__forceinline void UpdateSpdifMode()
{
int OPM=PlayMode;
if(Spdif.Out&0x4) // use 24/32bit PCM data streaming
{
PlayMode=8;
ConLog("* SPU2-X: WARNING: Possibly CDDA mode set!\n");
return;
}
if(Spdif.Out&SPDIF_OUT_BYPASS)
{
PlayMode=2;
if(!(Spdif.Mode&SPDIF_MODE_BYPASS_BITSTREAM))
PlayMode=4; //bitstream bypass
}
else
{
PlayMode=0; //normal processing
if(Spdif.Out&SPDIF_OUT_PCM)
{
PlayMode=1;
}
}
if(OPM!=PlayMode)
{
ConLog("* SPU2-X: Play Mode Set to %s (%d).\n",
(PlayMode==0) ? "Normal" : ((PlayMode==1) ? "PCM Clone" : ((PlayMode==2) ? "PCM Bypass" : "BitStream Bypass")),PlayMode);
}
}
void SndBuffer::_WriteSamples(StereoOut32 *bData, int nSamples)
{
m_predictData = 0;
// Problem:
// If the SPU2 gets out of sync with the SndOut device, the writepos of the
// circular buffer will overtake the readpos, leading to a prolonged period
// of hopscotching read/write accesses (ie, lots of staticy crap sound for
// several seconds).
//
// Compromise:
// When an overrun occurs, we adapt by discarding a portion of the buffer.
// The older portion of the buffer is discarded rather than incoming data,
// so that the overall audio synchronization is better.
int free = m_size - _GetApproximateDataInBuffer(); // -1, but the <= handles that
if( free <= nSamples )
{
// Disabled since the lock-free queue can't handle changing the read end from the write thread
#if 0
// Buffer overrun!
// Dump samples from the read portion of the buffer instead of dropping
// the newly written stuff.
s32 comp;
if( SynchMode == 0 ) // TimeStrech on
{
comp = timeStretchOverrun();
}
else
{
// Toss half the buffer plus whatever's being written anew:
comp = GetAlignedBufferSize( (m_size + nSamples ) / 16 );
if( comp > (m_size-SndOutPacketSize) ) comp = m_size-SndOutPacketSize;
}
_DropSamples_Internal(comp);
if( MsgOverruns() )
ConLog(" * SPU2 > Overrun Compensation (%d packets tossed)\n", comp / SndOutPacketSize );
lastPct = 0.0; // normalize the timestretcher
#else
if( MsgOverruns() )
ConLog(" * SPU2 > Overrun! 1 packet tossed)\n");
lastPct = 0.0; // normalize the timestretcher
return;
#endif
}
_WriteSamples_Safe(bData, nSamples);
}
// Core 1 Input is "CDDA mode" - Source audio data is 32 bits.
// PS2 note: Very! few PS2 games use this mode. Some PSX games used it, however no
// *known* PS2 game does since it was likely only available if the game was recorded to CD
// media (ie, not available in DVD mode, which almost all PS2 games use). Plus PS2 games
// generally prefer to use ADPCM streaming audio since they need as much storage space as
// possible for FMVs and high-def textures.
//
StereoOut32 V_Core::ReadInput_HiFi()
{
if (psxmode)
ConLog("ReadInput_HiFi!!!!!\n");
InputPosRead &= ~1;
//
//#ifdef PCM24_S1_INTERLEAVE
// StereoOut32 retval(
// *((s32*)(ADMATempBuffer+(InputPosRead<<1))),
// *((s32*)(ADMATempBuffer+(InputPosRead<<1)+2))
// );
//#else
// StereoOut32 retval(
// (s32&)(ADMATempBuffer[InputPosRead]),
// (s32&)(ADMATempBuffer[InputPosRead+0x200])
// );
//#endif
StereoOut32 retval(
(s32 &)(*GetMemPtr(0x2000 + (Index << 10) + InputPosRead)),
(s32 &)(*GetMemPtr(0x2200 + (Index << 10) + InputPosRead)));
if (Index == 1) {
// CDDA Mode:
// give 30 bit data (SndOut downsamples the rest of the way)
// HACKFIX: 28 bits seems better according to rama. I should take some time and do some
// bitcounting on this one. --air
retval.Left >>= 4;
retval.Right >>= 4;
}
EXPORT_C_(s32) SPU2init()
{
#define MAKESURE(a,b) \
/*fprintf(stderr,"%08p: %08p == %08p\n",&(regtable[a>>1]),regtable[a>>1],U16P(b));*/ \
assert(regtable[(a)>>1]==U16P(b))
MAKESURE(0x800,zero);
s32 c=0,v=0;
ReadSettings();
#ifdef SPU2_LOG
if(AccessLog())
{
spu2Log = _wfopen( AccessLogFileName, _T("w") );
setvbuf(spu2Log, NULL, _IONBF, 0);
FileLog("SPU2init\n");
}
#endif
srand((unsigned)time(NULL));
disableFreezes=false;
if (spu2init)
{
ConLog( " * SPU2: Already initialized - Ignoring SPU2init signal." );
return 0;
}
spu2init=true;
spu2regs = (short*)malloc(0x010000);
_spu2mem = (short*)malloc(0x200000);
// adpcm decoder cache:
// the cache data size is determined by taking the number of adpcm blocks
// (2MB / 16) and multiplying it by the decoded block size (28 samples).
// Thus: pcm_cache_data = 7,340,032 bytes (ouch!)
