCVSelect OpSize:#ResultSize, CondClass:$Cond, GPR:$TrueVal, GPR:$FalseVal": {
+        "Desc": ["Select based on value in NZCV flags",
+                 "op:",
+                 "Dest = Cond ? TrueVal : FalseVal"
+                ],
+        "DestSize": "ResultSize",
+        "EmitValidation": [
+          "ResultSize == FEXCore::IR::OpSize::i32Bit || ResultSize == FEXCore::IR::OpSize::i64Bit"
+        ]
+      },
       "GPR = Select OpSize:#ResultSize, OpSize:$CompareSize, CondClass:$Cond, SSA:$Cmp1, SSA:$Cmp2, GPR:$TrueVal, GPR:$FalseVal": {
         "Desc": ["Ternary selection of GPRs",
                  "op:",
@@ -1320,12 +1357,12 @@
         "DestSize": "DestElementSize"
       },
 
-      "GPR = FCmp u8:$ElementSize, FPR:$Scalar1, FPR:$Scalar2, u32:$Flags": {
-        "Desc": ["Does a scalar unordered compare and stores the asked for flags in to a GPR",
+      "FCmp u8:$ElementSize, FPR:$Scalar1, FPR:$Scalar2": {
+        "Desc": ["Does a scalar unordered compare and sets NZCV accordingly.",
+                 "NZCV follows Arm conventions, a separate AXFLAG instruction is required for x86",
                  "Ordering flag result is true if either float input is NaN"
                 ],
-        "ImplicitFlagClobber": true,
-        "DestSize": "4"
+        "HasSideEffects": true
       }
     },
     "VectorScalar": {
@@ -1581,7 +1618,6 @@
         "NumElements": "RegisterSize / ElementSize"
       },
       "FPR = VCMPEQZ u8:#RegisterSize, u8:#ElementSize, FPR:$Vector": {
-        "ImplicitFlagClobber": true,
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
@@ -1590,7 +1626,6 @@
                  "Each element is compared, if the result is true then the resulting element is ~0, else zero",
                  "Compares the vector against zero"
                 ],
-        "ImplicitFlagClobber": true,
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
@@ -1599,7 +1634,6 @@
                  "Each element is compared, if the result is true then the resulting element is ~0, else zero",
                  "Compares the vector against zero"
                 ],
-        "ImplicitFlagClobber": true,
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
@@ -1616,6 +1650,10 @@
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
+      "FPR = VUShraI u8:#RegisterSize, u8:#ElementSize, FPR:$DestVector, FPR:$Vector, u8:$BitShift": {
+        "DestSize": "RegisterSize",
+        "NumElements": "RegisterSize / ElementSize"
+      },
       "FPR = VSShrI u8:#RegisterSize, u8:#ElementSize, FPR:$Vector, u8:$BitShift": {
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
@@ -1848,12 +1886,10 @@
       },
 
       "FPR = VFMin u8:#RegisterSize, u8:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
-        "ImplicitFlagClobber": true,
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
       "FPR = VFMax u8:#RegisterSize, u8:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
-        "ImplicitFlagClobber": true,
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
@@ -1964,8 +2000,7 @@
 
       "FPR = VCMPEQ u8:#RegisterSize, u8:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
         "DestSize": "RegisterSize",
-        "NumElements": "RegisterSize / ElementSize",
-        "ImplicitFlagClobber": true
+        "NumElements": "RegisterSize / ElementSize"
       },
 
       "FPR = VCMPGT u8:#RegisterSize, u8:#ElementSize, FPR:$Vector1, FPR:$Vector2": {
@@ -1973,7 +2008,6 @@
                  "Each element is compared, if the result is true then the resulting element is ~0, else zero"
                 ],
 
-        "ImplicitFlagClobber": true,
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
@@ -2153,6 +2187,14 @@
         "Desc": "Assists in key generation",
         "DestSize": "16"
       },
+      "FPR = VSha1H FPR:$Src": {
+        "Desc": "Does vector scalar SHA1H instruction",
+        "DestSize": "FEXCore::IR::OpSize::i32Bit"
+      },
+      "FPR = VSha256U0 FPR:$Src1, FPR:$Src2": {
+        "Desc": "Does vector scalar VSha256U0 instruction",
+        "DestSize": "FEXCore::IR::OpSize::i128Bit"
+      },
       "GPR = CRC32 GPR:$Src1, GPR:$Src2, u8:$SrcSize": {
         "Desc": ["CRC32 using polynomial 0x1EDC6F41"
                 ],

FEXCore/Source/Interface/IR/IRDumper.cpp

@@ -44,6 +44,11 @@ static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const
 }
 
 static void PrintArg(fextl::stringstream *out, [[maybe_unused]] IRListView const* IR, CondClassType Arg) {
+  if (Arg == COND_AL) {
+    *out << "ALWAYS";
+    return;
+  }
+
   static constexpr std::array<std::string_view, 22> CondNames = {
     "EQ",
     "NEQ",

FEXCore/Source/Interface/IR/Passes/ConstProp.cpp

@@ -194,8 +194,6 @@ public:
 
 private:
   bool HandleConstantPools(IREmitter *IREmit, const IRListView& CurrentIR);
-  void CodeMotionAroundSelects(IREmitter *IREmit, const IRListView& CurrentIR);
-  void FCMPOptimization(IREmitter *IREmit, const IRListView& CurrentIR);
   void LoadMemStoreMemImmediatePooling(IREmitter *IREmit, const IRListView& CurrentIR);
   bool ZextAndMaskingElimination(IREmitter *IREmit, const IRListView& CurrentIR,
       OrderedNode* CodeNode, IROp_Header* IROp);
@@ -267,92 +265,6 @@ bool ConstProp::HandleConstantPools(IREmitter *IREmit, const IRListView& Current
   return Changed;
 }
 
-// Code motion around selects
-// Moves unary ops that depend on a select before the select, if both inputs are constants
-// assumes that unary ops without side effects on constants will be constprop'd
-void ConstProp::CodeMotionAroundSelects(IREmitter *IREmit, const IRListView& CurrentIR) {
-  // Code motion around selects
-  // Moves unary ops that depend on a select before the select, if both inputs are constants
-  // assumes that unary ops without side effects on constants will be constprop'd
-  for (auto [BlockNode, BlockIROp] : CurrentIR.GetBlocks()) {
-    auto BlockOp = BlockIROp->CW<FEXCore::IR::IROp_CodeBlock>();
-    for (auto [UnaryOpNode, UnaryOpHdr] : CurrentIR.GetCode(BlockNode)) {
-      if (IR::GetArgs(UnaryOpHdr->Op) == 1 && !HasSideEffects(UnaryOpHdr->Op)
-                                           && !ImplicitFlagClobber(UnaryOpHdr->Op)) {
-        // could be moved
-        auto SelectOpNode = IREmit->UnwrapNode(UnaryOpHdr->Args[0]);
-        auto SelectOpHdr = IREmit->GetOpHeader(UnaryOpHdr->Args[0]);
-        auto SelectOp = SelectOpHdr->CW<IR::IROp_Select>();
-
-        // the value isn't used after the select otherwise
-        // make sure the sizes match
-        if (SelectOpHdr->Size == UnaryOpHdr->Size && SelectOpHdr->Op == OP_SELECT && SelectOpNode->NumUses == 1
-            && IREmit->IsValueConstant(SelectOp->TrueVal)
-            && IREmit->IsValueConstant(SelectOp->FalseVal)) {
-
-          IREmit->SetWriteCursor(IREmit->UnwrapNode(SelectOpNode->Header.Previous));
-
-          size_t OpSize = FEXCore::IR::GetSize(UnaryOpHdr->Op);
-
-          /// copy for TrueVal ///
-          auto NewUnaryOp1 = IREmit->AllocateRawOp(OpSize);
-
-          // Copy over the op
-          memcpy(NewUnaryOp1.first, UnaryOpHdr, OpSize);
-
-          for (int i = 0; i < IR::GetArgs(NewUnaryOp1.first->Op); i++) {
-            NewUnaryOp1.first->Args[i] = IREmit->WrapNode(IREmit->Invalid());
-          }
-          // Set New Op to operate on the constant
-          IREmit->ReplaceNodeArgument(NewUnaryOp1, 0, IREmit->UnwrapNode(SelectOp->TrueVal));
-          // Make select use the operated constant
-          IREmit->ReplaceNodeArgument(SelectOpNode, 2, NewUnaryOp1);
-
-          /// copy for FalseVal ///
-          auto NewUnaryOp2 = IREmit->AllocateRawOp(OpSize);
-
-          // Copy over the op
-          memcpy(NewUnaryOp2.first, UnaryOpHdr, OpSize);
-
-          for (int i = 0; i < IR::GetArgs(NewUnaryOp2.first->Op); i++) {
-            NewUnaryOp2.first->Args[i] = IREmit->WrapNode(IREmit->Invalid());
-          }
-          // Set New Op to operate on the constant
-          IREmit->ReplaceNodeArgument(NewUnaryOp2, 0, IREmit->UnwrapNode(SelectOp->FalseVal));
-          // Make select use the operated constant
-          IREmit->ReplaceNodeArgument(SelectOpNode, 3, NewUnaryOp2);
-
-          // Replace uses of the defuct unary op w/ select
-          IREmit->ReplaceAllUsesWithRange(UnaryOpNode, SelectOpNode, IREmit->GetIterator(IREmit->WrapNode(UnaryOpNode)), IREmit->GetIterator(BlockOp->Last));
-        }
-      }
-    }
-  }
-}
-
-void ConstProp::FCMPOptimization(IREmitter *IREmit, const IRListView& CurrentIR) {
-  // Make all FCMPs set no flags
-  for (auto [CodeNode, IROp] : CurrentIR.GetAllCode()) {
-    if (IROp->Op == OP_FCMP) {
-      auto fcmp = IROp->CW<IR::IROp_FCmp>();
-      fcmp->Flags = 0;
-    }
-  }
-
-  // Set needed flags
-  for (auto [CodeNode, IROp] : CurrentIR.GetAllCode()) {
-    if (IROp->Op == OP_GETHOSTFLAG) {
-      auto ghf = IROp->CW<IR::IROp_GetHostFlag>();
-
-      auto fcmp = IREmit->GetOpHeader(ghf->Value)->CW<IR::IROp_FCmp>();
-      LOGMAN_THROW_AA_FMT(fcmp->Header.Op == OP_FCMP || fcmp->Header.Op == OP_F80CMP, "Unexpected OP_GETHOSTFLAG source");
-      if(fcmp->Header.Op == OP_FCMP) {
-        fcmp->Flags |= 1 << ghf->Flag;
-      }
-    }
-  }
-}
-
 // LoadMem / StoreMem imm pooling
 // If imms are close by, use address gen to generate the values instead of using a new imm
 void ConstProp::LoadMemStoreMemImmediatePooling(IREmitter *IREmit, const IRListView& CurrentIR) {
@@ -733,6 +645,8 @@ bool ConstProp::ConstantPropagation(IREmitter *IREmit, const IRListView& Current
       }
     break;
     }
+    /* TODO: restore this when we have rmif or something? */
+#if 0
     case OP_TESTNZ: {
       auto Op = IROp->CW<IR::IROp_TestNZ>();
       uint64_t Constant1{};
@@ -747,6 +661,7 @@ bool ConstProp::ConstantPropagation(IREmitter *IREmit, const IRListView& Current
       }
     break;
     }
+#endif
     case OP_OR: {
       auto Op = IROp->CW<IR::IROp_Or>();
       uint64_t Constant1{};
@@ -1005,37 +920,6 @@ bool ConstProp::ConstantPropagation(IREmitter *IREmit, const IRListView& Current
       }
       break;
     }
-    case OP_CONDJUMP: {
-      auto Op = IROp->CW<IR::IROp_CondJump>();
-
-      auto Select = IREmit->GetOpHeader(Op->Header.Args[0]);
-
-      uint64_t Constant;
-      // Fold the select into the CondJump if possible. Could handle more complex cases, too.
-      if (Op->Cond.Val == COND_NEQ && IREmit->IsValueConstant(Op->Cmp2, &Constant) && Constant == 0 &&  Select->Op == OP_SELECT) {
-
-        const auto SelectCmpClass = IREmit->WalkFindRegClass(Select->Args[0]);
-        if (SelectCmpClass == GPRPairClass) {
-          // If the comparison class is a GPRPair then don't fold the select since it isn't free.
-          break;
-        }
-        uint64_t Constant1{};
-        uint64_t Constant2{};
-
-        if (IREmit->IsValueConstant(Select->Args[2], &Constant1) && IREmit->IsValueConstant(Select->Args[3], &Constant2)) {
-          if (Constant1 == 1 && Constant2 == 0) {
-            auto slc = Select->C<IR::IROp_Select>();
-            IREmit->ReplaceNodeArgument(CodeNode, 0, IREmit->UnwrapNode(Select->Args[0]));
-            IREmit->ReplaceNodeArgument(CodeNode, 1, IREmit->UnwrapNode(Select->Args[1]));
-            Op->Cond = slc->Cond;
-            Op->CompareSize = slc->CompareSize;
-            Changed = true;
-          }
-        }
-      }
-      break;
-    }
-
     default:
       break;
     }
@@ -1102,16 +986,54 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
 
         break;
       }
+      case OP_CONDADDNZCV:
+      {
+        auto Op = IROp->C<IR::IROp_CondAddNZCV>();
+
+        uint64_t Constant2{};
+        if (IREmit->IsValueConstant(Op->Header.Args[1], &Constant2)) {
+          if (IsImmAddSub(Constant2)) {
+            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[1]));
+
+            IREmit->ReplaceNodeArgument(CodeNode, 1, CreateInlineConstant(IREmit, Constant2));
+
+            Changed = true;
+          }
+        }
+
+        uint64_t Constant1{};
+        if (IREmit->IsValueConstant(Op->Header.Args[0], &Constant1)) {
+          if (Constant1 == 0) {
+            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[0]));
+            IREmit->ReplaceNodeArgument(CodeNode, 0, CreateInlineConstant(IREmit, 0));
+            Changed = true;
+          }
+        }
+        break;
+      }
+      case OP_TESTNZ:
+      {
+        auto Op = IROp->C<IR::IROp_TestNZ>();
+
+        uint64_t Constant1{};
+        if (IREmit->IsValueConstant(Op->Header.Args[1], &Constant1)) {
+          if (IsImmLogical(Constant1, IROp->Size * 8)) {
+            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[1]));
+
+            IREmit->ReplaceNodeArgument(CodeNode, 1, CreateInlineConstant(IREmit, Constant1));
+
+            Changed = true;
+          }
+        }
+        break;
+      }
       case OP_SELECT:
       {
         auto Op = IROp->C<IR::IROp_Select>();
 
-        bool Bitwise = Op->Cond == COND_ANDZ ||
-                       Op->Cond == COND_ANDNZ;
-
         uint64_t Constant1{};
         if (IREmit->IsValueConstant(Op->Header.Args[1], &Constant1)) {
-          if (Bitwise ? IsImmLogical(Constant1, IROp->Size * 8) : IsImmAddSub(Constant1)) {
+          if (IsImmAddSub(Constant1)) {
             IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[1]));
 
             IREmit->ReplaceNodeArgument(CodeNode, 1, CreateInlineConstant(IREmit, Constant1));
@@ -1142,6 +1064,33 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
 
         break;
       }
+      case OP_NZCVSELECT:
+      {
+        auto Op = IROp->C<IR::IROp_NZCVSelect>();
+
+        uint64_t AllOnes = IROp->Size == 8 ? 0xffff'ffff'ffff'ffffull : 0xffff'ffffull;
+
+        // We always allow source 1 to be zero, but source 0 can only be a
+        // special 1/~0 constant if source 1 is 0.
+        uint64_t Constant0{};
+        uint64_t Constant1{};
+        if (IREmit->IsValueConstant(Op->Header.Args[1], &Constant1) &&
+            Constant1 == 0)
+        {
+          IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[1]));
+          IREmit->ReplaceNodeArgument(CodeNode, 1, CreateInlineConstant(IREmit, Constant1));
+
+          if (IREmit->IsValueConstant(Op->Header.Args[0], &Constant0) &&
+              (Constant0 == 1 || Constant0 == AllOnes))
+          {
+            IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Header.Args[0]));
+
+            IREmit->ReplaceNodeArgument(CodeNode, 0, CreateInlineConstant(IREmit, Constant0));
+          }
+        }
+
+        break;
+      }
       case OP_CONDJUMP:
       {
         auto Op = IROp->C<IR::IROp_CondJump>();
@@ -1269,6 +1218,35 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
         }
         break;
       }
+      case OP_MEMCPY:
+      {
+        auto Op = IROp->CW<IR::IROp_MemCpy>();
+
+        uint64_t Constant{};
+        if (IREmit->IsValueConstant(Op->Direction, &Constant)) {
+          IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Direction));
+
+          IREmit->ReplaceNodeArgument(CodeNode, Op->Direction_Index, CreateInlineConstant(IREmit, Constant & 1));
+
+          Changed = true;
+        }
+        break;
+      }
+      case OP_MEMSET:
+      {
+        auto Op = IROp->CW<IR::IROp_MemSet>();
+
+        uint64_t Constant{};
+        if (IREmit->IsValueConstant(Op->Direction, &Constant)) {
+          IREmit->SetWriteCursor(CurrentIR.GetNode(Op->Direction));
+
+          IREmit->ReplaceNodeArgument(CodeNode, Op->Direction_Index, CreateInlineConstant(IREmit, Constant & 1));
+
+          Changed = true;
+        }
+        break;
+      }
+
       default:
         break;
     }
@@ -1288,8 +1266,6 @@ bool ConstProp::Run(IREmitter *IREmit) {
     Changed = true;
   }
 
-  CodeMotionAroundSelects(IREmit, CurrentIR);
-  FCMPOptimization(IREmit, CurrentIR);
   LoadMemStoreMemImmediatePooling(IREmit, CurrentIR);
 
   for (auto [CodeNode, IROp] : CurrentIR.GetAllCode()) {

FEXCore/Source/Interface/IR/Passes/DeadContextStoreElimination.cpp

@@ -277,6 +277,24 @@ namespace {
       });
     }
 
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, pf_raw),
+        sizeof(FEXCore::Core::CPUState::pf_raw),
+      },
+      LastAccessType::NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, af_raw),
+        sizeof(FEXCore::Core::CPUState::af_raw),
+      },
+      LastAccessType::NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_MMS; ++i) {
       ContextClassification->emplace_back(ContextMemberInfo{
         ContextMemberClassification {
@@ -419,6 +437,10 @@ namespace {
       SetAccess(Offset++, LastAccessType::NONE);
     }
 
+    // PF/AF
+    SetAccess(Offset++, LastAccessType::NONE);
+    SetAccess(Offset++, LastAccessType::NONE);
+
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_MMS; ++i) {
       SetAccess(Offset++, LastAccessType::NONE);
     }

FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.cpp

@@ -6,6 +6,7 @@ $end_info$
 */
 
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
+#include "FEXCore/Core/X86Enums.h"
 #include "Interface/IR/Passes.h"
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IR.h>
@@ -518,12 +519,21 @@ namespace {
     const auto GetRegAndClassFromOffset = [&, this](uint32_t Offset) {
         const auto beginGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[0]);
         const auto endGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[16]);
+        const auto pf = offsetof(FEXCore::Core::CpuStateFrame, State.pf_raw);
+        const auto af = offsetof(FEXCore::Core::CpuStateFrame, State.af_raw);
 
         const auto [beginFpr, endFpr] = GetFPRBeginAndEnd();
 
-        LOGMAN_THROW_AA_FMT((Offset >= beginGpr && Offset < endGpr) || (Offset >= beginFpr && Offset < endFpr), "Unexpected Offset {}", Offset);
+        LOGMAN_THROW_AA_FMT((Offset >= beginGpr && Offset < endGpr) || (Offset >= beginFpr && Offset < endFpr) || (Offset == pf) || (Offset == af), "Unexpected Offset {}", Offset);
 
-        if (Offset >= beginGpr && Offset < endGpr) {
+        unsigned FlagOffset =
+          Graph->Set.Classes[GPRFixedClass.Val].PhysicalCount - 2;
+
+        if (Offset == pf) {
+          return PhysicalRegister(GPRFixedClass, FlagOffset);
+        } else if (Offset == af) {
+          return PhysicalRegister(GPRFixedClass, FlagOffset + 1);
+        } else if (Offset >= beginGpr && Offset < endGpr) {
           auto reg = (Offset - beginGpr) / Core::CPUState::GPR_REG_SIZE;
           return PhysicalRegister(GPRFixedClass, reg);
         } else if (Offset >= beginFpr && Offset < endFpr) {
@@ -544,12 +554,21 @@ namespace {
     const auto GetStaticMapFromOffset = [&](uint32_t Offset) -> LiveRange** {
         const auto beginGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[0]);
         const auto endGpr = offsetof(FEXCore::Core::CpuStateFrame, State.gregs[16]);
+        const auto pf = offsetof(FEXCore::Core::CpuStateFrame, State.pf_raw);
+        const auto af = offsetof(FEXCore::Core::CpuStateFrame, State.af_raw);
 
         const auto [beginFpr, endFpr] = GetFPRBeginAndEnd();
 
-        LOGMAN_THROW_AA_FMT((Offset >= beginGpr && Offset < endGpr) || (Offset >= beginFpr && Offset < endFpr), "Unexpected Offset {}", Offset);
+        LOGMAN_THROW_AA_FMT((Offset >= beginGpr && Offset < endGpr) || (Offset >= beginFpr && Offset < endFpr) || (Offset == pf) || (Offset == af), "Unexpected Offset {}", Offset);
 
-        if (Offset >= beginGpr && Offset < endGpr) {
+        unsigned FlagOffset =
+          Graph->Set.Classes[GPRFixedClass.Val].PhysicalCount - 2;
+
+        if (Offset == pf) {
+          return &StaticMaps[FlagOffset];
+        } else if (Offset == af) {
+          return &StaticMaps[FlagOffset + 1];
+        } else if (Offset >= beginGpr && Offset < endGpr) {
           auto reg = (Offset - beginGpr) / Core::CPUState::GPR_REG_SIZE;
           return &StaticMaps[reg];
         } else if (Offset >= beginFpr && Offset < endFpr) {

FEXCore/Source/Utils/Allocator/64BitAllocator.cpp

@@ -5,8 +5,8 @@
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/Utils/SignalScopeGuards.h>
 #include <FEXCore/fextl/sstream.h>
-#include <FEXCore/Utils/DeferredSignalMutex.h>
 #include <FEXHeaderUtils/Syscalls.h>
 #include <FEXHeaderUtils/TypeDefines.h>
 #include <FEXCore/fextl/memory.h>
@@ -272,7 +272,7 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
   size_t NumberOfPages = length / FHU::FEX_PAGE_SIZE;
 
   // This needs a mutex to be thread safe
-  FEXCore::ScopedPotentialDeferredSignalWithForkableMutex lk(AllocationMutex, TLSThread);
+  auto lk = FEXCore::GuardSignalDeferringSectionWithFallback(AllocationMutex, TLSThread);
 
   uint64_t AllocatedOffset{};
   LiveVMARegion *LiveRegion{};
@@ -460,7 +460,7 @@ int OSAllocator_64Bit::Munmap(void *addr, size_t length) {
   }
 
   // This needs a mutex to be thread safe
-  FEXCore::ScopedPotentialDeferredSignalWithForkableMutex lk(AllocationMutex, TLSThread);
+  auto lk = FEXCore::GuardSignalDeferringSectionWithFallback(AllocationMutex, TLSThread);
 
   length = FEXCore::AlignUp(length, FHU::FEX_PAGE_SIZE);
 
@@ -585,7 +585,7 @@ OSAllocator_64Bit::OSAllocator_64Bit() {
 
 OSAllocator_64Bit::~OSAllocator_64Bit() {
   // This needs a mutex to be thread safe
-  FEXCore::ScopedPotentialDeferredSignalWithForkableMutex lk(AllocationMutex, TLSThread);
+  auto lk = FEXCore::GuardSignalDeferringSectionWithFallback(AllocationMutex, TLSThread);
 
