Large-scale PowerPC recompiler rework

boost.natvis

@@ -0,0 +1,26 @@
+<?xml version='1.0' encoding='utf-8'?>
+<AutoVisualizer xmlns="http://schemas.microsoft.com/vstudio/debugger/natvis/2010">
+
+    <Type Name="boost::container::small_vector&lt;*&gt;">
+        <Expand>
+            <Item Name="[size]">m_holder.m_size</Item>
+            <ArrayItems>
+                <Size>m_holder.m_size</Size>
+                <ValuePointer>m_holder.m_start</ValuePointer>
+            </ArrayItems>
+        </Expand>
+    </Type>
+
+	<Type Name="boost::container::static_vector&lt;*&gt;">
+	<DisplayString>{{ size={m_holder.m_size} }}</DisplayString>
+	<Expand>
+	  <Item Name="[size]" ExcludeView="simple">m_holder.m_size</Item>
+	  <Item Name="[capacity]" ExcludeView="simple">static_capacity</Item>
+	  <ArrayItems>
+		<Size>m_holder.m_size</Size>
+		<ValuePointer>($T1*)m_holder.storage.data</ValuePointer>
+	  </ArrayItems>
+	</Expand>
+	</Type>
+
+</AutoVisualizer>

src/Cafe/CMakeLists.txt

@@ -67,24 +67,31 @@ add_library(CemuCafe
   HW/Espresso/Recompiler/PPCFunctionBoundaryTracker.h
   HW/Espresso/Recompiler/PPCRecompiler.cpp
   HW/Espresso/Recompiler/PPCRecompiler.h
-  HW/Espresso/Recompiler/PPCRecompilerImlAnalyzer.cpp
+  HW/Espresso/Recompiler/IML/IML.h
+  HW/Espresso/Recompiler/IML/IMLSegment.cpp
+  HW/Espresso/Recompiler/IML/IMLSegment.h
+  HW/Espresso/Recompiler/IML/IMLInstruction.cpp
+  HW/Espresso/Recompiler/IML/IMLInstruction.h
+  HW/Espresso/Recompiler/IML/IMLDebug.cpp
+  HW/Espresso/Recompiler/IML/IMLAnalyzer.cpp
+  HW/Espresso/Recompiler/IML/IMLOptimizer.cpp
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocator.cpp
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocator.h
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocatorRanges.cpp
+  HW/Espresso/Recompiler/IML/IMLRegisterAllocatorRanges.h
   HW/Espresso/Recompiler/PPCRecompilerImlGen.cpp
   HW/Espresso/Recompiler/PPCRecompilerImlGenFPU.cpp
   HW/Espresso/Recompiler/PPCRecompilerIml.h
-  HW/Espresso/Recompiler/PPCRecompilerImlOptimizer.cpp
-  HW/Espresso/Recompiler/PPCRecompilerImlRanges.cpp
-  HW/Espresso/Recompiler/PPCRecompilerImlRanges.h
-  HW/Espresso/Recompiler/PPCRecompilerImlRegisterAllocator2.cpp
-  HW/Espresso/Recompiler/PPCRecompilerImlRegisterAllocator.cpp
   HW/Espresso/Recompiler/PPCRecompilerIntermediate.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64AVX.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64BMI.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64FPU.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64Gen.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64GenFPU.cpp
-  HW/Espresso/Recompiler/PPCRecompilerX64.h
-  HW/Espresso/Recompiler/x64Emit.hpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64AVX.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64BMI.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64FPU.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64Gen.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64GenFPU.cpp
+  HW/Espresso/Recompiler/BackendX64/BackendX64.h
+  HW/Espresso/Recompiler/BackendX64/X64Emit.hpp
+  HW/Espresso/Recompiler/BackendX64/x86Emitter.h
   HW/Latte/Common/RegisterSerializer.cpp
   HW/Latte/Common/RegisterSerializer.h
   HW/Latte/Common/ShaderSerializer.cpp

src/Cafe/HW/Espresso/EspressoISA.h

@@ -91,13 +91,15 @@ namespace Espresso
 		BCCTR = 528
 	};
 
-	enum class OPCODE_31
+	enum class Opcode31
 	{
-
+		TW = 4,
+		MFTB = 371,
 	};
 
 	inline PrimaryOpcode GetPrimaryOpcode(uint32 opcode) { return (PrimaryOpcode)(opcode >> 26); };
 	inline Opcode19 GetGroup19Opcode(uint32 opcode) { return (Opcode19)((opcode >> 1) & 0x3FF); };
+	inline Opcode31 GetGroup31Opcode(uint32 opcode) { return (Opcode31)((opcode >> 1) & 0x3FF); };
 
 	struct BOField 
 	{
@@ -132,6 +134,12 @@ namespace Espresso
 		uint8 bo;
 	};
 
