Core/DSPCore: Extended opcode handling fixes

* Make writeToBackLog private to DSPIntExtOps.cpp 
  (JIT variants of 'l and 'ln are disabled and broken as is)
* Make zeroing of the backlog conditional on doing an interpreter fallback and
  do it at a few more places
* Fix selection of cleanup for extended opcodes.
* Fix the DSP unit tests to correctly emit the function prolog/epilog
  (else EBX wouldn't be saved)
* Add a few more DSP unit tests


git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6325 8ced0084-cf51-0410-be5f-012b33b47a6e

8 files changed, 269 insertions, 15 deletions

diff --git a/Source/Core/DSPCore/Src/DSPEmitter.cpp b/Source/Core/DSPCore/Src/DSPEmitter.cpp
index 7b0a27c362..0976d88d5a 100644
--- a/Source/Core/DSPCore/Src/DSPEmitter.cpp
+++ b/Source/Core/DSPCore/Src/DSPEmitter.cpp
@@ -93,6 +93,7 @@ void DSPEmitter::checkExceptions(u32 retval) {
 void DSPEmitter::EmitInstruction(UDSPInstruction inst)
 {
 	const DSPOPCTemplate *tinst = GetOpTemplate(inst);
+	bool ext_is_jit = false;
 
 	// Call extended
 	if (tinst->extended) {
@@ -100,15 +101,19 @@ void DSPEmitter::EmitInstruction(UDSPInstruction inst)
 			if (! extOpTable[inst & 0x7F]->jitFunc) {
 				// Fall back to interpreter
 				ABI_CallFunctionC16((void*)extOpTable[inst & 0x7F]->intFunc, inst);
+				ext_is_jit = false;
 			} else {
 				(this->*extOpTable[inst & 0x7F]->jitFunc)(inst);
+				ext_is_jit = true;
 			}
 		} else {
 			if (!extOpTable[inst & 0xFF]->jitFunc) {
 				// Fall back to interpreter
 				ABI_CallFunctionC16((void*)extOpTable[inst & 0xFF]->intFunc, inst);
+				ext_is_jit = false;
 			} else {
 				(this->*extOpTable[inst & 0xFF]->jitFunc)(inst);
+				ext_is_jit = true;
 			}
 		}
 	}
@@ -125,8 +130,10 @@ void DSPEmitter::EmitInstruction(UDSPInstruction inst)
 
 	// Backlog
 	if (tinst->extended) {
-		if (! extOpTable[inst & 0x7F]->jitFunc) {
-			ABI_CallFunction((void*)applyWriteBackLog);
+		if (!ext_is_jit) {
+			//need to call the online cleanup function because
+			//the writeBackLog gets populated at runtime
+			ABI_CallFunction((void*)::applyWriteBackLog);
 		} else {
 			popExtValueToReg();
 		}
diff --git a/Source/Core/DSPCore/Src/DSPEmitter.h b/Source/Core/DSPCore/Src/DSPEmitter.h
index 8ddd9aa368..95836de01c 100644
--- a/Source/Core/DSPCore/Src/DSPEmitter.h
+++ b/Source/Core/DSPCore/Src/DSPEmitter.h
@@ -64,6 +64,7 @@ public:
 	void popExtValueToReg();
 	void pushExtValueFromMem(u16 dreg, u16 sreg);
 	
+	void zeroWriteBackLog(const UDSPInstruction opc);
 	// Ext commands
 	void l(const UDSPInstruction opc);
 	void ln(const UDSPInstruction opc);
diff --git a/Source/Core/DSPCore/Src/DSPIntExtOps.cpp b/Source/Core/DSPCore/Src/DSPIntExtOps.cpp
index 0486d2a238..ce4f29585e 100644
--- a/Source/Core/DSPCore/Src/DSPIntExtOps.cpp
+++ b/Source/Core/DSPCore/Src/DSPIntExtOps.cpp
@@ -28,6 +28,13 @@
 // registers will wrap in odd ways, dictated by the corresponding wrapping
 // register, WR0-3.
 
