Merge pull request #4077 from Sonicadvance1/frontend_informed_support…

…savx SignalDelegator: Let the frontend inform AVX support
FEX-Emu · Sep 27, 2024 · 9100235 · 9100235
2 parents 4d62e75 + e7605c9
commit 9100235
Show file tree

Hide file tree

Showing 7 changed files with 30 additions and 40 deletions.
diff --git a/FEXCore/Source/Interface/Core/Core.cpp b/FEXCore/Source/Interface/Core/Core.cpp
@@ -318,8 +318,6 @@ bool ContextImpl::InitCore() {
 
   // Set up the SignalDelegator config since core is initialized.
   FEXCore::SignalDelegator::SignalDelegatorConfig SignalConfig {
-    .SupportsAVX = HostFeatures.SupportsAVX,
-
     .DispatcherBegin = Dispatcher->Start,
     .DispatcherEnd = Dispatcher->End,
 

diff --git a/FEXCore/include/FEXCore/Core/SignalDelegator.h b/FEXCore/include/FEXCore/Core/SignalDelegator.h
@@ -40,8 +40,6 @@ class SignalDelegator {
   virtual ~SignalDelegator() = default;
 
   struct SignalDelegatorConfig {
-    bool SupportsAVX {};
-
     // Dispatcher information
     uint64_t DispatcherBegin;
     uint64_t DispatcherEnd;

diff --git a/Source/Tools/FEXLoader/FEXLoader.cpp b/Source/Tools/FEXLoader/FEXLoader.cpp
@@ -498,16 +498,18 @@ int main(int argc, char** argv, char** const envp) {
   // System allocator is now system allocator or FEX
   FEXCore::Context::InitializeStaticTables(Loader.Is64BitMode() ? FEXCore::Context::MODE_64BIT : FEXCore::Context::MODE_32BIT);
 
+  bool SupportsAVX {};
   fextl::unique_ptr<FEXCore::Context::Context> CTX;
   {
     auto HostFeatures = FEX::FetchHostFeatures();
     CTX = FEXCore::Context::Context::CreateNewContext(HostFeatures);
+    SupportsAVX = HostFeatures.SupportsAVX;
   }
 
   // Setup TSO hardware emulation immediately after initializing the context.
   FEX::TSO::SetupTSOEmulation(CTX.get());
 
-  auto SignalDelegation = FEX::HLE::CreateSignalDelegator(CTX.get(), Program.ProgramName);
+  auto SignalDelegation = FEX::HLE::CreateSignalDelegator(CTX.get(), Program.ProgramName, SupportsAVX);
 
   auto SyscallHandler = Loader.Is64BitMode() ? FEX::HLE::x64::CreateHandler(CTX.get(), SignalDelegation.get()) :
                                                FEX::HLE::x32::CreateHandler(CTX.get(), SignalDelegation.get(), std::move(Allocator));

diff --git a/Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator.cpp b/Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator.cpp
@@ -118,7 +118,7 @@ void SignalDelegator::SpillSRA(FEXCore::Core::InternalThreadState* Thread, void*
     Thread->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRAIdxMap);
   }
 
-  if (Config.SupportsAVX) {
+  if (SupportsAVX) {
     // TODO: This doesn't save the upper 128-bits of the 256-bit registers.
     // This needs to be implemented still.
     for (size_t i = 0; i < Config.SRAFPRCount; i++) {
@@ -178,8 +178,6 @@ ArchHelpers::Context::ContextBackup* SignalDelegator::StoreThreadState(FEXCore::
 }
 
 void SignalDelegator::RestoreThreadState(FEXCore::Core::InternalThreadState* Thread, void* ucontext, RestoreType Type) {
-  const bool IsAVXEnabled = Config.SupportsAVX;
-
   uint64_t OldSP {};
   if (Type == RestoreType::TYPE_PAUSE) [[unlikely]] {
     OldSP = ArchHelpers::Context::GetSp(ucontext);
@@ -203,7 +201,7 @@ void SignalDelegator::RestoreThreadState(FEXCore::Core::InternalThreadState* Thr
       GuestSP += sizeof(siginfo_t);
       GuestSP = FEXCore::AlignUp(GuestSP, alignof(siginfo_t));
 