// Expanded: 16 bytes expands to 56 bytes [3.5:1 ratio]
// Resulting in 2MB * 3.5.
pcm_cache_data = (PcmCacheEntry*)calloc( pcm_BlockCount, sizeof(PcmCacheEntry) );
if( (spu2regs == NULL) || (_spu2mem == NULL) ||
(pcm_cache_data == NULL) )
{
SysMessage("SPU2: Error allocating Memory\n"); return -1;
}
for(int mem=0;mem<0x800;mem++)
{
u16 *ptr=regtable[mem>>1];
if(!ptr) {
regtable[mem>>1] = &(spu2Ru16(mem));
}
}
void SPU_ps1_write(u32 mem, u16 value)
{
bool show=true;
u32 reg = mem&0xffff;
if((reg>=0x1c00)&&(reg<0x1d80))
{
//voice values
u8 voice = ((reg-0x1c00)>>4);
u8 vval = reg&0xf;
switch(vval)
{
case 0: //VOLL (Volume L)
Cores[0].Voices[voice].Volume.Left.Mode = 0;
Cores[0].Voices[voice].Volume.Left.RegSet( value << 1 );
Cores[0].Voices[voice].Volume.Left.Reg_VOL = value;
break;
case 1: //VOLR (Volume R)
Cores[0].Voices[voice].Volume.Right.Mode = 0;
Cores[0].Voices[voice].Volume.Right.RegSet( value << 1 );
Cores[0].Voices[voice].Volume.Right.Reg_VOL = value;
break;
case 2: Cores[0].Voices[voice].Pitch = value; break;
case 3: Cores[0].Voices[voice].StartA = (u32)value<<8; break;
case 4: // ADSR1 (Envelope)
Cores[0].Voices[voice].ADSR.AttackMode = (value & 0x8000)>>15;
Cores[0].Voices[voice].ADSR.AttackRate = (value & 0x7F00)>>8;
Cores[0].Voices[voice].ADSR.DecayRate = (value & 0xF0)>>4;
Cores[0].Voices[voice].ADSR.SustainLevel = (value & 0xF);
Cores[0].Voices[voice].ADSR.Reg_ADSR1 = value;
break;
case 5: // ADSR2 (Envelope)
Cores[0].Voices[voice].ADSR.SustainMode = (value & 0xE000)>>13;
Cores[0].Voices[voice].ADSR.SustainRate = (value & 0x1FC0)>>6;
Cores[0].Voices[voice].ADSR.ReleaseMode = (value & 0x20)>>5;
Cores[0].Voices[voice].ADSR.ReleaseRate = (value & 0x1F);
Cores[0].Voices[voice].ADSR.Reg_ADSR2 = value;
break;
case 6:
Cores[0].Voices[voice].ADSR.Value = ((s32)value<<16) | value;
ConLog( "* SPU2: Mysterious ADSR Volume Set to 0x%x", value );
break;
case 7: Cores[0].Voices[voice].LoopStartA = (u32)value <<8; break;
jNO_DEFAULT;
}
}
void DspCloseLibrary()
#ifdef USE_A_THREAD
{
if(!dspPluginEnabled) return ;
PostThreadMessage(UpdateThreadId,WM_QUIT,0,0);
running=false;
if(WaitForSingleObject(hUpdateThread,1000)==WAIT_TIMEOUT)
{
ConLog("SPU2-X: WARNING: DSP Thread did not close itself in time. Assuming hung. Terminating.\n");
TerminateThread(hUpdateThread,1);
}
}
void V_Core::StartADMAWrite(u16 *pMem, u32 sz)
{
#ifndef ENABLE_NEW_IOPDMA_SPU2
int size = (sz)&(~511);
if(MsgAutoDMA()) ConLog("* SPU2-X: DMA%c AutoDMA Transfer of %d bytes to %x (%02x %x %04x).\n",
GetDmaIndexChar(), size<<1, TSA, DMABits, AutoDMACtrl, (~Regs.ATTR)&0x7fff);
InputDataProgress = 0;
if((AutoDMACtrl&(Index+1))==0)
{
TSA = 0x2000 + (Index<<10);
DMAICounter = size;
}
else if(size>=512)
{
InputDataLeft = size;
if(AdmaInProgress==0)
{
#ifdef PCM24_S1_INTERLEAVE
if((Index==1)&&((PlayMode&8)==8))
{
AutoDMAReadBuffer(Index,1);
}
else
{
AutoDMAReadBuffer(Index,0);
}
#else
if( ((PlayMode&4)==4) && (Index==0) )
Cores[0].InputPosRead=0;
AutoDMAReadBuffer(0);
#endif
if(size==512)
DMAICounter = size;
}
AdmaInProgress = 1;
}
else
{
InputDataLeft = 0;
DMAICounter = 1;
}
TADR = MADR + (size<<1);
#endif
}
void SndBuffer::timeStretchWrite()
{
bool progress = false;
// data prediction helps keep the tempo adjustments more accurate.
// The timestretcher returns packets in belated "clump" form.
// Meaning that most of the time we'll get nothing back, and then
// suddenly we'll get several chunks back at once. Thus we use
// data prediction to make the timestretcher more responsive.
PredictDataWrite( (int)( SndOutPacketSize / eTempo ) );
CvtPacketToFloat( sndTempBuffer );
pSoundTouch->putSamples( (float*)sndTempBuffer, SndOutPacketSize );
int tempProgress;
while( tempProgress = pSoundTouch->receiveSamples( (float*)sndTempBuffer, SndOutPacketSize),
tempProgress != 0 )
{
// Hint: It's assumed that pSoundTouch will return chunks of 128 bytes (it always does as
// long as the SSE optimizations are enabled), which means we can do our own SSE opts here.
CvtPacketToInt( sndTempBuffer, tempProgress );
_WriteSamples( sndTempBuffer, tempProgress );
progress = true;
}
#ifdef SPU2X_USE_OLD_STRETCHER
UpdateTempoChangeSoundTouch();
#else
UpdateTempoChangeSoundTouch2();
#endif
if( MsgOverruns() )
{
if( progress )
{
if( ++ts_stats_logcounter > 150 )
{
ts_stats_logcounter = 0;
ConLog( " * SPU2 > Timestretch Stats > %d percent stretched. Total stretchblocks = %d.\n",
( ts_stats_stretchblocks * 100 ) / ( ts_stats_normalblocks + ts_stats_stretchblocks ),
ts_stats_stretchblocks);
ts_stats_normalblocks = 0;
ts_stats_stretchblocks = 0;
}
}
}
}
// writes a signed value to the SPU2 ram
// Invalidates the ADPCM cache in the process.