+	// returns true if LK bit is set, only valid for branch instructions
+	inline bool DecodeLK(uint32 opcode)
+	{
+		return (opcode & 1) != 0;
+	}
+
 	inline void _decodeForm_I(uint32 opcode, uint32& LI, bool& AA, bool& LK)
 	{
 		LI = opcode & 0x3fffffc;
@@ -183,13 +191,7 @@ namespace Espresso
 		_decodeForm_D_branch(opcode, BD, BO, BI, AA, LK);
 	}
 
-	inline void decodeOp_BCLR(uint32 opcode, BOField& BO, uint32& BI, bool& LK)
-	{
-		// form XL (with BD field expected to be zero)
-		_decodeForm_XL(opcode, BO, BI, LK);
-	}
-
-	inline void decodeOp_BCCTR(uint32 opcode, BOField& BO, uint32& BI, bool& LK)
+	inline void decodeOp_BCSPR(uint32 opcode, BOField& BO, uint32& BI, bool& LK) // BCLR and BCSPR
 	{
 		// form XL (with BD field expected to be zero)
 		_decodeForm_XL(opcode, BO, BI, LK);

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterALU.hpp

@@ -3,12 +3,12 @@ static void PPCInterpreter_setXerOV(PPCInterpreter_t* hCPU, bool hasOverflow)
 {
 	if (hasOverflow)
 	{
-		hCPU->spr.XER |= XER_SO;
-		hCPU->spr.XER |= XER_OV;
+		hCPU->xer_so = 1;
+		hCPU->xer_ov = 1;
 	}
 	else
 	{
-		hCPU->spr.XER &= ~XER_OV;
+		hCPU->xer_ov = 0;
 	}
 }
 
@@ -246,7 +246,7 @@ static void PPCInterpreter_SUBFCO(PPCInterpreter_t* hCPU, uint32 opcode)
 	uint32 a = hCPU->gpr[rA];
 	uint32 b = hCPU->gpr[rB];
 	hCPU->gpr[rD] = ~a + b + 1;
-	// update xer
+	// update carry
 	if (ppc_carry_3(~a, b, 1))
 		hCPU->xer_ca = 1;
 	else
@@ -848,8 +848,7 @@ static void PPCInterpreter_CMP(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else 
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if ((hCPU->spr.XER & XER_SO) != 0)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 
@@ -871,8 +870,7 @@ static void PPCInterpreter_CMPL(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if ((hCPU->spr.XER & XER_SO) != 0)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 
@@ -895,8 +893,7 @@ static void PPCInterpreter_CMPI(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else 
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if (hCPU->spr.XER & XER_SO)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 
@@ -919,8 +916,7 @@ static void PPCInterpreter_CMPLI(PPCInterpreter_t* hCPU, uint32 opcode)
 		hCPU->cr[cr * 4 + CR_BIT_GT] = 1;
 	else
 		hCPU->cr[cr * 4 + CR_BIT_EQ] = 1;
-	if (hCPU->spr.XER & XER_SO)
-		hCPU->cr[cr * 4 + CR_BIT_SO] = 1;
+	hCPU->cr[cr * 4 + CR_BIT_SO] = hCPU->xer_so;
 	PPCInterpreter_nextInstruction(hCPU);
 }
 

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterFPU.cpp

@@ -32,7 +32,7 @@ espresso_frsqrte_entry_t frsqrteLookupTable[32] =
 	{0x20c1000, 0x35e},{0x1f12000, 0x332},{0x1d79000, 0x30a},{0x1bf4000, 0x2e6},
 };
 
-double frsqrte_espresso(double input)
+ATTR_MS_ABI double frsqrte_espresso(double input)
 {
 	unsigned long long x = *(unsigned long long*)&input;
 
@@ -111,7 +111,7 @@ espresso_fres_entry_t fresLookupTable[32] =
 	{0x88400, 0x11a},	{0x65000, 0x11a},	{0x41c00, 0x108},	{0x20c00, 0x106}
 };
 
-double fres_espresso(double input)
+ATTR_MS_ABI double fres_espresso(double input)
 {
 	// based on testing we know that fres uses only the first 15 bits of the mantissa
 	// seee eeee eeee mmmm mmmm mmmm mmmx xxxx ....		(s = sign, e = exponent, m = mantissa, x = not used)

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterInternal.h

@@ -50,9 +50,9 @@
 #define CR_BIT_EQ	2
 #define CR_BIT_SO	3
 
-#define XER_SO			(1<<31)	// summary overflow bit
-#define XER_OV			(1<<30)	// overflow bit
 #define XER_BIT_CA		(29)	// carry bit index. To accelerate frequent access, this bit is stored as a separate uint8
+#define XER_BIT_SO		(31)	// summary overflow, counterpart to CR SO
+#define XER_BIT_OV		(30)
 