-      if (IsAVXEnabled) {
+      if (SupportsAVX) {
         GuestSP += sizeof(FEXCore::x86_64::xstate);
         GuestSP = FEXCore::AlignUp(GuestSP, alignof(FEXCore::x86_64::xstate));
       } else {
@@ -221,7 +219,7 @@ void SignalDelegator::RestoreThreadState(FEXCore::Core::InternalThreadState* Thr
         GuestSP += sizeof(SigFrame_i32) - 8;
         GuestSP = FEXCore::AlignUp(GuestSP, alignof(SigFrame_i32));
 
-        if (IsAVXEnabled) {
+        if (SupportsAVX) {
           GuestSP += sizeof(FEXCore::x86::xstate);
           GuestSP = FEXCore::AlignUp(GuestSP, alignof(FEXCore::x86::xstate));
         } else {
@@ -238,7 +236,7 @@ void SignalDelegator::RestoreThreadState(FEXCore::Core::InternalThreadState* Thr
         GuestSP += sizeof(RTSigFrame_i32) - 4;
         GuestSP = FEXCore::AlignUp(GuestSP, alignof(RTSigFrame_i32));
 
-        if (IsAVXEnabled) {
+        if (SupportsAVX) {
           GuestSP += sizeof(FEXCore::x86::xstate);
           GuestSP = FEXCore::AlignUp(GuestSP, alignof(FEXCore::x86::xstate));
         } else {
@@ -833,9 +831,10 @@ void SignalDelegator::QueueSignal(pid_t tgid, pid_t tid, int Signal, siginfo_t*
   }
 }
 
-SignalDelegator::SignalDelegator(FEXCore::Context::Context* _CTX, const std::string_view ApplicationName)
+SignalDelegator::SignalDelegator(FEXCore::Context::Context* _CTX, const std::string_view ApplicationName, bool SupportsAVX)
   : CTX {_CTX}
-  , ApplicationName {ApplicationName} {
+  , ApplicationName {ApplicationName}
+  , SupportsAVX {SupportsAVX} {
   // Signal zero isn't real
   HostHandlers[0].Installed = true;
 
@@ -1295,7 +1294,8 @@ uint64_t SignalDelegator::GuestSignalFD(int fd, const uint64_t* set, size_t sigs
   return Result == -1 ? -errno : Result;
 }
 
-fextl::unique_ptr<FEX::HLE::SignalDelegator> CreateSignalDelegator(FEXCore::Context::Context* CTX, const std::string_view ApplicationName) {
-  return fextl::make_unique<FEX::HLE::SignalDelegator>(CTX, ApplicationName);
+fextl::unique_ptr<FEX::HLE::SignalDelegator>
+CreateSignalDelegator(FEXCore::Context::Context* CTX, const std::string_view ApplicationName, bool SupportsAVX) {
+  return fextl::make_unique<FEX::HLE::SignalDelegator>(CTX, ApplicationName, SupportsAVX);
 }
 } // namespace FEX::HLE
diff --git a/Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator.h b/Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator.h
@@ -50,7 +50,7 @@ class SignalDelegator final : public FEXCore::SignalDelegator, public FEXCore::A
 
   // 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(FEXCore::Context::Context* _CTX, const std::string_view ApplicationName, bool SupportsAVX);
   ~SignalDelegator() override;
 
   // Called from the signal trampoline function.
@@ -274,7 +274,9 @@ class SignalDelegator final : public FEXCore::SignalDelegator, public FEXCore::A
 
   std::mutex HostDelegatorMutex;
   std::mutex GuestDelegatorMutex;
+  bool SupportsAVX;
 };
 