__forceinline void spu2M_Write( u32 addr, s16 value )
{
// Make sure the cache is invalidated:
// (note to self : addr address WORDs, not bytes)
addr &= 0xfffff;
if( addr >= SPU2_DYN_MEMLINE )
{
const int cacheIdx = addr / pcm_WordsPerBlock;
pcm_cache_data[cacheIdx].Validated = false;
if(MsgToConsole()) ConLog( "* SPU2-X: PcmCache Block Clear at 0x%x (cacheIdx=0x%x)\n", addr, cacheIdx);
}
*GetMemPtr( addr ) = value;
}
void V_Core::WriteRegPS1( u32 mem, u16 value )
{
jASSUME( Index == 0 ); // Valid on Core 0 only!
bool show = true;
u32 reg = mem & 0xffff;
if((reg>=0x1c00)&&(reg<0x1d80))
{
//voice values
u8 voice = ((reg-0x1c00)>>4);
u8 vval = reg&0xf;
switch(vval)
{
case 0: //VOLL (Volume L)
Voices[voice].Volume.Left.Mode = 0;
Voices[voice].Volume.Left.RegSet( value << 1 );
Voices[voice].Volume.Left.Reg_VOL = value;
break;
case 1: //VOLR (Volume R)
Voices[voice].Volume.Right.Mode = 0;
Voices[voice].Volume.Right.RegSet( value << 1 );
Voices[voice].Volume.Right.Reg_VOL = value;
break;
case 2: Voices[voice].Pitch = value; break;
case 3: Voices[voice].StartA = (u32)value<<8; break;
case 4: // ADSR1 (Envelope)
Voices[voice].ADSR.regADSR1 = value;
break;
case 5: // ADSR2 (Envelope)
Voices[voice].ADSR.regADSR2 = value;
break;
case 6:
Voices[voice].ADSR.Value = ((s32)value<<16) | value;
ConLog( "* SPU2: Mysterious ADSR Volume Set to 0x%x", value );
break;
case 7: Voices[voice].LoopStartA = (u32)value <<8; break;
jNO_DEFAULT;
}
}
void StartADMAWrite(int core,u16 *pMem, u32 sz)
{
int size=(sz)&(~511);
if(MsgAutoDMA()) ConLog(" * SPU2: DMA%c AutoDMA Transfer of %d bytes to %x (%02x %x %04x).\n",
(core==0)?'4':'7',size<<1,Cores[core].TSA,Cores[core].DMABits,Cores[core].AutoDMACtrl,(~Cores[core].Regs.ATTR)&0x7fff);
Cores[core].InputDataProgress=0;
if((Cores[core].AutoDMACtrl&(core+1))==0)
{
Cores[core].TSA=0x2000+(core<<10);
Cores[core].DMAICounter=size;
}
else if(size>=512)
{
Cores[core].InputDataLeft=size;
if(Cores[core].AdmaInProgress==0)
{
#ifdef PCM24_S1_INTERLEAVE
if((core==1)&&((PlayMode&8)==8))
{
AutoDMAReadBuffer(core,1);
}
else
{
AutoDMAReadBuffer(core,0);
}
#else
if(((PlayMode&4)==4)&&(core==0))
Cores[0].InputPos=0;
AutoDMAReadBuffer(core,0);
#endif
if(size==512)
Cores[core].DMAICounter=size;
}
Cores[core].AdmaInProgress=1;
}
else
{
Cores[core].InputDataLeft=0;
Cores[core].DMAICounter=1;
}
Cores[core].TADR=Cores[core].MADR+(size<<1);
}
void V_Core::DoDMAwrite(u16* pMem, u32 size)
{
#ifndef ENABLE_NEW_IOPDMA_SPU2
DMAPtr = pMem;
if(size<2) {
//if(dma7callback) dma7callback();
Regs.STATX &= ~0x80;
//Regs.ATTR |= 0x30;
DMAICounter=1;
return;
}
if( IsDevBuild )
{
DebugCores[Index].lastsize = size;
DebugCores[Index].dmaFlag = 2;
}
TSA &= ~7;
bool adma_enable = ((AutoDMACtrl&(Index+1))==(Index+1));
if(adma_enable)
{
TSA&=0x1fff;
StartADMAWrite(pMem,size);
}
else
{
if(MsgDMA()) ConLog("* SPU2-X: DMA%c Transfer of %d bytes to %x (%02x %x %04x). IRQE = %d IRQA = %x \n",
GetDmaIndexChar(),size<<1,TSA,DMABits,AutoDMACtrl,(~Regs.ATTR)&0x7fff,
Cores[0].IRQEnable, Cores[0].IRQA);
PlainDMAWrite(pMem,size);
}
Regs.STATX &= ~0x80;
//Regs.ATTR |= 0x30;
#endif
}
// Returns TRUE if there is data to be output, or false if no data
// is available to be copied.
bool SndBuffer::CheckUnderrunStatus( int& nSamples, int& quietSampleCount )
{
quietSampleCount = 0;
int data = _GetApproximateDataInBuffer();
if( m_underrun_freeze )
{
int toFill = m_size / ( (SynchMode == 2) ? 32 : 400); // TimeStretch and Async off?
