Redesign x87 optimization framework

.clang-tidy

@@ -0,0 +1,8 @@
+Checks: '-*,clang-diagnostic-*,llvm-*,misc-*,-misc-const-correctness,-misc-unused-parameters,-misc-non-private-member-variables-in-classes,-misc-no-recursion,-misc-use-anonymous-namespace,readability-identifier-naming,-misc-include-cleaner'
+CheckOptions:
+  - key:             readability-identifier-naming.IgnoreMainLikeFunctions
+    value:           1
+  - key:             readability-redundant-member-init.IgnoreBaseInCopyConstructors
+    value:           1
+  - key:             modernize-use-default-member-init.UseAssignment
+    value:           1

FEXCore/Source/CMakeLists.txt

@@ -99,6 +99,7 @@ set (SRCS
   Interface/Core/OpcodeDispatcher/Flags.cpp
   Interface/Core/OpcodeDispatcher/Vector.cpp
   Interface/Core/OpcodeDispatcher/X87.cpp
+  Interface/Core/OpcodeDispatcher/X87New.cpp
   Interface/Core/OpcodeDispatcher/X87F64.cpp
   Interface/Core/OpcodeDispatcher.cpp
   Interface/Core/X86Tables.cpp
@@ -150,6 +151,7 @@ set (SRCS
   Interface/IR/Passes/DeadStoreElimination.cpp
   Interface/IR/Passes/RegisterAllocationPass.cpp
   Interface/IR/Passes/InlineCallOptimization.cpp
+  Interface/IR/Passes/x87StackOptimizationPass.cpp
   Utils/Telemetry.cpp
   Utils/Threads.cpp
   Utils/Profiler.cpp

FEXCore/Source/Interface/Core/Interpreter/Fallbacks/InterpreterFallbacks.cpp

@@ -89,7 +89,7 @@ bool InterpreterOps::GetFallbackHandler(bool SupportsPreserveAllABI, const IR::I
   uint8_t OpSize = IROp->Size;
   switch (IROp->Op) {
   case IR::OP_F80CVTTO: {
-    auto Op = IROp->C<IR::IROp_F80CVTTo>();
+    const auto* Op = IROp->C<IR::IROp_F80CVTTo>();
 
     switch (Op->SrcSize) {
     case 4: {
@@ -119,7 +119,7 @@ bool InterpreterOps::GetFallbackHandler(bool SupportsPreserveAllABI, const IR::I
     break;
   }
   case IR::OP_F80CVTINT: {
-    auto Op = IROp->C<IR::IROp_F80CVTInt>();
+    const auto* Op = IROp->C<IR::IROp_F80CVTInt>();
 
     switch (OpSize) {
     case 2: {
@@ -154,7 +154,7 @@ bool InterpreterOps::GetFallbackHandler(bool SupportsPreserveAllABI, const IR::I
     break;
   }
   case IR::OP_F80CMP: {
-    auto Op = IROp->C<IR::IROp_F80Cmp>();
+    const auto* Op = IROp->C<IR::IROp_F80Cmp>();
 
     static constexpr std::array handlers {
       &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>, &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>,
@@ -169,7 +169,7 @@ bool InterpreterOps::GetFallbackHandler(bool SupportsPreserveAllABI, const IR::I
   }
 
   case IR::OP_F80CVTTOINT: {
-    auto Op = IROp->C<IR::IROp_F80CVTToInt>();
+    const auto* Op = IROp->C<IR::IROp_F80CVTToInt>();
 
     switch (Op->SrcSize) {
     case 2: {

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

@@ -233,6 +233,5 @@ DEF_OP(RDRAND) {
 DEF_OP(Yield) {
   yield();
 }
-
 #undef DEF_OP
 } // namespace FEXCore::CPU

FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -737,10 +737,14 @@ class OpDispatchBuilder final : public IREmitter {
   template<bool Truncate>
   void FIST(OpcodeArgs);
 