+// Needs comments.
+inline static void writeToBackLog(int i, int idx, u16 value)
+{
+	writeBackLog[i] = value;
+	writeBackLogIdx[i] = idx;
+}
+
 namespace DSPInterpreter
 {
 
diff --git a/Source/Core/DSPCore/Src/DSPIntExtOps.h b/Source/Core/DSPCore/Src/DSPIntExtOps.h
index c154af24ee..46b809ca14 100644
--- a/Source/Core/DSPCore/Src/DSPIntExtOps.h
+++ b/Source/Core/DSPCore/Src/DSPIntExtOps.h
@@ -61,11 +61,4 @@ void nop(const UDSPInstruction opc);
 } // end namespace Ext
 } // end namespace DSPinterpeter
 
-// Needs comments.
-inline void writeToBackLog(int i, int idx, u16 value)
-{
-	writeBackLog[i] = value;
-	writeBackLogIdx[i] = idx;
-}
-
 #endif
diff --git a/Source/Core/DSPCore/Src/Jit/DSPJitExtOps.cpp b/Source/Core/DSPCore/Src/Jit/DSPJitExtOps.cpp
index fa5e1743cd..d151bf1194 100644
--- a/Source/Core/DSPCore/Src/Jit/DSPJitExtOps.cpp
+++ b/Source/Core/DSPCore/Src/Jit/DSPJitExtOps.cpp
@@ -19,8 +19,6 @@
 #include "x64Emitter.h"
 #include "ABI.h"
 
-#include "../DSPIntExtOps.h" // remove when getting rid of writebacklog
-// See docs in the interpeter
 using namespace Gen;
 
 // DR $arR
@@ -94,6 +92,7 @@ void DSPEmitter::l(const UDSPInstruction opc)
 	
 	if ((dreg >= DSP_REG_ACM0) && (g_dsp.r[DSP_REG_SR] & SR_40_MODE_BIT)) 
 	{
+/*
 		u16 val;
 		ext_dmem_read(sreg);
 		MOV(16, M(&val), R(EAX));
@@ -102,6 +101,7 @@ void DSPEmitter::l(const UDSPInstruction opc)
 		writeToBackLog(1, dreg,	val);
 		writeToBackLog(2, dreg - DSP_REG_ACM0 + DSP_REG_ACL0, 0);
 		increment_addr_reg(sreg);
+*/
 	}
 	else
 	{
@@ -127,6 +127,7 @@ void DSPEmitter::ln(const UDSPInstruction opc)
 
 	if ((dreg >= DSP_REG_ACM0) && (g_dsp.r[DSP_REG_SR] & SR_40_MODE_BIT)) 
 	{
+/*
 		u16 val;
 		ext_dmem_read(sreg);
 		MOV(16, M(&val), R(EAX));
@@ -134,6 +135,7 @@ void DSPEmitter::ln(const UDSPInstruction opc)
 		writeToBackLog(1, dreg,	val);
 		writeToBackLog(2, dreg - DSP_REG_ACM0 + DSP_REG_ACL0, 0);
 		increase_addr_reg(sreg);
+*/
 	}
 	else
 	{
@@ -700,8 +702,8 @@ void DSPEmitter::pushExtValueFromReg(u16 dreg, u16 sreg) {
 
 void DSPEmitter::pushExtValueFromMem(u16 dreg, u16 sreg) {
 	ext_dmem_read(sreg);
-	MOV(16, R(EBX), R(EAX));
-	
+	MOVZX(32, 16, EBX, R(EAX));
+
 	storeIndex = dreg;
 }
 
@@ -726,3 +728,26 @@ void DSPEmitter::popExtValueToReg() {
 	// TODO handle commands such as 'l
 }
 
+// This function is being called in the main op after all input regs were read
+// and before it writes into any regs. This way we can always use bitwise or to
+// apply the ext command output, because if the main op didn't change the value
+// then 0 | ext output = ext output and if it did then bitwise or is still the
+// right thing to do
+//this is only needed as long as we do fallback for ext ops
+void DSPEmitter::zeroWriteBackLog(const UDSPInstruction opc)
+{
+	const DSPOPCTemplate *tinst = GetOpTemplate(opc);
+
+	// Call extended
+	if (!tinst->extended)
+	    return;
+
+	if ((opc >> 12) == 0x3) {
+		if (! extOpTable[opc & 0x7F]->jitFunc)
+			ABI_CallFunction((void*)::zeroWriteBackLog);
+	} else {
+		if (! extOpTable[opc & 0xFF]->jitFunc)
+			ABI_CallFunction((void*)::zeroWriteBackLog);
+	}
+	return;
+}
diff --git a/Source/Core/DSPCore/Src/Jit/DSPJitMisc.cpp b/Source/Core/DSPCore/Src/Jit/DSPJitMisc.cpp
index dafa3bea8f..4a2e3c6540 100644
--- a/Source/Core/DSPCore/Src/Jit/DSPJitMisc.cpp
+++ b/Source/Core/DSPCore/Src/Jit/DSPJitMisc.cpp
@@ -88,6 +88,7 @@ void DSPEmitter::nx(const UDSPInstruction opc)
 void DSPEmitter::dar(const UDSPInstruction opc)
 {
 	//	g_dsp.r[opc & 0x3] = dsp_decrement_addr_reg(opc & 0x3);
+	zeroWriteBackLog(opc);
 	decrement_addr_reg(opc & 0x3);
 