   // Walk the pages and deallocate
   // First walk the live regions

FEXCore/include/FEXCore/Core/CPUBackend.h

@@ -155,13 +155,6 @@ namespace CPU {
      */
     [[nodiscard]] virtual void *MapRegion(void *HostPtr, uint64_t GuestPtr, uint64_t Size) = 0;
 
-    /**
-     * @brief This is post-setup initialization that is called just before code executino
-     *
-     * Guest memory is available at this point and ThreadState is valid
-     */
-    virtual void Initialize() {}
-
     /**
      * @brief Lets FEXCore know if this CPUBackend needs IR and DebugData for CompileCode
      *

FEXCore/include/FEXCore/Core/Context.h

@@ -116,16 +116,6 @@ namespace FEXCore::Context {
        */
       FEX_DEFAULT_VISIBILITY static fextl::unique_ptr<FEXCore::Context::Context> CreateNewContext();
 
-      /**
-       * @brief Post creation context initialization
-       * Once configurations have been set, do the post-creation initialization with that configuration
-       *
-       * @param CTX The context that we created
-       *
-       * @return true if we managed to initialize correctly
-       */
-      FEX_DEFAULT_VISIBILITY virtual bool InitializeContext() = 0;
-
       /**
        * @brief Allows setting up in memory code and other things prior to launchign code execution
        *
@@ -201,25 +191,6 @@ namespace FEXCore::Context {
       FEX_DEFAULT_VISIBILITY virtual void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) = 0;
       FEX_DEFAULT_VISIBILITY virtual void CompileRIPCount(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, uint64_t MaxInst) = 0;
 
-      /**
-       * @brief Gets the program exit status
-       *
-       *
-       * @param CTX The context that we created
-       *
-       * @return The program exit status
-       */
-      FEX_DEFAULT_VISIBILITY virtual int GetProgramStatus() const = 0;
-
-      /**
-       * @brief [[threadsafe]] Returns the ExitReason of the parent thread. Typically used for async result status
-       *
-       * @param CTX The context that we created
-       *
-       * @return The ExitReason for the parentthread
-       */
-      FEX_DEFAULT_VISIBILITY virtual ExitReason GetExitReason() = 0;
-
       /**
        * @brief [[theadsafe]] Checks if the Context is either done working or paused(in the case of single stepping)
        *
@@ -231,22 +202,6 @@ namespace FEXCore::Context {
        */
       FEX_DEFAULT_VISIBILITY virtual bool IsDone() const = 0;
 
-      /**
-       * @brief Gets a copy the CPUState of the parent thread
-       *
-       * @param CTX The context that we created
-       * @param State The state object to populate
-       */
-      FEX_DEFAULT_VISIBILITY virtual void GetCPUState(FEXCore::Core::CPUState *State) const = 0;
-
-      /**
-       * @brief Copies the CPUState provided to the parent thread
-       *
-       * @param CTX The context that we created
-       * @param State The satate object to copy from
-       */
-      FEX_DEFAULT_VISIBILITY virtual void SetCPUState(const FEXCore::Core::CPUState *State) = 0;
-
       /**
        * @brief Allows the frontend to pass in a custom CPUBackend creation factory
        *
@@ -270,12 +225,36 @@ namespace FEXCore::Context {
       ///< State reconstruction helpers
       ///< Reconstructs the guest RIP from the passed in thread context and related Host PC.
       FEX_DEFAULT_VISIBILITY virtual uint64_t RestoreRIPFromHostPC(FEXCore::Core::InternalThreadState *Thread, uint64_t HostPC) = 0;
-      ///< Reconstructs a compacted EFLAGS from FEX's internal EFLAG representation.
-      FEX_DEFAULT_VISIBILITY virtual uint32_t ReconstructCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread) = 0;
+      /**
+       * @brief Reconstructs a compacted EFLAGS from FEX's internal EFLAG representation.
+       *
+       * @param Thread The thread getting the state reconstructed
+       * @param WasInJIT If the code was in the JIT at the time.
+       * @param HostGPRs The host Arm64 GPRs at the point of state inside the JIT.
+       * @param PSTATE The Arm64 PState value.
+       *
+       * If WasInJIT is false then HostGPRs and PSTATE is ignored, with the assumption that the FEX JIT has already stored all state in to the
+       * ThreadState object.
+       *
+       * @return x86 EFLAGS reconstructed
+       */
+      FEX_DEFAULT_VISIBILITY virtual uint32_t ReconstructCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread, bool WasInJIT, uint64_t *HostGPRs, uint64_t PSTATE) = 0;
       ///< Sets FEX's internal EFLAGS representation to the passed in compacted form.
       FEX_DEFAULT_VISIBILITY virtual void SetFlagsFromCompactedEFLAGS(FEXCore::Core::InternalThreadState *Thread, uint32_t EFLAGS) = 0;
 
-      FEX_DEFAULT_VISIBILITY virtual FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) = 0;
+      /**
+       * @brief Create a new thread object that doesn't inherit any state.
+       * Used to create FEX thread objects in preparation for creating a true OS thread.
+       *
+       * @param InitialRIP The starting RIP of this thread
+       * @param StackPointer The starting RSP of this thread
+       * @param NewThreadState The thread state to inherit from if not nullptr.
+       * @param ParentTID The thread ID that the parent is inheriting from
+       *
+       * @return A new InternalThreadState object for using with a new guest thread.
+       */
+      FEX_DEFAULT_VISIBILITY virtual FEXCore::Core::InternalThreadState* CreateThread(uint64_t InitialRIP, uint64_t StackPointer, FEXCore::Core::CPUState *NewThreadState = nullptr, uint64_t ParentTID = 0) = 0;
+
       FEX_DEFAULT_VISIBILITY virtual void ExecutionThread(FEXCore::Core::InternalThreadState *Thread) = 0;
       FEX_DEFAULT_VISIBILITY virtual void InitializeThread(FEXCore::Core::InternalThreadState *Thread) = 0;
       FEX_DEFAULT_VISIBILITY virtual void RunThread(FEXCore::Core::InternalThreadState *Thread) = 0;

FEXCore/include/FEXCore/Core/CoreState.h

@@ -72,7 +72,7 @@ namespace FEXCore::Core {
   static_assert(std::is_trivially_copyable_v<NonAtomicRefCounter<uint64_t>>, "needs to be trivially copyable");
   static_assert(sizeof(NonAtomicRefCounter<uint64_t>) == sizeof(uint64_t), "Needs to be correct size");
 
-  struct FEX_PACKED CPUState {
+  struct CPUState {
     // Allows more efficient handling of the register
     // file in the event AVX is not supported.
     union XMMRegs {
@@ -102,6 +102,8 @@ namespace FEXCore::Core {
     uint64_t InlineJITBlockHeader{};
     XMMRegs xmm{};
     uint8_t flags[48]{};
+    uint64_t pf_raw{};
+    uint64_t af_raw{};
     uint64_t mm[8][2]{};
 
     // 32bit x86 state
@@ -335,7 +337,4 @@ namespace FEXCore::Core {
   static_assert(sizeof(CpuStateFrame::SynchronousFaultData) == 8, "This needs to be 8 bytes");
   static_assert(std::alignment_of_v<CpuStateFrame::SynchronousFaultDataStruct> == 8, "This needs to be 8 bytes");
   static_assert(offsetof(CpuStateFrame, SynchronousFaultData) % 8 == 0, "This needs to be aligned");
-
-  FEX_DEFAULT_VISIBILITY std::string_view const& GetFlagName(unsigned Flag);
-  FEX_DEFAULT_VISIBILITY std::string_view const& GetGRegName(unsigned Reg);
 }

FEXCore/include/FEXCore/Core/SignalDelegator.h

@@ -35,8 +35,6 @@ namespace Core {
 #endif
   };
 }
-  using HostSignalDelegatorFunction = std::function<bool(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext)>;
-
   class SignalDelegator {
   public:
     virtual ~SignalDelegator() = default;
@@ -49,16 +47,6 @@ namespace Core {
     virtual void RegisterTLSState(FEXCore::Core::InternalThreadState *Thread) = 0;
     virtual void UninstallTLSState(FEXCore::Core::InternalThreadState *Thread) = 0;
 
-    /**
-     * @brief Registers a signal handler for the host to handle a signal
-     *
-     * It's a process level signal handler so one must be careful
-     */
-    void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
-
-    // Called from the thunk handler to handle the signal
-    void HandleSignal(int Signal, void *Info, void *UContext);
-
     /**
      * @brief Check to ensure the XID handler is still set to the FEX handler
      *
@@ -67,12 +55,6 @@ namespace Core {
      */
     virtual void CheckXIDHandler() = 0;
 
-    constexpr static size_t MAX_SIGNALS {64};
-
-    // Use the last signal just so we are less likely to ever conflict with something that the guest application is using
-    // 64 is used internally by Valgrind
-    constexpr static size_t SIGNAL_FOR_PAUSE {63};
-
     struct SignalDelegatorConfig {
       bool StaticRegisterAllocation{};
       bool SupportsAVX{};
@@ -124,31 +106,5 @@ namespace Core {
 
   protected:
     SignalDelegatorConfig Config;
-
-    virtual FEXCore::Core::InternalThreadState *GetTLSThread() = 0;
-    virtual void HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) = 0;
-
-    /**
-     * @brief Registers a signal handler for the host to handle a signal
-     *
-     * It's a process level signal handler so one must be careful
-     */
-    virtual void FrontendRegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) = 0;
-    virtual void FrontendRegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) = 0;
-
-  private:
-    struct HostSignalHandler {
-      fextl::vector<FEXCore::HostSignalDelegatorFunction> Handlers{};
-      FEXCore::HostSignalDelegatorFunction FrontendHandler{};
-    };
-    std::array<HostSignalHandler, MAX_SIGNALS + 1> HostHandlers{};
-
-  protected:
-    void SetHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
-      HostHandlers[Signal].Handlers.push_back(std::move(Func));
-    }
-    void SetFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
-      HostHandlers[Signal].FrontendHandler = std::move(Func);
-    }
   };
 }

FEXCore/include/FEXCore/IR/IREmitter.h

@@ -334,7 +334,7 @@ friend class FEXCore::IR::PassManager;
       return Ptr;
     }
 
-    virtual void SaveNZCV() {
+    virtual void SaveNZCV(IROps Op) {
       // Overriden by dispatcher, stubbed for IR tests
     }
 

FEXCore/include/FEXCore/Utils/DeferredSignalMutex.h

@@ -1,338 +0,0 @@
-// SPDX-License-Identifier: MIT
-#pragma once
-
-#include <FEXCore/Debug/InternalThreadState.h>
-
-#include <atomic>
-#include <cstdint>
-#include <mutex>
-#include <shared_mutex>
-#include <signal.h>
-#ifndef _WIN32
-#include <sys/syscall.h>
-#endif
-#include <unistd.h>
-
-namespace FEXCore {
-#ifndef _WIN32
-  // Replacement for std::mutexes to deal with unlocking issues in the face of Linux fork() semantics.
-  //
-  // A fork() only clones the parent's calling thread. Other threads are silently dropped, which permanently leaves any mutexes owned by them locked.
-  // To address this issue, ForkableUniqueMutex and ForkableSharedMutex provide a way to forcefully remove any dangling locks and reset the mutexes to their default state.
-  class ForkableUniqueMutex final {
-    public:
-      ForkableUniqueMutex()
-        : Mutex (PTHREAD_MUTEX_INITIALIZER) {
-      }
-
-      // Move-only type
-      ForkableUniqueMutex(const ForkableUniqueMutex&) = delete;
-      ForkableUniqueMutex& operator=(const ForkableUniqueMutex&) = delete;
-      ForkableUniqueMutex(ForkableUniqueMutex &&rhs) = default;
-      ForkableUniqueMutex& operator=(ForkableUniqueMutex &&) = default;
-
-      void lock() {
-        [[maybe_unused]] const auto Result = pthread_mutex_lock(&Mutex);
-        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to lock with {}", __func__, Result);
-      }
-      void unlock() {
-        [[maybe_unused]] const auto Result = pthread_mutex_unlock(&Mutex);
-        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to unlock with {}", __func__, Result);
-      }
-      // Initialize the internal pthread object to its default initializer state.
-      // Should only ever be used in the child process when a Linux fork() has occured.
-      void StealAndDropActiveLocks() {
-        Mutex = PTHREAD_MUTEX_INITIALIZER;
-      }
-    private:
-      pthread_mutex_t Mutex;
-  };
-
-  class ForkableSharedMutex final {
-    public:
-      ForkableSharedMutex()
-        : Mutex (PTHREAD_RWLOCK_INITIALIZER) {
-      }
-
-      // Move-only type
-      ForkableSharedMutex(const ForkableSharedMutex&) = delete;
-      ForkableSharedMutex& operator=(const ForkableSharedMutex&) = delete;
-      ForkableSharedMutex(ForkableSharedMutex &&rhs) = default;
-      ForkableSharedMutex& operator=(ForkableSharedMutex &&) = default;
-
-      void lock() {
-        [[maybe_unused]] const auto Result = pthread_rwlock_wrlock(&Mutex);
-        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to lock with {}", __func__, Result);
-      }
-      void unlock() {
-        [[maybe_unused]] const auto Result = pthread_rwlock_unlock(&Mutex);
-        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to unlock with {}", __func__, Result);
-      }
-      void lock_shared() {
-        [[maybe_unused]] const auto Result = pthread_rwlock_rdlock(&Mutex);
-        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to lock with {}", __func__, Result);
-      }
-
-      void unlock_shared() {
-        unlock();
-      }
-
-      bool try_lock() {
-        const auto Result = pthread_rwlock_trywrlock(&Mutex);
-        return Result == 0;
-      }
-
-      bool try_lock_shared() {
-        const auto Result = pthread_rwlock_tryrdlock(&Mutex);
-        return Result == 0;
-      }
-      // Initialize the internal pthread object to its default initializer state.
-      // Should only ever be used in the child process when a Linux fork() has occured.
-      void StealAndDropActiveLocks() {
-        Mutex = PTHREAD_RWLOCK_INITIALIZER;
-      }
-    private:
-      pthread_rwlock_t Mutex;
-  };
-#else
-  // Windows doesn't support forking, so these can be standard mutexes.
-  class ForkableUniqueMutex final {
-    public:
-      ForkableUniqueMutex() = default;
-
-      // Non-moveable
-      ForkableUniqueMutex(const ForkableUniqueMutex&) = delete;
-      ForkableUniqueMutex& operator=(const ForkableUniqueMutex&) = delete;
-      ForkableUniqueMutex(ForkableUniqueMutex &&rhs) = delete;
-      ForkableUniqueMutex& operator=(ForkableUniqueMutex &&) = delete;
-
-      void lock() {
-        Mutex.lock();
-      }
-      void unlock() {
-        Mutex.unlock();
-      }
-      // Initialize the internal pthread object to its default initializer state.
-      // Should only ever be used in the child process when a Linux fork() has occured.
-      void StealAndDropActiveLocks() {
-        LogMan::Msg::AFmt("{} is unsupported on WIN32 builds!", __func__);
-      }
-    private:
-      std::mutex Mutex;
-  };
-
-  class ForkableSharedMutex final {
-    public:
-      ForkableSharedMutex() = default;
-
-      // Non-moveable
-      ForkableSharedMutex(const ForkableSharedMutex&) = delete;
-      ForkableSharedMutex& operator=(const ForkableSharedMutex&) = delete;
-      ForkableSharedMutex(ForkableSharedMutex &&rhs) = delete;
-      ForkableSharedMutex& operator=(ForkableSharedMutex &&) = delete;
-
-      void lock() {
-        Mutex.lock();
-      }
-      void unlock() {
-        Mutex.unlock();
-      }
-      void lock_shared() {
-        Mutex.lock_shared();
-      }
-
-      void unlock_shared() {
-        Mutex.unlock_shared();
-      }
-
-      bool try_lock() {
-        return Mutex.try_lock();
-      }
-
-      bool try_lock_shared() {
-        return Mutex.try_lock_shared();
-      }
-      // Initialize the internal pthread object to its default initializer state.
-      // Should only ever be used in the child process when a Linux fork() has occured.
-      void StealAndDropActiveLocks() {
-        LogMan::Msg::AFmt("{} is unsupported on WIN32 builds!", __func__);
-      }
-    private:
-      std::shared_mutex Mutex;
-  };
-#endif
-
-  template<typename MutexType, void (MutexType::*lock_fn)(), void (MutexType::*unlock_fn)()>
-  class ScopedDeferredSignalWithMutexBase final {
-    public:
-
-      ScopedDeferredSignalWithMutexBase(MutexType &_Mutex, FEXCore::Core::InternalThreadState *Thread)
-        : Mutex {&_Mutex}
-        , Thread {Thread} {
-        // Needs to be atomic so that operations can't end up getting reordered around this.
-        Thread->CurrentFrame->State.DeferredSignalRefCount.Increment(1);
-        // Lock the mutex
-        (Mutex->*lock_fn)();
-      }
-
-      // No copy or assignment possible
-      ScopedDeferredSignalWithMutexBase(const ScopedDeferredSignalWithMutexBase&) = delete;
-      ScopedDeferredSignalWithMutexBase& operator=(ScopedDeferredSignalWithMutexBase&) = delete;
-
-      // Only move
-      ScopedDeferredSignalWithMutexBase(ScopedDeferredSignalWithMutexBase &&rhs)
-        : Mutex {rhs.Mutex}
-        , Thread {rhs.Thread} {
-        rhs.Mutex = nullptr;
-      }
-
-      ~ScopedDeferredSignalWithMutexBase() {
-        if (Mutex != nullptr) {
-          // Unlock the mutex
-          (Mutex->*unlock_fn)();
-
-#ifdef _M_X86_64
-          // Needs to be atomic so that operations can't end up getting reordered around this.
-          // Without this, the recount and the signal access could get reordered.
-          auto Result = Thread->CurrentFrame->State.DeferredSignalRefCount.Decrement(1);
-
-          // X86-64 must do an additional check around the store.
-          if ((Result - 1) == 0) {
-            // Must happen after the refcount store
-            Thread->CurrentFrame->State.DeferredSignalFaultAddress->Store(0);
-          }
-#else
-          Thread->CurrentFrame->State.DeferredSignalRefCount.Decrement(1);
-          Thread->CurrentFrame->State.DeferredSignalFaultAddress->Store(0);
-#endif
-        }
-      }
-    private:
-      MutexType *Mutex;
-      FEXCore::Core::InternalThreadState *Thread;
-  };
-
-  using ScopedDeferredSignalWithMutex      = ScopedDeferredSignalWithMutexBase<std::mutex, &std::mutex::lock, &std::mutex::unlock>;
-  using ScopedDeferredSignalWithSharedLock = ScopedDeferredSignalWithMutexBase<std::shared_mutex, &std::shared_mutex::lock_shared, &std::shared_mutex::unlock_shared>;
-  using ScopedDeferredSignalWithUniqueLock = ScopedDeferredSignalWithMutexBase<std::shared_mutex, &std::shared_mutex::lock, &std::shared_mutex::unlock>;
-
-  // Forkable variant
-  using ScopedDeferredSignalWithForkableMutex      = ScopedDeferredSignalWithMutexBase<
-    FEXCore::ForkableUniqueMutex,
-    &FEXCore::ForkableUniqueMutex::lock,
-    &FEXCore::ForkableUniqueMutex::unlock>;
-  using ScopedDeferredSignalWithForkableSharedLock = ScopedDeferredSignalWithMutexBase<
-    FEXCore::ForkableSharedMutex,
-    &FEXCore::ForkableSharedMutex::lock_shared,
-    &FEXCore::ForkableSharedMutex::unlock_shared>;
-  using ScopedDeferredSignalWithForkableUniqueLock = ScopedDeferredSignalWithMutexBase<
-    FEXCore::ForkableSharedMutex,
-    &FEXCore::ForkableSharedMutex::lock,
-    &FEXCore::ForkableSharedMutex::unlock>;
-
-  class ScopedSignalMasker final {
-    public:
-      ScopedSignalMasker() = default;
-
-      void Mask(uint64_t Mask) {
-#ifndef _WIN32
-        // Mask all signals, storing the original incoming mask
-        ::syscall(SYS_rt_sigprocmask, SIG_SETMASK, &Mask, &OriginalMask, sizeof(OriginalMask));
-#endif
-      }
-
-      // Move-only type
-      ScopedSignalMasker(const ScopedSignalMasker&) = delete;
-      ScopedSignalMasker& operator=(ScopedSignalMasker&) = delete;
-      ScopedSignalMasker(ScopedSignalMasker &&rhs) = default;
-      ScopedSignalMasker& operator=(ScopedSignalMasker &&) = default;
-
-      void Unmask() {
-#ifndef _WIN32
-        ::syscall(SYS_rt_sigprocmask, SIG_SETMASK, &OriginalMask, nullptr, sizeof(OriginalMask));
-#endif
-      }
-    private:
-#ifndef _WIN32
-      uint64_t OriginalMask{};
-#endif
-  };
-
-  template<typename MutexType, void (MutexType::*lock_fn)(), void (MutexType::*unlock_fn)()>
-  class ScopedPotentialDeferredSignalWithMutexBase final {
-    public:
-      ScopedPotentialDeferredSignalWithMutexBase(MutexType &_Mutex, FEXCore::Core::InternalThreadState *Thread, uint64_t Mask = ~0ULL)
-        : Mutex {&_Mutex}
-        , Thread {Thread} {
-        if (Thread) {
-          Thread->CurrentFrame->State.DeferredSignalRefCount.Increment(1);
-        }
-        else {
-          Masker.Mask(Mask);
-        }
-        // Lock the mutex
-        (Mutex->*lock_fn)();
-      }
-
-      // No copy or assignment possible
-      ScopedPotentialDeferredSignalWithMutexBase(const ScopedPotentialDeferredSignalWithMutexBase&) = delete;
-      ScopedPotentialDeferredSignalWithMutexBase& operator=(ScopedPotentialDeferredSignalWithMutexBase&) = delete;
-
-      // Only move
-      ScopedPotentialDeferredSignalWithMutexBase(ScopedPotentialDeferredSignalWithMutexBase &&rhs)
-        : Mutex {rhs.Mutex}
-        , Thread {rhs.Thread} {
-        rhs.Mutex = nullptr;
-      }
-
-      ~ScopedPotentialDeferredSignalWithMutexBase() {
-        if (Mutex != nullptr) {
-          // Unlock the mutex
-          (Mutex->*unlock_fn)();
-
-          if (Thread) {
-#ifdef _M_X86_64
-            // Needs to be atomic so that operations can't end up getting reordered around this.
-            // Without this, the refcount and the signal access could get reordered.
-            auto Result = Thread->CurrentFrame->State.DeferredSignalRefCount.Decrement(1);
-
-            // X86-64 must do an additional check around the store.
-            if ((Result - 1) == 0) {
-              // Must happen after the refcount store
-              Thread->CurrentFrame->State.DeferredSignalFaultAddress->Store(0);
-            }
-#else
-            Thread->CurrentFrame->State.DeferredSignalRefCount.Decrement(1);
-            Thread->CurrentFrame->State.DeferredSignalFaultAddress->Store(0);
-#endif
-          }
-          else {
-            // Unmask back to the original signal mask
-            Masker.Unmask();
-          }
-        }
-      }
-    private:
-      MutexType *Mutex;
-      ScopedSignalMasker Masker;
-      FEXCore::Core::InternalThreadState *Thread;
-  };
-
-  using ScopedPotentialDeferredSignalWithMutex      = ScopedPotentialDeferredSignalWithMutexBase<std::mutex, &std::mutex::lock, &std::mutex::unlock>;
-  using ScopedPotentialDeferredSignalWithSharedLock = ScopedPotentialDeferredSignalWithMutexBase<std::shared_mutex, &std::shared_mutex::lock_shared, &std::shared_mutex::unlock_shared>;
-  using ScopedPotentialDeferredSignalWithUniqueLock = ScopedPotentialDeferredSignalWithMutexBase<std::shared_mutex, &std::shared_mutex::lock, &std::shared_mutex::unlock>;
-
-  // Forkable variant
-  using ScopedPotentialDeferredSignalWithForkableMutex      = ScopedPotentialDeferredSignalWithMutexBase<
-    FEXCore::ForkableUniqueMutex,
-    &FEXCore::ForkableUniqueMutex::lock,
-    &FEXCore::ForkableUniqueMutex::unlock>;
-  using ScopedPotentialDeferredSignalWithForkableSharedLock = ScopedPotentialDeferredSignalWithMutexBase<
-    FEXCore::ForkableSharedMutex,
-    &FEXCore::ForkableSharedMutex::lock_shared,
-    &FEXCore::ForkableSharedMutex::unlock_shared>;
-  using ScopedPotentialDeferredSignalWithForkableUniqueLock = ScopedPotentialDeferredSignalWithMutexBase<
-    FEXCore::ForkableSharedMutex,
-    &FEXCore::ForkableSharedMutex::lock,
-    &FEXCore::ForkableSharedMutex::unlock>;
-}