+  // OpResult is used for Stack operations,
+  // describes if the result of the operation is stored in ST(0) or ST(i),
+  // where ST(i) is one of the arguments to the operation.
   enum class OpResult {
     RES_ST0,
     RES_STI,
   };
+
   template<size_t width, bool Integer, OpResult ResInST0>
   void FADD(OpcodeArgs);
   template<size_t width, bool Integer, OpResult ResInST0>
@@ -2070,6 +2074,7 @@ class OpDispatchBuilder final : public IREmitter {
 
   bool Multiblock {};
   uint64_t Entry;
+  IROp_IRHeader *CurrentHeader{};
 
   OrderedNode* _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode* Addr, OrderedNode* Value, uint8_t Align = 1) {
     if (CTX->IsAtomicTSOEnabled()) {

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

@@ -123,6 +123,7 @@ OrderedNode* OpDispatchBuilder::ReconstructX87StateFromFSW(OrderedNode* FSW) {
   return Top;
 }
 
+#if 0
 template<size_t width>
 void OpDispatchBuilder::FLD(OpcodeArgs) {
   // Update TOP
@@ -157,9 +158,13 @@ void OpDispatchBuilder::FLD(OpcodeArgs) {
   //_StoreContext(converted, 16, offsetof(FEXCore::Core::CPUState, mm[7][0]));
 }
 
-template void OpDispatchBuilder::FLD<32>(OpcodeArgs);
-template void OpDispatchBuilder::FLD<64>(OpcodeArgs);
-template void OpDispatchBuilder::FLD<80>(OpcodeArgs);
+template
+void OpDispatchBuilder::FLD<32>(OpcodeArgs);
+template
+void OpDispatchBuilder::FLD<64>(OpcodeArgs);
+template
+void OpDispatchBuilder::FLD<80>(OpcodeArgs);
+#endif
 
 void OpDispatchBuilder::FBLD(OpcodeArgs) {
   // Update TOP
@@ -254,6 +259,7 @@ void OpDispatchBuilder::FILD(OpcodeArgs) {
   _StoreContextIndexed(converted, top, 16, MMBaseOffset(), 16, FPRClass);
 }
 
+#if 0
 template<size_t width>
 void OpDispatchBuilder::FST(OpcodeArgs) {
   auto orig_top = GetX87Top();
@@ -274,9 +280,13 @@ void OpDispatchBuilder::FST(OpcodeArgs) {
   }
 }
 
-template void OpDispatchBuilder::FST<32>(OpcodeArgs);
-template void OpDispatchBuilder::FST<64>(OpcodeArgs);
-template void OpDispatchBuilder::FST<80>(OpcodeArgs);
+template
+void OpDispatchBuilder::FST<32>(OpcodeArgs);
+template
+void OpDispatchBuilder::FST<64>(OpcodeArgs);
+template
+void OpDispatchBuilder::FST<80>(OpcodeArgs);
+#endif
 
 template<bool Truncate>
 void OpDispatchBuilder::FIST(OpcodeArgs) {
@@ -300,7 +310,8 @@ void OpDispatchBuilder::FIST(OpcodeArgs) {
 template void OpDispatchBuilder::FIST<false>(OpcodeArgs);
 template void OpDispatchBuilder::FIST<true>(OpcodeArgs);
 
-template<size_t width, bool Integer, OpDispatchBuilder::OpResult ResInST0>
+#if 0
+template <size_t width, bool Integer, OpDispatchBuilder::OpResult ResInST0>
 void OpDispatchBuilder::FADD(OpcodeArgs) {
   auto top = GetX87Top();
   OrderedNode* StackLocation = top;
@@ -351,9 +362,13 @@ template void OpDispatchBuilder::FADD<64, false, OpDispatchBuilder::OpResult::RE
 template void OpDispatchBuilder::FADD<80, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
 template void OpDispatchBuilder::FADD<80, false, OpDispatchBuilder::OpResult::RES_STI>(OpcodeArgs);
 
-template void OpDispatchBuilder::FADD<16, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-template void OpDispatchBuilder::FADD<32, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+template
+void OpDispatchBuilder::FADD<16, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+template
+void OpDispatchBuilder::FADD<32, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+#endif
 