-fextl::unique_ptr<FEX::HLE::SignalDelegator> CreateSignalDelegator(FEXCore::Context::Context* CTX, const std::string_view ApplicationName);
+fextl::unique_ptr<FEX::HLE::SignalDelegator>
+CreateSignalDelegator(FEXCore::Context::Context* CTX, const std::string_view ApplicationName, bool SupportsAVX);
 } // namespace FEX::HLE
diff --git a/Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator/GuestFramesManagement.cpp b/Source/Tools/LinuxEmulation/LinuxSyscalls/SignalDelegator/GuestFramesManagement.cpp
@@ -110,8 +110,6 @@ static void SetXStateInfo(T* xstate, bool is_avx_enabled) {
 
 void SignalDelegator::RestoreFrame_x64(FEXCore::Core::InternalThreadState* Thread, ArchHelpers::Context::ContextBackup* Context,
                                        FEXCore::Core::CpuStateFrame* Frame, void* ucontext) {
-  const bool IsAVXEnabled = Config.SupportsAVX;
-
   auto* guest_uctx = reinterpret_cast<FEXCore::x86_64::ucontext_t*>(Context->UContextLocation);
   [[maybe_unused]] auto* guest_siginfo = reinterpret_cast<siginfo_t*>(Context->SigInfoLocation);
 
@@ -155,7 +153,7 @@ void SignalDelegator::RestoreFrame_x64(FEXCore::Core::InternalThreadState* Threa
     // Copy float registers
     memcpy(Frame->State.mm, fpstate->_st, sizeof(Frame->State.mm));
 
-    if (IsAVXEnabled) {
+    if (SupportsAVX) {
       CTX->SetXMMRegistersFromState(Thread, fpstate->_xmm, xstate->ymmh.ymmh_space);
     } else {
       CTX->SetXMMRegistersFromState(Thread, fpstate->_xmm, nullptr);
@@ -176,8 +174,6 @@ void SignalDelegator::RestoreFrame_x64(FEXCore::Core::InternalThreadState* Threa
 
 void SignalDelegator::RestoreFrame_ia32(FEXCore::Core::InternalThreadState* Thread, ArchHelpers::Context::ContextBackup* Context,
                                         FEXCore::Core::CpuStateFrame* Frame, void* ucontext) {
-  const bool IsAVXEnabled = Config.SupportsAVX;
-
   SigFrame_i32* guest_uctx = reinterpret_cast<SigFrame_i32*>(Context->UContextLocation);
   // If the guest modified the RIP then we need to take special precautions here
   if (Context->OriginalRIP != guest_uctx->sc.ip || Context->FaultToTopAndGeneratedException) {
@@ -228,7 +224,7 @@ void SignalDelegator::RestoreFrame_ia32(FEXCore::Core::InternalThreadState* Thre
     }
 
     // Extended XMM state
-    if (IsAVXEnabled) {
+    if (SupportsAVX) {
       CTX->SetXMMRegistersFromState(Thread, fpstate->_xmm, xstate->ymmh.ymmh_space);
     } else {
       CTX->SetXMMRegistersFromState(Thread, fpstate->_xmm, nullptr);
@@ -249,8 +245,6 @@ void SignalDelegator::RestoreFrame_ia32(FEXCore::Core::InternalThreadState* Thre
 
 void SignalDelegator::RestoreRTFrame_ia32(FEXCore::Core::InternalThreadState* Thread, ArchHelpers::Context::ContextBackup* Context,
                                           FEXCore::Core::CpuStateFrame* Frame, void* ucontext) {
-  const bool IsAVXEnabled = Config.SupportsAVX;
-
   RTSigFrame_i32* guest_uctx = reinterpret_cast<RTSigFrame_i32*>(Context->UContextLocation);
   // If the guest modified the RIP then we need to take special precautions here
   if (Context->OriginalRIP != guest_uctx->uc.uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP] || Context->FaultToTopAndGeneratedException) {
@@ -302,7 +296,7 @@ void SignalDelegator::RestoreRTFrame_ia32(FEXCore::Core::InternalThreadState* Th
     }
 