FEXCore/include/FEXCore/Utils/SignalScopeGuards.h

@@ -0,0 +1,243 @@
+// SPDX-License-Identifier: MIT
+#pragma once
+
+#include <FEXCore/Debug/InternalThreadState.h>
+
+#include <atomic>
+#include <cstdint>
+#include <mutex>
+#include <optional>
+#include <signal.h>
+#ifndef _WIN32
+#include <sys/syscall.h>
+#endif
+#include <unistd.h>
+#include <variant>
+
+namespace FEXCore {
+#ifndef _WIN32
+  // Replacement for std::mutexes to deal with unlocking issues in the face of Linux fork() semantics.
+  //
+  // A fork() only clones the parent's calling thread. Other threads are silently dropped, which permanently leaves any mutexes owned by them locked.
+  // To address this issue, ForkableUniqueMutex and ForkableSharedMutex provide a way to forcefully remove any dangling locks and reset the mutexes to their default state.
+  class ForkableUniqueMutex final {
+    public:
+      ForkableUniqueMutex()
+        : Mutex (PTHREAD_MUTEX_INITIALIZER) {
+      }
+
+      // Move-only type
+      ForkableUniqueMutex(const ForkableUniqueMutex&) = delete;
+      ForkableUniqueMutex& operator=(const ForkableUniqueMutex&) = delete;
+      ForkableUniqueMutex(ForkableUniqueMutex &&rhs) = default;
+      ForkableUniqueMutex& operator=(ForkableUniqueMutex &&) = default;
+
+      void lock() {
+        [[maybe_unused]] const auto Result = pthread_mutex_lock(&Mutex);
+        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to lock with {}", __func__, Result);
+      }
+      void unlock() {
+        [[maybe_unused]] const auto Result = pthread_mutex_unlock(&Mutex);
+        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to unlock with {}", __func__, Result);
+      }
+      // Initialize the internal pthread object to its default initializer state.
+      // Should only ever be used in the child process when a Linux fork() has occured.
+      void StealAndDropActiveLocks() {
+        Mutex = PTHREAD_MUTEX_INITIALIZER;
+      }
+    private:
+      pthread_mutex_t Mutex;
+  };
+
+  class ForkableSharedMutex final {
+    public:
+      ForkableSharedMutex()
+        : Mutex (PTHREAD_RWLOCK_INITIALIZER) {
+      }
+
+      // Move-only type
+      ForkableSharedMutex(const ForkableSharedMutex&) = delete;
+      ForkableSharedMutex& operator=(const ForkableSharedMutex&) = delete;
+      ForkableSharedMutex(ForkableSharedMutex &&rhs) = default;
+      ForkableSharedMutex& operator=(ForkableSharedMutex &&) = default;
+
+      void lock() {
+        [[maybe_unused]] const auto Result = pthread_rwlock_wrlock(&Mutex);
+        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to lock with {}", __func__, Result);
+      }
+      void unlock() {
+        [[maybe_unused]] const auto Result = pthread_rwlock_unlock(&Mutex);
+        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to unlock with {}", __func__, Result);
+      }
+      void lock_shared() {
+        [[maybe_unused]] const auto Result = pthread_rwlock_rdlock(&Mutex);
+        LOGMAN_THROW_A_FMT(Result == 0, "{} failed to lock with {}", __func__, Result);
+      }
+
+      void unlock_shared() {
+        unlock();
+      }
+
+      bool try_lock() {
+        const auto Result = pthread_rwlock_trywrlock(&Mutex);
+        return Result == 0;
+      }
+
+      bool try_lock_shared() {
+        const auto Result = pthread_rwlock_tryrdlock(&Mutex);
+        return Result == 0;
+      }
+      // Initialize the internal pthread object to its default initializer state.
+      // Should only ever be used in the child process when a Linux fork() has occured.
+      void StealAndDropActiveLocks() {
+        Mutex = PTHREAD_RWLOCK_INITIALIZER;
+      }
+    private:
+      pthread_rwlock_t Mutex;
+  };
+#else
+  // Windows doesn't support forking, so these can be standard mutexes.
+  class ForkableUniqueMutex final : public std::mutex {
+    public:
+      void StealAndDropActiveLocks() {
+        LogMan::Msg::AFmt("{} is unsupported on WIN32 builds!", __func__);
+      }
+  };
+
+  class ForkableSharedMutex final : public std::shared_mutex {
+    public:
+      void StealAndDropActiveLocks() {
+        LogMan::Msg::AFmt("{} is unsupported on WIN32 builds!", __func__);
+      }
+  };
+#endif
+
+  // Helper class to manage deferred signal refcounting within a block scope
+  class DeferredSignalRefCountGuard final {
+    public:
+      explicit DeferredSignalRefCountGuard(FEXCore::Core::InternalThreadState *Thread) : Thread(Thread) {
+        // Needs to be atomic so that operations can't end up getting reordered around this.
+        Thread->CurrentFrame->State.DeferredSignalRefCount.Increment(1);
+      }
+
+      // Move-only type
+      DeferredSignalRefCountGuard(const DeferredSignalRefCountGuard&) = delete;
+      DeferredSignalRefCountGuard& operator=(DeferredSignalRefCountGuard&) = delete;
+      DeferredSignalRefCountGuard(DeferredSignalRefCountGuard&& rhs) : Thread(rhs.Thread) {
+        rhs.Thread = nullptr;
+      }
+
+      ~DeferredSignalRefCountGuard() {
+        if (Thread) {
+#ifdef _M_X86_64
+          // Needs to be atomic so that operations can't end up getting reordered around this.
+          // Without this, the refcount and the signal access could get reordered.
+          auto Result = Thread->CurrentFrame->State.DeferredSignalRefCount.Decrement(1);
+
+          // X86-64 must do an additional check around the store.
+          if ((Result - 1) == 0) {
+            // Must happen after the refcount store
+            Thread->CurrentFrame->State.DeferredSignalFaultAddress->Store(0);
+          }
+#else
+          Thread->CurrentFrame->State.DeferredSignalRefCount.Decrement(1);
+          Thread->CurrentFrame->State.DeferredSignalFaultAddress->Store(0);
+#endif
+        }
+      }
+    private:
+      FEXCore::Core::InternalThreadState *Thread;
+  };
+
+#ifndef _WIN32
+  // Helper class to mask POSIX signals within a block scope
+  class ScopedSignalMasker final {
+    public:
+      explicit ScopedSignalMasker(uint64_t Mask) : OriginalMask(0) {
+        // Mask all signals, storing the original incoming mask
+        ::syscall(SYS_rt_sigprocmask, SIG_SETMASK, &Mask, &*OriginalMask, sizeof(*OriginalMask));
+      }
+
+      // Move-only type
+      ScopedSignalMasker(const ScopedSignalMasker&) = delete;
+      ScopedSignalMasker& operator=(ScopedSignalMasker&) = delete;
+      ScopedSignalMasker(ScopedSignalMasker&& rhs) : OriginalMask(rhs.OriginalMask) {
+        rhs.OriginalMask.reset();
+      }
+
+      ~ScopedSignalMasker() {
+        if (OriginalMask) {
+          ::syscall(SYS_rt_sigprocmask, SIG_SETMASK, &OriginalMask, nullptr, sizeof(*OriginalMask));
+        }
+      }
+    private:
+      std::optional<uint64_t> OriginalMask{};
+  };
+#endif
+
+  /**
+   * @brief Produces a wrapper object around a scoped lock of the given mutex
+   * while ensuring POSIX signals are masked while the mutex is locked
+   *
+   * Use this to prevent reentrancy issues of C++ mutexes with certain signal handlers.
+   * Common examples of such issues are:
+   * - C++ mutexes not unlocking due to a signal handler calling longjmp from within a scope owning the mutex
+   * - The signal handler itself using a mutex that would be re-locked if the handler gets invoked
+   *   again before unlocking
+   *
+   * Ownership of the returned object may be moved, but it is NOT SAFE to move across threads.
+   */
+  template<template<typename> class LockType = std::unique_lock, typename MutexType>
+  [[nodiscard]] static auto MaskSignalsAndLockMutex(MutexType& mutex, uint64_t Mask = ~0ULL) {
+#ifndef _WIN32
+      // Signals are masked first, and then the lock is acquired
+      struct {
+          ScopedSignalMasker mask;
+          LockType<MutexType> lock;
+      } scope_guard { ScopedSignalMasker { Mask }, LockType<MutexType> { mutex } };
+      return scope_guard;
+#else
+      // TODO: Doesn't block signals which may or may not cause issues.
+      return LockType<MutexType> { mutex };
+#endif
+  }
+
+  /**
+   * @brief Produces a wrapper object around a scoped lock of the given mutex
+   * while bumping the Thread's deferred signal refcount while the mutex is
+   * locked.
+   */
+  template<template<typename> class LockType = std::unique_lock, typename MutexType>
+  [[nodiscard]] static auto GuardSignalDeferringSection(MutexType& mutex, FEXCore::Core::InternalThreadState *Thread, uint64_t Mask = ~0ULL) {
+    // Refcount is incremented first, and then the lock is acquired.
+    struct {
+      std::optional<DeferredSignalRefCountGuard> refcount;
+      LockType<MutexType> lock;
+    } scope_guard = { DeferredSignalRefCountGuard { Thread }, LockType<MutexType> { mutex } };
+    return scope_guard;
+  }
+
+  // Like GuardSignalDeferringSection but falls back to masking signals when Thread is nullptr
+  template<template<typename> class LockType = std::unique_lock, typename MutexType>
+  [[nodiscard]] static auto GuardSignalDeferringSectionWithFallback(MutexType& mutex, FEXCore::Core::InternalThreadState *Thread, uint64_t Mask = ~0ULL) {
+#ifndef _WIN32
+    using ExtraGuard = std::variant<ScopedSignalMasker, DeferredSignalRefCountGuard>;
+#else
+    using ExtraGuard = std::variant<std::monostate, DeferredSignalRefCountGuard>;
+#endif
+
+    struct {
+      ExtraGuard refcount_or_mask;
+      LockType<MutexType> lock;
+    } scope_guard {
+        Thread ? ExtraGuard { DeferredSignalRefCountGuard { Thread } }
+#ifndef _WIN32
+                : ExtraGuard { ScopedSignalMasker { Mask } }
+#else
+                : ExtraGuard { }
+#endif
+    };
+    scope_guard.lock = LockType<MutexType> { mutex };
+    return scope_guard;
+  }
+}

FEXCore/unittests/Emitter/ALU_Tests.cpp

@@ -1712,6 +1712,12 @@ TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: Evaluate into flags") {
 TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: Carry flag invert") {
   TEST_SINGLE(cfinv(), "cfinv");
 }
+TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: Arm to eXternal FLAG") {
+  TEST_SINGLE(axflag(), "axflag");
+}
+TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: eXternal to Arm FLAG") {
+  TEST_SINGLE(xaflag(), "xaflag");
+}
 TEST_CASE_METHOD(TestDisassembler, "Emitter: ALU: Conditional compare - register") {
   TEST_SINGLE(ccmn(Size::i32Bit, Reg::r29, Reg::r28, StatusFlags::None, Condition::CC_AL), "ccmn w29, w28, #nzcv, al");
   TEST_SINGLE(ccmn(Size::i32Bit, Reg::r29, Reg::r28, StatusFlags::Flag_N, Condition::CC_AL), "ccmn w29, w28, #Nzcv, al");

FEXHeaderUtils/FEXHeaderUtils/ScopedSignalMask.h

@@ -1,123 +0,0 @@
-// SPDX-License-Identifier: MIT
-#pragma once
-
-#include <FEXCore/Utils/DeferredSignalMutex.h>
-
-#include <atomic>
-#include <cstdint>
-#include <mutex>
-#include <shared_mutex>
-#ifndef _WIN32
-#include <signal.h>
-#include <sys/syscall.h>
-#endif
-#include <unistd.h>
-
-namespace FHU {
-  /**
-   * @brief A drop-in replacement for std::lock_guard that masks POSIX signals while the mutex is locked
-   *
-   * Use this class to prevent reentrancy issues of C++ mutexes with certain signal handlers.
-   * Common examples of such issues are:
-   * - C++ mutexes not unlocking due to a signal handler longjmping out of a scope owning the mutex
-   * - The signal handler itself using a mutex that would be re-locked if the handler gets invoked
-   *   again before unlocking
-   *
-   * Ownership of this object may be moved, but it is NOT SAFE to move across threads.
-   *
-   * Constructor order:
-   * 1) Mask signals
-   * 2) Lock Mutex
-   *
-   * Destructor Order:
-   * 1) Unlock Mutex
-   * 2) Unmask signals
-   */
-#ifndef _WIN32
-  template<typename MutexType, void (MutexType::*lock_fn)(), void (MutexType::*unlock_fn)()>
-  class ScopedSignalMaskWithMutexBase final {
-    public:
-
-      ScopedSignalMaskWithMutexBase(MutexType &_Mutex, uint64_t Mask = ~0ULL)
-        : Mutex {&_Mutex} {
-        // Mask all signals, storing the original incoming mask
-        ::syscall(SYS_rt_sigprocmask, SIG_SETMASK, &Mask, &OriginalMask, sizeof(OriginalMask));
-
-        // Lock the mutex
-        (Mutex->*lock_fn)();
-      }
-
-      // No copy or assignment possible
-      ScopedSignalMaskWithMutexBase(const ScopedSignalMaskWithMutexBase&) = delete;
-      ScopedSignalMaskWithMutexBase& operator=(ScopedSignalMaskWithMutexBase&) = delete;
-
-      // Only move
-      ScopedSignalMaskWithMutexBase(ScopedSignalMaskWithMutexBase &&rhs)
-       : OriginalMask {rhs.OriginalMask}, Mutex {rhs.Mutex} {
-        rhs.Mutex = nullptr;
-      }
-
-      ~ScopedSignalMaskWithMutexBase() {
-        if (Mutex != nullptr) {
-          // Unlock the mutex
-          (Mutex->*unlock_fn)();
-
-          // Unmask back to the original signal mask
-          ::syscall(SYS_rt_sigprocmask, SIG_SETMASK, &OriginalMask, nullptr, sizeof(OriginalMask));
-        }
-      }
-    private:
-      uint64_t OriginalMask{};
-      MutexType *Mutex;
-  };
-#else
-  // TODO: Doesn't block signals which may or may not cause issues.
-  template<typename MutexType, void (MutexType::*lock_fn)(), void (MutexType::*unlock_fn)()>
-  class ScopedSignalMaskWithMutexBase final {
-    public:
-
-      ScopedSignalMaskWithMutexBase(MutexType &_Mutex, [[maybe_unused]] uint64_t Mask = ~0ULL)
-        : Mutex {&_Mutex} {
-        // Lock the mutex
-        (Mutex->*lock_fn)();
-      }
-
-      // No copy or assignment possible
-      ScopedSignalMaskWithMutexBase(const ScopedSignalMaskWithMutexBase&) = delete;
-      ScopedSignalMaskWithMutexBase& operator=(ScopedSignalMaskWithMutexBase&) = delete;
-
-      // Only move
-      ScopedSignalMaskWithMutexBase(ScopedSignalMaskWithMutexBase &&rhs)
-       : Mutex {rhs.Mutex} {
-        rhs.Mutex = nullptr;
-      }
-
-      ~ScopedSignalMaskWithMutexBase() {
-        if (Mutex != nullptr) {
-          // Unlock the mutex
-          (Mutex->*unlock_fn)();
-        }
-      }
-    private:
-      MutexType *Mutex;
-  };
-
-#endif
-
-  using ScopedSignalMaskWithMutex = ScopedSignalMaskWithMutexBase<std::mutex, &std::mutex::lock, &std::mutex::unlock>;
-  using ScopedSignalMaskWithSharedLock = ScopedSignalMaskWithMutexBase<std::shared_mutex, &std::shared_mutex::lock_shared, &std::shared_mutex::unlock_shared>;
-  using ScopedSignalMaskWithUniqueLock = ScopedSignalMaskWithMutexBase<std::shared_mutex, &std::shared_mutex::lock, &std::shared_mutex::unlock>;
-
-  using ScopedSignalMaskWithForkableMutex = ScopedSignalMaskWithMutexBase<
-    FEXCore::ForkableUniqueMutex,
-    &FEXCore::ForkableUniqueMutex::lock,
-    &FEXCore::ForkableUniqueMutex::unlock>;
-  using ScopedSignalMaskWithForkableSharedLock = ScopedSignalMaskWithMutexBase<
-    FEXCore::ForkableSharedMutex,
-    &FEXCore::ForkableSharedMutex::lock_shared,
-    &FEXCore::ForkableSharedMutex::unlock_shared>;
-  using ScopedSignalMaskWithForkableUniqueLock = ScopedSignalMaskWithMutexBase<
-    FEXCore::ForkableSharedMutex,
-    &FEXCore::ForkableSharedMutex::lock,
-    &FEXCore::ForkableSharedMutex::unlock>;
-}

Scripts/InstructionCountParser.py

@@ -14,14 +14,12 @@ logger.setLevel(logging.ERROR)
 @dataclass
 class TestData:
     name: str
-    optimal: int
     expectedinstructioncount: int
     code: bytes
     instructions: list
-    def __init__(self, Name, Optimal, ExpectedInstructionCount, Code, Instructions):
+    def __init__(self, Name, ExpectedInstructionCount, Code, Instructions):
         self.name = Name
         self.expectedinstructioncount = ExpectedInstructionCount
-        self.optimal = Optimal
         self.code = Code
         self.instructions = Instructions
 
@@ -29,10 +27,6 @@ class TestData:
     def Name(self):
         return self.name
 
-    @property
-    def Optimal(self):
-        return self.optimal
-
     @property
     def ExpectedInstructionCount(self):
         return self.expectedinstructioncount
@@ -58,6 +52,7 @@ class HostFeatures(Flag) :
     FEATURE_RPRES  = (1 << 7)
     FEATURE_FLAGM  = (1 << 8)
     FEATURE_FLAGM2 = (1 << 9)
+    FEATURE_CRYPTO = (1 << 10)
 
 HostFeaturesLookup = {
     "SVE128"  : HostFeatures.FEATURE_SVE128,
@@ -70,6 +65,7 @@ HostFeaturesLookup = {
     "RPRES"   : HostFeatures.FEATURE_RPRES,
     "FLAGM"   : HostFeatures.FEATURE_FLAGM,
     "FLAGM2"  : HostFeatures.FEATURE_FLAGM2,
+    "CRYPTO"  : HostFeatures.FEATURE_CRYPTO,
 }
 
 def GetHostFeatures(data):
@@ -112,15 +108,10 @@ def parse_json_data(json_filepath, json_filename, json_data, output_binary_path)
 
     for key, items in json_data["Instructions"].items():
         ExpectedInstructionCount = 0
-        Optimal = 0
         Instructions = []
         if ("ExpectedInstructionCount" in items):
             ExpectedInstructionCount = int(items["ExpectedInstructionCount"])
 
-        if ("Optimal" in items):
-                if items["Optimal"].upper() == "YES":
-                    Optimal = 1
-
         if ("Skip" in items):
                 if items["Skip"].upper() == "YES":
                     continue
@@ -163,7 +154,7 @@ def parse_json_data(json_filepath, json_filename, json_data, output_binary_path)
         with open(tmp_asm_out, "rb") as tmp_asm_out_file:
             binary_hex = tmp_asm_out_file.read()
 
-        TestDataMap[TestName] = TestData(key, Optimal, ExpectedInstructionCount, binary_hex, Instructions)
+        TestDataMap[TestName] = TestData(key, ExpectedInstructionCount, binary_hex, Instructions)
 
         os.remove(tmp_asm)
         os.remove(tmp_asm_out)
@@ -181,7 +172,6 @@ def parse_json_data(json_filepath, json_filename, json_data, output_binary_path)
         # };
         # struct TestInfo {
         #   char InstName[128];
-        #   uint64_t Optimal;
         #   int64_t ExpectedInstructionCount;
         #   uint64_t CodeSize;
         #   uint64_t x86InstCount;
@@ -208,7 +198,6 @@ def parse_json_data(json_filepath, json_filename, json_data, output_binary_path)
     # Add each test
     for key, item in TestDataMap.items():
         MemData += struct.pack('128s', item.Name.encode("ascii"))
-        MemData += struct.pack('Q', item.Optimal)
         MemData += struct.pack('q', item.ExpectedInstructionCount)
         MemData += struct.pack('Q', len(item.Code))
         MemData += struct.pack('Q', len(item.Instructions))

Source/Common/FEXServerClient.cpp

@@ -104,6 +104,17 @@ namespace FEXServerClient {
     return GetServerLockFolder() + "RootFS.lock";
   }
 
+  fextl::string GetTempFolder() {
+    auto XDGRuntimeEnv = getenv("XDG_RUNTIME_DIR");
+    if (XDGRuntimeEnv) {
+      // If the XDG runtime directory works then use that.
+      return XDGRuntimeEnv;
+    }
+    // Fallback to `/tmp/` if XDG_RUNTIME_DIR doesn't exist.
+    // Might not be ideal but we don't have much of a choice.
+    return fextl::string{std::filesystem::temp_directory_path().string()};
+  }
+
   fextl::string GetServerMountFolder() {
     // We need a FEXServer mount directory that has some tricky requirements.
     // - We don't want to use `/tmp/` if possible.
@@ -119,17 +130,7 @@ namespace FEXServerClient {
     //   - If this path doesn't exist then fallback to `/tmp/` as a last resort.
     //   - pressure-vessel explicitly creates an internal XDG_RUNTIME_DIR inside its chroot.
     //     - This is okay since pressure-vessel rbinds the FEX rootfs from the host to `/run/pressure-vessel/interpreter-root`.
-    fextl::string Folder{};
-    auto XDGRuntimeEnv = getenv("XDG_RUNTIME_DIR");
-    if (XDGRuntimeEnv) {
-      // If the XDG runtime directory works then use that.
-      Folder = XDGRuntimeEnv;
-    }
-    else {
-      // Fallback to `/tmp/` if XDG_RUNTIME_DIR doesn't exist.
-      // Might not be ideal but we don't have much of a choice.
-      Folder = std::filesystem::temp_directory_path().string();
-    }
+    auto Folder = GetTempFolder();
 
     if (FEXCore::Config::FindContainer() == "pressure-vessel") {
       // In pressure-vessel the mount point changes location.