toFill = GetAlignedBufferSize( toFill );
// toFill is now aligned to a SndOutPacket
if( data < toFill )
{
quietSampleCount = nSamples;
return false;
}
m_underrun_freeze = false;
if( MsgOverruns() )
ConLog(" * SPU2 > Underrun compensation (%d packets buffered)\n", toFill / SndOutPacketSize );
lastPct = 0.0; // normalize timestretcher
}
else if( data < nSamples )
{
nSamples = data;
quietSampleCount = SndOutPacketSize - data;
m_underrun_freeze = true;
if( SynchMode == 0 ) // TimeStrech on
timeStretchUnderrun();
return nSamples != 0;
}
return true;
}
s32 V_Core::NewDmaWrite(u32* data, u32 bytesLeft, u32* bytesProcessed)
{
#ifdef ENABLE_NEW_IOPDMA_SPU2
bool DmaStarting = !DmaStarted;
DmaStarted = true;
if(bytesLeft<2)
{
// execute interrupt code early
NewDmaInterrupt();
*bytesProcessed = bytesLeft;
return 0;
}
if( IsDevBuild )
{
DebugCores[Index].lastsize = bytesLeft;
DebugCores[Index].dmaFlag = 1;
}
TSA &= ~7;
bool adma_enable = ((AutoDMACtrl&(Index+1))==(Index+1));
if(adma_enable)
{
TSA&=0x1fff;
if(MsgAutoDMA() && DmaStarting) ConLog("* SPU2-X: DMA%c AutoDMA Transfer of %d bytes to %x (%02x %x %04x).\n",
GetDmaIndexChar(), bytesLeft<<1, TSA, DMABits, AutoDMACtrl, (~Regs.ATTR)&0x7fff);
u32 processed = 0;
while((AutoDmaFree>0)&&(bytesLeft>=0x400))
{
// copy block
LogAutoDMA( Index ? ADMA7LogFile : ADMA4LogFile );
// HACKFIX!! DMAPtr can be invalid after a savestate load, so the savestate just forces it
// to NULL and we ignore it here. (used to work in old VM editions of PCSX2 with fixed
// addressing, but new PCSX2s have dynamic memory addressing).
s16* mptr = (s16*)data;
if(false)//(mode)
{
//memcpy((ADMATempBuffer+(InputPosWrite<<1)),mptr,0x400);
memcpy(GetMemPtr(0x2000+(Index<<10)+InputPosWrite),mptr,0x400);
mptr+=0x200;
// Flag interrupt? If IRQA occurs between start and dest, flag it.
// Important: Test both core IRQ settings for either DMA!
u32 dummyTSA = 0x2000+(Index<<10)+InputPosWrite;
u32 dummyTDA = 0x2000+(Index<<10)+InputPosWrite+0x200;
for( int i=0; i<2; i++ )
{
if( Cores[i].IRQEnable && (Cores[i].IRQA >= dummyTSA) && (Cores[i].IRQA < dummyTDA) )
{
SetIrqCall(i);
}
}
}
else
{
//memcpy((ADMATempBuffer+InputPosWrite),mptr,0x200);
memcpy(GetMemPtr(0x2000+(Index<<10)+InputPosWrite),mptr,0x200);
mptr+=0x100;
// Flag interrupt? If IRQA occurs between start and dest, flag it.
// Important: Test both core IRQ settings for either DMA!
u32 dummyTSA = 0x2000+(Index<<10)+InputPosWrite;
u32 dummyTDA = 0x2000+(Index<<10)+InputPosWrite+0x100;
for( int i=0; i<2; i++ )
{
if( Cores[i].IRQEnable && (Cores[i].IRQA >= dummyTSA) && (Cores[i].IRQA < dummyTDA) )
{
SetIrqCall(i);
}
}
//memcpy((ADMATempBuffer+InputPosWrite+0x200),mptr,0x200);
memcpy(GetMemPtr(0x2200+(Index<<10)+InputPosWrite),mptr,0x200);
mptr+=0x100;
// Flag interrupt? If IRQA occurs between start and dest, flag it.
// Important: Test both core IRQ settings for either DMA!
dummyTSA = 0x2200+(Index<<10)+InputPosWrite;
dummyTDA = 0x2200+(Index<<10)+InputPosWrite+0x100;
for( int i=0; i<2; i++ )
{
if( Cores[i].IRQEnable && (Cores[i].IRQA >= dummyTSA) && (Cores[i].IRQA < dummyTDA) )
{
SetIrqCall(i);
}
}
}
// See ReadInput at mixer.cpp for explanation on the commented out lines
//
InputPosWrite = (InputPosWrite + 0x100) & 0x1ff;
AutoDmaFree -= 0x200;
processed += 0x400;
bytesLeft -= 0x400;
}
if(processed==0)
{
*bytesProcessed = 0;
return 768*15; // pause a bit
}
else
{
*bytesProcessed = processed;
return 0; // auto pause
}
}
else
{
if(MsgDMA() && DmaStarting) ConLog("* SPU2-X: DMA%c Transfer of %d bytes to %x (%02x %x %04x).\n",
GetDmaIndexChar(),bytesLeft,TSA,DMABits,AutoDMACtrl,(~Regs.ATTR)&0x7fff);
if(bytesLeft> 2048)
bytesLeft = 2048;
// TODO: Sliced transfers?
PlainDMAWrite((u16*)data,bytesLeft/2);
}
Regs.STATX &= ~0x80;
Regs.STATX |= 0x400;
#endif
*bytesProcessed = bytesLeft;
return 0;
}
//actual stretch algorithm implementation
void SndBuffer::UpdateTempoChangeSoundTouch2()
{
long targetSamplesReservoir = 48 * SndOutLatencyMS; //48000*SndOutLatencyMS/1000
//base aim at buffer filled %
float baseTargetFullness = (double)targetSamplesReservoir; ///(double)m_size;//0.05;
//state vars
static bool inside_hysteresis; //=false;
static int hys_ok_count; //=0;
static float dynamicTargetFullness; //=baseTargetFullness;
if (gRequestStretcherReset >= STRETCHER_RESET_THRESHOLD) {
ConLog("______> stretch: Reset.\n");
inside_hysteresis = false;
hys_ok_count = 0;
dynamicTargetFullness = baseTargetFullness;
}
int data = _GetApproximateDataInBuffer();
float bufferFullness = (float)data; ///(float)m_size;
#ifdef NEWSTRETCHER_USE_DYNAMIC_TUNING
{ //test current iterations/sec every 0.5s, and change algo params accordingly if different than previous IPS more than 30%
static long iters = 0;
static wxDateTime last = wxDateTime::UNow();
wxDateTime unow = wxDateTime::UNow();
wxTimeSpan delta = unow.Subtract(last);
if (delta.GetMilliseconds() > 500) {
int pot_targetIPS = 1000.0 / delta.GetMilliseconds().ToDouble() * iters;
if (!IsInRange(pot_targetIPS, int((float)targetIPS / 1.3f), int((float)targetIPS * 1.3f))) {
if (MsgOverruns())
ConLog("Stretcher: setting iters/sec from %d to %d\n", targetIPS, pot_targetIPS);
targetIPS = pot_targetIPS;
AVERAGING_WINDOW = GetClamped((int)(50.0f * (float)targetIPS / 750.0f), 3, (int)AVERAGING_BUFFER_SIZE);
}
last = unow;
iters = 0;
}
iters++;
}
#endif
//Algorithm params: (threshold params (hysteresis), etc)
const float hys_ok_factor = 1.04f;
const float hys_bad_factor = 1.2f;
int hys_min_ok_count = GetClamped((int)(50.0 * (float)targetIPS / 750.0), 2, 100); //consecutive iterations within hys_ok before going to 1:1 mode
int compensationDivider = GetClamped((int)(100.0 * (float)targetIPS / 750), 15, 150);
float tempoAdjust = bufferFullness / dynamicTargetFullness;
float avgerage = addToAvg(tempoAdjust);
tempoAdjust = avgerage;
// Dampen the adjustment to avoid overshoots (this means the average will compensate to the other side).
