Files
itgmania212121/src/RageSoundReader_Resample_Good.cpp
T
Devin J. Pohly feb919f0bf Revert memory leak commits
5f7001e: "Added a new branch"
01456ed: "Fixed a lot of memory leaks"
dac4493: "Fixed all remaining memory leaks that I could figure out"
0792db7: "Removed the smnew macro and the call to _CrtSetDbgFlag()"

Some of these caused destructor-time problems due to static initialization
order fiasco and related issues.  Notably, the program would no longer exit on
OSX and had to be killed.

There were probably legitimate fixes in here, but since these are monolithic
commits it's too much work to extract them now.  Let's reapply them
individually and in the forward direction.
2013-04-27 00:05:14 -04:00

765 lines
21 KiB
C++

/*
* This implements audio resampling, using the method described at:
* http://www.dspguru.com/info/faqs/mrfaq.htm
*
* Each conversion ratio uses some memory, but the resulting table is
* shared, so the memory overhead per stream is negligible.
*/
#include "global.h"
#include "RageSoundReader_Resample_Good.h"
#include "RageLog.h"
#include "RageUtil.h"
#include "RageMath.h"
#include "RageThreads.h"
#include <numeric>
/* Filter length. This must be a power of 2. */
#define L 8
namespace
{
float sincf( float f )
{
if( f == 0 )
return 1;
return sinf(f)/f;
}
/* Modified Bessel function I0. From Abramowitz and Stegun "Handbook of Mathematical
* Functions", "Modified Bessel Functions I and K". */
float BesselI0( float fX )
{
float fAbsX = fabsf( fX );
if( fAbsX < 3.75f )
{
float y = fX / 3.75f;
y *= y;
float fRet = 1.0f+y*(+3.5156229f+y*(+3.0899424f+y*(+1.2067492f+y*(+0.2659732f+y*(+0.0360768f+y*+0.0045813f)))));
return fRet;
}
else
{
float y = 3.75f/fAbsX;
float fRet = (exp(fAbsX)/sqrt(fAbsX)) *
(+0.39894228f+y*(+0.01328592f+y*(+0.00225319f+y*(-0.00157565f+y*(0.00916281f+
y*(-0.02057706f+y*(+0.02635537f+y*(-0.01647633f+y*+0.00392377f))))))));
return fRet;
}
}
/*
* Kaiser window:
*
* K(n) = I0( B*sqrt(1-(n/p)^2) )
* -----------------------
* I0(B)
*
* where B is the beta parameter, p is len/2, and n is in [-len/2,+len/2].
*/
void ApplyKaiserWindow( float *pBuf, int iLen, float fBeta )
{
const float fDenom = BesselI0(fBeta);
float p = (iLen-1)/2.0f;
for( int n = 0; n < iLen; ++n )
{
float fN1 = fabsf((n-p)/p);
float fNum = fBeta * sqrtf( max(1-fN1*fN1, 0) );
fNum = BesselI0( fNum );
float fVal = fNum/fDenom;
pBuf[n] *= fVal;
}
}
void MultiplyVector( float *pStart, float *pEnd, float f )
{
for( ; pStart != pEnd; ++pStart )
*pStart *= f;
}
void GenerateSincLowPassFilter( float *pFIR, int iWinSize, float fCutoff )
{
float p = (iWinSize-1)/2.0f;
for( int n = 0; n < iWinSize; ++n )
{
float fN1 = (n-p);
float fVal = sincf(2*PI*fCutoff * fN1)*(2*fCutoff);
// printf( "n %i, %f, %f -> %f\n", n, p, fN1, fVal );
pFIR[n] = fVal;
}
#if 0
float *pFIRp = pFIR+iWinSize/2;
for(int i=-iWinSize/2;i<=iWinSize/2;i++)
{
float ff = sinc(2*M_PI*fCutoff * (i + 0.0))*(2*fCutoff);
printf( "%i: %f\n", i, ff );
pFIRp[i]=ff;
}
for( int i=0; i < iWinSize; i++ )
printf( "sinc: %i: %f\n", i, pFIR[i] );
#endif
}
void NormalizeVector( float *pBuf, int iSize )
{
float fTotal = accumulate( &pBuf[0], &pBuf[iSize], 0.0f );
MultiplyVector( &pBuf[0], &pBuf[iSize], 1/fTotal );
}
int GCD( int i1, int i2 )
{
while(1)
{
unsigned iRem = i2 % i1;
if( iRem == 0 )
return i1;
i2 = i1;
i1 = iRem;
}
return i1;
}
}
#if 0
void RunFIRFilter( float *pIn, float *pOut, int iInputValues, float *pFIR, int iWinSize )
{
for( int i = 0; i < iInputValues; ++i )
{
float fSum = 0;