Source/Common/FEXServerClient.h

@@ -50,6 +50,7 @@ namespace FEXServerClient {
   fextl::string GetServerLockFolder();
   fextl::string GetServerLockFile();
   fextl::string GetServerRootFSLockFile();
+  fextl::string GetTempFolder();
   fextl::string GetServerMountFolder();
   fextl::string GetServerSocketName();
   int GetServerFD();

Source/Tools/CodeSizeValidation/Main.cpp

@@ -227,7 +227,6 @@ void AssertHandler(char const *Message) {
 
 struct TestInfo {
   char TestInst[128];
-  uint64_t Optimal;
   int64_t ExpectedInstructionCount;
   uint64_t CodeSize;
   uint64_t x86InstCount;
@@ -276,8 +275,8 @@ static bool TestInstructions(FEXCore::Context::Context *CTX, FEXCore::Core::Inte
 
     LogMan::Msg::IFmt("Testing instruction '{}': {} host instructions", CurrentTest->TestInst, INSTStats->first.HostCodeInstructions);
 
-    // Show the code if we know the implementation isn't optimal or if the count of instructions changed to something we didn't expect.
-    bool ShouldShowCode = CurrentTest->Optimal == 0 ||
+    // Show the code if the count of instructions changed to something we didn't expect.
+    bool ShouldShowCode =
       INSTStats->first.HostCodeInstructions != CurrentTest->ExpectedInstructionCount;
 
     if (ShouldShowCode) {
@@ -467,6 +466,7 @@ int main(int argc, char **argv, char **const envp) {
     FEATURE_RPRES  = (1U << 7),
     FEATURE_FLAGM  = (1U << 8),
     FEATURE_FLAGM2 = (1U << 9),
+    FEATURE_CRYPTO = (1U << 10),
   };
 
   uint64_t SVEWidth = 0;
@@ -503,11 +503,12 @@ int main(int argc, char **argv, char **const envp) {
   if (TestHeaderData->EnabledHostFeatures & FEATURE_FLAGM2) {
     HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::ENABLEFLAGM2);
   }
+  if (TestHeaderData->EnabledHostFeatures & FEATURE_CRYPTO) {
+    HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::ENABLECRYPTO);
+  }
 
   // Always enable ARMv8.1 LSE atomics.
   HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::ENABLEATOMICS);
-  // Always enable crypto extensions.
-  HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::ENABLECRYPTO);
 
   if (TestHeaderData->DisabledHostFeatures & FEATURE_SVE128) {
     HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::DISABLESVE);
@@ -539,6 +540,9 @@ int main(int argc, char **argv, char **const envp) {
   if (TestHeaderData->DisabledHostFeatures & FEATURE_FLAGM2) {
     HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::DISABLEFLAGM2);
   }
+  if (TestHeaderData->DisabledHostFeatures & FEATURE_CRYPTO) {
+    HostFeatureControl |= static_cast<uint64_t>(FEXCore::Config::HostFeatures::DISABLECRYPTO);
+  }
 
   FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_HOSTFEATURES, fextl::fmt::format("{}", HostFeatureControl));
   FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_FORCESVEWIDTH, fextl::fmt::format("{}", SVEWidth));
@@ -549,7 +553,6 @@ int main(int argc, char **argv, char **const envp) {
   // Create FEXCore context.
   auto CTX = FEXCore::Context::Context::CreateNewContext();
 
-  CTX->InitializeContext();
   auto SignalDelegation = FEX::DummyHandlers::CreateSignalDelegator();
   auto SyscallHandler = FEX::DummyHandlers::CreateSyscallHandler();
 

Source/Tools/CommonTools/DummyHandlers.h

@@ -34,21 +34,11 @@ class DummySignalDelegator final : public FEXCore::SignalDelegator, public FEXCo
   }
 
   protected:
-    // Called from the thunk handler to handle the signal
-    void HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *Info, void *UContext) override {}
-
     void RegisterTLSState(FEXCore::Core::InternalThreadState *Thread) override;
     void UninstallTLSState(FEXCore::Core::InternalThreadState *Thread) override;
 
-    FEXCore::Core::InternalThreadState *GetTLSThread() override;
-
-    /**
-     * @brief Registers a signal handler for the host to handle a signal
-     *
-     * It's a process level signal handler so one must be careful
-     */
-    void FrontendRegisterHostSignalHandler(int Signal, FEXCore::HostSignalDelegatorFunction Func, bool Required) override {}
-    void FrontendRegisterFrontendHostSignalHandler(int Signal, FEXCore::HostSignalDelegatorFunction Func, bool Required) override {}
+  private:
+    FEXCore::Core::InternalThreadState *GetTLSThread();
 };
 
 fextl::unique_ptr<FEXCore::HLE::SyscallHandler> CreateSyscallHandler();

Source/Tools/FEXLoader/AOT/AOTGenerator.cpp

@@ -106,8 +106,7 @@ void AOTGenSection(FEXCore::Context::Context *CTX, ELFCodeLoader::LoadedSection
       setpriority(PRIO_PROCESS, FHU::Syscalls::gettid(), 19);
 
       // Setup thread - Each compilation thread uses its own backing FEX thread
-      FEXCore::Core::CPUState state;
-      auto Thread = CTX->CreateThread(&state, FHU::Syscalls::gettid());
+      auto Thread = CTX->CreateThread(0, 0);
       fextl::set<uint64_t> ExternalBranchesLocal;
       CTX->ConfigureAOTGen(Thread, &ExternalBranchesLocal, SectionMaxAddress);
 

Source/Tools/FEXLoader/ArchHelpers/MContext.h

@@ -143,6 +143,10 @@ static inline uint64_t GetArmReg(void* ucontext, uint32_t id) {
   return GetMContext(ucontext)->regs[id];
 }
 
+static inline uint64_t GetArmPState(void* ucontext) {
+  return GetMContext(ucontext)->pstate;
+}
+
 static inline uint64_t *GetArmGPRs(void* ucontext) {
   return reinterpret_cast<uint64_t*>(GetMContext(ucontext)->regs);
 }
@@ -313,6 +317,10 @@ static inline __uint128_t GetArmFPR(void* ucontext, uint32_t id) {
   ERROR_AND_DIE_FMT("Not implemented for x86 host");
 }
 
+static inline uint64_t GetArmPState(void* ucontext) {
+  ERROR_AND_DIE_FMT("Not implemented for x86 host");
+}
+
 static inline uint64_t *GetArmGPRs(void* ucontext) {
   ERROR_AND_DIE_FMT("Not implemented for x86 host");
 }

Source/Tools/FEXLoader/CMakeLists.txt

@@ -37,6 +37,7 @@ if (NOT MINGW_BUILD)
         ${PTHREAD_LIB}
         fmt::fmt
     )
+    target_compile_options(${NAME} PRIVATE ${FEX_TUNE_COMPILE_FLAGS})
     target_compile_definitions(${NAME} PRIVATE -DFEXLOADER_AS_INTERPRETER=${AsInterpreter})
 
     if (CMAKE_BUILD_TYPE MATCHES "RELEASE")

Source/Tools/FEXLoader/FEXLoader.cpp

@@ -12,6 +12,7 @@ $end_info$
 #include "Common/Config.h"
 #include "ELFCodeLoader.h"
 #include "VDSO_Emulation.h"
+#include "LinuxSyscalls/GdbServer.h"
 #include "LinuxSyscalls/LinuxAllocator.h"
 #include "LinuxSyscalls/Syscalls.h"
 #include "LinuxSyscalls/Utils/Threads.h"
@@ -443,7 +444,6 @@ int main(int argc, char **argv, char **const envp) {
   FEXCore::Context::InitializeStaticTables(Loader.Is64BitMode() ? FEXCore::Context::MODE_64BIT : FEXCore::Context::MODE_32BIT);
 
   auto CTX = FEXCore::Context::Context::CreateNewContext();
-  CTX->InitializeContext();
 
   // Setup TSO hardware emulation immediately after initializing the context.
   FEX::TSO::SetupTSOEmulation(CTX.get());
@@ -474,7 +474,14 @@ int main(int argc, char **argv, char **const envp) {
 
   CTX->SetSignalDelegator(SignalDelegation.get());
   CTX->SetSyscallHandler(SyscallHandler.get());
-  CTX->InitCore(Loader.DefaultRIP(), Loader.GetStackPointer());
+
+  FEX_CONFIG_OPT(GdbServer, GDBSERVER);
+  fextl::unique_ptr<FEX::GdbServer> DebugServer;
+  if (GdbServer) {
+    DebugServer = fextl::make_unique<FEX::GdbServer>(CTX.get(), SignalDelegation.get(), SyscallHandler.get());
+  }
+
+  auto ParentThread = CTX->InitCore(Loader.DefaultRIP(), Loader.GetStackPointer());
 
   // Pass in our VDSO thunks
   CTX->AppendThunkDefinitions(FEX::VDSO::GetVDSOThunkDefinitions());
@@ -550,8 +557,9 @@ int main(int argc, char **argv, char **const envp) {
     }
   }
 
-  auto ProgramStatus = CTX->GetProgramStatus();
+  auto ProgramStatus = ParentThread->StatusCode;
 
+  DebugServer.reset();
   SyscallHandler.reset();
   SignalDelegation.reset();
 

Source/Tools/FEXLoader/IRLoader.cpp

@@ -160,7 +160,6 @@ int main(int argc, char **argv, char **const envp)
 
   FEXCore::Context::InitializeStaticTables();
   auto CTX = FEXCore::Context::Context::CreateNewContext();
-  CTX->InitializeContext();
 
   auto SignalDelegation = FEX::HLE::CreateSignalDelegator(CTX.get(), {});
 
@@ -179,7 +178,7 @@ int main(int argc, char **argv, char **const envp)
 
   if (Loader.LoadIR(CTX.get()))
   {
-    CTX->InitCore(Loader.DefaultRIP(), Loader.GetStackPointer());
+    auto ParentThread = CTX->InitCore(Loader.DefaultRIP(), Loader.GetStackPointer());
 
     auto ShutdownReason = FEXCore::Context::ExitReason::EXIT_SHUTDOWN;
 
@@ -211,10 +210,7 @@ int main(int argc, char **argv, char **const envp)
     LogMan::Msg::DFmt("Reason we left VM: {}", FEXCore::ToUnderlying(ShutdownReason));
 
     // Just re-use compare state. It also checks against the expected values in config.
-    FEXCore::Core::CPUState State;
-    CTX->GetCPUState(&State);
-
-    const bool Passed = Loader.CompareStates(&State, SupportsAVX);
+    const bool Passed = Loader.CompareStates(&ParentThread->CurrentFrame->State, SupportsAVX);
 
     LogMan::Msg::IFmt("Passed? {}\n", Passed ? "Yes" : "No");
 

Source/Tools/FEXLoader/LinuxSyscalls/CMakeLists.txt

@@ -1,6 +1,7 @@
 add_compile_options(-fno-operator-names)
 
 set (SRCS
+  GdbServer.cpp
   EmulatedFiles/EmulatedFiles.cpp
   FileManagement.cpp
   LinuxAllocator.cpp

Source/Tools/FEXLoader/LinuxSyscalls/EmulatedFiles/EmulatedFiles.cpp

@@ -14,6 +14,7 @@ $end_info$
 #include <FEXCore/Core/CodeLoader.h>
 #include <FEXCore/Core/Context.h>
 #include <FEXCore/Core/CPUID.h>
+#include <FEXCore/Utils/CPUInfo.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/string.h>
@@ -46,9 +47,22 @@ namespace FEX::EmulatedFile {
    *
    * @return A temporary file that we can use
    */
-  static int GenTmpFD() {
-    int fd = open("/tmp", O_RDWR | O_TMPFILE | O_EXCL, S_IRUSR | S_IWUSR);
-    return fd;
+  static int GenTmpFD(const char *pathname, int flags) {
+    uint32_t memfd_flags {MFD_ALLOW_SEALING};
+    if (flags & O_CLOEXEC) memfd_flags |= MFD_CLOEXEC;
+
+    return memfd_create(pathname, memfd_flags);
+  }
+
+  // Seal the tmpfd features by sealing them all.
+  // Makes the tmpfd read-only.
+  static void SealTmpFD(int fd) {
+    fcntl(fd, F_ADD_SEALS,
+      F_SEAL_SEAL |
+      F_SEAL_SHRINK |
+      F_SEAL_GROW |
+      F_SEAL_WRITE |
+      F_SEAL_FUTURE_WRITE);
   }
 
   fextl::string GenerateCPUInfo(FEXCore::Context::Context *ctx, uint32_t CPUCores) {
@@ -621,21 +635,23 @@ namespace FEX::EmulatedFile {
   }
 
   EmulatedFDManager::EmulatedFDManager(FEXCore::Context::Context *ctx)
-    : CTX {ctx} {
+    : CTX {ctx}
+    , ThreadsConfig { FEXCore::CPUInfo::CalculateNumberOfCPUs() } {
     FDReadCreators["/proc/cpuinfo"] = [&](FEXCore::Context::Context *ctx, int32_t fd, const char *pathname, int32_t flags, mode_t mode) -> int32_t {
       // Only allow a single thread to initialize the cpu_info.
       // Jit in-case multiple threads try to initialize at once.
       // Check if deferred cpuinfo initialization has occured.
-      std::call_once(cpu_info_initialized, [&]() { cpu_info = GenerateCPUInfo(ctx, ThreadsConfig()); });
+      std::call_once(cpu_info_initialized, [&]() { cpu_info = GenerateCPUInfo(ctx, ThreadsConfig); });
 
-      int FD = GenTmpFD();
+      int FD = GenTmpFD(pathname, flags);
       write(FD, (void*)&cpu_info.at(0), cpu_info.size());
       lseek(FD, 0, SEEK_SET);
+      SealTmpFD(FD);
       return FD;
     };
 
     FDReadCreators["/proc/sys/kernel/osrelease"] = [&](FEXCore::Context::Context *ctx, int32_t fd, const char *pathname, int32_t flags, mode_t mode) -> int32_t {
-      int FD = GenTmpFD();
+      int FD = GenTmpFD(pathname, flags);
       uint32_t GuestVersion = FEX::HLE::_SyscallHandler->GetGuestKernelVersion();
       char Tmp[64]{};
       snprintf(Tmp, sizeof(Tmp), "%d.%d.%d\n",
@@ -645,11 +661,12 @@ namespace FEX::EmulatedFile {
       // + 1 to ensure null at the end
       write(FD, Tmp, strlen(Tmp) + 1);
       lseek(FD, 0, SEEK_SET);
+      SealTmpFD(FD);
       return FD;
     };
 
     FDReadCreators["/proc/version"] = [&](FEXCore::Context::Context *ctx, int32_t fd, const char *pathname, int32_t flags, mode_t mode) -> int32_t {
-      int FD = GenTmpFD();
+      int FD = GenTmpFD(pathname, flags);
       // UTS version NEEDS to be in a format that can pass to `date -d`
       // Format of this is Linux version <Release> (<Compile By>@<Compile Host>) (<Linux Compiler>) #<version> {SMP, PREEMPT, PREEMPT_RT} <UTS version>\n"
       const char kernel_version[] = "Linux version %d.%d.%d (FEX@FEX) (clang) #" GIT_DESCRIBE_STRING " SMP " __DATE__ " " __TIME__ "\n";
@@ -662,13 +679,15 @@ namespace FEX::EmulatedFile {
       // + 1 to ensure null at the end
       write(FD, Tmp, strlen(Tmp) + 1);
       lseek(FD, 0, SEEK_SET);
+      SealTmpFD(FD);
       return FD;
     };
 
     auto NumCPUCores = [&](FEXCore::Context::Context *ctx, int32_t fd, const char *pathname, int32_t flags, mode_t mode) -> int32_t {
-      int FD = GenTmpFD();
+      int FD = GenTmpFD(pathname, flags);
       write(FD, (void*)&cpus_online.at(0), cpus_online.size());
       lseek(FD, 0, SEEK_SET);
+      SealTmpFD(FD);
       return FD;
     };
 
@@ -681,7 +700,7 @@ namespace FEX::EmulatedFile {
     FDReadCreators["/proc/self/auxv"] = &EmulatedFDManager::ProcAuxv;
 
     auto cmdline_handler = [&](FEXCore::Context::Context *ctx, int32_t fd, const char *pathname, int32_t flags, mode_t mode) -> int32_t {
-      int FD = GenTmpFD();
+      int FD = GenTmpFD(pathname, flags);
       auto CodeLoader = FEX::HLE::_SyscallHandler->GetCodeLoader();
       auto Args = CodeLoader->GetApplicationArguments();
       char NullChar{};
@@ -695,6 +714,7 @@ namespace FEX::EmulatedFile {
 
       // One additional null terminator to finish the list
       lseek(FD, 0, SEEK_SET);
+      SealTmpFD(FD);
       return FD;
     };
 
@@ -702,9 +722,8 @@ namespace FEX::EmulatedFile {
     fextl::string procCmdLine = fextl::fmt::format("/proc/{}/cmdline", getpid());
     FDReadCreators[procCmdLine] = cmdline_handler;
 
-    uint64_t CPUCores = ThreadsConfig();
-    if (CPUCores > 1) {
-      cpus_online = fextl::fmt::format("0-{}", CPUCores - 1);
+    if (ThreadsConfig > 1) {
+      cpus_online = fextl::fmt::format("0-{}", ThreadsConfig - 1);
     }
     else {
       cpus_online = "0";
@@ -721,6 +740,7 @@ namespace FEX::EmulatedFile {
     auto Creator = FDReadCreators.end();
     if (pathname) {
       Creator = FDReadCreators.find(pathname);
+      Path = pathname;
     }
 
     if (Creator == FDReadCreators.end()) {
@@ -788,9 +808,10 @@ namespace FEX::EmulatedFile {
       return -1;
     }
 
-    int FD = GenTmpFD();
+    int FD = GenTmpFD(pathname, flags);
     write(FD, (void*)auxvBase, auxvSize);
     lseek(FD, 0, SEEK_SET);
+    SealTmpFD(FD);
     return FD;
   }
 }

Source/Tools/FEXLoader/LinuxSyscalls/EmulatedFiles/EmulatedFiles.h

@@ -34,6 +34,6 @@ namespace FEX::EmulatedFile {
       fextl::unordered_map<fextl::string, FDReadStringFunc> FDReadCreators;
 
       static int32_t ProcAuxv(FEXCore::Context::Context* ctx, int32_t fd, const char* pathname, int32_t flags, mode_t mode);
-      FEX_CONFIG_OPT(ThreadsConfig, THREADS);
+      const uint32_t ThreadsConfig;
   };
 }

Source/Tools/FEXLoader/LinuxSyscalls/GdbServer.cpp

@@ -12,6 +12,7 @@ $end_info$
 #include <memory>
 #include <optional>
 
+#include <Common/FEXServerClient.h>
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CodeLoader.h>
 #include <FEXCore/Core/Context.h>
@@ -32,6 +33,7 @@ $end_info$
 #include <FEXCore/fextl/sstream.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/vector.h>
+#include <FEXHeaderUtils/Filesystem.h>
 
 #include <atomic>
 #include <cstring>
@@ -42,18 +44,72 @@ $end_info$
 #endif
 #include <errno.h>
 #include <fcntl.h>
-#include <fmt/format.h>
+#include <poll.h>
 #include <signal.h>
 #include <stddef.h>
 #include <string_view>
 #include <sys/stat.h>
+#include <sys/un.h>
+#include <sys/utsname.h>
 #include <unistd.h>
 #include <utility>
 
-#include "GdbServer.h"
+#include "LinuxSyscalls/GdbServer.h"
 
-namespace FEXCore
+namespace FEX
 {
+
+constexpr std::array<std::string_view const, 22> FlagNames = {
+  "CF",
+  "",
+  "PF",
+  "",
+  "AF",
+  "",
+  "ZF",
+  "SF",
+  "TF",
+  "IF",
+  "DF",
+  "OF",
+  "IOPL",
+  "",
+  "NT",
+  "",
+  "RF",
+  "VM",
+  "AC",
+  "VIF",
+  "VIP",
+  "ID",
+};
+
+static std::string_view const& GetFlagName(unsigned Flag) {
+  return FlagNames[Flag];
+}
+
+static std::string_view const GetGRegName(unsigned Reg) {
+  switch (Reg) {
+    case FEXCore::X86State::REG_RAX: return "rax";
+    case FEXCore::X86State::REG_RBX: return "rbx";
+    case FEXCore::X86State::REG_RCX: return "rcx";
+    case FEXCore::X86State::REG_RDX: return "rdx";
+    case FEXCore::X86State::REG_RSP: return "rsp";
+    case FEXCore::X86State::REG_RBP: return "rbp";
+    case FEXCore::X86State::REG_RSI: return "rsi";
+    case FEXCore::X86State::REG_RDI: return "rdi";
+    case FEXCore::X86State::REG_R8:  return "r8";
+    case FEXCore::X86State::REG_R9:  return "r9";
+    case FEXCore::X86State::REG_R10: return "r10";
+    case FEXCore::X86State::REG_R11: return "r11";
+    case FEXCore::X86State::REG_R12: return "r12";
+    case FEXCore::X86State::REG_R13: return "r13";
+    case FEXCore::X86State::REG_R14: return "r14";
+    case FEXCore::X86State::REG_R15: return "r15";
+    default: FEX_UNREACHABLE;
+  }
+}
+
 #ifndef _WIN32
 void GdbServer::Break(int signal) {
   std::lock_guard lk(sendMutex);
@@ -70,7 +126,12 @@ void GdbServer::WaitForThreadWakeup() {
   ThreadBreakEvent.Wait();
 }
 
-GdbServer::GdbServer(FEXCore::Context::Context *ctx, SignalDelegator *SignalDelegation, FEXCore::HLE::SyscallHandler *const SyscallHandler)
+GdbServer::~GdbServer() {
+  CoreShuttingDown = true;
+  close(ListenSocket);
+}
+
+GdbServer::GdbServer(FEXCore::Context::Context *ctx, FEX::HLE::SignalDelegator *SignalDelegation, FEXCore::HLE::SyscallHandler *const SyscallHandler)
   : CTX(ctx)
   , SyscallHandler {SyscallHandler} {
   // Pass all signals by default
@@ -88,7 +149,7 @@ GdbServer::GdbServer(FEXCore::Context::Context *ctx, SignalDelegator *SignalDele
 
   // This is a total hack as there is currently no way to resume once hitting a segfault
   // But it's semi-useful for debugging.
-  for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
+  for (uint32_t Signal = 0; Signal <= FEX::HLE::SignalDelegator::MAX_SIGNALS; ++Signal) {
     SignalDelegation->RegisterHostSignalHandler(Signal, [this] (FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) {
       if (PassSignals[Signal]) {
         // Pass signal to the guest
@@ -140,6 +201,10 @@ static fextl::string encodeHex(const unsigned char *data, size_t length) {
   return ss.str();
 }
 
+static fextl::string encodeHex(std::string_view str) {
+  return encodeHex(reinterpret_cast<const unsigned char*>(str.data()), str.size());
+}
+
 static fextl::string getThreadName(uint32_t ThreadID) {
   const auto ThreadFile = fextl::fmt::format("/proc/{}/task/{}/comm", getpid(), ThreadID);
   fextl::string ThreadName;
@@ -254,15 +319,15 @@ struct X80Float {
 };
 
 struct FEX_PACKED GDBContextDefinition {
-  uint64_t gregs[Core::CPUState::NUM_GPRS];
+  uint64_t gregs[FEXCore::Core::CPUState::NUM_GPRS];
   uint64_t rip;
   uint32_t eflags;
   uint32_t cs, ss, ds, es, fs, gs;
-  X80Float mm[Core::CPUState::NUM_MMS];
+  X80Float mm[FEXCore::Core::CPUState::NUM_MMS];
   uint32_t fctrl;
   uint32_t fstat;
   uint32_t dummies[6];
-  uint64_t xmm[Core::CPUState::NUM_XMMS][4];
+  uint64_t xmm[FEXCore::Core::CPUState::NUM_XMMS][4];
   uint32_t mxcsr;
 };
 
@@ -293,9 +358,9 @@ fextl::string GdbServer::readRegs() {
   memcpy(&GDB.gregs[0], &state.gregs[0], sizeof(GDB.gregs));
   memcpy(&GDB.rip, &state.rip, sizeof(GDB.rip));
 
-  GDB.eflags = CTX->ReconstructCompactedEFLAGS(CurrentThread);
+  GDB.eflags = CTX->ReconstructCompactedEFLAGS(CurrentThread, false, nullptr, 0);
 
-  for (size_t i = 0; i < Core::CPUState::NUM_MMS; ++i) {
+  for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_MMS; ++i) {
     memcpy(&GDB.mm[i], &state.mm[i], sizeof(GDB.mm));
   }
 
@@ -349,7 +414,7 @@ GdbServer::HandledPacketType GdbServer::readReg(const fextl::string& packet) {
     return {encodeHex((unsigned char *)(&state.rip), sizeof(uint64_t)), HandledPacketType::TYPE_ACK};
   }
   else if (addr == offsetof(GDBContextDefinition, eflags)) {
-    uint32_t eflags = CTX->ReconstructCompactedEFLAGS(CurrentThread);
+    uint32_t eflags = CTX->ReconstructCompactedEFLAGS(CurrentThread, false, nullptr, 0);
 
     return {encodeHex((unsigned char *)(&eflags), sizeof(uint32_t)), HandledPacketType::TYPE_ACK};
   }
@@ -383,9 +448,9 @@ GdbServer::HandledPacketType GdbServer::readReg(const fextl::string& packet) {
   }
   else if (addr >= offsetof(GDBContextDefinition, xmm[0][0]) &&
            addr < offsetof(GDBContextDefinition, xmm[16][0])) {
-    const auto XmmIndex = (addr - offsetof(GDBContextDefinition, xmm[0][0])) / Core::CPUState::XMM_AVX_REG_SIZE;
+    const auto XmmIndex = (addr - offsetof(GDBContextDefinition, xmm[0][0])) / FEXCore::Core::CPUState::XMM_AVX_REG_SIZE;
     const auto *Data = (unsigned char *)&state.xmm.avx.data[XmmIndex][0];
-    return {encodeHex(Data, Core::CPUState::XMM_AVX_REG_SIZE), HandledPacketType::TYPE_ACK};
+    return {encodeHex(Data, FEXCore::Core::CPUState::XMM_AVX_REG_SIZE), HandledPacketType::TYPE_ACK};
   }
   else if (addr == offsetof(GDBContextDefinition, mxcsr)) {
     uint32_t Empty{};
@@ -409,7 +474,7 @@ fextl::string buildTargetXML() {
       xml << "<flags id='fex_eflags' size='4'>\n";
       // flags register
       for(int i = 0; i < 22; i++) {
-          auto name = FEXCore::Core::GetFlagName(i);
+          auto name = GetFlagName(i);
           if (name.empty()) {
             continue;
           }
@@ -427,8 +492,8 @@ fextl::string buildTargetXML() {
       // We want to just memcpy our x86 state to gdb, so we tell it the ordering.
 