 }
@@ -98,6 +99,7 @@ void DSPEmitter::dar(const UDSPInstruction opc)
 void DSPEmitter::iar(const UDSPInstruction opc)
 {
 	//	g_dsp.r[opc & 0x3] = dsp_increment_addr_reg(opc & 0x3);
+	zeroWriteBackLog(opc);
 	increment_addr_reg(opc & 0x3);
 }
 
@@ -109,6 +111,7 @@ void DSPEmitter::subarn(const UDSPInstruction opc)
 {
 	//	u8 dreg = opc & 0x3;
 	//	g_dsp.r[dreg] = dsp_decrease_addr_reg(dreg, (s16)g_dsp.r[DSP_REG_IX0 + dreg]);
+	zeroWriteBackLog(opc);
 	decrease_addr_reg(opc & 0x3);
 }
 
@@ -123,6 +126,7 @@ void DSPEmitter::addarn(const UDSPInstruction opc)
 	//	g_dsp.r[dreg] = dsp_increase_addr_reg(dreg, (s16)g_dsp.r[DSP_REG_IX0 + sreg]);
 	
 	// From looking around it is always called with the matching index register
+	zeroWriteBackLog(opc);
 	increase_addr_reg(opc & 0x3);
 }
 
@@ -180,7 +184,7 @@ void DSPEmitter::sbset(const UDSPInstruction opc)
 // but it's harder to know exactly what effect they have.
 void DSPEmitter::srbith(const UDSPInstruction opc)
 {
-	ABI_CallFunction((void *)zeroWriteBackLog);
+	zeroWriteBackLog(opc);
 	switch ((opc >> 8) & 0xf)
 	{
 	// M0/M2 change the multiplier mode (it can multiply by 2 for free).
diff --git a/Source/UnitTests/AudioJitTests.cpp b/Source/UnitTests/AudioJitTests.cpp
index cfabaeeb6c..4cef5931fa 100644
--- a/Source/UnitTests/AudioJitTests.cpp
+++ b/Source/UnitTests/AudioJitTests.cpp
@@ -28,6 +28,15 @@ void nx_nr()
 	tester.Report();
 }
 
