From a9e0f21373d354cf6d9ab2882ca42a53a7002063 Mon Sep 17 00:00:00 2001
From: Sintendo <bram.speeckaert@gmail.com>
Date: Fri, 17 Sep 2021 22:59:00 +0200
Subject: [PATCH] Jit64: Split arithcx into addcx and subfcx

JitArm64 also opts to separate the two. The shared logic makes less
sense once we start adding more optimizations.
---
 Source/Core/Core/PowerPC/Jit64/Jit.h          |  3 +-
 .../Core/Core/PowerPC/Jit64/Jit64_Tables.cpp  |  8 +--
 .../Core/Core/PowerPC/Jit64/Jit_Integer.cpp   | 54 ++++++++++++-------
 3 files changed, 42 insertions(+), 23 deletions(-)

diff --git a/Source/Core/Core/PowerPC/Jit64/Jit.h b/Source/Core/Core/PowerPC/Jit64/Jit.h
index 35f198dc6b..7da679b86c 100644
--- a/Source/Core/Core/PowerPC/Jit64/Jit.h
+++ b/Source/Core/Core/PowerPC/Jit64/Jit.h
@@ -150,7 +150,7 @@ public:
   void DynaRunTable63(UGeckoInstruction inst);
 
   void addx(UGeckoInstruction inst);
-  void arithcx(UGeckoInstruction inst);
+  void addcx(UGeckoInstruction inst);
   void mulli(UGeckoInstruction inst);
   void mulhwXx(UGeckoInstruction inst);
   void mullwx(UGeckoInstruction inst);
@@ -233,6 +233,7 @@ public:
 
   void subfic(UGeckoInstruction inst);
   void subfx(UGeckoInstruction inst);
+  void subfcx(UGeckoInstruction inst);
 
   void twX(UGeckoInstruction inst);
 
diff --git a/Source/Core/Core/PowerPC/Jit64/Jit64_Tables.cpp b/Source/Core/Core/PowerPC/Jit64/Jit64_Tables.cpp
index 24f1398b73..35b17bcdb7 100644
--- a/Source/Core/Core/PowerPC/Jit64/Jit64_Tables.cpp
+++ b/Source/Core/Core/PowerPC/Jit64/Jit64_Tables.cpp
@@ -150,8 +150,8 @@ constexpr std::array<GekkoOPTemplate, 13> s_table19{{
 constexpr std::array<GekkoOPTemplate, 107> s_table31{{
     {266, &Jit64::addx},      // addx
     {778, &Jit64::addx},      // addox
-    {10, &Jit64::arithcx},    // addcx
-    {522, &Jit64::arithcx},   // addcox
+    {10, &Jit64::addcx},    // addcx
+    {522, &Jit64::addcx},   // addcox
     {138, &Jit64::arithXex},  // addex
     {650, &Jit64::arithXex},  // addeox
     {234, &Jit64::arithXex},  // addmex
@@ -170,8 +170,8 @@ constexpr std::array<GekkoOPTemplate, 107> s_table31{{
     {616, &Jit64::negx},      // negox
     {40, &Jit64::subfx},      // subfx
     {552, &Jit64::subfx},     // subfox
-    {8, &Jit64::arithcx},     // subfcx
-    {520, &Jit64::arithcx},   // subfcox
+    {8, &Jit64::subfcx},     // subfcx
+    {520, &Jit64::subfcx},   // subfcox
     {136, &Jit64::arithXex},  // subfex
     {648, &Jit64::arithXex},  // subfeox
     {232, &Jit64::arithXex},  // subfmex
diff --git a/Source/Core/Core/PowerPC/Jit64/Jit_Integer.cpp b/Source/Core/Core/PowerPC/Jit64/Jit_Integer.cpp
index be61b048cf..d85fd1a3a9 100644
--- a/Source/Core/Core/PowerPC/Jit64/Jit_Integer.cpp
+++ b/Source/Core/Core/PowerPC/Jit64/Jit_Integer.cpp
@@ -1806,11 +1806,39 @@ void Jit64::arithXex(UGeckoInstruction inst)
     ComputeRC(d);
 }
 
-void Jit64::arithcx(UGeckoInstruction inst)
+void Jit64::addcx(UGeckoInstruction inst)
+{
+  INSTRUCTION_START
+  JITDISABLE(bJITIntegerOff);
+  int a = inst.RA, b = inst.RB, d = inst.RD;
+
+  {
+    RCOpArg Ra = gpr.Use(a, RCMode::Read);
+    RCOpArg Rb = gpr.Use(b, RCMode::Read);
+    RCX64Reg Rd = gpr.Bind(d, RCMode::Write);
+    RegCache::Realize(Ra, Rb, Rd);
+
+    if (d == a)
+    {
+      ADD(32, Rd, Rb);
+    }
+    else
+    {
+      if (d != b)
+        MOV(32, Rd, Rb);
+      ADD(32, Rd, Ra);
+    }
+  }
+
+  FinalizeCarryOverflow(inst.OE);
+  if (inst.Rc)
+    ComputeRC(d);
+}
+
+void Jit64::subfcx(UGeckoInstruction inst)
 {
   INSTRUCTION_START
   JITDISABLE(bJITIntegerOff);
-  bool add = !!(inst.SUBOP10 & 2);  // add or sub
   int a = inst.RA, b = inst.RB, d = inst.RD;
 
   {
@@ -1821,30 +1849,20 @@ void Jit64::arithcx(UGeckoInstruction inst)
 
     if (d == a && d != b)
     {
-      if (add)
-      {
-        ADD(32, Rd, Rb);
-      }
-      else
-      {
-        // special case, because sub isn't reversible
-        MOV(32, R(RSCRATCH), Ra);
-        MOV(32, Rd, Rb);
-        SUB(32, Rd, R(RSCRATCH));
-      }
+      // special case, because sub isn't reversible
+      MOV(32, R(RSCRATCH), Ra);
+      MOV(32, Rd, Rb);
+      SUB(32, Rd, R(RSCRATCH));
     }
     else
     {
       if (d != b)
         MOV(32, Rd, Rb);
-      if (add)
-        ADD(32, Rd, Ra);
-      else
-        SUB(32, Rd, Ra);
+      SUB(32, Rd, Ra);
     }
   }
 
-  FinalizeCarryOverflow(inst.OE, !add);
+  FinalizeCarryOverflow(inst.OE, true);
   if (inst.Rc)
     ComputeRC(d);
 }