       // GPRs
-      for (uint32_t i = 0; i < Core::CPUState::NUM_GPRS; i++) {
-        reg(FEXCore::Core::GetGRegName(i), "int64", 64);
+      for (uint32_t i = 0; i < FEXCore::Core::CPUState::NUM_GPRS; i++) {
+        reg(GetGRegName(i), "int64", 64);
       }
 
       reg("rip", "code_ptr", 64);
@@ -484,7 +549,7 @@ fextl::string buildTargetXML() {
         )";
 
       // SSE regs
-      for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+      for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_XMMS; i++) {
           reg(fextl::fmt::format("xmm{}", i), "vec128", 128);
       }
 
@@ -510,7 +575,7 @@ fextl::string buildTargetXML() {
           <field name="uint128" type="uint128"/>
         </union>
         )";
-    for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+    for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_XMMS; i++) {
         reg(fmt::format("ymm{}h", i), "vec128", 128);
     }
     xml << "</feature>\n";
@@ -815,7 +880,6 @@ GdbServer::HandledPacketType GdbServer::handleMemory(const fextl::string &packet
   }
 }
 
-
 GdbServer::HandledPacketType GdbServer::handleQuery(const fextl::string &packet) {
   const auto match = [&](const char *str) -> bool { return packet.rfind(str, 0) == 0; };
   const auto MatchStr = [](const fextl::string &Str, const char *str) -> bool { return Str.rfind(str, 0) == 0; };
@@ -867,12 +931,17 @@ GdbServer::HandledPacketType GdbServer::handleQuery(const fextl::string &packet)
     SupportedFeatures += "QNonStop+;";
 
     SupportedFeatures += "qXfer:osdata:read+;";
+    SupportedFeatures += "QStartNoAckMode+;";
 
-    // Causes GDB to crash?
-    // SupportedFeatures += "QStartNoAckMode+;";
+    // TODO: Support breakpoints
+    // SupportedFeatures += "swbreak+;";
+    // SupportedFeatures += "hwbreak+;";
+    // SupportedFeatures += "BreakpointCommands+;";
+
+    // TODO: If we want to support conditional breakpoints then we need to support single stepping.
+    // SupportedFeatures += "ConditionalBreakpoints+;";
 
     for (auto &Feature : Features) {
-
       if (MatchStr(Feature, "swbreak+")) {
         SupportedFeatures += "swbreak+;";
       }
@@ -972,13 +1041,68 @@ GdbServer::HandledPacketType GdbServer::handleQuery(const fextl::string &packet)
     // We now have a semi-colon deliminated list of signals to pass to the guest process
     for (fextl::string tmp; std::getline(ss, tmp, ';'); ) {
       uint32_t Signal = std::stoi(tmp.c_str(), nullptr, 16);
-      if (Signal < SignalDelegator::MAX_SIGNALS) {
+      if (Signal < FEX::HLE::SignalDelegator::MAX_SIGNALS) {
         PassSignals[Signal] = true;
       }
     }
 
     return {"OK", HandledPacketType::TYPE_ACK};
   }
+
+  // lldb specific queries
+  if (match("qHostInfo")) {
+    // Returns Key:Value pairs separated by ;
+    // eg:
+    // triple:7838365f36342d70632d6c696e75782d676e75;
+    // ptrsize:8;
+    // distribution_id:7562756e7475;
+    // watchpoint_exceptions_received:after;
+    // endian:little;
+    // os_version:6.3.3;
+    // os_build:362e332e332d3036303330332d67656e65726963;
+    // os_kernel:2332303233303531373133333620534d5020505245454d50545f44594e414d494320576564204d61792031372031333a34353a3139205554432032303233;
+    // hostname:7279616e682d545235303030;
+    fextl::string HostFeatures{};
+
+    // 64-bit always returned for the host environment.
+    // qProcessInfo will return i386 or not.
+    HostFeatures += fextl::fmt::format("triple:{};", encodeHex("x86_64-pc-linux-gnu"));
+    HostFeatures += "ptrsize:8;";
+
+    // Always little-endian.
+    HostFeatures += "endian:little;";
+
+    struct utsname buf{};
+    if (uname(&buf) != -1) {
+      uint32_t Major{};
+      uint32_t Minor{};
+      uint32_t Patch{};
+
+      // Parse kernel version in the form of `<Major>.<Minor>.<Patch>[Optional Data]`
+      const auto End = buf.release + sizeof(buf.release);
+      auto Results = std::from_chars(buf.release, End, Major, 10);
+      Results = std::from_chars(Results.ptr + 1, End, Minor, 10);
+      Results = std::from_chars(Results.ptr + 1, End, Patch, 10);
+
+      HostFeatures += fextl::fmt::format("os_version:{}.{}.{};", Major, Minor, Patch);
+
+      // os_build returns the release untouched.
+      HostFeatures += fextl::fmt::format("os_build:{};", encodeHex(buf.release));
+      HostFeatures += fextl::fmt::format("os_kernel:{};", encodeHex(buf.version));
+      HostFeatures += fextl::fmt::format("hostname:{};", encodeHex(buf.nodename));
+    }
+
+    // TODO: distribution_id should be fetched with `lsb_release -i`
+    // TODO: watchpoint_exceptions_received is unsupported
+    return {std::move(HostFeatures), HandledPacketType::TYPE_ACK};
+  }
+  if (match("qGetWorkingDir")) {
+    char Tmp[PATH_MAX];
+    if (getcwd(Tmp, PATH_MAX)) {
+      return {encodeHex(Tmp), HandledPacketType::TYPE_ACK};
+    }
+    return {"E00", HandledPacketType::TYPE_ACK};
+  }
   return {"", HandledPacketType::TYPE_UNKNOWN};
 }
 
@@ -1221,11 +1345,45 @@ void GdbServer::SendPacketPair(const HandledPacketType& response) {
   }
 }
 
+GdbServer::WaitForConnectionResult GdbServer::WaitForConnection() {
+  while (!CoreShuttingDown.load()) {
+    struct pollfd PollFD {
+      .fd = ListenSocket,
+      .events = POLLIN | POLLPRI | POLLRDHUP,
+      .revents = 0,
+    };
+    int Result = ppoll(&PollFD, 1, nullptr, nullptr);
+    if (Result > 0) {
+      if (PollFD.revents & POLLIN) {
+        CommsStream = OpenSocket();
+        return WaitForConnectionResult::CONNECTION;
+      }
+      else if (PollFD.revents & (POLLHUP | POLLERR | POLLNVAL)) {
+        // Listen socket error or shutting down
+        LogMan::Msg::EFmt("[GdbServer] gdbserver shutting down: {}");
+        return WaitForConnectionResult::ERROR;
+      }
+    }
+    else if (Result == -1) {
+      LogMan::Msg::EFmt("[GdbServer] poll failure: {}", errno);
+    }
+  }
+
+  LogMan::Msg::EFmt("[GdbServer] Shutting Down");
+  return WaitForConnectionResult::ERROR;
+}
+
 void GdbServer::GdbServerLoop() {
   OpenListenSocket();
+  if (ListenSocket == -1) {
+    // Couldn't open socket, just exit.
+    return;
+  }
 
   while (!CoreShuttingDown.load()) {
-    CommsStream = OpenSocket();
+    if (WaitForConnection() == WaitForConnectionResult::ERROR) {
+      break;
+    }
 
     HandledPacketType response{};
 
@@ -1275,9 +1433,11 @@ void GdbServer::GdbServerLoop() {
   }
 
   close(ListenSocket);
+  unlink(GdbUnixSocketPath.c_str());
 }
 static void* ThreadHandler(void *Arg) {
-  FEXCore::GdbServer *This = reinterpret_cast<FEXCore::GdbServer*>(Arg);
+  FEXCore::Threads::SetThreadName("FEX:gdbserver");
+  auto This = reinterpret_cast<FEX::GdbServer*>(Arg);
   This->GdbServerLoop();
   return nullptr;
 }
@@ -1289,38 +1449,48 @@ void GdbServer::StartThread() {
 }
 
 void GdbServer::OpenListenSocket() {
-  // getaddrinfo allocates memory that can't be removed.
-  FEXCore::Allocator::YesIKnowImNotSupposedToUseTheGlibcAllocator glibc;
-  struct addrinfo hints, *res;
-
-  memset(&hints, 0, sizeof(hints));
-  hints.ai_family = AF_UNSPEC;
-  hints.ai_socktype = SOCK_STREAM;
-  hints.ai_flags = AI_PASSIVE;
-
-  if(getaddrinfo(NULL, "8086", &hints, &res) < 0) {
-    perror("getaddrinfo");
+  const auto GdbUnixPath = fextl::fmt::format("{}/FEX_gdbserver/", FEXServerClient::GetTempFolder());
+  if (FHU::Filesystem::CreateDirectory(GdbUnixPath) == FHU::Filesystem::CreateDirectoryResult::ERROR) {
+    LogMan::Msg::EFmt("[GdbServer] Couldn't create gdbserver folder {}", GdbUnixPath);
+    return;
   }
 
-  int on = 1;
+  GdbUnixSocketPath = fextl::fmt::format("{}{}-gdb", GdbUnixPath, ::getpid());
 
-  ListenSocket = socket(res->ai_family, res->ai_socktype, res->ai_protocol);
-  if (ListenSocket < 0) {
-    perror("socket");
+  ListenSocket = socket(AF_UNIX, SOCK_STREAM | SOCK_CLOEXEC, 0);
+  if (ListenSocket == -1) {
+    LogMan::Msg::EFmt("[GdbServer] Couldn't open AF_UNIX socket {} {}", errno, strerror(errno));
+    return;
   }
-  if(setsockopt(ListenSocket, SOL_SOCKET, SO_REUSEADDR, (char*)&on, sizeof(on)) < 0) {
-    perror("setsockopt");
-    close(ListenSocket);
+
+  struct sockaddr_un addr{};
+  addr.sun_family = AF_UNIX;
+  strncpy(addr.sun_path, GdbUnixSocketPath.data(), sizeof(addr.sun_path));
+  size_t SizeOfAddr = offsetof(sockaddr_un, sun_path) + GdbUnixSocketPath.size();
+
+  // Bind the socket to the path
+  int Result{};
+  for (int attempt = 0; attempt < 2; ++attempt) {
+    Result = bind(ListenSocket, reinterpret_cast<struct sockaddr*>(&addr), SizeOfAddr);
+    if (Result == 0) {
+      break;
+    }
+
+    // This can happen periodically with execve. unlink the path and try again.
+    // The PID is reused but FEX likely started a gdbserver thread for the PID before execve.
+    unlink(GdbUnixSocketPath.c_str());
   }
 
-  if (bind(ListenSocket, res->ai_addr, res->ai_addrlen) < 0) {
-    perror("bind");
+  if (Result != 0) {
+    LogMan::Msg::EFmt("[GdbServer] Couldn't bind AF_UNIX socket '{}': {} {}\n", addr.sun_path, errno, strerror(errno));
     close(ListenSocket);
+    ListenSocket = -1;
+    return;
   }
 
   listen(ListenSocket, 1);
-
-  freeaddrinfo(res);
+  LogMan::Msg::IFmt("[GdbServer] Waiting for connection on {}", GdbUnixSocketPath);
+  LogMan::Msg::IFmt("[GdbServer] gdb-multiarch -ex \"target extended-remote {}\"", GdbUnixSocketPath);
 }
 
 fextl::unique_ptr<std::iostream> GdbServer::OpenSocket() {
@@ -1328,7 +1498,6 @@ fextl::unique_ptr<std::iostream> GdbServer::OpenSocket() {
   struct sockaddr_storage their_addr{};
   socklen_t addr_size{};
 
-  LogMan::Msg::IFmt("GdbServer, waiting for connection on localhost:8086");
   int new_fd = accept(ListenSocket, (struct sockaddr *)&their_addr, &addr_size);
 
   return fextl::make_unique<FEXCore::Utils::NetStream>(new_fd);

Source/Tools/FEXLoader/LinuxSyscalls/GdbServer.h

@@ -19,11 +19,14 @@ $end_info$
 #include <mutex>
 #include <stdint.h>
 
-namespace FEXCore {
+#include "LinuxSyscalls/SignalDelegator.h"
+
+namespace FEX {
 
 class GdbServer {
 public:
-    GdbServer(FEXCore::Context::Context *ctx, SignalDelegator *SignalDelegation, FEXCore::HLE::SyscallHandler *const SyscallHandler);
+    GdbServer(FEXCore::Context::Context *ctx, FEX::HLE::SignalDelegator *SignalDelegation, FEXCore::HLE::SyscallHandler *const SyscallHandler);
+    ~GdbServer();
 
     // Public for threading
     void GdbServerLoop();
@@ -36,6 +39,11 @@ private:
     void Break(int signal);
 
     void OpenListenSocket();
+    enum class WaitForConnectionResult {
+      CONNECTION,
+      ERROR,
+    };
+    WaitForConnectionResult WaitForConnection();
     fextl::unique_ptr<std::iostream> OpenSocket();
     void StartThread();
     fextl::string ReadPacket(std::iostream &stream);
@@ -90,9 +98,10 @@ private:
     fextl::string LibraryMapString{};
 
     // Used to keep track of which signals to pass to the guest
-    std::array<bool, SignalDelegator::MAX_SIGNALS + 1> PassSignals{};
+    std::array<bool, FEX::HLE::SignalDelegator::MAX_SIGNALS + 1> PassSignals{};
     uint32_t CurrentDebuggingThread{};
     int ListenSocket{};
+    fextl::string GdbUnixSocketPath{};
     FEX_CONFIG_OPT(Filename, APP_FILENAME);
     FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
 };

Source/Tools/FEXLoader/LinuxSyscalls/SignalDelegator.cpp

@@ -150,6 +150,38 @@ namespace FEX::HLE {
     return SigInfoLayout::LAYOUT_KILL;
   }
 
+  void SignalDelegator::HandleSignal(int Signal, void *Info, void *UContext) {
+    // Let the host take first stab at handling the signal
+    auto Thread = GetTLSThread();
+
+    if (!Thread) {
+      LogMan::Msg::AFmt("[{}] Thread has received a signal and hasn't registered itself with the delegate! Programming error!", FHU::Syscalls::gettid());
+    }
+    else {
+      SignalHandler &Handler = HostHandlers[Signal];
+      for (auto &HandlerFunc : Handler.Handlers) {
+        if (HandlerFunc(Thread, Signal, Info, UContext)) {
+          // If the host handler handled the fault then we can continue now
+          return;
+        }
+      }
+
+      if (Handler.FrontendHandler &&
+          Handler.FrontendHandler(Thread, Signal, Info, UContext)) {
+        return;
+      }
+
+      // Now let the frontend handle the signal
+      // It's clearly a guest signal and this ends up being an OS specific issue
+      HandleGuestSignal(Thread, Signal, Info, UContext);
+    }
+  }
+
+  void SignalDelegator::RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
+    SetHostSignalHandler(Signal, Func, Required);
+    FrontendRegisterHostSignalHandler(Signal, Func, Required);
+  }
+
   void SignalDelegator::SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) {
 #ifdef _M_ARM_64
     for (size_t i = 0; i < Config.SRAGPRCount; i++) {
@@ -1137,12 +1169,13 @@ namespace FEX::HLE {
     ++Thread->CurrentFrame->SignalHandlerRefCounter;
 
     uint64_t OldPC = ArchHelpers::Context::GetPc(ucontext);
+    const bool WasInJIT = Thread->CPUBackend->IsAddressInCodeBuffer(OldPC);
 
     // Spill the SRA regardless of signal handler type
     // We are going to be returning to the top of the dispatcher which will fill again
     // Otherwise we might load garbage
     if (Config.StaticRegisterAllocation) {
-      if (Thread->CPUBackend->IsAddressInCodeBuffer(OldPC)) {
+      if (WasInJIT) {
         uint32_t IgnoreMask{};
 #ifdef _M_ARM_64
         if (Frame->InSyscallInfo != 0) {
@@ -1207,7 +1240,7 @@ namespace FEX::HLE {
     // Backup where we think the RIP currently is
     ContextBackup->OriginalRIP = CTX->RestoreRIPFromHostPC(Thread, ArchHelpers::Context::GetPc(ucontext));
     // Calculate eflags upfront.
-    uint32_t eflags = CTX->ReconstructCompactedEFLAGS(Thread);
+    uint32_t eflags = CTX->ReconstructCompactedEFLAGS(Thread, WasInJIT, ArchHelpers::Context::GetArmGPRs(ucontext), ArchHelpers::Context::GetArmPState(ucontext));
 
     if (Is64BitMode) {
       NewGuestSP = SetupFrame_x64(Thread, ContextBackup, Frame, Signal, HostSigInfo, ucontext, GuestAction, GuestStack, NewGuestSP, eflags);
@@ -1810,7 +1843,7 @@ namespace FEX::HLE {
     ThreadData.Thread = nullptr;
   }
 
-  void SignalDelegator::FrontendRegisterHostSignalHandler(int Signal, FEXCore::HostSignalDelegatorFunction Func, bool Required) {
+  void SignalDelegator::FrontendRegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
     // Linux signal handlers are per-process rather than per thread
     // Multiple threads could be calling in to this
     std::lock_guard lk(HostDelegatorMutex);
@@ -1818,7 +1851,7 @@ namespace FEX::HLE {
     InstallHostThunk(Signal);
   }
 
-  void SignalDelegator::FrontendRegisterFrontendHostSignalHandler(int Signal, FEXCore::HostSignalDelegatorFunction Func, bool Required) {
+  void SignalDelegator::FrontendRegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
     // Linux signal handlers are per-process rather than per thread
     // Multiple threads could be calling in to this
     std::lock_guard lk(HostDelegatorMutex);

Source/Tools/FEXLoader/LinuxSyscalls/SignalDelegator.h

@@ -40,20 +40,36 @@ namespace FEX::HLE {
 
   class SignalDelegator final : public FEXCore::SignalDelegator, public FEXCore::Allocator::FEXAllocOperators {
   public:
+    constexpr static size_t MAX_SIGNALS {64};
+
+    // Use the last signal just so we are less likely to ever conflict with something that the guest application is using
+    // 64 is used internally by Valgrind
+    constexpr static size_t SIGNAL_FOR_PAUSE {63};
+
     // Returns true if the host handled the signal
     // Arguments are the same as sigaction handler
     SignalDelegator(FEXCore::Context::Context *_CTX, const std::string_view ApplicationName);
     ~SignalDelegator() override;
 
+    // Called from the signal trampoline function.
+    void HandleSignal(int Signal, void *Info, void *UContext);
+
     void RegisterTLSState(FEXCore::Core::InternalThreadState *Thread) override;
     void UninstallTLSState(FEXCore::Core::InternalThreadState *Thread) override;
 
+    /**
+     * @brief Registers a signal handler for the host to handle a signal
+     *
+     * It's a process level signal handler so one must be careful
+     */
+    void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
+
     /**
      * @brief Registers a signal handler for the host to handle a signal specifically for guest handling
      *
      * It's a process level signal handler so one must be careful
      */
-    void RegisterHostSignalHandlerForGuest(int Signal, FEX::HLE::HostSignalDelegatorFunctionForGuest Func);
+    void RegisterHostSignalHandlerForGuest(int Signal, HostSignalDelegatorFunctionForGuest Func);
     void RegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
 
     /**
@@ -107,21 +123,27 @@ namespace FEX::HLE {
     FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
 
     void SaveTelemetry();
-  protected:
-    // Called from the thunk handler to handle the signal
-    void HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *Info, void *UContext) override;
+  private:
+    FEXCore::Core::InternalThreadState *GetTLSThread();
 
-    FEXCore::Core::InternalThreadState *GetTLSThread() override;
+    // Called from the thunk handler to handle the signal
+    void HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *Info, void *UContext);
 
     /**
      * @brief Registers a signal handler for the host to handle a signal
      *
      * It's a process level signal handler so one must be careful
      */
-    void FrontendRegisterHostSignalHandler(int Signal, FEXCore::HostSignalDelegatorFunction Func, bool Required) override;
-    void FrontendRegisterFrontendHostSignalHandler(int Signal, FEXCore::HostSignalDelegatorFunction Func, bool Required) override;
+    void FrontendRegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
+    void FrontendRegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
+
+    void SetHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
+      HostHandlers[Signal].Handlers.push_back(std::move(Func));
+    }
+    void SetFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
+      HostHandlers[Signal].FrontendHandler = std::move(Func);
+    }
 
-  private:
     FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
     FEX_CONFIG_OPT(Core, CORE);
     fextl::string const ApplicationName;
@@ -155,6 +177,10 @@ namespace FEX::HLE {
       FEX::HLE::HostSignalDelegatorFunctionForGuest GuestHandler{};
       GuestSigAction GuestAction{};
       DefaultBehaviour DefaultBehaviour {DEFAULT_TERM};
+
+      // Callbacks
+      fextl::vector<HostSignalDelegatorFunction> Handlers{};
+      HostSignalDelegatorFunction FrontendHandler{};
     };
 
     std::array<SignalHandler, MAX_SIGNALS + 1> HostHandlers{};

Source/Tools/FEXLoader/LinuxSyscalls/Syscalls.cpp

@@ -525,6 +525,14 @@ static uint64_t Clone3Handler(FEXCore::Core::CpuStateFrame *Frame, FEX::HLE::clo
 uint64_t CloneHandler(FEXCore::Core::CpuStateFrame *Frame, FEX::HLE::clone3_args *args) {
   uint64_t flags = args->args.flags;
 
+  if (flags & CLONE_CLEAR_SIGHAND) {
+    // CLONE_CLEAR_SIGHAND was added in kernel 5.5. FEX doesn't properly support this.
+    // glibc started using this flag in 2.38 as an optimization for posix_spawn.
+    // If clone returns EINVAL or ENOSYS then it will fallback to the non-optimized path.
+    LogMan::Msg::IFmt("CLONE_CLEAR_SIGHAND passed to clone3. Returning EINVAL.");
+    return -EINVAL;
+  }
+
   auto HasUnhandledFlags = [](FEX::HLE::clone3_args *args) -> bool {
     constexpr uint64_t UNHANDLED_FLAGS =
       CLONE_NEWNS |

Source/Tools/FEXLoader/LinuxSyscalls/Syscalls.h

@@ -16,7 +16,7 @@ $end_info$
 #include <FEXCore/HLE/SourcecodeResolver.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/Utils/CompilerDefs.h>
-#include <FEXCore/Utils/DeferredSignalMutex.h>
+#include <FEXCore/Utils/SignalScopeGuards.h>
 #include <FEXCore/fextl/fmt.h>
 #include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory.h>
@@ -175,7 +175,6 @@ public:
   FEX_CONFIG_OPT(IsInterpreterInstalled, INTERPRETER_INSTALLED);
   FEX_CONFIG_OPT(Filename, APP_FILENAME);
   FEX_CONFIG_OPT(RootFSPath, ROOTFS);
-  FEX_CONFIG_OPT(ThreadsConfig, THREADS);
   FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
   FEX_CONFIG_OPT(SMCChecks, SMCCHECKS);
 