// This is different than simply bigger averaging window since bigger window also has bigger "momentum",
// so it's slower to slow down when it gets close to the equilibrium state and can therefore resonate.
// The dampening (sqrt was chosen for no very good reason) manages to mostly prevent that.
tempoAdjust = sqrt(tempoAdjust);
tempoAdjust = GetClamped(tempoAdjust, 0.05f, 10.0f);
if (tempoAdjust < 1)
baseTargetFullness /= sqrt(tempoAdjust); // slightly increase latency when running slow.
dynamicTargetFullness += (baseTargetFullness / tempoAdjust - dynamicTargetFullness) / (double)compensationDivider;
if (IsInRange(tempoAdjust, 0.9f, 1.1f) && IsInRange(dynamicTargetFullness, baseTargetFullness * 0.9f, baseTargetFullness * 1.1f))
dynamicTargetFullness = baseTargetFullness;
if (!inside_hysteresis) {
if (IsInRange(tempoAdjust, 1.0f / hys_ok_factor, hys_ok_factor))
hys_ok_count++;
else
hys_ok_count = 0;
if (hys_ok_count >= hys_min_ok_count) {
inside_hysteresis = true;
if (MsgOverruns())
ConLog("======> stretch: None (1:1)\n");
}
} else if (!IsInRange(tempoAdjust, 1.0f / hys_bad_factor, hys_bad_factor)) {
if (MsgOverruns())
ConLog("~~~~~~> stretch: Dynamic\n");
inside_hysteresis = false;
hys_ok_count = 0;
}
if (inside_hysteresis)
tempoAdjust = 1.0;
if (MsgOverruns()) {
static int iters = 0;
static wxDateTime last = wxDateTime::UNow();
wxDateTime unow = wxDateTime::UNow();
wxTimeSpan delta = unow.Subtract(last);
if (delta.GetMilliseconds() > 1000) { //report buffers state and tempo adjust every second
ConLog("buffers: %4d ms (%3.0f%%), tempo: %f, comp: %2.3f, iters: %d, (N-IPS:%d -> avg:%d, minokc:%d, div:%d) reset:%d\n",
(int)(data / 48), (double)(100.0 * bufferFullness / baseTargetFullness), (double)tempoAdjust, (double)(dynamicTargetFullness / baseTargetFullness), iters, (int)targetIPS, AVERAGING_WINDOW, hys_min_ok_count, compensationDivider, gRequestStretcherReset);
last = unow;
iters = 0;
}
iters++;
}
pSoundTouch->setTempo(tempoAdjust);
if (gRequestStretcherReset >= STRETCHER_RESET_THRESHOLD)
gRequestStretcherReset = 0;
return;
}
s32 Init()
{
numBuffers = Config_WaveOut.NumBuffers;
MMRESULT woores;
if (Test())
return -1;
// TODO : Use dsound to determine the speaker configuration, and expand audio from there.
#if 0
int speakerConfig;
//if( StereoExpansionEnabled )
speakerConfig = 2; // better not mess with this in wavout :p (rama)
// Any windows driver should support stereo at the software level, I should think!
pxAssume( speakerConfig > 1 );
LPTHREAD_START_ROUTINE threadproc;
switch( speakerConfig )
{
case 2:
ConLog( "* SPU2 > Using normal 2 speaker stereo output.\n" );
threadproc = (LPTHREAD_START_ROUTINE)&RThread<StereoOut16>;
speakerConfig = 2;
break;
case 4:
ConLog( "* SPU2 > 4 speaker expansion enabled [quadraphenia]\n" );
threadproc = (LPTHREAD_START_ROUTINE)&RThread<StereoQuadOut16>;
speakerConfig = 4;
break;
case 6:
case 7:
ConLog( "* SPU2 > 5.1 speaker expansion enabled.\n" );
threadproc = (LPTHREAD_START_ROUTINE)&RThread<Stereo51Out16>;
speakerConfig = 6;
break;
default:
ConLog( "* SPU2 > 7.1 speaker expansion enabled.\n" );
threadproc = (LPTHREAD_START_ROUTINE)&RThread<Stereo51Out16>;
speakerConfig = 8;
break;
}
#endif
wformat.wFormatTag = WAVE_FORMAT_PCM;
wformat.nSamplesPerSec = SampleRate;
wformat.wBitsPerSample = 16;
wformat.nChannels = 2;
wformat.nBlockAlign = ((wformat.wBitsPerSample * wformat.nChannels) / 8);
wformat.nAvgBytesPerSec = (wformat.nSamplesPerSec * wformat.nBlockAlign);
wformat.cbSize = 0;
qbuffer = new StereoOut16[BufferSize * numBuffers];
woores = waveOutOpen(&hwodevice, WAVE_MAPPER, &wformat, 0, 0, 0);
if (woores != MMSYSERR_NOERROR) {
waveOutGetErrorText(woores, (wchar_t *)&ErrText, 255);
SysMessage("WaveOut Error: %s", ErrText);
return -1;
}
const int BufferSizeBytes = wformat.nBlockAlign * BufferSize;
for (u32 i = 0; i < numBuffers; i++) {
whbuffer[i].dwBufferLength = BufferSizeBytes;
whbuffer[i].dwBytesRecorded = BufferSizeBytes;
whbuffer[i].dwFlags = 0;
whbuffer[i].dwLoops = 0;
whbuffer[i].dwUser = 0;
whbuffer[i].lpData = (LPSTR)QBUFFER(i);
whbuffer[i].lpNext = 0;
whbuffer[i].reserved = 0;
waveOutPrepareHeader(hwodevice, whbuffer + i, sizeof(WAVEHDR));
whbuffer[i].dwFlags |= WHDR_DONE; //avoid deadlock
}
// Start Thread
// [Air]: The waveout code does not use wait objects, so setting a time critical
// priority level is a bad idea. Standard priority will do fine. The buffer will get the
// love it needs and won't suck resources idling pointlessly. Just don't try to
// run it in uber-low-latency mode.