+#if 0
 template<size_t width, bool Integer, OpDispatchBuilder::OpResult ResInST0>
 void OpDispatchBuilder::FMUL(OpcodeArgs) {
   auto top = GetX87Top();
@@ -407,9 +422,12 @@ template void OpDispatchBuilder::FMUL<64, false, OpDispatchBuilder::OpResult::RE
 template void OpDispatchBuilder::FMUL<80, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
 template void OpDispatchBuilder::FMUL<80, false, OpDispatchBuilder::OpResult::RES_STI>(OpcodeArgs);
 
-template void OpDispatchBuilder::FMUL<16, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-template void OpDispatchBuilder::FMUL<32, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-
+template
+void OpDispatchBuilder::FMUL<16, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+template
+void OpDispatchBuilder::FMUL<32, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+#endif
+#if 0
 template<size_t width, bool Integer, bool reverse, OpDispatchBuilder::OpResult ResInST0>
 void OpDispatchBuilder::FDIV(OpcodeArgs) {
   auto top = GetX87Top();
@@ -478,9 +496,12 @@ template void OpDispatchBuilder::FDIV<80, false, true, OpDispatchBuilder::OpResu
 template void OpDispatchBuilder::FDIV<16, true, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
 template void OpDispatchBuilder::FDIV<16, true, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
 
-template void OpDispatchBuilder::FDIV<32, true, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-template void OpDispatchBuilder::FDIV<32, true, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-
+template
+void OpDispatchBuilder::FDIV<32, true, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+template
+void OpDispatchBuilder::FDIV<32, true, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+#endif
+#if 0
 template<size_t width, bool Integer, bool reverse, OpDispatchBuilder::OpResult ResInST0>
 void OpDispatchBuilder::FSUB(OpcodeArgs) {
   auto top = GetX87Top();
@@ -549,9 +570,11 @@ template void OpDispatchBuilder::FSUB<80, false, true, OpDispatchBuilder::OpResu
 template void OpDispatchBuilder::FSUB<16, true, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
 template void OpDispatchBuilder::FSUB<16, true, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
 
-template void OpDispatchBuilder::FSUB<32, true, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-template void OpDispatchBuilder::FSUB<32, true, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
-
+template
+void OpDispatchBuilder::FSUB<32, true, false, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+template
+void OpDispatchBuilder::FSUB<32, true, true, OpDispatchBuilder::OpResult::RES_ST0>(OpcodeArgs);
+#endif
 void OpDispatchBuilder::FCHS(OpcodeArgs) {
   auto top = GetX87Top();
   auto a = _LoadContextIndexed(top, 16, MMBaseOffset(), 16, FPRClass);
@@ -911,7 +934,7 @@ void OpDispatchBuilder::X87ATAN(OpcodeArgs) {
   // Write to ST[TOP]
   _StoreContextIndexed(result, top, 16, MMBaseOffset(), 16, FPRClass);
 }
-
+#if 0
 void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
   const auto Size = GetSrcSize(Op);
   OrderedNode* Mem = MakeSegmentAddress(Op, Op->Src[0]);
@@ -929,7 +952,8 @@ void OpDispatchBuilder::X87LDENV(OpcodeArgs) {
     SetX87FTW(_LoadMem(GPRClass, Size, MemLocation, Size));
   }
 }
-
+#endif
+#if 0
 void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
   // 14 bytes for 16bit
   // 2 Bytes : FCW
@@ -995,7 +1019,7 @@ void OpDispatchBuilder::X87FNSTENV(OpcodeArgs) {
     _StoreMem(GPRClass, Size, MemLocation, ZeroConst, Size);
   }
 }
-
+#endif
 void OpDispatchBuilder::X87FLDCW(OpcodeArgs) {
   OrderedNode* NewFCW = LoadSource(GPRClass, Op, Op->Src[0], Op->Flags);
   _StoreContext(2, GPRClass, NewFCW, offsetof(FEXCore::Core::CPUState, FCW));