Merge pull request #11242 from Sintendo/arm64cmp

JitArm64: Optimize cmp

2 files changed, 48 insertions, 32 deletions

diff --git a/Source/Core/Common/Arm64Emitter.h b/Source/Core/Common/Arm64Emitter.h
index caf997a29d..19d63a8dae 100644
--- a/Source/Core/Common/Arm64Emitter.h
+++ b/Source/Core/Common/Arm64Emitter.h
@@ -309,6 +309,18 @@ enum class ShiftType
   ROR = 3,
 };
 
+enum class ExtendSpecifier
+{
+  UXTB = 0x0,
+  UXTH = 0x1,
+  UXTW = 0x2, /* Also LSL on 32bit width */
+  UXTX = 0x3, /* Also LSL on 64bit width */
+  SXTB = 0x4,
+  SXTH = 0x5,
+  SXTW = 0x6,
+  SXTX = 0x7,
+};
+
 enum class IndexType
 {
   Unsigned,
@@ -405,18 +417,6 @@ private:
     Width64Bit,
   };
 
-  enum class ExtendSpecifier
-  {
-    UXTB = 0x0,
-    UXTH = 0x1,
-    UXTW = 0x2, /* Also LSL on 32bit width */
-    UXTX = 0x3, /* Also LSL on 64bit width */
-    SXTB = 0x4,
-    SXTH = 0x5,
-    SXTW = 0x6,
-    SXTX = 0x7,
-  };
-
   enum class TypeSpecifier
   {
     ExtendedReg,
@@ -463,6 +463,15 @@ public:
     }
     m_shifttype = ShiftType::LSL;
   }
+  ArithOption(ARM64Reg Rd, ExtendSpecifier extend_type, u32 shift = 0)
+  {
+    m_destReg = Rd;
+    m_width = Is64Bit(Rd) ? WidthSpecifier::Width64Bit : WidthSpecifier::Width32Bit;
+    m_extend = extend_type;
+    m_type = TypeSpecifier::ExtendedReg;
+    m_shifttype = ShiftType::LSL;
+    m_shift = shift;
+  }
   ArithOption(ARM64Reg Rd, ShiftType shift_type, u32 shift)
   {
     m_destReg = Rd;
diff --git a/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp b/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
index 2fc1524809..f315efb976 100644
--- a/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
+++ b/Source/Core/Core/PowerPC/JitArm64/JitArm64_Integer.cpp
@@ -578,25 +578,29 @@ void JitArm64::cmp(UGeckoInstruction inst)
     s64 A = static_cast<s32>(gpr.GetImm(a));
     s64 B = static_cast<s32>(gpr.GetImm(b));
     MOVI2R(CR, A - B);
-    return;
   }
-
-  if (gpr.IsImm(b) && !gpr.GetImm(b))
+  else if (gpr.IsImm(a) && !gpr.GetImm(a))
+  {
+    NEG(EncodeRegTo32(CR), gpr.R(b));
+    SXTW(CR, EncodeRegTo32(CR));
+  }
+  else if (gpr.IsImm(a) && gpr.GetImm(a) == 0xFFFFFFFF)
+  {
+    MVN(EncodeRegTo32(CR), gpr.R(b));
+    SXTW(CR, EncodeRegTo32(CR));
+  }
+  else if (gpr.IsImm(b) && !gpr.GetImm(b))
   {
     SXTW(CR, gpr.R(a));
-    return;
   }
+  else
+  {
+    ARM64Reg RA = gpr.R(a);
+    ARM64Reg RB = gpr.R(b);
 
-  ARM64Reg WA = gpr.GetReg();
-  ARM64Reg XA = EncodeRegTo64(WA);
-  ARM64Reg RA = gpr.R(a);
-  ARM64Reg RB = gpr.R(b);
-
-  SXTW(XA, RA);
-  SXTW(CR, RB);
-  SUB(CR, XA, CR);
-
-  gpr.Unlock(WA);
+    SXTW(CR, RA);
+    SUB(CR, CR, RB, ArithOption(RB, ExtendSpecifier::SXTW));
+  }
 }
 
 void JitArm64::cmpl(UGeckoInstruction inst)
@@ -615,16 +619,19 @@ void JitArm64::cmpl(UGeckoInstruction inst)
     u64 A = gpr.GetImm(a);
     u64 B = gpr.GetImm(b);
     MOVI2R(CR, A - B);
-    return;
   }
-
-  if (gpr.IsImm(b) && !gpr.GetImm(b))
+  else if (gpr.IsImm(a) && !gpr.GetImm(a))
+  {
+    NEG(CR, EncodeRegTo64(gpr.R(b)));
+  }
+  else if (gpr.IsImm(b) && !gpr.GetImm(b))
   {
     MOV(EncodeRegTo32(CR), gpr.R(a));
-    return;
   }
-
-  SUB(gpr.CR(crf), EncodeRegTo64(gpr.R(a)), EncodeRegTo64(gpr.R(b)));
+  else
+  {
+    SUB(CR, EncodeRegTo64(gpr.R(a)), EncodeRegTo64(gpr.R(b)));
+  }
 }
 
 void JitArm64::cmpi(UGeckoInstruction inst)