Fixed the DSP JIT timing when it is executing in a thread. This allows the Zelda ucode games to run when the "DSPLLE on thread" option has been enabled. However, the DSP Interpreter still hangs when this option is enabled.

* Tightened the timing between the CPU and the DSP in thread mode so that it works closer to how the non-threaded mode works.  The CPU now waits for all of the DSP cycles to be exhausted before adding more cycles.
* DSP Idle skipping has been disabled as it messes up the timing when it is running in a thread.
* Checked for external interrupt requests before entering the dispatcher and inside the dispatcher loop
* Added a critical section around the mailbox read high function

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@7040 8ced0084-cf51-0410-be5f-012b33b47a6e

Source/Core/Core/Src/DSP/DSPCore.cpp

@@ -250,10 +250,6 @@ int DSPCore_RunCycles(int cycles)
 {
 	if (dspjit)
 	{
-		cyclesLeft = cycles;
-		DSPCompiledCode pExecAddr = (DSPCompiledCode)dspjit->enterDispatcher;
-		pExecAddr();
-
 		if (g_dsp.external_interrupt_waiting)
 		{
 			DSPCore_CheckExternalInterrupt();
@@ -261,6 +257,10 @@ int DSPCore_RunCycles(int cycles)
 			DSPCore_SetExternalInterrupt(false);
 		}
 
+		cyclesLeft = cycles;
+		DSPCompiledCode pExecAddr = (DSPCompiledCode)dspjit->enterDispatcher;
+		pExecAddr();
+
 		return cyclesLeft;
 	}
 

Source/Core/Core/Src/DSP/DSPEmitter.cpp

@@ -20,6 +20,7 @@
 #include "DSPEmitter.h"
 #include "DSPMemoryMap.h"
 #include "DSPCore.h"
+#include "DSPHost.h"
 #include "DSPInterpreter.h"
 #include "DSPAnalyzer.h"
 #include "Jit/DSPJitUtil.h"
@@ -269,7 +270,7 @@ void DSPEmitter::Compile(u16 start_addr)
 			DSPJitRegCache c(gpr);
 			HandleLoop();
 			SaveDSPRegs();
-			if (DSPAnalyzer::code_flags[start_addr] & DSPAnalyzer::CODE_IDLE_SKIP)
+			if (!DSPHost_OnThread() && DSPAnalyzer::code_flags[start_addr] & DSPAnalyzer::CODE_IDLE_SKIP)
 			{
 				MOV(16, R(EAX), Imm16(DSP_IDLE_SKIP_CYCLES));
 			}
@@ -307,7 +308,7 @@ void DSPEmitter::Compile(u16 start_addr)
 				DSPJitRegCache c(gpr);
 				//don't update g_dsp.pc -- the branch insn already did
 				SaveDSPRegs();
-				if (DSPAnalyzer::code_flags[start_addr] & DSPAnalyzer::CODE_IDLE_SKIP)
+				if (!DSPHost_OnThread() && DSPAnalyzer::code_flags[start_addr] & DSPAnalyzer::CODE_IDLE_SKIP)
 				{
 					MOV(16, R(EAX), Imm16(DSP_IDLE_SKIP_CYCLES));
 				}
@@ -373,7 +374,7 @@ void DSPEmitter::Compile(u16 start_addr)
 	}
 
 	SaveDSPRegs();
-	if (DSPAnalyzer::code_flags[start_addr] & DSPAnalyzer::CODE_IDLE_SKIP)
+	if (!DSPHost_OnThread() && DSPAnalyzer::code_flags[start_addr] & DSPAnalyzer::CODE_IDLE_SKIP)
 	{
 		MOV(16, R(EAX), Imm16(DSP_IDLE_SKIP_CYCLES));
 	}
@@ -401,6 +402,13 @@ void DSPEmitter::CompileDispatcher()
 
 	const u8 *dispatcherLoop = GetCodePtr();
 
+	FixupBranch exceptionExit;
+	if (DSPHost_OnThread())
+	{
+		CMP(8, M(&g_dsp.external_interrupt_waiting), Imm8(0));
+		exceptionExit = J_CC(CC_NE);
+	}
+
 	// Check for DSP halt
 #ifdef _M_IX86
 	TEST(8, M(&g_dsp.cr), Imm8(CR_HALT));
@@ -440,6 +448,10 @@ void DSPEmitter::CompileDispatcher()
 	
 	// DSP gave up the remaining cycles.
 	SetJumpTarget(_halt);
+	if (DSPHost_OnThread())
+	{
+		SetJumpTarget(exceptionExit);
+	}
 	//MOV(32, M(&cyclesLeft), Imm32(0));
 	ABI_PopAllCalleeSavedRegsAndAdjustStack();
 	RET();

Source/Core/Core/Src/DSP/DSPHWInterface.cpp

@@ -99,7 +99,12 @@ void gdsp_mbox_write_l(u8 mbx, u16 val)
 
 u16 gdsp_mbox_read_h(u8 mbx)
 {
-	return g_dsp.mbox[mbx][0];  // TODO: mask away the top bit?
+	if (DSPHost_OnThread())
+		g_CriticalSection.Enter();
+	u16 val =  g_dsp.mbox[mbx][0];  // TODO: mask away the top bit?
+	if (DSPHost_OnThread())
+		g_CriticalSection.Leave();
+	return val;
 }
 
 

Source/Core/Core/Src/HW/DSPLLE/DSPLLE.cpp

@@ -86,12 +86,15 @@ void DSPLLE::dsp_thread(DSPLLE *lpParameter)
 	{
 		int cycles = (int)dsp_lle->m_cycle_count;
 		if (cycles > 0) {
-			if (dspjit) 
+			if (dspjit)
+			{
 				DSPCore_RunCycles(cycles);
+			}
 			else
+			{
 				DSPInterpreter::RunCycles(cycles);
-
+			}
-			Common::AtomicAdd(dsp_lle->m_cycle_count, -cycles);
+			Common::AtomicStore(dsp_lle->m_cycle_count, 0);
 		}
 		// yield?
 	}
@@ -261,8 +264,10 @@ void DSPLLE::DSP_WriteMailBoxLow(bool _CPUMailbox, u16 _uLowMail)
 
 void DSPLLE::DSP_Update(int cycles)
 {
-	unsigned int dsp_cycles = cycles / 6;  //(jit?20:6);
+	int dsp_cycles = cycles / 6;
 
+	if (dsp_cycles <= 0)
+		return;
 // Sound stream update job has been handled by AudioDMA routine, which is more efficient
 /*
 	// This gets called VERY OFTEN. The soundstream update might be expensive so only do it 200 times per second or something.
@@ -289,10 +294,10 @@ void DSPLLE::DSP_Update(int cycles)
 	}
 	else
 	{
-		// Wait for dsp thread to catch up reasonably. Note: this logic should be thought through.
+		// Wait for dsp thread to complete its cycle. Note: this logic should be thought through.
-		while (m_cycle_count > dsp_cycles)
+		while (m_cycle_count != 0)
 			;
-		Common::AtomicAdd(m_cycle_count, dsp_cycles);
+		Common::AtomicStore(m_cycle_count, dsp_cycles);
 	}
 }