 // FPSCR
 #define FPSCR_VXSNAN	(1<<24)
@@ -118,7 +118,8 @@
 
 static inline void ppc_update_cr0(PPCInterpreter_t* hCPU, uint32 r)
 {
-	hCPU->cr[CR_BIT_SO] = (hCPU->spr.XER&XER_SO) ? 1 : 0;
+	cemu_assert_debug(hCPU->xer_so <= 1);
+	hCPU->cr[CR_BIT_SO] = hCPU->xer_so;
 	hCPU->cr[CR_BIT_LT] = ((r != 0) ? 1 : 0) & ((r & 0x80000000) ? 1 : 0);
 	hCPU->cr[CR_BIT_EQ] = (r == 0);
 	hCPU->cr[CR_BIT_GT] = hCPU->cr[CR_BIT_EQ] ^ hCPU->cr[CR_BIT_LT] ^ 1;  // this works because EQ and LT can never be set at the same time. So the only case where GT becomes 1 is when LT=0 and EQ=0
@@ -190,8 +191,8 @@ inline double roundTo25BitAccuracy(double d)
 	return *(double*)&v;
 }
 
-double fres_espresso(double input);
-double frsqrte_espresso(double input);
+ATTR_MS_ABI double fres_espresso(double input);
+ATTR_MS_ABI double frsqrte_espresso(double input);
 
 void fcmpu_espresso(PPCInterpreter_t* hCPU, int crfD, double a, double b);
 

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterLoadStore.hpp

@@ -85,7 +85,8 @@ static void PPCInterpreter_STWCX(PPCInterpreter_t* hCPU, uint32 Opcode)
 			ppc_setCRBit(hCPU, CR_BIT_GT, 0);
 			ppc_setCRBit(hCPU, CR_BIT_EQ, 1);
 		}
-		ppc_setCRBit(hCPU, CR_BIT_SO, (hCPU->spr.XER&XER_SO) != 0 ? 1 : 0);
+		cemu_assert_debug(hCPU->xer_so <= 1);
+		ppc_setCRBit(hCPU, CR_BIT_SO, hCPU->xer_so);
 		// remove reservation
 		hCPU->reservedMemAddr = 0;
 		hCPU->reservedMemValue = 0;

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterMain.cpp

@@ -63,16 +63,24 @@ void PPCInterpreter_setDEC(PPCInterpreter_t* hCPU, uint32 newValue)
 uint32 PPCInterpreter_getXER(PPCInterpreter_t* hCPU)
 {
 	uint32 xerValue = hCPU->spr.XER;
-	xerValue &= ~(1<<XER_BIT_CA);
-	if( hCPU->xer_ca )
-		xerValue |= (1<<XER_BIT_CA);
+	xerValue &= ~(1 << XER_BIT_CA);
+	xerValue &= ~(1 << XER_BIT_SO);
+	xerValue &= ~(1 << XER_BIT_OV);
+	if (hCPU->xer_ca)
+		xerValue |= (1 << XER_BIT_CA);
+	if (hCPU->xer_so)
+		xerValue |= (1 << XER_BIT_SO);
+	if (hCPU->xer_ov)
+		xerValue |= (1 << XER_BIT_OV);
 	return xerValue;
 }
 
 void PPCInterpreter_setXER(PPCInterpreter_t* hCPU, uint32 v)
 {
 	hCPU->spr.XER = v;
-	hCPU->xer_ca = (v>>XER_BIT_CA)&1;
+	hCPU->xer_ca = (v >> XER_BIT_CA) & 1;
+	hCPU->xer_so = (v >> XER_BIT_SO) & 1;
+	hCPU->xer_ov = (v >> XER_BIT_OV) & 1;
 }
 
 uint32 PPCInterpreter_getCoreIndex(PPCInterpreter_t* hCPU)

src/Cafe/HW/Espresso/Interpreter/PPCInterpreterOPC.cpp

@@ -5,7 +5,6 @@
 #include "Cafe/OS/libs/coreinit/coreinit_CodeGen.h"
 
 #include "../Recompiler/PPCRecompiler.h"
-#include "../Recompiler/PPCRecompilerX64.h"
 
 #include <float.h>
 #include "Cafe/HW/Latte/Core/LatteBufferCache.h"

src/Cafe/HW/Espresso/PPCState.h

@@ -49,6 +49,8 @@ struct PPCInterpreter_t
 	uint32 fpscr;
 	uint8 cr[32]; // 0 -> bit not set, 1 -> bit set (upper 7 bits of each byte must always be zero) (cr0 starts at index 0, cr1 at index 4 ..)
 	uint8 xer_ca;  // carry from xer
+	uint8 xer_so;
+	uint8 xer_ov;
 	uint8 LSQE;
 	uint8 PSE;
 	// thread remaining cycles
@@ -67,7 +69,8 @@ struct PPCInterpreter_t
 	uint32 reservedMemValue;
 	// temporary storage for recompiler
 	FPR_t temporaryFPR[8];
-	uint32 temporaryGPR[4];
+	uint32 temporaryGPR[4]; // deprecated, refactor backend dependency on this away
+	uint32 temporaryGPR_reg[4];
 	// values below this are not used by Cafe OS usermode
 	struct  
 	{