Source/Tools/FEXLoader/LinuxSyscalls/Syscalls/Info.cpp

@@ -89,21 +89,8 @@ namespace FEX::HLE {
 
     REGISTER_SYSCALL_IMPL_FLAGS(getcpu, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
       [](FEXCore::Core::CpuStateFrame *Frame, unsigned *cpu, unsigned *node, struct getcpu_cache *tcache) -> uint64_t {
-      uint32_t LocalCPU{};
-      uint32_t LocalNode{};
       // tcache is ignored
-      uint64_t Result = ::syscall(SYSCALL_DEF(getcpu), cpu ? &LocalCPU : nullptr, node ? &LocalNode : nullptr, nullptr);
-      if (Result == 0) {
-        if (cpu) {
-          // Ensure we don't return a number over our number of emulated cores
-          *cpu = LocalCPU % FEX::HLE::_SyscallHandler->ThreadsConfig();
-        }
-
-        if (node) {
-          // Just claim we are part of node zero
-          *node = 0;
-        }
-      }
+      uint64_t Result = ::syscall(SYSCALL_DEF(getcpu), cpu, node, nullptr);
       SYSCALL_ERRNO();
     });
 

Source/Tools/FEXLoader/LinuxSyscalls/Syscalls/Sched.cpp

@@ -80,35 +80,14 @@ namespace FEX::HLE {
 
     REGISTER_SYSCALL_IMPL_FLAGS(sched_setaffinity, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY | SyscallFlags::NOSIDEEFFECTS,
       [](FEXCore::Core::CpuStateFrame *Frame, pid_t pid, size_t cpusetsize, const unsigned long *mask) -> uint64_t {
-      return 0;
+      uint64_t Result = ::syscall(SYSCALL_DEF(sched_setaffinity), pid, cpusetsize, mask);
+      SYSCALL_ERRNO();
     });
 
     REGISTER_SYSCALL_IMPL_FLAGS(sched_getaffinity, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,
       [](FEXCore::Core::CpuStateFrame *Frame, pid_t pid, size_t cpusetsize, unsigned char *mask) -> uint64_t {
-      uint64_t Cores = FEX::HLE::_SyscallHandler->ThreadsConfig();
-
-      // Bytes need to round up to size of uint64_t
-      uint64_t Bytes = FEXCore::AlignUp(Cores, sizeof(uint64_t));
-
-      // cpusetsize needs to be 8byte aligned
-      if (cpusetsize & (sizeof(uint64_t) - 1)) {
-        return -EINVAL;
-      }
-
-      // If we don't have enough bytes to store the resulting structure
-      // then we need to return -EINVAL
-      if (cpusetsize < Bytes) {
-        return -EINVAL;
-      }
-
-      memset(mask, 0, Bytes);
-
-      for (uint64_t i = 0; i < Cores; ++i) {
-        mask[i / 8] |= (1 << (i % 8));
-      }
-
-      // Returns the number of bytes written in to mask
-      return Bytes;
+      uint64_t Result = ::syscall(SYSCALL_DEF(sched_getaffinity), pid, cpusetsize, mask);
+      SYSCALL_ERRNO();
     });
 
     REGISTER_SYSCALL_IMPL_PASS_FLAGS(sched_setattr, SyscallFlags::OPTIMIZETHROUGH | SyscallFlags::NOSYNCSTATEONENTRY,

Source/Tools/FEXLoader/LinuxSyscalls/Syscalls/Thread.cpp

@@ -58,7 +58,7 @@ namespace FEX::HLE {
       NewThreadState.gregs[FEXCore::X86State::REG_RSP] = args->args.stack;
     }
 
-    auto NewThread = CTX->CreateThread(&NewThreadState, args->args.parent_tid);
+    auto NewThread = CTX->CreateThread(0, 0, &NewThreadState, args->args.parent_tid);
     CTX->InitializeThread(NewThread);
 
     if (FEX::HLE::_SyscallHandler->Is64BitMode()) {
@@ -131,7 +131,7 @@ namespace FEX::HLE {
       }
 
       // Overwrite thread
-      NewThread = CTX->CreateThread(&NewThreadState, GuestArgs->parent_tid);
+      NewThread = CTX->CreateThread(0, 0, &NewThreadState, GuestArgs->parent_tid);
 
       // CLONE_PARENT_SETTID, CLONE_CHILD_SETTID, CLONE_CHILD_CLEARTID, CLONE_PIDFD will be handled by kernel
       // Call execution thread directly since we already are on the new thread

Source/Tools/FEXLoader/LinuxSyscalls/SyscallsSMCTracking.cpp

@@ -16,11 +16,10 @@ $end_info$
 #include "LinuxSyscalls/Syscalls.h"
 
 #include <FEXHeaderUtils/TypeDefines.h>
-#include <FEXHeaderUtils/ScopedSignalMask.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
-#include <FEXCore/Utils/DeferredSignalMutex.h>
+#include <FEXCore/Utils/SignalScopeGuards.h>
 
 namespace FEX::HLE {
 
@@ -55,7 +54,7 @@ bool SyscallHandler::HandleSegfault(FEXCore::Core::InternalThreadState *Thread,
 
   {
     // Can't use the deferred signal lock in the SIGSEGV handler.
-    FHU::ScopedSignalMaskWithForkableSharedLock lk(_SyscallHandler->VMATracking.Mutex);
+    auto lk = FEXCore::MaskSignalsAndLockMutex<std::shared_lock>(_SyscallHandler->VMATracking.Mutex);
 
     auto VMATracking = &_SyscallHandler->VMATracking;
 
@@ -112,7 +111,7 @@ void SyscallHandler::MarkGuestExecutableRange(FEXCore::Core::InternalThreadState
       return;
     }
 
-    FEXCore::ScopedDeferredSignalWithForkableSharedLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection<std::shared_lock>(VMATracking.Mutex, Thread);
 
     // Find the first mapping at or after the range ends, or ::end().
     // Top points to the address after the end of the range
@@ -167,7 +166,7 @@ void SyscallHandler::MarkGuestExecutableRange(FEXCore::Core::InternalThreadState
 
 // Used for AOT
 FEXCore::HLE::AOTIRCacheEntryLookupResult SyscallHandler::LookupAOTIRCacheEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestAddr) {
-  FEXCore::ScopedDeferredSignalWithForkableSharedLock lk(VMATracking.Mutex, Thread);
+  auto lk = FEXCore::GuardSignalDeferringSection<std::shared_lock>(VMATracking.Mutex, Thread);
 
   // Get the first mapping after GuestAddr, or end
   // GuestAddr is inclusive
@@ -194,8 +193,8 @@ void SyscallHandler::TrackMmap(FEXCore::Core::InternalThreadState *Thread, uintp
   {
     // NOTE: Frontend calls this with a nullptr Thread during initialization, but
     //       providing this code with a valid Thread object earlier would allow
-    //       us to be more optimal by using ScopedDeferredSignalWithUniqueLock instead
-    FEXCore::ScopedPotentialDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    //       us to be more optimal by using GuardSignalDeferringSection instead
+    auto lk = FEXCore::GuardSignalDeferringSectionWithFallback(VMATracking.Mutex, Thread);
 
     static uint64_t AnonSharedId = 1;
 
@@ -244,9 +243,9 @@ void SyscallHandler::TrackMunmap(FEXCore::Core::InternalThreadState *Thread, uin
 
   {
     // Frontend calls this with nullptr Thread during initialization.
-    // This is why `ScopedPotentialDeferredSignalWithUniqueLock` is used here.
+    // This is why `GuardSignalDeferringSectionWithFallback` is used here.
     // To be more optimal the frontend should provide this code with a valid Thread object earlier.
-    FEXCore::ScopedPotentialDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSectionWithFallback(VMATracking.Mutex, Thread);
 
     VMATracking.ClearUnsafe(CTX, Base, Size);
   }
@@ -260,7 +259,7 @@ void SyscallHandler::TrackMprotect(FEXCore::Core::InternalThreadState *Thread, u
   Size = FEXCore::AlignUp(Size, FHU::FEX_PAGE_SIZE);
 
   {
-    FEXCore::ScopedDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
 
     VMATracking.ChangeUnsafe(Base, Size, VMAProt::fromProt(Prot));
   }
@@ -275,7 +274,7 @@ void SyscallHandler::TrackMremap(FEXCore::Core::InternalThreadState *Thread, uin
   NewSize = FEXCore::AlignUp(NewSize, FHU::FEX_PAGE_SIZE);
 
   {
-    FEXCore::ScopedDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
 
     const auto OldVMA = VMATracking.LookupVMAUnsafe(OldAddress);
 
@@ -333,7 +332,7 @@ void SyscallHandler::TrackShmat(FEXCore::Core::InternalThreadState *Thread, int
   uint64_t Length = stat.shm_segsz;
 
   {
-    FEXCore::ScopedDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
 
     // TODO
     MRID mrid{SpecialDev::SHM, static_cast<uint64_t>(shmid)};
@@ -355,7 +354,7 @@ void SyscallHandler::TrackShmat(FEXCore::Core::InternalThreadState *Thread, int
 void SyscallHandler::TrackShmdt(FEXCore::Core::InternalThreadState *Thread, uintptr_t Base) {
   uintptr_t Length = 0;
   {
-    FEXCore::ScopedDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
 
     Length = VMATracking.ClearShmUnsafe(CTX, Base);
   }
@@ -369,7 +368,7 @@ void SyscallHandler::TrackShmdt(FEXCore::Core::InternalThreadState *Thread, uint
 void SyscallHandler::TrackMadvise(FEXCore::Core::InternalThreadState *Thread, uintptr_t Base, uintptr_t Size, int advice) {
   Size = FEXCore::AlignUp(Size, FHU::FEX_PAGE_SIZE);
   {
-    FEXCore::ScopedDeferredSignalWithForkableUniqueLock lk(VMATracking.Mutex, Thread);
+    auto lk = FEXCore::GuardSignalDeferringSection(VMATracking.Mutex, Thread);
     // TODO
   }
 }

Source/Tools/FEXLoader/TestHarnessRunner.cpp

@@ -245,8 +245,6 @@ int main(int argc, char **argv, char **const envp) {
 
   auto CTX = FEXCore::Context::Context::CreateNewContext();
 
-  CTX->InitializeContext();
-
 #ifndef _WIN32
   auto SignalDelegation = FEX::HLE::CreateSignalDelegator(CTX.get(), {});
 #else
@@ -303,9 +301,9 @@ int main(int argc, char **argv, char **const envp) {
     CTX->SetSignalDelegator(SignalDelegation.get());
     CTX->SetSyscallHandler(SyscallHandler.get());
 
-    bool Result1 = CTX->InitCore(Loader.DefaultRIP(), Loader.GetStackPointer());
+    auto ParentThread = CTX->InitCore(Loader.DefaultRIP(), Loader.GetStackPointer());
 
-    if (!Result1) {
+    if (!ParentThread) {
       return 1;
     }
 
@@ -315,7 +313,7 @@ int main(int argc, char **argv, char **const envp) {
     }
 
     // Just re-use compare state. It also checks against the expected values in config.
-    CTX->GetCPUState(&State);
+    memcpy(&State, &ParentThread->CurrentFrame->State, sizeof(State));
 
     SyscallHandler.reset();
   }

Source/Tools/FEXRootFSFetcher/Main.cpp

@@ -212,6 +212,7 @@ namespace WorkingAppsTester {
 
   // EroFS specific
   static bool Has_EroFSFuse {false};
+  static bool Has_EroFSFsck {false};
 
   void CheckCurl() {
     // Check if curl exists on the host
@@ -295,11 +296,23 @@ namespace WorkingAppsTester {
     Has_EroFSFuse = Result != -1;
   }
 
+  void CheckEroFSFsck() {
+    std::vector<const char*> ExecveArgs = {
+      "fsck.erofs",
+      "-V",
+      nullptr,
+    };
+
+    int32_t Result = Exec::ExecAndWaitForResponseRedirect(ExecveArgs[0], const_cast<char* const*>(ExecveArgs.data()), -1, -1);
+    Has_EroFSFsck = Result != -1;
+  }
+
   void Init() {
     CheckCurl();
     CheckSquashfuse();
     CheckUnsquashfs();
     CheckEroFSFuse();
+    CheckEroFSFsck();
   }
 }
 
@@ -476,12 +489,9 @@ namespace WebFileFetcher {
   const static std::string DownloadURL = "https://rootfs.fex-emu.gg/RootFS_links.json";
 
   std::string DownloadToString(const std::string &URL) {
-    std::string BigArgs =
-    fmt::format("curl {}", URL);
     std::vector<const char*> ExecveArgs = {
-      "/bin/sh",
-      "-c",
-      BigArgs.c_str(),
+      "curl",
+      URL.c_str(),
       nullptr,
     };
 
@@ -492,12 +502,11 @@ namespace WebFileFetcher {
     auto filename = URL.substr(URL.find_last_of('/') + 1);
     auto PathName = Path + filename;
 
-    std::string BigArgs =
-    fmt::format("curl {} -o {}", URL, PathName);
     std::vector<const char*> ExecveArgs = {
-      "/bin/sh",
-      "-c",
-      BigArgs.c_str(),
+      "curl",
+      URL.c_str(),
+      "-o",
+      PathName.c_str(),
       nullptr,
     };
 
@@ -1091,7 +1100,7 @@ namespace UnSquash {
     bool Extract = true;
     std::error_code ec;
     if (std::filesystem::exists(TargetFolder, ec)) {
-      fextl::string Question = FolderName + " Already exists. Overwrite?";
+      fextl::string Question = "Target folder \"" + FolderName + "\" already exists. Overwrite?";
       if (AskForConfirmation(Question)) {
         if (std::filesystem::remove_all(TargetFolder, ec) != ~0ULL) {
           Extract = true;
@@ -1114,6 +1123,40 @@ namespace UnSquash {
 
     return false;
   }
+
+  bool ExtractEroFS(const fextl::string &Path, const fextl::string &RootFS, const fextl::string &FolderName) {
+    auto TargetFolder = Path + FolderName;
+
+    bool Extract = true;
+    std::error_code ec;
+    if (std::filesystem::exists(TargetFolder, ec)) {
+      fextl::string Question = "Target folder \"" + FolderName + "\" already exists. Overwrite?";
+      if (AskForConfirmation(Question)) {
+        if (std::filesystem::remove_all(TargetFolder, ec) != ~0ULL) {
+          Extract = true;
+        }
+        if (ec) {
+          ExecWithInfo("Couldn't remove previous directory. Won't extract.");
+        }
+      }
+    }
+
+    if (Extract) {
+      ExecWithInfo("Extracting Erofs. This might take a few minutes.");
+
+      const auto ExtractOption = fmt::format("--extract={}", TargetFolder);
+      const std::vector<const char*> ExecveArgs = {
+        "fsck.erofs",
+        ExtractOption.c_str(),
+        RootFS.c_str(),
+        nullptr,
+      };
+
+      return Exec::ExecAndWaitForResponse(ExecveArgs[0], const_cast<char* const*>(ExecveArgs.data())) == 0;
+    }
+
+    return false;
+  }
 }
 
 int main(int argc, char **argv, char **const envp) {
@@ -1240,62 +1283,91 @@ int main(int argc, char **argv, char **const envp) {
         }
       }
 
+      struct ExtractStrings {
+        char const *ExtractOrAsIs;
+        char const *AsIsSinceMounterNonFunctional;
+        char const *AsIsSinceExtractorNonFunctional;
+        char const *AsIsSinceNothingWorks;
+      };
+
+      ArgOptions::CompressedImageOption UseImageAs {ArgOptions::CompressedUsageOption};
+      bool HasExtractor{};
+      bool HasMounter{};
+      std::function<bool (const fextl::string &Path, const fextl::string &RootFS, const fextl::string &FolderName)> ExtractHelper;
+      ExtractStrings ExtractingStrings;
+      if (Target.Type == WebFileFetcher::FileTargets::FileType::TYPE_SQUASHFS) {
+        HasExtractor = WorkingAppsTester::Has_Unsquashfs;
+        HasMounter = WorkingAppsTester::Has_Squashfuse;
+        ExtractHelper = UnSquash::UnsquashRootFS;
+        ExtractingStrings =
+          {
+            "Do you wish to extract the squashfs file or use it as-is?",
+            "Squashfuse doesn't work. Do you wish to extract the squashfs file?",
+            "Unsquashfs doesn't work. Do you want to use the squashfs file as-is?",
+            "Unsquashfs and squashfuse isn't working. Leaving rootfs as-is",
+          };
+      }
+      else if (Target.Type == WebFileFetcher::FileTargets::FileType::TYPE_EROFS) {
+        HasExtractor = WorkingAppsTester::Has_EroFSFsck;
+        HasMounter = WorkingAppsTester::Has_EroFSFuse;
+        ExtractHelper = UnSquash::ExtractEroFS;
+        ExtractingStrings =
+          {
+            "Do you wish to extract the erofs file or use it as-is?",
+            "erofsfuse doesn't work. Do you wish to extract the erofs file?",
+            "Extracting erofs doesn't work. Do you want to use the erofs file as-is?",
+            "Extracting erofs and erofsfuse isn't working. Leaving rootfs as-is",
+          };
+      }
+
+      int32_t Result{};
       std::vector<fextl::string> Args = {
         "Extract",
         "As-Is",
       };
 
-      ArgOptions::CompressedImageOption UseImageAs {ArgOptions::CompressedUsageOption};
-      if (Target.Type == WebFileFetcher::FileTargets::FileType::TYPE_SQUASHFS) {
-        int32_t Result{};
-        if (UseImageAs == ArgOptions::CompressedImageOption::OPTION_ASK) {
-          if (WorkingAppsTester::Has_Unsquashfs) {
-            if (WorkingAppsTester::Has_Squashfuse) {
-              Result = AskForConfirmationList("Do you wish to extract the squashfs file or use it as-is?", Args);
-              if (Result == 0) {
-                UseImageAs = ArgOptions::CompressedImageOption::OPTION_EXTRACT;
-              }
-              else if (Result == 1) {
-                UseImageAs = ArgOptions::CompressedImageOption::OPTION_ASIS;
-              }
+      if (UseImageAs == ArgOptions::CompressedImageOption::OPTION_ASK) {
+        if (HasExtractor) {
+          if (HasMounter) {
+            Result = AskForConfirmationList(ExtractingStrings.ExtractOrAsIs, Args);
+            if (Result == 0) {
+              UseImageAs = ArgOptions::CompressedImageOption::OPTION_EXTRACT;
             }
-            else {
-              Args.pop_back();
-              Result = AskForConfirmationList("Squashfuse doesn't work. Do you wish to extract the squashfs file?", Args);
-              if (Result == 0) {
-                UseImageAs = ArgOptions::CompressedImageOption::OPTION_EXTRACT;
-              }
-            }
-          }
-          else {
-            if (WorkingAppsTester::Has_Squashfuse) {
-              Args.erase(Args.begin());
-              Result = AskForConfirmationList("Unsquashfs doesn't work. Do you want to use the squashfs file as-is?", Args);
-              if (Result == 0) {
-                // We removed an argument, Just change "As-Is" from 0 to 1 for later logic to work
-                UseImageAs = ArgOptions::CompressedImageOption::OPTION_ASIS;
-              }
-            }
-            else {
-              Args.erase(Args.begin());
-              ExecWithInfo("Unsquashfs and squashfuse isn't working. Leaving rootfs as-is");
+            else if (Result == 1) {
               UseImageAs = ArgOptions::CompressedImageOption::OPTION_ASIS;
             }
           }
+          else {
+            Args.pop_back();
+            Result = AskForConfirmationList(ExtractingStrings.AsIsSinceMounterNonFunctional, Args);
+            if (Result == 0) {
+              UseImageAs = ArgOptions::CompressedImageOption::OPTION_EXTRACT;
+            }
+          }
         }
-
-        if (UseImageAs == ArgOptions::CompressedImageOption::OPTION_EXTRACT) {
-          auto FolderName = filename.substr(0, filename.find_last_of('.'));
-          if (UnSquash::UnsquashRootFS(RootFS, PathName, FolderName)) {
-            // Remove the .sqsh suffix since we extracted to that
-            filename = FolderName;
+        else {
+          if (HasMounter) {
+            Args.erase(Args.begin());
+            Result = AskForConfirmationList(ExtractingStrings.AsIsSinceExtractorNonFunctional, Args);
+            if (Result == 0) {
+              // We removed an argument, Just change "As-Is" from 0 to 1 for later logic to work
+              UseImageAs = ArgOptions::CompressedImageOption::OPTION_ASIS;
+            }
+          }
+          else {
+            Args.erase(Args.begin());
+            ExecWithInfo(ExtractingStrings.AsIsSinceNothingWorks);
+            UseImageAs = ArgOptions::CompressedImageOption::OPTION_ASIS;
           }
         }
       }
-      else if (Target.Type == WebFileFetcher::FileTargets::FileType::TYPE_EROFS) {
-        // Once erofs tooling is available for easy extraction, offer the same settings to extract as squashfs.
-        // Currently this is unavailable, would need a mount + copy + unmount dance.
-        UseImageAs = ArgOptions::CompressedImageOption::OPTION_ASIS;
+
+      if (UseImageAs == ArgOptions::CompressedImageOption::OPTION_EXTRACT) {
+        auto FolderName = filename.substr(0, filename.find_last_of('.'));
+        if (ExtractHelper(RootFS, PathName, FolderName)) {
+          // Remove the image file suffix since we extracted to that.
+          filename = FolderName;
+        }
       }
 
       if (AskForConfirmation("Do you wish to set this RootFS as default?")) {

Source/Windows/Defs/wow64.def

@@ -22,12 +22,13 @@ EXPORTS
   Wow64PassExceptionToGuest @16
   Wow64PrepareForDebuggerAttach @17 PRIVATE
   Wow64PrepareForException @18
-  Wow64RaiseException @19
-  Wow64ShallowThunkAllocObjectAttributes32TO64_FNC @20 PRIVATE
-  Wow64ShallowThunkAllocSecurityQualityOfService32TO64_FNC @21 PRIVATE
-  Wow64ShallowThunkSIZE_T32TO64 @22 PRIVATE
-  Wow64ShallowThunkSIZE_T64TO32 @23 PRIVATE
-  Wow64SuspendLocalThread @24
-  Wow64SystemServiceEx @25
-  Wow64ValidateUserCallTarget @26 PRIVATE
-  Wow64ValidateUserCallTargetFilter @27 PRIVATE
+  Wow64ProcessPendingCrossProcessItems @19
+  Wow64RaiseException @20
+  Wow64ShallowThunkAllocObjectAttributes32TO64_FNC @21 PRIVATE
+  Wow64ShallowThunkAllocSecurityQualityOfService32TO64_FNC @22 PRIVATE
+  Wow64ShallowThunkSIZE_T32TO64 @23 PRIVATE
+  Wow64ShallowThunkSIZE_T64TO32 @24 PRIVATE
+  Wow64SuspendLocalThread @25 PRIVATE
+  Wow64SystemServiceEx @26
+  Wow64ValidateUserCallTarget @27 PRIVATE
+  Wow64ValidateUserCallTargetFilter @28 PRIVATE

Source/Windows/WOW64/Module.cpp

@@ -35,6 +35,7 @@ $end_info$
 #include <atomic>
 #include <mutex>
 #include <utility>
+#include <unordered_set>
 #include <ntstatus.h>
 #include <windef.h>
 #include <winternl.h>
@@ -94,6 +95,7 @@ namespace {
   SYSTEM_CPU_INFORMATION CpuInfo{};
 
   std::mutex ThreadSuspendLock;
+  std::unordered_set<DWORD> InitializedWOWThreads; // Set of TIDs, `ThreadSuspendLock` must be locked when accessing
 
   std::pair<NTSTATUS, TLS> GetThreadTLS(HANDLE Thread) {
     THREAD_BASIC_INFORMATION Info;
@@ -179,7 +181,7 @@ namespace Context {
     Context->Esp = State.gregs[FEXCore::X86State::REG_RSP];
 