     // Extended XMM state
-    if (IsAVXEnabled) {
+    if (SupportsAVX) {
       CTX->SetXMMRegistersFromState(Thread, fpstate->_xmm, xstate->ymmh.ymmh_space);
     } else {
       CTX->SetXMMRegistersFromState(Thread, fpstate->_xmm, nullptr);
@@ -329,8 +323,6 @@ uint64_t SignalDelegator::SetupFrame_x64(FEXCore::Core::InternalThreadState* Thr
   // 32-bit doesn't have a redzone
   NewGuestSP -= 128;
 
-  const bool IsAVXEnabled = Config.SupportsAVX;
-
   // On 64-bit the kernel sets up the siginfo_t and ucontext_t regardless of SA_SIGINFO set.
   // This allows the application to /always/ get the siginfo and ucontext even if it didn't set this flag.
   //
@@ -345,7 +337,7 @@ uint64_t SignalDelegator::SetupFrame_x64(FEXCore::Core::InternalThreadState* Thr
 
   uint64_t HostStackLocation = NewGuestSP;
 
-  if (IsAVXEnabled) {
+  if (SupportsAVX) {
     NewGuestSP -= sizeof(FEXCore::x86_64::xstate);
     NewGuestSP = FEXCore::AlignDown(NewGuestSP, alignof(FEXCore::x86_64::xstate));
   } else {
@@ -378,7 +370,7 @@ uint64_t SignalDelegator::SetupFrame_x64(FEXCore::Core::InternalThreadState* Thr
   // Pointer to where the fpreg memory is
   guest_uctx->uc_mcontext.fpregs = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(FPStateLocation);
   auto* xstate = reinterpret_cast<FEXCore::x86_64::xstate*>(FPStateLocation);
-  SetXStateInfo(xstate, IsAVXEnabled);
+  SetXStateInfo(xstate, SupportsAVX);
 
   guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP] = ContextBackup->OriginalRIP;
   guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_EFL] = eflags;
@@ -427,7 +419,7 @@ uint64_t SignalDelegator::SetupFrame_x64(FEXCore::Core::InternalThreadState* Thr
   // Copy float registers
   memcpy(fpstate->_st, Frame->State.mm, sizeof(Frame->State.mm));
 
-  if (IsAVXEnabled) {
+  if (SupportsAVX) {
     CTX->ReconstructXMMRegisters(Thread, fpstate->_xmm, xstate->ymmh.ymmh_space);
   } else {
     CTX->ReconstructXMMRegisters(Thread, fpstate->_xmm, nullptr);
@@ -472,15 +464,14 @@ uint64_t SignalDelegator::SetupFrame_ia32(FEXCore::Core::InternalThreadState* Th
                                           FEXCore::Core::CpuStateFrame* Frame, int Signal, siginfo_t* HostSigInfo, void* ucontext,
                                           GuestSigAction* GuestAction, stack_t* GuestStack, uint64_t NewGuestSP, const uint32_t eflags) {
 
-  const bool IsAVXEnabled = Config.SupportsAVX;
   const uint64_t SignalReturn = reinterpret_cast<uint64_t>(VDSOPointers.VDSO_kernel_sigreturn);
 
   NewGuestSP -= sizeof(uint64_t);
   NewGuestSP = FEXCore::AlignDown(NewGuestSP, alignof(uint64_t));
 
   uint64_t HostStackLocation = NewGuestSP;
 
-  if (IsAVXEnabled) {
+  if (SupportsAVX) {
     NewGuestSP -= sizeof(FEXCore::x86::xstate);
     NewGuestSP = FEXCore::AlignDown(NewGuestSP, alignof(FEXCore::x86::xstate));
   } else {
@@ -505,7 +496,7 @@ uint64_t SignalDelegator::SetupFrame_ia32(FEXCore::Core::InternalThreadState* Th
   // Pointer to where the fpreg memory is
   guest_uctx->sc.fpstate = static_cast<uint32_t>(FPStateLocation);
   auto* xstate = reinterpret_cast<FEXCore::x86::xstate*>(FPStateLocation);
-  SetXStateInfo(xstate, IsAVXEnabled);
+  SetXStateInfo(xstate, SupportsAVX);
 
   guest_uctx->sc.cs = Frame->State.cs_idx;
   guest_uctx->sc.ds = Frame->State.ds_idx;
@@ -549,7 +540,7 @@ uint64_t SignalDelegator::SetupFrame_ia32(FEXCore::Core::InternalThreadState* Th
   // Extended XMM state
   fpstate->status = FEXCore::x86::fpstate_magic::MAGIC_XFPSTATE;
 