const float *pInData = &pIn[i];
for( int j = 0; j < iWinSize; ++j )
{
float in = pInData[j];
fSum += in*pFIR[j];
printf( "%i: in %f * %f, += %f\n", j, pInData[j], pFIR[j], in*pFIR[j] );
}
pOut[i] = fSum;
}
}
#endif
template<typename T>
class AlignedBuffer
{
public:
AlignedBuffer( int iSize )
{
m_iSize = iSize;
m_pBuf = new T[m_iSize];
}
AlignedBuffer( const AlignedBuffer &cpy )
{
m_iSize = cpy.m_iSize;
m_pBuf = new T[m_iSize];
memcpy( m_pBuf, cpy.m_pBuf, sizeof(T)*m_iSize );
}
~AlignedBuffer()
{
delete [] m_pBuf;
}
operator T*() { return m_pBuf; }
operator const T*() const { return m_pBuf; }
private:
T& operator=( T &rhs );
int m_iSize;
T *m_pBuf;
};
struct PolyphaseFilter
{
struct State
{
State( int iUpFactor ):
m_fBuf( L * 2 )
{
m_iPolyIndex = iUpFactor-1;
m_iFilled = 0;
m_iBufNext = 0;
}
int m_iPolyIndex;
int m_iFilled;
/* This buffer is duplicated. If the circular buffer is size L, the actual buffer
* is size L*2, and data at buf[N] is also at buf[N+L]. That way, we can access
* up to buf[N*2-1] without having to wrap. */
AlignedBuffer<float> m_fBuf;
int m_iBufNext;
};
friend struct State;
PolyphaseFilter( int iUpFactor ):
m_pPolyphase( L*iUpFactor )
{
m_iUpFactor = iUpFactor;
}
void Generate( const float *pFIR );
int RunPolyphaseFilter( State &State, const float *pIn, int iSamplesIn, int iDownFactor,
float *pOut, int iSamplesOut, int iSampleStride ) const;
int GetLatency() const { return L/2; }
int NumInputsForOutputSamples( const State &State, int iOut, int iDownFactor ) const;
private:
AlignedBuffer<float> m_pPolyphase;
int m_iUpFactor;
};
/*
* Convert an FIR filter to a polyphase filter.
*
* pFIR is the input FIR filter, which has iL*iUpFactor values.
* iL is the number of real samples each output sample looks at.
* iUpFactor is the actual upsampling factor; the amount of zero-stuffing between each real sample.
* pOutput is the 2D output polyphase filter, with iL*iL values.
*
* With an upsampling factor (iUpFactor) of 3, and a sinc filter length of 12 (iL*iUpFactor),
*
* input first output sample (before decimation)
* sample second output sample
* third output sample
*
* 0 0
* 0 1 0
* 1592 2 1 0
* 0 3 2 1
* 0 4 3 2
* 1623 5 4 3
* 0 6 5 4
* 0 7 6 5
* 1682 8 7 6
* 0 9 8 7
* 0 10 9 8
* 1730 11 10 9
* 0 11 10
* 0 11
*
* first row: 2, 5, 8, 11
* second: 1, 4, 7, 10
* third: 0, 3, 6, 9
* Read a new sample after passing the last line.
*/
void PolyphaseFilter::Generate( const float *pFIR )
{
float *pOutput=m_pPolyphase;
int iInputSize = L*m_iUpFactor;
for( int iRow = 0; iRow < m_iUpFactor; ++iRow )
{
int iInputOffset = (m_iUpFactor-iRow-1) % m_iUpFactor;
for( int iCol = 0; iCol < L; ++iCol )
{
*pOutput = pFIR[iInputOffset];
++pOutput;
iInputOffset += m_iUpFactor;
iInputOffset %= iInputSize;
}
}
}
/*
* We only want one boundary check when running the filter; either on the
* number of inputs used, or the number of outputs produced. Otherwise, we'll
* have to maintain two counters, and check two values per iteration.
*
* First, call NumInputsForOutputSamples(out), to find out how many inputs to supply to get
* the desired number of outputs. Then, pass the data, the input count
* and the output count to RunPolyphaseFilter.
*
* - When downsampling, we use the number of inputs as the boundary. For example,
* if the ratio is 1:3 (downsample x3), and the user gives us 10 samples, then we
* process until we've consumed all of the input. (This will result in exactly
* the number of samples the user asked for with NumInputsForOutputSamples.)