waveout_running = true;
thread = CreateThread(NULL, 0, (LPTHREAD_START_ROUTINE)RThread<StereoOut16>, this, 0, &tid);
return 0;
}
void DoDMAWrite(int core,u16 *pMem,u32 size)
{
// Perform an alignment check.
// Not really important. Everything should work regardless,
// but it could be indicative of an emulation foopah elsewhere.
#if 0
uptr pa = ((uptr)pMem)&7;
uptr pm = Cores[core].TSA&0x7;
if( pa )
{
fprintf(stderr, "* SPU2 DMA Write > Missaligned SOURCE! Core: %d TSA: 0x%x TDA: 0x%x Size: 0x%x\n", core, Cores[core].TSA, Cores[core].TDA, size);
}
if( pm )
{
fprintf(stderr, "* SPU2 DMA Write > Missaligned TARGET! Core: %d TSA: 0x%x TDA: 0x%x Size: 0x%x\n", core, Cores[core].TSA, Cores[core].TDA, size );
}
#endif
if(core==0)
DMA4LogWrite(pMem,size<<1);
else
DMA7LogWrite(pMem,size<<1);
if(MsgDMA()) ConLog(" * SPU2: DMA%c Transfer of %d bytes to %x (%02x %x %04x).\n",(core==0)?'4':'7',size<<1,Cores[core].TSA,Cores[core].DMABits,Cores[core].AutoDMACtrl,(~Cores[core].Regs.ATTR)&0x7fff);
Cores[core].TSA &= 0xfffff;
u32 buff1end = Cores[core].TSA + size;
u32 buff2end=0;
if( buff1end > 0x100000 )
{
buff2end = buff1end - 0x100000;
buff1end = 0x100000;
}
const int cacheIdxStart = Cores[core].TSA / pcm_WordsPerBlock;
const int cacheIdxEnd = (buff1end+pcm_WordsPerBlock-1) / pcm_WordsPerBlock;
PcmCacheEntry* cacheLine = &pcm_cache_data[cacheIdxStart];
PcmCacheEntry& cacheEnd = pcm_cache_data[cacheIdxEnd];
do
{
cacheLine->Validated = false;
cacheLine++;
} while ( cacheLine != &cacheEnd );
#if 0
// Pcm Cache Invalidation!
// It's a requirement that we mask bits for the blocks that are written to *only*,
// because doing anything else can cause the cache to fail, thanks to the progressive
// nature of the SPU2's ADPCM encoding. (the same thing that makes it impossible
// to use SSE optimizations on it).
u8* cache = (u8*)pcm_cache_flags;
// Step 1: Clear bits in the front remainder.
const int pcmTSA = Cores[core].TSA / pcm_WordsPerBlock;
const int pcmTDA = buff1end / pcm_WordsPerBlock;
const int remFront = pcmTSA & 31;
const int remBack = ((buff1end+pcm_WordsPerBlock-1)/pcm_WordsPerBlock) & 31; // round up to get the end remainder
int flagTSA = pcmTSA / 32;
if( remFront )
{
// need to clear some upper bits of this u32
uint mask = (1ul<<remFront)-1;
cache[flagTSA++] &= mask;
}
// Step 2: Clear the middle run
const int flagClearLen = pcmTDA-pcmTSA;
memset( &cache[flagTSA], 0, flagClearLen );
// Step 3: Clear bits in the end remainder.
if( remBack )
{
// need to clear some lower bits in this u32
uint mask = ~(1ul<<remBack)-1;
cache[flagTSA + flagClearLen] &= mask;
}
#endif
//ConLog( " * SPU2 : Cache Clear Range! TSA=0x%x, TDA=0x%x (low8=0x%x, high8=0x%x, len=0x%x)\n",
// Cores[core].TSA, buff1end, flagTSA, flagTDA, clearLen );
// First Branch needs cleared:
// It starts at TSA and goes to buff1end.
const u32 buff1size = (buff1end-Cores[core].TSA);
memcpy( GetMemPtr( Cores[core].TSA ), pMem, buff1size*2 );
if( buff2end > 0 )
{
// second branch needs copied:
// It starts at the beginning of memory and moves forward to buff2end
// endpoint cache should be irrelevant, since it's almost certainly dynamic
// memory below 0x2800 (registers and such)
//const u32 endpt2 = (buff2end + roundUp) / indexer_scalar;
//memset( pcm_cache_flags, 0, endpt2 );
memcpy( GetMemPtr( 0 ), &pMem[buff1size], buff2end*2 );
Cores[core].TDA = (buff2end+1) & 0xfffff;
if(Cores[core].IRQEnable)
{
// Flag interrupt?
// If IRQA occurs between start and dest, flag it.
// Since the buffer wraps, the conditional might seem odd, but it works.
if( ( Cores[core].IRQA >= Cores[core].TSA ) ||
( Cores[core].IRQA <= Cores[core].TDA ) )
{
Spdif.Info=4<<core;
SetIrqCall();
}
}
}
else
{
// Buffer doesn't wrap/overflow!