+void nx_mv()
+{
+	DSPJitTester tester(0x8000, 0x0010);
+	tester.AddTestData(DSP_REG_ACL0);
+	tester.AddTestData(DSP_REG_AXL0);
+	tester.TestAll(true);
+	tester.Report();
+}
+
 void dar()
 {
 	DSPJitTester tester(0x0004);
@@ -81,14 +90,205 @@ void nx_s()
 {
 	DSPJitTester tester(0x8000, 0x0020);
 	tester.AddTestData(DSP_REG_AR0);
+	tester.AddTestData(DSP_REG_WR0);
 	tester.AddTestData(DSP_REG_ACL0);
 	tester.TestAll(true);
 	tester.Report();
 }
+
+void nx_sn() 
+{
+	DSPJitTester tester(0x8000, 0x0024);
+	tester.AddTestData(DSP_REG_AR0);
+	tester.AddTestData(DSP_REG_WR0);
+	tester.AddTestData(DSP_REG_IX0);
+	tester.AddTestData(DSP_REG_ACL0);
+	tester.TestAll(true);
+	tester.Report();
+}
+
+void nx_l() 
+{
+
+	DSPJitTester tester(0x8000, 0x0040);
+	tester.AddTestData(DSP_REG_AR0);
+	tester.AddTestData(DSP_REG_WR0);
+	tester.AddTestData(DSP_REG_AXL0);
+	tester.TestAll(true);
+	tester.Report();
+}
+
+void set16_l() 
+{
+
+	DSPJitTester tester(0x8e00, 0x0070);
+	tester.AddTestData(DSP_REG_SR, 0);
+	tester.AddTestData(DSP_REG_SR, SR_40_MODE_BIT);
+	tester.AddTestData(DSP_REG_AR0);
+	tester.AddTestData(DSP_REG_WR0);
+	tester.AddTestData(DSP_REG_ACM0);
+	tester.TestAll(true);
+	tester.Report();
+}
+
+void nx_ln() 
+{
+	DSPJitTester tester(0x8000, 0x0044);
+	tester.AddTestData(DSP_REG_AR0);
+	tester.AddTestData(DSP_REG_WR0);
+	tester.AddTestData(DSP_REG_IX0);
+	tester.AddTestData(DSP_REG_AXL0);
+	tester.TestAll(true);
+	tester.Report();
+}
+
+void nx_ls() 
+{
+	DSPJitTester tester1(0x8000, 0x0080);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x0080);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_lsn() 
+{
+	DSPJitTester tester1(0x8000, 0x0084);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.AddTestData(DSP_REG_IX0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x0084);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_lsm() 
+{
+	DSPJitTester tester1(0x8000, 0x0088);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x0088);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.AddTestData(DSP_REG_IX3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_lsnm() 
+{
+	DSPJitTester tester1(0x8000, 0x008c);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.AddTestData(DSP_REG_IX0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x008c);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.AddTestData(DSP_REG_IX3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_sl() 
+{
+	DSPJitTester tester1(0x8000, 0x0082);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x0082);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_sln() 
+{
+	DSPJitTester tester1(0x8000, 0x0086);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.AddTestData(DSP_REG_IX0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x0086);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_slm() 
+{
+	DSPJitTester tester1(0x8000, 0x008a);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x008a);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.AddTestData(DSP_REG_IX3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
+void nx_slnm() 
+{
+	DSPJitTester tester1(0x8000, 0x008e);
+	tester1.AddTestData(DSP_REG_ACM0);
+	tester1.AddTestData(DSP_REG_AR0);
+	tester1.AddTestData(DSP_REG_WR0);
+	tester1.AddTestData(DSP_REG_IX0);
+	tester1.TestAll(true);
+	tester1.Report();
+
+	DSPJitTester tester2(0x8000, 0x008e);
+	tester2.AddTestData(DSP_REG_ACM0);
+	tester2.AddTestData(DSP_REG_AR3);
+	tester2.AddTestData(DSP_REG_WR3);
+	tester2.AddTestData(DSP_REG_IX3);
+	tester2.TestAll(true);
+	tester2.Report();
+}
+
 void AudioJitTests()
 {
 	DSPJitTester::Initialize();
-
 	dar();
 	iar();
 	subarn();
@@ -99,7 +299,22 @@ void AudioJitTests()
 	nx_ir();
 	nx_dr();
 	nx_nr();
+	nx_mv();
+
+	set16_l();
+
 	nx_s();
+	nx_sn();
+	nx_l();
+	nx_ln();
+	nx_ls();
+	nx_lsn();
+	nx_lsm();
+	nx_lsnm();
+	nx_sl();
+	nx_sln();
+	nx_slm();
+	nx_slnm();
 }
 
 //required to be able to link against DSPCore
diff --git a/Source/UnitTests/DSPJitTester.cpp b/Source/UnitTests/DSPJitTester.cpp
index 2012ee09b8..c0b95ca35e 100644
--- a/Source/UnitTests/DSPJitTester.cpp
+++ b/Source/UnitTests/DSPJitTester.cpp
@@ -41,7 +41,9 @@ SDSP DSPJitTester::RunJit(SDSP dsp_settings)
 	ResetJit();
 	memcpy(&g_dsp, &dsp_settings, sizeof(SDSP));
 	const u8* code = jit.GetCodePtr();
+	jit.ABI_PushAllCalleeSavedRegsAndAdjustStack();
 	jit.EmitInstruction(instruction);
+	jit.ABI_PopAllCalleeSavedRegsAndAdjustStack();
 	jit.RET();
 	((void(*)())code)();