*
* - When upsampling, we use the number of outputs as the boundary. For example,
* if the ratio is 3:1 (upsample x3), and the user wants 8 samples to be output,
* we'll have been given 3 samples as input. Process until we've produced 8
* samples.
*
* In both cases, we have overlap. In the first, it's possible that we could
* have consumed an additional input without producing an output. In the second,
* it's possible that we could have produced an additional output without
* consuming an input.
*/
int PolyphaseFilter::RunPolyphaseFilter(
State &State,
const float *pIn, int iSamplesIn, int iDownFactor,
float *pOut, int iSamplesOut,
int iSampleStride ) const
{
ASSERT( iSamplesIn >= 0 );
float *pOutOrig = pOut;
const float *pInEnd = pIn + iSamplesIn*iSampleStride;
const float *pOutEnd = pOut + iSamplesOut*iSampleStride;
int iFilled = State.m_iFilled;
int iPolyIndex = State.m_iPolyIndex;
while( pOut != pOutEnd )
{
if( iFilled < L )
{
if( pIn == pInEnd )
break;
State.m_fBuf[State.m_iBufNext] = *pIn;
State.m_fBuf[State.m_iBufNext + L] = *pIn;
++State.m_iBufNext;
State.m_iBufNext &= L-1;
pIn += iSampleStride;
++iFilled;
continue;
}
while( pOut != pOutEnd )
{
const float *pCurPoly = &m_pPolyphase[iPolyIndex*L];
const float *pInData = &State.m_fBuf[State.m_iBufNext];
float fTot = 0;
for( int j = 0; j < L; ++j )
fTot += pInData[j]*pCurPoly[j];
*pOut = fTot;
pOut += iSampleStride;
iPolyIndex += iDownFactor;
if( iPolyIndex >= m_iUpFactor )
break;
}
iFilled -= iPolyIndex/m_iUpFactor;
iPolyIndex %= m_iUpFactor;
}
State.m_iFilled = iFilled;
State.m_iPolyIndex = iPolyIndex;
int iRetSamples = pOut - pOutOrig;
int iRetFrames = iRetSamples / iSampleStride;
return iRetFrames;
}
/*
* Return the number of input samples needed to produce the given number of output
* samples. This is dependent on the number of bytes in the buffer and the current
* position of the stream.
*/
int PolyphaseFilter::NumInputsForOutputSamples( const State &State, int iOut, int iDownFactor ) const
{
int iIn = 0;
int iFilled = State.m_iFilled;
int iPolyIndex = State.m_iPolyIndex;
#if 0
while( iOut > 0 )
{
if( iFilled < L )
{
int iToFill = L-iFilled;
iIn += iToFill;
iFilled += iToFill;
}
while( iFilled == L && iOut )
{
--iOut;
iPolyIndex += iDownFactor;
if( iPolyIndex >= m_iUpFactor )
break;
}
iFilled -= iPolyIndex/m_iUpFactor;
iPolyIndex %= m_iUpFactor;
}
#endif
if( iOut > 0 )
{
if( iFilled < L )
{
int iToFill = L-iFilled;
iIn += iToFill;
}
// The -1 here is because we don't refill m_fBuf after writing the last output.
iPolyIndex += iDownFactor*(iOut-1);
iIn += iPolyIndex/m_iUpFactor;
}
return iIn;
}
/** @brief Utilities for working with the PolyphaseFilter cache. */
namespace PolyphaseFilterCache
{
/* Cache filter data, and reuse it without copying. All operations after creation
* are const, so this doesn't cause thread-safety problems. */
typedef map<pair<int,float>, PolyphaseFilter *> FilterMap;
static RageMutex PolyphaseFiltersLock("PolyphaseFiltersLock");
static FilterMap g_mapPolyphaseFilters;
const PolyphaseFilter *MakePolyphaseFilter( int iUpFactor, float fCutoffFrequency )
{
PolyphaseFiltersLock.Lock();
pair<int,float> params( make_pair(iUpFactor, fCutoffFrequency) );
FilterMap::const_iterator it = g_mapPolyphaseFilters.find(params);
if( it != g_mapPolyphaseFilters.end() )
{
/* We already have a filter for this upsampling factor and cutoff; use it. */
PolyphaseFilter *pPolyphase = it->second;
PolyphaseFiltersLock.Unlock();
return pPolyphase;
}
int iWinSize = L*iUpFactor;
float *pFIR = new float[iWinSize];
GenerateSincLowPassFilter( pFIR, iWinSize, fCutoffFrequency );
ApplyKaiserWindow( pFIR, iWinSize, 8 );
NormalizeVector( pFIR, iWinSize );
MultiplyVector( &pFIR[0], &pFIR[iWinSize], (float) iUpFactor );
PolyphaseFilter *pPolyphase = new PolyphaseFilter( iUpFactor );
pPolyphase->Generate( pFIR );
delete [] pFIR;
g_mapPolyphaseFilters[params] = pPolyphase;
PolyphaseFiltersLock.Unlock();
return pPolyphase;
}
const PolyphaseFilter *FindNearestPolyphaseFilter( int iUpFactor, float fCutoffFrequency )
{
/* Find a cached filter with the same iUpFactor and a nearby cutoff frequency.
