/* Copyright 2009 DeSmuME team
This file is part of DeSmuME
DeSmuME is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
DeSmuME is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with DeSmuME; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <queue>
#include <vector>
#include <cassert>
#include "../types.h"
#include "metaspu.h"
//for pcsx2 method
//(havent bothered to get it compiling in gcc yet)
#ifdef _MSC_VER
#include "SndOut.h"
#endif
/*template<typename T> inline T _abs(T val)
{
if(val<0) return -val;
else return val;
}*/
/*template<typename T> inline T moveValueTowards(T val, T target, T incr)
{
incr = _abs(incr);
T delta = _abs(target-val);
if(val<target) val += incr;
else if(val>target) val -= incr;
T newDelta = _abs(target-val);
if(newDelta >= delta)
val = target;
return val;
}*/
class ZeromusSynchronizer : public ISynchronizingAudioBuffer
{
public:
ZeromusSynchronizer()
: mixqueue_go(false)
,
#ifdef NDEBUG
adjustobuf(200,1000)
#else
adjustobuf(22000,44000)
#endif
{
}
bool mixqueue_go;
virtual void enqueue_samples(int16_t* buf, int samples_provided)
{
for(int i=0;i<samples_provided;i++) {
int16_t left = *buf++;
int16_t right = *buf++;
adjustobuf.enqueue(left,right);
}
}
//returns the number of samples actually supplied, which may not match the number requested
virtual int output_samples(int16_t* buf, int samples_requested)
{
int done = 0;
if(!mixqueue_go) {
if(adjustobuf.size > 200)
mixqueue_go = true;
}
else
{
for(int i=0;i<samples_requested;i++) {
if(adjustobuf.size==0) {
mixqueue_go = false;
break;
}
done++;
int16_t left, right;
adjustobuf.dequeue(left,right);
*buf++ = left;
*buf++ = right;
}
}
return done;
}
private:
class Adjustobuf
{
public:
Adjustobuf(int _minLatency, int _maxLatency)
: minLatency(_minLatency)
, maxLatency(_maxLatency)
, size(0)
{
rollingTotalSize = 0;
targetLatency = (maxLatency + minLatency)/2;
rate = 1.0f;
cursor = 0.0f;
curr[0] = curr[1] = 0;
kAverageSize = 80000;
}
float rate, cursor;
int minLatency, targetLatency, maxLatency;
std::queue<int16_t> buffer;
int size;
int16_t curr[2];
std::queue<int> statsHistory;
void enqueue(int16_t left, int16_t right)
{
buffer.push(left);
buffer.push(right);
size++;
}
int64_t rollingTotalSize;
uint32_t kAverageSize;
void addStatistic()
{
statsHistory.push(size);
rollingTotalSize += size;
if(statsHistory.size()>kAverageSize)
{
rollingTotalSize -= statsHistory.front();
statsHistory.pop();
float averageSize = (float)(rollingTotalSize / kAverageSize);
//static int ctr=0; ctr++; if((ctr&127)==0) printf("avg size: %f curr size: %d rate: %f\n",averageSize,size,rate);
{
float targetRate;
if(averageSize < targetLatency)
{
targetRate = 1.0f - (targetLatency-averageSize)/kAverageSize;
}
else if(averageSize > targetLatency) {
targetRate = 1.0f + (averageSize-targetLatency)/kAverageSize;
} else targetRate = 1.0f;
//rate = moveValueTowards(rate,targetRate,0.001f);
rate = targetRate;
}
}
}
void dequeue(int16_t& left, int16_t& right)
{
left = right = 0;
addStatistic();
if(size==0) { return; }
cursor += rate;
while(cursor>1.0f) {
cursor -= 1.0f;
if(size>0) {
curr[0] = buffer.front(); buffer.pop();
curr[1] = buffer.front(); buffer.pop();
size--;
}
}
left = curr[0];
right = curr[1];
}
} adjustobuf;
};
class NitsujaSynchronizer : public ISynchronizingAudioBuffer
{
private:
struct ssamp
{
int16_t l, r;
ssamp() {}
ssamp(int16_t ll, int16_t rr) : l(ll), r(rr) {}
};
std::vector<ssamp> sampleQueue;
// returns values going between 0 and y-1 in a saw wave pattern, based on x
static inline int pingpong(int x, int y)
{
x %= 2*y;
if(x >= y)
x = 2*y - x - 1;
return x;
// in case we want to switch to odd buffer sizes for more sharpness
//x %= 2*(y-1);
//if(x >= y)
// x = 2*(y-1) - x;
//return x;
}
static inline ssamp crossfade (ssamp lhs, ssamp rhs, int cur, int start, int end)
{
if(cur <= start)
return lhs;
if(cur >= end)
return rhs;
// in case we want sine wave interpolation instead of linear here