// Just set the TDA and check for an IRQ...
Cores[core].TDA = buff1end;
if(Cores[core].IRQEnable)
{
// Flag interrupt?
// If IRQA occurs between start and dest, flag it:
if( ( Cores[core].IRQA >= Cores[core].TSA ) &&
( Cores[core].IRQA <= Cores[core].TDA ) )
{
Spdif.Info=4<<core;
SetIrqCall();
}
}
}
Cores[core].TSA=Cores[core].TDA&0xFFFF0;
Cores[core].DMAICounter=size;
Cores[core].TADR=Cores[core].MADR+(size<<1);
}
void V_Core::Init( int index )
{
ConLog( "* SPU2-X: Init SPU2 core %d \n", index );
memset( this, 0, sizeof(V_Core) );
const int c = Index = index;
Regs.STATX = 0;
Regs.ATTR = 0;
ExtVol = V_VolumeLR::Max;
InpVol = V_VolumeLR::Max;
FxVol = V_VolumeLR(0);
MasterVol = V_VolumeSlideLR(0,0);
memset( &DryGate, -1, sizeof(DryGate) );
memset( &WetGate, -1, sizeof(WetGate) );
DryGate.ExtL = 0;
DryGate.ExtR = 0;
if (!c)
{
WetGate.ExtL = 0;
WetGate.ExtL = 0;
}
Regs.MMIX = c ? 0xFFC : 0xFF0; // PS2 confirmed (f3c and f30 after BIOS ran, ffc and ff0 after sdinit)
Regs.VMIXL = 0xFFFFFF;
Regs.VMIXR = 0xFFFFFF;
Regs.VMIXEL = 0xFFFFFF;
Regs.VMIXER = 0xFFFFFF;
EffectsStartA = c ? 0xFFFF8 : 0xEFFF8;
EffectsEndA = c ? 0xFFFFF : 0xEFFFF;
ExtEffectsStartA = EffectsStartA;
ExtEffectsEndA = EffectsEndA;
FxEnable = 0; // Uninitialized it's 0 for both cores. Resetting libs however may set this to 0 or 1.
// These are real PS2 values, mainly constant apart from a few bits: 0x3220EAA4, 0x40505E9C.
// These values mean nothing. They do not reflect the actual address the SPU2 is testing,
// it would seem that reading the IRQA register returns the last written value, not the
// value of the internal register. Rewriting the registers with their current values changes
// whether interrupts fire (they do while uninitialised, but do not when rewritten).
// The exact boot value is unknown and probably unknowable, but it seems to be somewhere
// in the input or output areas, so we're using 0x800.
// F1 2005 is known to rely on an uninitialised IRQA being an address which will be hit.
IRQA = 0x800;
IRQEnable = 0; // PS2 confirmed
for( uint v=0; v<NumVoices; ++v )
{
VoiceGates[v].DryL = -1;
VoiceGates[v].DryR = -1;
VoiceGates[v].WetL = -1;
VoiceGates[v].WetR = -1;
Voices[v].Volume = V_VolumeSlideLR(0,0); // V_VolumeSlideLR::Max;
Voices[v].SCurrent = 28;
Voices[v].ADSR.Value = 0;
Voices[v].ADSR.Phase = 0;
Voices[v].Pitch = 0x3FFF;
Voices[v].NextA = 0x2801;
Voices[v].StartA = 0x2800;
Voices[v].LoopStartA = 0x2800;
}
DMAICounter = 0;
AdmaInProgress = 0;
Regs.STATX = 0x80;
Regs.ENDX = 0xffffff; // PS2 confirmed
RevBuffers.NeedsUpdated = true;
UpdateEffectsBufferSize();
}
void V_Core::AnalyzeReverbPreset()
{
ConLog("Reverb Parameter Update for Core %d:\n", Index);
ConLog("----------------------------------------------------------\n");
ConLog(" IN_COEF_L, IN_COEF_R 0x%08x, 0x%08x\n", Revb.IN_COEF_L, Revb.IN_COEF_R);
ConLog(" FB_SRC_A, FB_SRC_B 0x%08x, 0x%08x\n", Revb.FB_SRC_A, Revb.FB_SRC_B);
ConLog(" FB_ALPHA, FB_X 0x%08x, 0x%08x\n", Revb.FB_ALPHA, Revb.FB_X);
ConLog(" ACC_COEF_A 0x%08x\n", Revb.ACC_COEF_A);
ConLog(" ACC_COEF_B 0x%08x\n", Revb.ACC_COEF_B);
ConLog(" ACC_COEF_C 0x%08x\n", Revb.ACC_COEF_C);
ConLog(" ACC_COEF_D 0x%08x\n", Revb.ACC_COEF_D);
ConLog(" ACC_SRC_A0, ACC_SRC_A1 0x%08x, 0x%08x\n", Revb.ACC_SRC_A0, Revb.ACC_SRC_A1);
ConLog(" ACC_SRC_B0, ACC_SRC_B1 0x%08x, 0x%08x\n", Revb.ACC_SRC_B0, Revb.ACC_SRC_B1);
ConLog(" ACC_SRC_C0, ACC_SRC_C1 0x%08x, 0x%08x\n", Revb.ACC_SRC_C0, Revb.ACC_SRC_C1);
ConLog(" ACC_SRC_D0, ACC_SRC_D1 0x%08x, 0x%08x\n", Revb.ACC_SRC_D0, Revb.ACC_SRC_D1);
ConLog(" IIR_SRC_A0, IIR_SRC_A1 0x%08x, 0x%08x\n", Revb.IIR_SRC_A0, Revb.IIR_SRC_A1);
ConLog(" IIR_SRC_B0, IIR_SRC_B1 0x%08x, 0x%08x\n", Revb.IIR_SRC_B0, Revb.IIR_SRC_B1);
ConLog(" IIR_DEST_A0, IIR_DEST_A1 0x%08x, 0x%08x\n", Revb.IIR_DEST_A0, Revb.IIR_DEST_A1);
ConLog(" IIR_DEST_B0, IIR_DEST_B1 0x%08x, 0x%08x\n", Revb.IIR_DEST_B0, Revb.IIR_DEST_B1);
ConLog(" IIR_ALPHA, IIR_COEF 0x%08x, 0x%08x\n", Revb.IIR_ALPHA, Revb.IIR_COEF);
ConLog(" MIX_DEST_A0 0x%08x\n", Revb.MIX_DEST_A0);
ConLog(" MIX_DEST_A1 0x%08x\n", Revb.MIX_DEST_A1);
ConLog(" MIX_DEST_B0 0x%08x\n", Revb.MIX_DEST_B0);
ConLog(" MIX_DEST_B1 0x%08x\n", Revb.MIX_DEST_B1);
ConLog(" EffectsBufferSize 0x%x\n", EffectsBufferSize);
ConLog("----------------------------------------------------------\n");
}
static int luasrc_ConLog (lua_State *L) {
ConLog(luaL_checkstring(L, 1));
return 0;
}
void __fastcall TimeUpdate(u32 cClocks)
{
u32 dClocks = cClocks-lClocks;
// [Air]: Sanity Check
// If for some reason our clock value seems way off base, just mix
// out a little bit, skip the rest, and hope the ship "rights" itself later on.