     Context->Eip = State.rip;
-    Context->EFlags = CTX->ReconstructCompactedEFLAGS(Thread);
+    Context->EFlags = CTX->ReconstructCompactedEFLAGS(Thread, false, nullptr, 0);
 
     Context->SegEs = State.es_idx;
     Context->SegCs = State.cs_idx;
@@ -460,6 +462,7 @@ public:
       const uint64_t EntryRAX = Frame->State.gregs[FEXCore::X86State::REG_RAX];
 
       Context::UnlockJITContext();
+      Wow64ProcessPendingCrossProcessItems();
       ReturnRAX = static_cast<uint64_t>(Wow64SystemServiceEx(static_cast<UINT>(EntryRAX),
                                                              reinterpret_cast<UINT *>(ReturnRSP + 4)));
       Context::LockJITContext();
@@ -512,7 +515,6 @@ void BTCpuProcessInit() {
   SyscallHandler = fextl::make_unique<WowSyscallHandler>();
 
   CTX = FEXCore::Context::Context::CreateNewContext();
-  CTX->InitializeContext();
   CTX->SetSignalDelegator(SignalDelegator.get());
   CTX->SetSyscallHandler(SyscallHandler.get());
   CTX->InitCore(0, 0);
@@ -552,8 +554,10 @@ void BTCpuProcessInit() {
 }
 
 NTSTATUS BTCpuThreadInit() {
-  GetTLS().ThreadState() = CTX->CreateThread(nullptr, 0);
+  GetTLS().ThreadState() = CTX->CreateThread(0, 0);
 
+  std::scoped_lock Lock(ThreadSuspendLock);
+  InitializedWOWThreads.emplace(GetCurrentThreadId());
   return STATUS_SUCCESS;
 }
 
@@ -563,6 +567,17 @@ NTSTATUS BTCpuThreadTerm(HANDLE Thread) {
     return Err;
   }
 
+  {
+    THREAD_BASIC_INFORMATION Info;
+    if (NTSTATUS Err = NtQueryInformationThread(Thread, ThreadBasicInformation, &Info, sizeof(Info), nullptr); Err) {
+      return Err;
+    }
+
+    const auto ThreadTID = reinterpret_cast<uint64_t>(Info.ClientId.UniqueThread);
+    std::scoped_lock Lock(ThreadSuspendLock);
+    InitializedWOWThreads.erase(ThreadTID);
+  }
+
   CTX->DestroyThread(TLS.ThreadState());
   return STATUS_SUCCESS;
 }
@@ -665,6 +680,11 @@ NTSTATUS BTCpuSuspendLocalThread(HANDLE Thread, ULONG *Count) {
   }
 
   std::scoped_lock Lock(ThreadSuspendLock);
+
+  // If the thread hasn't yet been initialized, suspend it without special handling as it wont yet have entered the JIT
+  if (!InitializedWOWThreads.contains(ThreadTID))
+    return NtSuspendThread(Thread, Count);
+
   // If CONTROL_IN_JIT is unset at this point, then it can never be set (and thus the JIT cannot be reentered) as
   // CONTROL_PAUSED has been set, as such, while this may redundantly request interrupts in rare cases it will never
   // miss them

Source/Windows/include/winternl.h

@@ -90,6 +90,7 @@ typedef enum _MEMORY_INFORMATION_CLASS {
 } MEMORY_INFORMATION_CLASS;
 
 NTSTATUS WINAPI Wow64SystemServiceEx(UINT,UINT*);
+void WINAPI Wow64ProcessPendingCrossProcessItems(void);
 
 NTSTATUS WINAPI RtlWow64SetThreadContext(HANDLE,const WOW64_CONTEXT*);
 NTSTATUS WINAPI RtlWow64GetThreadContext(HANDLE,WOW64_CONTEXT*);

ThunkLibs/libX11/libX11_Guest.cpp

@@ -232,14 +232,6 @@ extern "C" {
       return rv;
     }
 
-    static void LockMutexFunction(LockInfoPtr) {
-      fprintf(stderr, "libX11: LockMutex\n");
-    }
-
-    static void UnlockMutexFunction(LockInfoPtr) {
-      fprintf(stderr, "libX11: LockMutex\n");
-    }
-
   int XFree(void* ptr) {
     // This function must be able to handle both guest heap pointers *and* host heap pointers,
     // so it only forwards to the native host library for the latter.
@@ -368,8 +360,8 @@ extern "C" {
     return fexfn_pack_XUnregisterIMInstantiateCallback(dpy, rdb, res_name, res_class, AllocateHostTrampolineForGuestFunction(callback), client_data);
   }
 
-  void (*_XLockMutex_fn)(LockInfoPtr) = LockMutexFunction;
-  void (*_XUnlockMutex_fn)(LockInfoPtr) = UnlockMutexFunction;
+  void (*_XLockMutex_fn)(LockInfoPtr) = nullptr;
+  void (*_XUnlockMutex_fn)(LockInfoPtr) = nullptr;
   LockInfoPtr _Xglobal_lock = (LockInfoPtr)0x4142434445464748ULL;
 }
 

docs/SourceOutline.md

@@ -1,4 +1,4 @@
-# FEX-2311
+# FEX-2312.1
 
 ## FEXCore
 See [FEXCore/Readme.md](../FEXCore/Readme.md) for more details
@@ -64,6 +64,10 @@ Metadata that drives the frontend x86/64 decoding
 - [X87.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher/X87.cpp): Handles x86/64 x87 to IR
 - [X87F64.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp): Handles x86/64 x87 to IR
 - [OpcodeDispatcher.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_BACKUP_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_BACKUP_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_BASE_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_BASE_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_LOCAL_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_LOCAL_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_REMOTE_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_REMOTE_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
 
 
 
@@ -77,10 +81,6 @@ Logic that binds various parts together
 Emulation mainloop related glue logic
 - [Core.cpp](../FEXCore/Source/Interface/Core/Core.cpp): Glues Frontend, OpDispatcher and IR Opts & Compilation, LookupCache, Dispatcher and provides the Execution loop entrypoint
 
-#### gdbserver
-- [GdbServer.cpp](../FEXCore/Source/Interface/Core/GdbServer.cpp): Provides a gdb interface to the guest state
-- [GdbServer.h](../FEXCore/Source/Interface/Core/GdbServer.h)
-
 #### log-manager
 - [LogManager.cpp](../FEXCore/Source/Utils/LogManager.cpp)
 
@@ -143,6 +143,10 @@ Text -> IR
 - [X87.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher/X87.cpp): Handles x86/64 x87 to IR
 - [X87F64.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp): Handles x86/64 x87 to IR
 - [OpcodeDispatcher.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_BACKUP_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_BACKUP_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_BASE_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_BASE_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_LOCAL_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_LOCAL_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
+- [OpcodeDispatcher_REMOTE_124790.cpp](../FEXCore/Source/Interface/Core/OpcodeDispatcher_REMOTE_124790.cpp): Handles x86/64 ops to IR, no-pf opt, local-flags opt
 
 ## ThunkLibs
 See [ThunkLibs/README.md](../ThunkLibs/README.md) for more details

unittests/32Bit_ASM/FEX_bugs/IMUL_garbagedata.asm

@@ -0,0 +1,51 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "RAX": "0x000000007dbf2800",
+    "RDX": "0x0000000000000000",
+    "RBX": "0x000000000000004f",
+    "RCX": "0x000000000000004f",
+    "RBP": "0x0000000000009e4f",
+    "RSI": "0x0000000000009e4f",
+    "RSP": "0x000000000000004f"
+  },
+  "Mode": "32BIT"
+}
+%endif
+
+; FEX had a bug where smaller than 64-bit imul could leave garbage data in the upper 32-bits of the 32-bit result.
+; This would cause subsequent instructions after the imul to receive garbage bits.
+; In particular this would feed in to address calculation in DXVK with "Dungeon Defenders" doing address calculation.
+; The address calculation did something similar to:
+;   xor edx, edx
+;   mov eax, 0x7dbf2800
+;   imul ebx, ebx, 0xaaaaaaab
+;   div ebx
+; Divide expected 0x4f but received 0xffffffb1'0000'004f
+
+; Dividend
+xor edx, edx
+mov eax, 0x7dbf2800
+
+; Multiply starting value
+mov ebx, 0xED
+
+jmp .test
+
+.test:
+
+; imul 1-src
+mov edi, 0xaaaaaaab
+imul di, bx
+mov esp, 0xaaaaaaab
+imul esp, ebx
+
+; imul 2-src 8-bit check
+imul bp, bx, 0xab
+imul esi, ebx, 0xab
+
+; imul 2-src 16-bit check
+imul cx, bx, 0xaaab
+imul ebx, ebx, 0xaaaaaaab
+
+hlt

unittests/ASM/FEX_bugs/BZHI_Sign.asm

@@ -0,0 +1,45 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "RAX": "0x4",
+    "RBX": "0xFFFFFFFFFFFFFFF4",
+    "RCX": "0x0",
+    "RDX": "0x1337"
+  }
+}
+%endif
+
+; FEX had a bug where bzhi would fail to update SF. Test that bzhi correctly
+; sets ZF/SF correctly based on the result.
+
+mov rcx, 4
+mov rbx, -12
+
+; Result is 0x4
+bzhi rax, rbx, rcx
+mov rdx, 0xdead1
+jz .fail
+mov rdx, 0xdead2
+js .fail
+
+; Result is -12
+mov rcx, 64
+bzhi rdx, rbx, rcx
+mov rdx, 0xdead3
+jz .fail
+mov rdx, 0xdead4
+jns .fail
+
+; Result is 0x00
+mov rdx, 0
+bzhi rcx, rbx, rdx
+mov rdx, 0xdead5
+jnz .fail
+mov rdx, 0xdead6
+js .fail
+
+mov rdx, 0x1337
+hlt
+
+.fail:
+hlt

unittests/ASM/FEX_bugs/ShiftConstantBug.asm

@@ -0,0 +1,15 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "RAX": "0x500000020"
+  }
+}
+%endif
+
+; FEX had a bug in its `TestNZ` opcode where it would try to load a constant in to the tst instruction
+; If the constant didn't fit in a logical encoding it would generate invalid instructions and also crash.
+; This snippet of code was found in libGLX.so.0.0.0 when trying to load steamwebhelper.
+mov     eax, 0x28000001
+shl     rax, 0x5
+
+hlt

unittests/ASM/FEX_bugs/nzcv_spill_enderlilies.asm

@@ -0,0 +1,49 @@
+%ifdef CONFIG
+{
+  "RegData": {
+    "RAX": "0",
+    "XMM0": ["0", "0"]
+  }
+}
+%endif
+
+; FEX-Emu has a bug around NZCV flags getting spilled and filled.
+; The bug comes down to NZCV actually being 32-bit but our IR incorrectly assumed that all flags were 8-bit.
+; Once a spill situation happened, it would only store and reload the lower 8-bits of the NZCV flag which wasn't correct.
+; This caused this code to infinite loop and read past memory and crash.
+
+; Code found from Ender Lilies in their `sha1_block_data_order` function which is significantly longer than this snippit.
+lea rsi, [rel .data_vecs]
+mov rax, 1
+
+; Break visibility
+jmp loop_top
+loop_top:
+
+; Decrement counter.
+dec     rax
+
+; Load rsi + 0x40 in to rbx
+lea     rbx, [rsi+0x40]
+
+; Move rbx in to rsi, incrementing the pointer by 64-bytes if rax isn't zero.
+cmovne  rsi, rbx
+
+; Do a sha1rnds4, which uses enough temporaries to spill NZCV which picks up a crash.
+sha1rnds4 xmm0, xmm0, 0x0
+
+; This memory access will crash once we loop too many times.
+movdqu  xmm0, [rsi]
+
+; Jump back to the top
+jne     loop_top
+
+hlt
+
+.data_vecs:
+dq 0, 0, 0, 0
+dq 0, 0, 0, 0
+dq 0, 0, 0, 0
+dq 0, 0, 0, 0
+dq 0, 0, 0, 0
+dq 0, 0, 0, 0

unittests/InstructionCountCI/AFP/H0F3A.json

@@ -12,7 +12,6 @@
   "Instructions": {
     "roundss xmm0, xmm1, 00000000b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "Nearest rounding",
         "0x66 0x0f 0x3a 0x0a"
@@ -23,7 +22,6 @@
     },
     "roundss xmm0, xmm1, 00000001b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "-inf rounding",
         "0x66 0x0f 0x3a 0x0a"
@@ -34,7 +32,6 @@
     },
     "roundss xmm0, xmm1, 00000010b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "+inf rounding",
         "0x66 0x0f 0x3a 0x0a"
@@ -45,7 +42,6 @@
     },
     "roundss xmm0, xmm1, 00000011b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "truncate rounding",
         "0x66 0x0f 0x3a 0x0a"
@@ -56,7 +52,6 @@
     },
     "roundss xmm0, xmm1, 00000100b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "host rounding mode rounding",
         "0x66 0x0f 0x3a 0x0a"
@@ -67,7 +62,6 @@
     },
     "roundsd xmm0, xmm1, 00000000b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "Nearest rounding",
         "0x66 0x0f 0x3a 0x0b"
@@ -78,7 +72,6 @@
     },
     "roundsd xmm0, xmm1, 00000001b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "-inf rounding",
         "0x66 0x0f 0x3a 0x0b"
@@ -89,7 +82,6 @@
     },
     "roundsd xmm0, xmm1, 00000010b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "+inf rounding",
         "0x66 0x0f 0x3a 0x0b"
@@ -100,7 +92,6 @@
     },
     "roundsd xmm0, xmm1, 00000011b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "truncate rounding",
         "0x66 0x0f 0x3a 0x0b"
@@ -111,7 +102,6 @@
     },
     "roundsd xmm0, xmm1, 00000100b": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "host rounding mode rounding",
         "0x66 0x0f 0x3a 0x0b"

unittests/InstructionCountCI/AFP/SVE256/Secondary.json

@@ -11,7 +11,6 @@
   "Instructions": {
     "cvtpi2ps xmm0, [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x2a"
       ],
@@ -22,12 +21,11 @@
     },
     "cvtpi2ps xmm0, mm0": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x2a"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #752]",
+        "ldr d2, [x28, #768]",
         "scvtf v16.2s, v2.2s"
       ]
     }

unittests/InstructionCountCI/AFP/SVE256/Secondary_REP.json

@@ -11,7 +11,6 @@
   "Instructions": {
     "cvtsi2ss xmm0, eax": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x2a"
       ],
@@ -21,7 +20,6 @@
     },
     "cvtsi2ss xmm0, dword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x2a"
       ],
@@ -32,7 +30,6 @@
     },
     "cvtsi2ss xmm0, qword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x2a"
       ],
@@ -43,7 +40,6 @@
     },
     "sqrtss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": "0xf3 0x0f 0x51",
       "ExpectedArm64ASM": [
         "fsqrt s16, s17"
@@ -51,7 +47,6 @@
     },
     "rsqrtss xmm0, xmm1": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "FEAT_FPRES could make this more optimal",
         "0xf3 0x0f 0x52"
@@ -64,7 +59,6 @@
     },
     "rcpss xmm0, xmm1": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "FEAT_FPRES could make this more optimal",
         "0xf3 0x0f 0x53"
@@ -76,7 +70,6 @@
     },
     "addss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x58"
       ],
@@ -86,7 +79,6 @@
     },
     "mulss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x59"
       ],
@@ -96,7 +88,6 @@
     },
     "cvtss2sd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": "0xf3 0x0f 0x5a",
       "ExpectedArm64ASM": [
         "fcvt d16, s17"
@@ -104,7 +95,6 @@
     },
     "cvtss2sd xmm0, [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": "0xf3 0x0f 0x5a",
       "ExpectedArm64ASM": [
         "ldr d2, [x4]",
@@ -113,7 +103,6 @@
     },
     "subss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5c"
       ],
@@ -123,7 +112,6 @@
     },
     "minss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5d"
       ],
@@ -133,7 +121,6 @@
     },
     "divss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5e"
       ],
@@ -143,7 +130,6 @@
     },
     "maxss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5f"
       ],
@@ -153,7 +139,6 @@
     },
     "cmpss xmm0, xmm1, 0": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -163,7 +148,6 @@
     },
     "cmpss xmm0, xmm1, 1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -173,7 +157,6 @@
     },
     "cmpss xmm0, xmm1, 2": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -183,7 +166,6 @@
     },
     "cmpss xmm0, xmm1, 3": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -198,7 +180,6 @@
     },
     "cmpss xmm0, xmm1, 4": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -211,7 +192,6 @@
     },
     "cmpss xmm0, xmm1, 5": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -223,7 +203,6 @@
     },
     "cmpss xmm0, xmm1, 6": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -235,7 +214,6 @@
     },
     "cmpss xmm0, xmm1, 7": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],

unittests/InstructionCountCI/AFP/SVE256/Secondary_REPNE.json

@@ -11,7 +11,6 @@
   "Instructions": {
     "cvtsi2sd xmm0, eax": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -21,7 +20,6 @@
     },
     "cvtsi2sd xmm0, dword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -32,7 +30,6 @@
     },
     "cvtsi2sd xmm0, rax": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -42,7 +39,6 @@
     },
     "cvtsi2sd xmm0, qword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -53,7 +49,6 @@
     },
     "sqrtsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x51"
       ],
@@ -63,7 +58,6 @@
     },
     "addsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x58"
       ],
@@ -73,7 +67,6 @@
     },
     "mulsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x59"
       ],
@@ -83,7 +76,6 @@
     },
     "cvtsd2ss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5a"
       ],
@@ -93,7 +85,6 @@
     },
     "cvtsd2ss xmm0, [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5a"
       ],
@@ -104,7 +95,6 @@
     },
     "subsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5c"
       ],
@@ -114,7 +104,6 @@
     },
     "minsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5d"
       ],
@@ -124,7 +113,6 @@
     },
     "divsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5e"
       ],
@@ -134,7 +122,6 @@
     },
     "maxsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5f"
       ],
@@ -144,7 +131,6 @@
     },
     "cmpsd xmm0, xmm1, 0": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -154,7 +140,6 @@
     },
     "cmpsd xmm0, xmm1, 1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -164,7 +149,6 @@
     },
     "cmpsd xmm0, xmm1, 2": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -174,7 +158,6 @@
     },
     "cmpsd xmm0, xmm1, 3": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -189,7 +172,6 @@
     },
     "cmpsd xmm0, xmm1, 4": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -202,7 +184,6 @@
     },
     "cmpsd xmm0, xmm1, 5": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -214,7 +195,6 @@
     },
     "cmpsd xmm0, xmm1, 6": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -226,7 +206,6 @@
     },
     "cmpsd xmm0, xmm1, 7": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],

unittests/InstructionCountCI/AFP/Secondary.json

@@ -12,7 +12,6 @@
   "Instructions": {
     "cvtpi2ps xmm0, [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x2a"
       ],
@@ -23,12 +22,11 @@
     },
     "cvtpi2ps xmm0, mm0": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x2a"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #752]",
+        "ldr d2, [x28, #768]",
         "scvtf v16.2s, v2.2s"
       ]
     }

unittests/InstructionCountCI/AFP/Secondary_REP.json

@@ -12,7 +12,6 @@
   "Instructions": {
     "cvtsi2ss xmm0, eax": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x2a"
       ],
@@ -22,7 +21,6 @@
     },
     "cvtsi2ss xmm0, dword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x2a"
       ],
@@ -33,7 +31,6 @@
     },
     "cvtsi2ss xmm0, qword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x2a"
       ],
@@ -44,7 +41,6 @@
     },
     "sqrtss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": "0xf3 0x0f 0x51",
       "ExpectedArm64ASM": [
         "fsqrt s16, s17"
@@ -52,7 +48,6 @@
     },
     "rsqrtss xmm0, xmm1": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "FEAT_FPRES could make this more optimal",
         "0xf3 0x0f 0x52"
@@ -65,7 +60,6 @@
     },
     "rcpss xmm0, xmm1": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "FEAT_FPRES could make this more optimal",
         "0xf3 0x0f 0x53"
@@ -77,7 +71,6 @@
     },
     "addss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x58"
       ],
@@ -87,7 +80,6 @@
     },
     "mulss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x59"
       ],
@@ -97,7 +89,6 @@
     },
     "cvtss2sd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": "0xf3 0x0f 0x5a",
       "ExpectedArm64ASM": [
         "fcvt d16, s17"
@@ -105,7 +96,6 @@
     },
     "cvtss2sd xmm0, [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": "0xf3 0x0f 0x5a",
       "ExpectedArm64ASM": [
         "ldr d2, [x4]",
@@ -114,7 +104,6 @@
     },
     "subss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5c"
       ],
@@ -124,7 +113,6 @@
     },
     "minss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5d"
       ],
@@ -134,7 +122,6 @@
     },
     "divss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5e"
       ],
@@ -144,7 +131,6 @@
     },
     "maxss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0x5f"
       ],
@@ -154,7 +140,6 @@
     },
     "cmpss xmm0, xmm1, 0": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -164,7 +149,6 @@
     },
     "cmpss xmm0, xmm1, 1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -174,7 +158,6 @@
     },
     "cmpss xmm0, xmm1, 2": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -184,7 +167,6 @@
     },
     "cmpss xmm0, xmm1, 3": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -198,7 +180,6 @@
     },
     "cmpss xmm0, xmm1, 4": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -210,7 +191,6 @@
     },
     "cmpss xmm0, xmm1, 5": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -222,7 +202,6 @@
     },
     "cmpss xmm0, xmm1, 6": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],
@@ -234,7 +213,6 @@
     },
     "cmpss xmm0, xmm1, 7": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "0xf3 0x0f 0xc2"
       ],

unittests/InstructionCountCI/AFP/Secondary_REPNE.json

@@ -12,7 +12,6 @@
   "Instructions": {
     "cvtsi2sd xmm0, eax": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -22,7 +21,6 @@
     },
     "cvtsi2sd xmm0, dword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -33,7 +31,6 @@
     },
     "cvtsi2sd xmm0, rax": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -43,7 +40,6 @@
     },
     "cvtsi2sd xmm0, qword [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x2a"
       ],
@@ -54,7 +50,6 @@
     },
     "sqrtsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x51"
       ],
@@ -64,7 +59,6 @@
     },
     "addsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x58"
       ],
@@ -74,7 +68,6 @@
     },
     "mulsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x59"
       ],
@@ -84,7 +77,6 @@
     },
     "cvtsd2ss xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5a"
       ],
@@ -94,7 +86,6 @@
     },
     "cvtsd2ss xmm0, [rax]": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5a"
       ],
@@ -105,7 +96,6 @@
     },
     "subsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5c"
       ],
@@ -115,7 +105,6 @@
     },
     "minsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5d"
       ],
@@ -125,7 +114,6 @@
     },
     "divsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5e"
       ],
@@ -135,7 +123,6 @@
     },
     "maxsd xmm0, xmm1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0x5f"
       ],
@@ -145,7 +132,6 @@
     },
     "cmpsd xmm0, xmm1, 0": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -155,7 +141,6 @@
     },
     "cmpsd xmm0, xmm1, 1": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -165,7 +150,6 @@
     },
     "cmpsd xmm0, xmm1, 2": {
       "ExpectedInstructionCount": 1,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -175,7 +159,6 @@
     },
     "cmpsd xmm0, xmm1, 3": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -189,7 +172,6 @@
     },
     "cmpsd xmm0, xmm1, 4": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -201,7 +183,6 @@
     },
     "cmpsd xmm0, xmm1, 5": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -213,7 +194,6 @@
     },
     "cmpsd xmm0, xmm1, 6": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],
@@ -225,7 +205,6 @@
     },
     "cmpsd xmm0, xmm1, 7": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "0xf2 0x0f 0xc2"
       ],

