OpcodeDispatcher: absorb invert into PF calculation

with xorn

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>

FEXCore/Source/Interface/Core/OpcodeDispatcher.cpp

@@ -775,24 +775,20 @@ void OpDispatchBuilder::CALLAbsoluteOp(OpcodeArgs) {
   _ExitFunction(JMPPCOffset); // If we get here then leave the function now
 }
 
-OrderedNode *OpDispatchBuilder::SelectBit(OrderedNode *Cmp, bool TrueIsNonzero, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue) {
+OrderedNode *OpDispatchBuilder::SelectBit(OrderedNode *Cmp, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue) {
   uint64_t TrueConst, FalseConst;
   if (IsValueConstant(WrapNode(TrueValue), &TrueConst) &&
       IsValueConstant(WrapNode(FalseValue), &FalseConst) &&
       TrueConst == 1 &&
       FalseConst == 0) {
 
-      if (!TrueIsNonzero)
-        Cmp = _Not(OpSize::i32Bit, Cmp);
-
       return _And(ResultSize, Cmp, _Constant(1));
   }
 
   SaveNZCV();
   _TestNZ(OpSize::i32Bit, Cmp, _Constant(1));
-  return _NZCVSelect(ResultSize,
-                 TrueIsNonzero ? CondClassType{COND_NEQ} : CondClassType{COND_EQ},
-                 TrueValue, FalseValue);
+  return _NZCVSelect(ResultSize, CondClassType{COND_NEQ},
+                     TrueValue, FalseValue);
 }
 
 std::pair<bool, CondClassType> OpDispatchBuilder::DecodeNZCVCondition(uint8_t OP) const {
@@ -857,10 +853,10 @@ OrderedNode *OpDispatchBuilder::SelectCC(uint8_t OP, IR::OpSize ResultSize, Orde
     }
     case 0xA: { // JP - Jump if PF == 1
       // Raw value contains inverted PF in bottom bit
-      return SelectBit(LoadPFRaw(), false, ResultSize, TrueValue, FalseValue);
+      return SelectBit(LoadPFRaw(true), ResultSize, TrueValue, FalseValue);
     }
     case 0xB: { // JNP - Jump if PF == 0
-      return SelectBit(LoadPFRaw(), true, ResultSize, TrueValue, FalseValue);
+      return SelectBit(LoadPFRaw(false), ResultSize, TrueValue, FalseValue);
     }
     default:
       LOGMAN_MSG_A_FMT("Unknown CC Op: 0x{:x}\n", OP);

FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -1627,7 +1627,7 @@ private:
   }
 
   std::pair<bool, CondClassType> DecodeNZCVCondition(uint8_t OP) const;
-  OrderedNode *SelectBit(OrderedNode *Cmp, bool Invert, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
+  OrderedNode *SelectBit(OrderedNode *Cmp, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
   OrderedNode *SelectCC(uint8_t OP, IR::OpSize ResultSize, OrderedNode *TrueValue, OrderedNode *FalseValue);
 
   /**
@@ -1761,7 +1761,7 @@ private:
   /**
    * @name These functions are used by the deferred flag handling while it is calculating and storing flags in to RFLAGs.
    * @{ */
-  OrderedNode *LoadPFRaw();
+  OrderedNode *LoadPFRaw(bool Invert);
   OrderedNode *LoadAF();
   void FixupAF();
   void SetAFAndFixup(OrderedNode *AF);

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

@@ -179,7 +179,7 @@ OrderedNode *OpDispatchBuilder::GetPackedRFLAG(uint32_t FlagsMask) {
   // instead.
   if (FlagsMask & (1 << FEXCore::X86State::RFLAG_PF_RAW_LOC)) {
     // Set every bit except the bottommost.
-    auto OnesInvPF = _Or(OpSize::i64Bit, LoadPFRaw(), _Constant(~1ull));
+    auto OnesInvPF = _Or(OpSize::i64Bit, LoadPFRaw(false), _Constant(~1ull));
 
     // Rotate the bottom bit to the appropriate location for PF, so we get
     // something like 111P1111. Then invert that to get 000p0000. Then OR that
@@ -237,7 +237,7 @@ void OpDispatchBuilder::CalculateOF(uint8_t SrcSize, OrderedNode *Res, OrderedNo
   SetRFLAG<FEXCore::X86State::RFLAG_OF_RAW_LOC>(Anded, SrcSize * 8 - 1, true);
 }
 
-OrderedNode *OpDispatchBuilder::LoadPFRaw() {
+OrderedNode *OpDispatchBuilder::LoadPFRaw(bool Invert) {
   // 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);
@@ -245,7 +245,11 @@ OrderedNode *OpDispatchBuilder::LoadPFRaw() {
   // Cascade to calculate parity of bottom 8-bits to bottom bit.
   Result = _XorShift(OpSize::i32Bit, Result, Result, ShiftType::LSR, 4);
   Result = _XorShift(OpSize::i32Bit, Result, Result, ShiftType::LSR, 2);
-  Result = _XorShift(OpSize::i32Bit, Result, Result, ShiftType::LSR, 1);
+
+  if (Invert)
+    Result = _XornShift(OpSize::i32Bit, Result, Result, ShiftType::LSR, 1);
+  else
+    Result = _XorShift(OpSize::i32Bit, Result, Result, ShiftType::LSR, 1);
 
   return Result;
 }