if( dClocks > TickInterval*SanityInterval )
{
ConLog( " * SPU2 > TimeUpdate Sanity Check (Tick Delta: %d) (PS2 Ticks: %d)\n", dClocks/TickInterval, cClocks/TickInterval );
dClocks = TickInterval*SanityInterval;
lClocks = cClocks-dClocks;
}
//UpdateDebugDialog();
//Update Mixing Progress
while(dClocks>=TickInterval)
{
if(has_to_call_irq)
{
ConLog(" * SPU2: Irq Called (%04x).\n",Spdif.Info);
has_to_call_irq=false;
if(_irqcallback) _irqcallback();
}
if(Cores[0].InitDelay>0)
{
Cores[0].InitDelay--;
if(Cores[0].InitDelay==0)
{
Cores[0].Reset();
}
}
if(Cores[1].InitDelay>0)
{
Cores[1].InitDelay--;
if(Cores[1].InitDelay==0)
{
Cores[1].Reset();
}
}
//Update DMA4 interrupt delay counter
if(Cores[0].DMAICounter>0)
{
Cores[0].DMAICounter-=TickInterval;
if(Cores[0].DMAICounter<=0)
{
Cores[0].MADR=Cores[0].TADR;
Cores[0].DMAICounter=0;
if(dma4callback) dma4callback();
}
else {
Cores[0].MADR+=TickInterval<<1;
}
}
//Update DMA7 interrupt delay counter
if(Cores[1].DMAICounter>0)
{
Cores[1].DMAICounter-=TickInterval;
if(Cores[1].DMAICounter<=0)
{
Cores[1].MADR=Cores[1].TADR;
Cores[1].DMAICounter=0;
//ConLog( "* SPU2 > DMA 7 Callback! %d\n", Cycles );
if(dma7callback) dma7callback();
}
else {
Cores[1].MADR+=TickInterval<<1;
}
}
dClocks-=TickInterval;
lClocks+=TickInterval;
Cycles++;
Mix();
}
}
__forceinline void TimeUpdate(u32 cClocks)
{
u32 dClocks = cClocks - lClocks;
// Sanity Checks:
// It's not totally uncommon for the IOP's clock to jump backwards a cycle or two, and in
// such cases we just want to ignore the TimeUpdate call.
if( dClocks > (u32)-15 ) return;
// But if for some reason our clock value seems way off base (typically due to bad dma
// timings from PCSX2), just mix out a little bit, skip the rest, and hope the ship
// "rights" itself later on.
if( dClocks > (u32)(TickInterval*SanityInterval) )
{
if(MsgToConsole()) ConLog( " * SPU2 > TimeUpdate Sanity Check (Tick Delta: %d) (PS2 Ticks: %d)\n", dClocks/TickInterval, cClocks/TickInterval );
dClocks = TickInterval * SanityInterval;
lClocks = cClocks - dClocks;
}
// Visual debug display showing all core's activity! Disabled via #define on release builds.
#ifdef __WIN32__
UpdateDebugDialog();
#endif
if( SynchMode == 1 ) // AsyncMix on
SndBuffer::UpdateTempoChangeAsyncMixing();
else TickInterval = 768; // Reset to default, in case the user hotswitched from async to something else.
//Update Mixing Progress
while(dClocks>=TickInterval)
{
if(has_to_call_irq)
{
//ConLog("* SPU2-X: Irq Called (%04x) at cycle %d.\n", Spdif.Info, Cycles);
has_to_call_irq=false;
if(_irqcallback) _irqcallback();
}
#ifndef ENABLE_NEW_IOPDMA_SPU2
//Update DMA4 interrupt delay counter
if(Cores[0].DMAICounter>0)
{
Cores[0].DMAICounter-=TickInterval;
if(Cores[0].DMAICounter<=0)
{
Cores[0].MADR=Cores[0].TADR;
Cores[0].DMAICounter=0;
if(dma4callback) dma4callback();
}
else {
Cores[0].MADR+=TickInterval<<1;
}
}
//Update DMA7 interrupt delay counter
if(Cores[1].DMAICounter>0)
{
Cores[1].DMAICounter-=TickInterval;
if(Cores[1].DMAICounter<=0)
{
Cores[1].MADR=Cores[1].TADR;
Cores[1].DMAICounter=0;
//ConLog( "* SPU2 > DMA 7 Callback! %d\n", Cycles );
if(dma7callback) dma7callback();
}
else {
Cores[1].MADR+=TickInterval<<1;
}
}
#endif
dClocks -= TickInterval;
lClocks += TickInterval;
Cycles++;
// Note: IOP does not use MMX regs, so no need to save them.
//SaveMMXRegs();
Mix();
//RestoreMMXRegs();
}
}