//float ang = 3.14159f * (float)(cur - start) / (float)(end - start);
//cur = start + (int)((1-cosf(ang))*0.5f * (end - start));
int inNum = cur - start;
int outNum = end - cur;
int denom = end - start;
int lrv = ((int)lhs.l * outNum + (int)rhs.l * inNum) / denom;
int rrv = ((int)lhs.r * outNum + (int)rhs.r * inNum) / denom;
return ssamp(lrv,rrv);
}
static inline void emit_sample(int16_t*& outbuf, ssamp sample)
{
*outbuf++ = sample.l;
*outbuf++ = sample.r;
}
static inline void emit_samples(int16_t*& outbuf, const ssamp* samplebuf, int samples)
{
for(int i=0;i<samples;i++)
emit_sample(outbuf,samplebuf[i]);
}
public:
NitsujaSynchronizer()
{}
virtual void enqueue_samples(int16_t* buf, int samples_provided)
{
for(int i=0;i<samples_provided;i++)
{
sampleQueue.push_back(ssamp(buf[0],buf[1]));
buf += 2;
}
}
virtual int output_samples(int16_t* buf, int samples_requested)
{
int audiosize = samples_requested;
int queued = sampleQueue.size();
// I am too lazy to deal with odd numbers
audiosize &= ~1;
queued &= ~1;
if(queued > 0x200 && audiosize > 0) // is there any work to do?
{
// are we going at normal speed?
// or more precisely, are the input and output queues/buffers of similar size?
if(queued > 900 || audiosize > queued * 2)
{
// not normal speed. we have to resample it somehow in this case.
if(audiosize <= queued)
{
// fast forward speed
// this is the easy case, just crossfade it and it sounds ok
for(int i = 0; i < audiosize; i++)
{
int j = i + queued - audiosize;
ssamp outsamp = crossfade(sampleQueue[i],sampleQueue[j], i,0,audiosize);
emit_sample(buf,outsamp);
}
}
else
{
// slow motion speed
// here we take a very different approach,
// instead of crossfading it, we select a single sample from the queue
// and make sure that the index we use to select a sample is constantly moving
// and that it starts at the first sample in the queue and ends on the last one.
//
// hopefully the index doesn't move discontinuously or we'll get slight crackling
// (there might still be a minor bug here that causes this occasionally)
//
// here's a diagram of how the index we sample from moves:
//
// queued (this axis represents the index we sample from. the top means the end of the queue)
// ^
// | --> audiosize (this axis represents the output index we write to, right meaning forward in output time/position)
// | A C C end
// A A B C C C
// A A A B C C C
// A A A B C C
// A A C
// start
//
// yes, this means we are spending some stretches of time playing the sound backwards,
// but the stretches are short enough that this doesn't sound weird.
// this lets us avoid most crackling problems due to the endpoints matching up.
// first calculate a shorter-than-full window
// that has minimal slope at the endpoints
// (to further reduce crackling, especially in sine waves)
int beststart = 0, extraAtEnd = 0;
{
int bestend = queued;
static const int worstdiff = 99999999;
int beststartdiff = worstdiff;
int bestenddiff = worstdiff;
for(int i = 0; i < 128; i+=2)
{
int diff = abs(sampleQueue[i].l - sampleQueue[i+1].l) + abs(sampleQueue[i].r - sampleQueue[i+1].r);
if(diff < beststartdiff)
{
beststartdiff = diff;
beststart = i;
}
}
for(int i = queued-3; i > queued-3-128; i-=2)
{
int diff = abs(sampleQueue[i].l - sampleQueue[i+1].l) + abs(sampleQueue[i].r - sampleQueue[i+1].r);
if(diff < bestenddiff)
{
bestenddiff = diff;
bestend = i+1;
}
}
extraAtEnd = queued - bestend;
queued = bestend - beststart;
int oksize = queued;
while(oksize + queued*2 + beststart + extraAtEnd <= samples_requested)
oksize += queued*2;
audiosize = oksize;
for(int x = 0; x < beststart; x++)
{
emit_sample(buf,sampleQueue[x]);
}
sampleQueue.erase(sampleQueue.begin(), sampleQueue.begin() + beststart);
}
int midpointX = audiosize >> 1;
int midpointY = queued >> 1;
// all we need to do here is calculate the X position of the leftmost "B" in the above diagram.