-  if (IsAVXEnabled) {
+  if (SupportsAVX) {
     CTX->ReconstructXMMRegisters(Thread, fpstate->_xmm, xstate->ymmh.ymmh_space);
   } else {
     CTX->ReconstructXMMRegisters(Thread, fpstate->_xmm, nullptr);
@@ -601,15 +592,14 @@ uint64_t SignalDelegator::SetupRTFrame_ia32(FEXCore::Core::InternalThreadState*
                                             FEXCore::Core::CpuStateFrame* Frame, int Signal, siginfo_t* HostSigInfo, void* ucontext,
                                             GuestSigAction* GuestAction, stack_t* GuestStack, uint64_t NewGuestSP, const uint32_t eflags) {
 
-  const bool IsAVXEnabled = Config.SupportsAVX;
   const uint64_t SignalReturn = reinterpret_cast<uint64_t>(VDSOPointers.VDSO_kernel_rt_sigreturn);
 
   NewGuestSP -= sizeof(uint64_t);
   NewGuestSP = FEXCore::AlignDown(NewGuestSP, alignof(uint64_t));
 
   uint64_t HostStackLocation = NewGuestSP;
 
-  if (IsAVXEnabled) {
+  if (SupportsAVX) {
     NewGuestSP -= sizeof(FEXCore::x86::xstate);
     NewGuestSP = FEXCore::AlignDown(NewGuestSP, alignof(FEXCore::x86::xstate));
   } else {
@@ -638,7 +628,7 @@ uint64_t SignalDelegator::SetupRTFrame_ia32(FEXCore::Core::InternalThreadState*
   // Pointer to where the fpreg memory is
   guest_uctx->uc.uc_mcontext.fpregs = static_cast<uint32_t>(FPStateLocation);
   auto* xstate = reinterpret_cast<FEXCore::x86::xstate*>(FPStateLocation);
-  SetXStateInfo(xstate, IsAVXEnabled);
+  SetXStateInfo(xstate, SupportsAVX);
 
   guest_uctx->uc.uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS] = Frame->State.cs_idx;
   guest_uctx->uc.uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds_idx;
@@ -684,7 +674,7 @@ uint64_t SignalDelegator::SetupRTFrame_ia32(FEXCore::Core::InternalThreadState*
   // Extended XMM state
   fpstate->status = FEXCore::x86::fpstate_magic::MAGIC_XFPSTATE;
 
-  if (IsAVXEnabled) {
+  if (SupportsAVX) {
     CTX->ReconstructXMMRegisters(Thread, fpstate->_xmm, xstate->ymmh.ymmh_space);
   } else {
     CTX->ReconstructXMMRegisters(Thread, fpstate->_xmm, nullptr);

diff --git a/Source/Tools/TestHarnessRunner/TestHarnessRunner.cpp b/Source/Tools/TestHarnessRunner/TestHarnessRunner.cpp
@@ -257,7 +257,7 @@ int main(int argc, char** argv, char** const envp) {
   auto CTX = FEXCore::Context::Context::CreateNewContext(HostFeatures);
 
 #ifndef _WIN32
-  auto SignalDelegation = FEX::HLE::CreateSignalDelegator(CTX.get(), {});
+  auto SignalDelegation = FEX::HLE::CreateSignalDelegator(CTX.get(), {}, HostFeatures.SupportsAVX);
 #else
   // Enable exit on HLT while Wine's longjump is broken.
   //