unittests/InstructionCountCI/AFP/VEX_map1.json

@@ -11,7 +11,6 @@
   "Instructions": {
     "vsqrtss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x51 128-bit"
       ],
@@ -22,7 +21,6 @@
     },
     "vsqrtsd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x51 128-bit"
       ],
@@ -33,7 +31,6 @@
     },
     "vrsqrtss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "FEAT_FPRES could make this more optimal",
         "Map 1 0b10 0x52 128-bit"
@@ -47,7 +44,6 @@
     },
     "vrcpss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "FEAT_FPRES could make this more optimal",
         "Map 1 0b10 0x53 128-bit"
@@ -60,7 +56,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x00": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -71,7 +66,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x01": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -82,7 +76,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x02": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -93,7 +86,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x03": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -108,7 +100,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x04": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -121,7 +112,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x05": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -134,7 +124,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x06": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -147,7 +136,6 @@
     },
     "vcmpss xmm0, xmm1, xmm2, 0x07": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0xC2 128-bit"
       ],
@@ -161,7 +149,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x00": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -172,7 +159,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x01": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -183,7 +169,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x02": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -194,7 +179,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x03": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -209,7 +193,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x04": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -222,7 +205,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x05": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -235,7 +217,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x06": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -248,7 +229,6 @@
     },
     "vcmpsd xmm0, xmm1, xmm2, 0x07": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0xC2 128-bit"
       ],
@@ -262,7 +242,6 @@
     },
     "vcvtsi2ss xmm0, xmm1, eax": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x2A 128-bit"
       ],
@@ -273,7 +252,6 @@
     },
     "vcvtsi2ss xmm0, xmm1, rax": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x2A 128-bit"
       ],
@@ -284,7 +262,6 @@
     },
     "vcvtsi2sd xmm0, xmm1, eax": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x2A 128-bit"
       ],
@@ -295,7 +272,6 @@
     },
     "vcvtsi2sd xmm0, xmm1, rax": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x2A 128-bit"
       ],
@@ -306,7 +282,6 @@
     },
     "vmulss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x59 128-bit"
       ],
@@ -317,7 +292,6 @@
     },
     "vmulsd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x59 128-bit"
       ],
@@ -328,7 +302,6 @@
     },
     "vcvtss2sd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x5a 128-bit"
       ],
@@ -339,7 +312,6 @@
     },
     "vcvtsd2ss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x5a 128-bit"
       ],
@@ -350,7 +322,6 @@
     },
     "vsubss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x5c 128-bit"
       ],
@@ -361,7 +332,6 @@
     },
     "vsubsd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x5c 128-bit"
       ],
@@ -372,7 +342,6 @@
     },
     "vminss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x5d 128-bit"
       ],
@@ -383,7 +352,6 @@
     },
     "vminsd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x5d 128-bit"
       ],
@@ -394,7 +362,6 @@
     },
     "vdivss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x5e 128-bit"
       ],
@@ -405,7 +372,6 @@
     },
     "vdivsd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x5e 128-bit"
       ],
@@ -416,7 +382,6 @@
     },
     "vmaxss xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b10 0x5f 128-bit"
       ],
@@ -427,7 +392,6 @@
     },
     "vmaxsd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "Map 1 0b11 0x5f 128-bit"
       ],
@@ -438,7 +402,6 @@
     },
     "vminps xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "No",
       "Comment": [
         "Map 1 0b00 0x5d 128-bit"
       ],
@@ -450,7 +413,6 @@
     },
     "vminps ymm0, ymm1, ymm2": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "No",
       "Comment": [
         "Map 1 0b00 0x5d 256-bit"
       ],
@@ -464,7 +426,6 @@
     },
     "vminpd xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "No",
       "Comment": [
         "Map 1 0b01 0x5d 128-bit"
       ],
@@ -476,7 +437,6 @@
     },
     "vminpd ymm0, ymm1, ymm2": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "No",
       "Comment": [
         "Map 1 0b01 0x5d 256-bit"
       ],

unittests/InstructionCountCI/AFP/VEX_map3.json

@@ -11,7 +11,6 @@
   "Instructions": {
     "vroundss xmm0, xmm1, 00000000b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "nearest rounding",
         "Map 3 0b01 0x0a 128-bit"
@@ -23,7 +22,6 @@
     },
     "vroundss xmm0, xmm1, 00000001b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "-inf rounding",
         "Map 3 0b01 0x0a 128-bit"
@@ -35,7 +33,6 @@
     },
     "vroundss xmm0, xmm1, 00000010b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "+inf rounding",
         "Map 3 0b01 0x0a 128-bit"
@@ -47,7 +44,6 @@
     },
     "vroundss xmm0, xmm1, 00000011b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "truncate rounding",
         "Map 3 0b01 0x0a 128-bit"
@@ -59,7 +55,6 @@
     },
     "vroundss xmm0, xmm1, 00000100b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "host mode rounding",
         "Map 3 0b01 0x0a 128-bit"
@@ -71,7 +66,6 @@
     },
     "vroundsd xmm0, xmm1, 00000000b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "nearest rounding",
         "Map 3 0b01 0x0b 128-bit"
@@ -83,7 +77,6 @@
     },
     "vroundsd xmm0, xmm1, 00000001b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "-inf rounding",
         "Map 3 0b01 0x0b 128-bit"
@@ -95,7 +88,6 @@
     },
     "vroundsd xmm0, xmm1, 00000010b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "+inf rounding",
         "Map 3 0b01 0x0b 128-bit"
@@ -107,7 +99,6 @@
     },
     "vroundsd xmm0, xmm1, 00000011b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "truncate rounding",
         "Map 3 0b01 0x0b 128-bit"
@@ -119,7 +110,6 @@
     },
     "vroundsd xmm0, xmm1, 00000100b": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": [
         "host mode rounding",
         "Map 3 0b01 0x0b 128-bit"

unittests/InstructionCountCI/Crypto/H0F38.json

@@ -0,0 +1,138 @@
+{
+  "Features": {
+    "Bitness": 64,
+    "EnabledHostFeatures": [
+      "CRYPTO"
+    ],
+    "DisabledHostFeatures": [
+      "SVE128",
+      "SVE256",
+      "AFP"
+    ]
+  },
+  "Instructions": {
+    "sha1nexte xmm0, xmm1": {
+      "ExpectedInstructionCount": 6,
+      "Comment": [
+        "0x66 0x0f 0x38 0xc8"
+      ],
+      "ExpectedArm64ASM": [
+        "dup v2.4s, v16.s[3]",
+        "unimplemented (Unimplemented)",
+        "dup v2.4s, v2.s[0]",
+        "add v2.4s, v17.4s, v2.4s",
+        "mov v16.16b, v17.16b",
+        "mov v16.s[3], v2.s[3]"
+      ]
+    },
+    "sha256msg1 xmm0, xmm1": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0x66 0x0f 0x38 0xcc"
+      ],
+      "ExpectedArm64ASM": [
+        "unimplemented (Unimplemented)"
+      ]
+    },
+    "aesimc xmm0, xmm1": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0x66 0x0f 0x38 0xdb"
+      ],
+      "ExpectedArm64ASM": [
+        "unimplemented (Unimplemented)"
+      ]
+    },
+    "aesenc xmm0, xmm1": {
+      "ExpectedInstructionCount": 4,
+      "Comment": [
+        "0x66 0x0f 0x38 0xdc"
+      ],
+      "ExpectedArm64ASM": [
+        "movi v2.2d, #0x0",
+        "unimplemented (Unimplemented)",
+        "unimplemented (Unimplemented)",
+        "eor v16.16b, v16.16b, v17.16b"
+      ]
+    },
+    "aesenclast xmm0, xmm1": {
+      "ExpectedInstructionCount": 3,
+      "Comment": [
+        "0x66 0x0f 0x38 0xdd"
+      ],
+      "ExpectedArm64ASM": [
+        "movi v2.2d, #0x0",
+        "unimplemented (Unimplemented)",
+        "eor v16.16b, v16.16b, v17.16b"
+      ]
+    },
+    "aesdec xmm0, xmm1": {
+      "ExpectedInstructionCount": 4,
+      "Comment": [
+        "0x66 0x0f 0x38 0xde"
+      ],
+      "ExpectedArm64ASM": [
+        "movi v2.2d, #0x0",
+        "unimplemented (Unimplemented)",
+        "unimplemented (Unimplemented)",
+        "eor v16.16b, v16.16b, v17.16b"
+      ]
+    },
+    "aesdeclast xmm0, xmm1": {
+      "ExpectedInstructionCount": 3,
+      "Comment": [
+        "0x66 0x0f 0x38 0xdf"
+      ],
+      "ExpectedArm64ASM": [
+        "movi v2.2d, #0x0",
+        "unimplemented (Unimplemented)",
+        "eor v16.16b, v16.16b, v17.16b"
+      ]
+    },
+    "crc32 eax, bl": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0xf2 0x0f 0x38 0xf0"
+      ],
+      "ExpectedArm64ASM": [
+        "crc32cb w4, w4, w7"
+      ]
+    },
+    "crc32 eax, bx": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0xf2 0x0f 0x38 0xf1"
+      ],
+      "ExpectedArm64ASM": [
+        "crc32ch w4, w4, w7"
+      ]
+    },
+    "crc32 eax, ebx": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0xf2 0x0f 0x38 0xf1"
+      ],
+      "ExpectedArm64ASM": [
+        "crc32cw w4, w4, w7"
+      ]
+    },
+    "crc32 rax, bl": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0xf2 0x0f 0x38 0xf0"
+      ],
+      "ExpectedArm64ASM": [
+        "crc32cb w4, w4, w7"
+      ]
+    },
+    "crc32 rax, rbx": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0xf2 0x0f 0x38 0xf1"
+      ],
+      "ExpectedArm64ASM": [
+        "crc32cx w4, w4, x7"
+      ]
+    }
+  }
+}

unittests/InstructionCountCI/Crypto/H0F3A.json

@@ -0,0 +1,82 @@
+{
+  "Features": {
+    "Bitness": 64,
+    "EnabledHostFeatures": [
+      "CRYPTO"
+    ],
+    "DisabledHostFeatures": [
+      "SVE128",
+      "SVE256",
+      "AFP"
+    ]
+  },
+  "Instructions": {
+    "pclmulqdq xmm0, xmm1, 00000b": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0x66 0x0f 0x3a 0x44"
+      ],
+      "ExpectedArm64ASM": [
+        "unallocated (Unallocated)"
+      ]
+    },
+    "pclmulqdq xmm0, xmm1, 00001b": {
+      "ExpectedInstructionCount": 2,
+      "Comment": [
+        "0x66 0x0f 0x3a 0x44"
+      ],
+      "ExpectedArm64ASM": [
+        "dup v0.2d, v16.d[1]",
+        "unallocated (Unallocated)"
+      ]
+    },
+    "pclmulqdq xmm0, xmm1, 10000b": {
+      "ExpectedInstructionCount": 2,
+      "Comment": [
+        "0x66 0x0f 0x3a 0x44"
+      ],
+      "ExpectedArm64ASM": [
+        "dup v0.2d, v17.d[1]",
+        "unallocated (Unallocated)"
+      ]
+    },
+    "pclmulqdq xmm0, xmm1, 10001b": {
+      "ExpectedInstructionCount": 1,
+      "Comment": [
+        "0x66 0x0f 0x3a 0x44"
+      ],
+      "ExpectedArm64ASM": [
+        "unallocated (Unallocated)"
+      ]
+    },
+    "aeskeygenassist xmm0, xmm1, 0": {
+      "ExpectedInstructionCount": 5,
+      "Comment": [
+        "0x66 0x0f 0x3a 0xdf"
+      ],
+      "ExpectedArm64ASM": [
+        "ldr q2, [x28, #2080]",
+        "movi v3.2d, #0x0",
+        "mov v16.16b, v17.16b",
+        "unimplemented (Unimplemented)",
+        "tbl v16.16b, {v16.16b}, v2.16b"
+      ]
+    },
+    "aeskeygenassist xmm0, xmm1, 0xFF": {
+      "ExpectedInstructionCount": 8,
+      "Comment": [
+        "0x66 0x0f 0x3a 0xdf"
+      ],
+      "ExpectedArm64ASM": [
+        "ldr q2, [x28, #2080]",
+        "movi v3.2d, #0x0",
+        "mov v16.16b, v17.16b",
+        "unimplemented (Unimplemented)",
+        "tbl v16.16b, {v16.16b}, v2.16b",
+        "mov x0, #0xff00000000",
+        "dup v1.2d, x0",
+        "eor v16.16b, v16.16b, v1.16b"
+      ]
+    }
+  }
+}

unittests/InstructionCountCI/DDD.json

@@ -16,322 +16,294 @@
   "Instructions": {
     "pi2fw mm0, mm1": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x0c"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "uzp1 v2.4h, v2.4h, v2.4h",
         "sxtl v2.4s, v2.4h",
         "scvtf v2.2s, v2.2s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pi2fd mm0, mm1": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x0d"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "scvtf v2.2s, v2.2s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pf2iw mm0, mm1": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x1c"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "fcvtzs v2.2s, v2.2s",
         "uzp1 v2.4h, v2.4h, v2.4h",
         "sxtl v2.4s, v2.4h",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pf2id mm0, mm1": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x1d"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "fcvtzs v2.2s, v2.2s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfrcpv mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x86"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "fmov v0.4s, #0x70 (1.0000)",
         "fdiv v2.4s, v0.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfrsqrtv mm0, mm1": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x87"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "fmov v0.4s, #0x70 (1.0000)",
         "fsqrt v1.4s, v2.4s",
         "fdiv v2.4s, v0.4s, v1.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfnacc mm0, mm1": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0x8a",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #752]",
-        "ldr d3, [x28, #768]",
+        "ldr d2, [x28, #768]",
+        "ldr d3, [x28, #784]",
         "uzp1 v4.2s, v2.2s, v3.2s",
         "uzp2 v2.2s, v2.2s, v3.2s",
         "fsub v2.4s, v4.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfpnacc mm0, mm1": {
       "ExpectedInstructionCount": 7,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0x8e",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #752]",
-        "ldr d3, [x28, #768]",
+        "ldr d2, [x28, #768]",
+        "ldr d3, [x28, #784]",
         "dup v4.2s, v2.s[1]",
         "fsub s2, s2, s4",
         "faddp v3.4s, v3.4s, v3.4s",
         "mov v2.s[1], v3.s[0]",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfcmpge mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0x90",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fcmge v2.4s, v3.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfmin mm0, mm1": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0x94",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fcmgt v0.4s, v3.4s, v2.4s",
         "bif v2.16b, v3.16b, v0.16b",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfrcp mm0, mm1": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x96"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "fmov s0, #0x70 (1.0000)",
         "fdiv s2, s0, s2",
         "dup v2.2s, v2.s[0]",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfrsqrt mm0, mm1": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "Yes",
       "Comment": [
         "0x0f 0x0f 0x97"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "fmov s0, #0x70 (1.0000)",
         "fsqrt s1, s2",
         "fdiv s2, s0, s1",
         "dup v2.2s, v2.s[0]",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfsub mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0x9a",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fsub v2.4s, v3.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfadd mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0x9e",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fadd v2.4s, v3.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfcmpgt mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xa0",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fcmgt v2.4s, v3.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfmax mm0, mm1": {
       "ExpectedInstructionCount": 5,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xa4",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fcmgt v0.4s, v3.4s, v2.4s",
         "bit v2.16b, v3.16b, v0.16b",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfrcpit1 mm0, mm1": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xa6",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "str d2, [x28, #752]"
+        "ldr d2, [x28, #784]",
+        "str d2, [x28, #768]"
       ]
     },
     "pfrcpit1 mm0, mm0": {
       "ExpectedInstructionCount": 0,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xa6",
       "ExpectedArm64ASM": []
     },
     "pfrsqit1 mm0, mm1": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xa7",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "str d2, [x28, #752]"
+        "ldr d2, [x28, #784]",
+        "str d2, [x28, #768]"
       ]
     },
     "pfrsqit1 mm0, mm0": {
       "ExpectedInstructionCount": 0,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xa7",
       "ExpectedArm64ASM": []
     },
     "pfsubr mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xaa",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fsub v2.4s, v2.4s, v3.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfcmpeq mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xb0",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fcmeq v2.4s, v3.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfmul mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xb4",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "fmul v2.4s, v3.4s, v2.4s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pfrcpit2 mm0, mm1": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xb6",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "str d2, [x28, #752]"
+        "ldr d2, [x28, #784]",
+        "str d2, [x28, #768]"
       ]
     },
     "pfrcpit2 mm0, mm0": {
       "ExpectedInstructionCount": 0,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xb6",
       "ExpectedArm64ASM": []
     },
     "db 0x0f, 0x0f, 0xc1, 0xb7": {
       "ExpectedInstructionCount": 7,
-      "Optimal": "Yes",
       "Comment": [
         "nasm doesn't support emitting this instruction",
         "pmulhrw mm0, mm1",
         "0x0f 0x0f 0xb7"
       ],
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #752]",
-        "ldr d3, [x28, #768]",
+        "ldr d2, [x28, #768]",
+        "ldr d3, [x28, #784]",
         "smull v2.4s, v2.4h, v3.4h",
         "movi v3.4s, #0x80, lsl #8",
         "add v2.4s, v2.4s, v3.4s",
         "shrn v2.4h, v2.4s, #16",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pswapd mm0, mm1": {
       "ExpectedInstructionCount": 3,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xbb",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
+        "ldr d2, [x28, #784]",
         "rev64 v2.2s, v2.2s",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     },
     "pavgusb mm0, mm1": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "Yes",
       "Comment": "0x0f 0x0f 0xbf",
       "ExpectedArm64ASM": [
-        "ldr d2, [x28, #768]",
-        "ldr d3, [x28, #752]",
+        "ldr d2, [x28, #784]",
+        "ldr d3, [x28, #768]",
         "urhadd v2.16b, v3.16b, v2.16b",
-        "str d2, [x28, #752]"
+        "str d2, [x28, #768]"
       ]
     }
   }

unittests/InstructionCountCI/FEXOpt/MultiInst.json

@@ -15,7 +15,6 @@
   "Instructions": {
     "push ax, bx": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "No",
       "Comment": [
         "Mergable 16-bit pushes. May or may not be an optimization."
       ],
@@ -30,7 +29,6 @@
     },
     "push rax, rbx": {
       "ExpectedInstructionCount": 2,
-      "Optimal": "No",
       "Comment": [
         "Mergable 64-bit pushes"
       ],
@@ -45,7 +43,6 @@
     },
     "adds xmm0, xmm1, xmm2": {
       "ExpectedInstructionCount": 4,
-      "Optimal": "No",
       "Comment": [
         "Redundant scalar adds that can get eliminated without AFP."
       ],
@@ -62,7 +59,6 @@
     },
     "positive movsb": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -82,7 +78,6 @@
     },
     "positive movsw": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -102,7 +97,6 @@
     },
     "positive movsd": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -122,7 +116,6 @@
     },
     "positive movsq": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -142,7 +135,6 @@
     },
     "negative movsb": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -162,7 +154,6 @@
     },
     "negative movsw": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -182,7 +173,6 @@
     },
     "negative movsd": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -202,7 +192,6 @@
     },
     "negative movsq": {
       "ExpectedInstructionCount": 6,
-      "Optimal": "No",
       "Comment": [
         "When direction flag is a compile time constant we can optimize",
         "loads and stores can turn in to post-increment when known"
@@ -219,6 +208,444 @@
         "sub x10, x10, #0x8 (8)",
         "sub x11, x11, #0x8 (8)"
       ]
+    },
+    "positive rep movsb": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep movsb"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldrb w3, [x2], #1",
+        "strb w3, [x1], #1",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "add x22, x0, x2",
+        "add x23, x1, x2",
+        "mov x11, x22",
+        "mov x10, x23",
+        "mov x5, x20"
+      ]
+    },
+    "positive rep movsw": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep movsw"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldrh w3, [x2], #2",
+        "strh w3, [x1], #2",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "add x22, x0, x2, lsl #1",
+        "add x23, x1, x2, lsl #1",
+        "mov x11, x22",
+        "mov x10, x23",
+        "mov x5, x20"
+      ]
+    },
+    "positive rep movsd": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep movsd"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldr w3, [x2], #4",
+        "str w3, [x1], #4",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "add x22, x0, x2, lsl #2",
+        "add x23, x1, x2, lsl #2",
+        "mov x11, x22",
+        "mov x10, x23",
+        "mov x5, x20"
+      ]
+    },
+    "positive rep movsq": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep movsq"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldr x3, [x2], #8",
+        "str x3, [x1], #8",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "add x22, x0, x2, lsl #3",
+        "add x23, x1, x2, lsl #3",
+        "mov x11, x22",
+        "mov x10, x23",
+        "mov x5, x20"
+      ]
+    },
+    "negative rep movsb": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep movsb"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldrb w3, [x2], #-1",
+        "strb w3, [x1], #-1",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "sub x20, x0, x2",
+        "sub x21, x1, x2",
+        "mov x11, x20",
+        "mov x10, x21",
+        "mov w5, #0x0"
+      ]
+    },
+    "negative rep movsw": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep movsw"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldrh w3, [x2], #-2",
+        "strh w3, [x1], #-2",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "sub x20, x0, x2, lsl #1",
+        "sub x21, x1, x2, lsl #1",
+        "mov x11, x20",
+        "mov x10, x21",
+        "mov w5, #0x0"
+      ]
+    },
+    "negative rep movsd": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep movsd"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldr w3, [x2], #-4",
+        "str w3, [x1], #-4",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "sub x20, x0, x2, lsl #2",
+        "sub x21, x1, x2, lsl #2",
+        "mov x11, x20",
+        "mov x10, x21",
+        "mov w5, #0x0"
+      ]
+    },
+    "negative rep movsq": {
+      "ExpectedInstructionCount": 18,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep movsq"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "mov x2, x10",
+        "cbz x0, #+0x14",
+        "ldr x3, [x2], #-8",
+        "str x3, [x1], #-8",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0xc",
+        "mov x0, x11",
+        "mov x1, x10",
+        "mov x2, x5",
+        "sub x20, x0, x2, lsl #3",
+        "sub x21, x1, x2, lsl #3",
+        "mov x11, x20",
+        "mov x10, x21",
+        "mov w5, #0x0"
+      ]
+    },
+    "positive rep stosb": {
+      "ExpectedInstructionCount": 11,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep stosb"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "uxtb w21, w4",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "strb w21, [x1], #1",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "add x11, x11, x5",
+        "mov x5, x20"
+      ]
+    },
+    "positive rep stosw": {
+      "ExpectedInstructionCount": 11,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep stosw"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "uxth w21, w4",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "strh w21, [x1], #2",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "add x11, x11, x5, lsl #1",
+        "mov x5, x20"
+      ]
+    },
+    "positive rep stosd": {
+      "ExpectedInstructionCount": 11,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep stosd"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "mov w21, w4",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "str w21, [x1], #4",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "add x11, x11, x5, lsl #2",
+        "mov x5, x20"
+      ]
+    },
+    "positive rep stosq": {
+      "ExpectedInstructionCount": 10,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "cld",
+        "rep stosq"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x0",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "str x4, [x1], #8",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "add x11, x11, x5, lsl #3",
+        "mov x5, x20"
+      ]
+    },
+    "negative rep stosb": {
+      "ExpectedInstructionCount": 11,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep stosb"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "uxtb w20, w4",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "strb w20, [x1], #-1",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "sub x11, x11, x5",
+        "mov w5, #0x0"
+      ]
+    },
+    "negative rep stosw": {
+      "ExpectedInstructionCount": 11,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep stosw"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "uxth w20, w4",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "strh w20, [x1], #-2",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "sub x11, x11, x5, lsl #1",
+        "mov w5, #0x0"
+      ]
+    },
+    "negative rep stosd": {
+      "ExpectedInstructionCount": 11,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep stosd"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "mov w20, w4",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "str w20, [x1], #-4",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "sub x11, x11, x5, lsl #2",
+        "mov w5, #0x0"
+      ]
+    },
+    "negative rep stosq": {
+      "ExpectedInstructionCount": 10,
+      "Comment": [
+        "When direction flag is a compile time constant we can optimize",
+        "loads and stores can turn in to post-increment when known"
+      ],
+      "x86Insts": [
+        "std",
+        "rep stosq"
+      ],
+      "ExpectedArm64ASM": [
+        "mov w20, #0x1",
+        "strb w20, [x28, #714]",
+        "mov x0, x5",
+        "mov x1, x11",
+        "cbz x0, #+0x10",
+        "str x4, [x1], #-8",
+        "sub x0, x0, #0x1 (1)",
+        "cbnz x0, #-0x8",
+        "sub x11, x11, x5, lsl #3",
+        "mov w5, #0x0"
+      ]
     }
   }
 }