// TODO: we should calculate it with a simple equation like
// midpointXOffset = min(something,somethingElse);
// but it's a little difficult to work it out exactly
// so here's a stupid search for the value for now:
int prevA = 999999;
int midpointXOffset = queued/2;
while(true)
{
int a = abs(pingpong(midpointX - midpointXOffset, queued) - midpointY) - midpointXOffset;
if(((a > 0) != (prevA > 0) || (a < 0) != (prevA < 0)) && prevA != 999999)
{
if((a + prevA)&1) // there's some sort of off-by-one problem with this search since we're moving diagonally...
midpointXOffset++; // but this fixes it most of the time...
break; // found it
}
prevA = a;
midpointXOffset--;
if(midpointXOffset < 0)
{
midpointXOffset = 0;
break; // failed to find it. the two sides probably meet exactly in the center.
}
}
int leftMidpointX = midpointX - midpointXOffset;
int rightMidpointX = midpointX + midpointXOffset;
int leftMidpointY = pingpong(leftMidpointX, queued);
int rightMidpointY = (queued-1) - pingpong((int)audiosize-1 - rightMidpointX + queued*2, queued);
// output the left almost-half of the sound (section "A")
for(int x = 0; x < leftMidpointX; x++)
{
int i = pingpong(x, queued);
emit_sample(buf,sampleQueue[i]);
}
// output the middle stretch (section "B")
int y = leftMidpointY;
int dyMidLeft = (leftMidpointY < midpointY) ? 1 : -1;
int dyMidRight = (rightMidpointY > midpointY) ? 1 : -1;
for(int x = leftMidpointX; x < midpointX; x++, y+=dyMidLeft)
emit_sample(buf,sampleQueue[y]);
for(int x = midpointX; x < rightMidpointX; x++, y+=dyMidRight)
emit_sample(buf,sampleQueue[y]);
// output the end of the queued sound (section "C")
for(int x = rightMidpointX; x < audiosize; x++)
{
int i = (queued-1) - pingpong((int)audiosize-1 - x + queued*2, queued);
emit_sample(buf,sampleQueue[i]);
}
for(int x = 0; x < extraAtEnd; x++)
{
int i = queued + x;
emit_sample(buf,sampleQueue[i]);
}
queued += extraAtEnd;
audiosize += beststart + extraAtEnd;
} //end else
sampleQueue.erase(sampleQueue.begin(), sampleQueue.begin() + queued);
return audiosize;
}
else
{
// normal speed
// just output the samples straightforwardly.
//
// at almost-full speeds (like 50/60 FPS)
// what will happen is that we rapidly fluctuate between entering this branch
// and entering the "slow motion speed" branch above.
// but that's ok! because all of these branches sound similar enough that we can get away with it.
// so the two cases actually complement each other.
if(audiosize >= queued)
{
emit_samples(buf,&sampleQueue[0],queued);
sampleQueue.erase(sampleQueue.begin(), sampleQueue.begin() + queued);
return queued;
}
else
{
emit_samples(buf,&sampleQueue[0],audiosize);
sampleQueue.erase(sampleQueue.begin(), sampleQueue.begin()+audiosize);
return audiosize;
}
} //end normal speed
} //end if there is any work to do
else
{
return 0;
}
} //output_samples
private:
}; //NitsujaSynchronizer
#ifdef _MSC_VER
class PCSX2Synchronizer : public ISynchronizingAudioBuffer
{
public:
std::queue<int16_t> readySamples;
PCSX2Synchronizer()
{
SndBuffer::Init();
}
virtual void enqueue_samples(int16_t* buf, int samples_provided)
{
for(int i=0;i<samples_provided;i++)
{
StereoOut32 so32(buf[0],buf[1]);
SndBuffer::Write(so32);
buf++;
buf++;
}
}
virtual int output_samples(int16_t* buf, int samples_requested)
{
for(int i=0;i<samples_requested;i++) {
if(readySamples.size()==0) {
//SndOutPacketSize
StereoOut16 temp[SndOutPacketSize*2];
SndBuffer::ReadSamples( temp );
for(int i=0;i<SndOutPacketSize;i++) {
readySamples.push(temp[i].Left);
readySamples.push(temp[i].Right);
}
}
*buf++ = readySamples.front(); readySamples.pop();
*buf++ = readySamples.front(); readySamples.pop();
}
return samples_requested;
}
};
#endif
ISynchronizingAudioBuffer* metaspu_construct(ESynchMethod method)
{
switch(method)
{
case ESynchMethod_N: return new NitsujaSynchronizer();
case ESynchMethod_Z: return new ZeromusSynchronizer();
#ifdef _MSC_VER
case ESynchMethod_P: return new PCSX2Synchronizer();
#endif
default: return NULL;
}
}