* Round the cutoff down, if possible; it's better to filter out too much than
* too little. */
PolyphaseFiltersLock.Lock();
pair<int,float> params( make_pair(iUpFactor, fCutoffFrequency + 0.0001f) );
FilterMap::const_iterator it = g_mapPolyphaseFilters.upper_bound( params );
if( it != g_mapPolyphaseFilters.begin() )
--it;
ASSERT( it->first.first == iUpFactor );
PolyphaseFilter *pPolyphase = it->second;
PolyphaseFiltersLock.Unlock();
return pPolyphase;
}
}
/*
* Interface to PolyphaseFilter, providing a simple resampling interface. This handles
* reuse of PolyphaseFilters. This does not handle delay, flushing, or multiple channels.
*/
class RageSoundResampler_Polyphase
{
public:
/* Note that going outside of [iMinDownFactor,iMaxDownFactor] while resampling isn't
* fatal. It'll only cause aliasing, by not having a LPF that's low enough, or cause
* too much filtering, by not having a LPF that's high enough. */
RageSoundResampler_Polyphase( int iUpFactor, int iMinDownFactor, int iMaxDownFactor )
{
/* Cache filters between iMinDownFactor and iMaxDownFactor. Do them in
* iFilterIncrement increments; we'll round down to the closest match
* when filtering. This will only cause the low-pass filter to be rounded;
* the conversion ratio will always be exact. */
m_iUpFactor = iUpFactor;
m_pPolyphase = NULL;
int iFilterIncrement = max( (iMaxDownFactor - iMinDownFactor)/10, 1 );
for( int iDownFactor = iMinDownFactor; iDownFactor <= iMaxDownFactor; iDownFactor += iFilterIncrement )
{
float fCutoffFrequency = GetCutoffFrequency( iDownFactor );
PolyphaseFilterCache::MakePolyphaseFilter( m_iUpFactor, fCutoffFrequency );
}
SetDownFactor( iUpFactor );
m_pState = new PolyphaseFilter::State( iUpFactor );
}
~RageSoundResampler_Polyphase()
{
delete m_pState;
}
void SetDownFactor( int iDownFactor )
{
m_iDownFactor = iDownFactor;
m_pPolyphase = GetFilter( m_iDownFactor );
}
int Run( const float *pIn, int iSamplesIn, float *pOut, int iSamplesOut, int iSampleStride ) const
{
return m_pPolyphase->RunPolyphaseFilter( *m_pState, pIn, iSamplesIn, m_iDownFactor, pOut, iSamplesOut, iSampleStride );
}
void Reset()
{
delete m_pState;
m_pState = new PolyphaseFilter::State( m_iUpFactor );
}
int NumInputsForOutputSamples( int iOut ) const { return m_pPolyphase->NumInputsForOutputSamples(*m_pState, iOut, m_iDownFactor); }
int GetLatency() const { return m_pPolyphase->GetLatency(); }
int GetFilled() const { return m_pState->m_iFilled; }
RageSoundResampler_Polyphase( const RageSoundResampler_Polyphase &cpy )
{
m_pPolyphase = cpy.m_pPolyphase; // don't copy
m_pState = new PolyphaseFilter::State(*cpy.m_pState);
m_iUpFactor = cpy.m_iUpFactor;
m_iDownFactor = cpy.m_iDownFactor;
}
private:
float GetCutoffFrequency( int iDownFactor ) const
{
/*
* If we're upsampling, we want the low-pass filter to cut off at the
* nyquist frequency of the original sample.
*
* If we're downsampling, we want the low-pass filter to cut off at the
* nyquist frequency of the new sample.
*/
float fCutoffFrequency;
fCutoffFrequency = 1.0f / (2*m_iUpFactor);
fCutoffFrequency = min( fCutoffFrequency, 1.0f / (2*iDownFactor) );
return fCutoffFrequency;
}
const PolyphaseFilter *GetFilter( int iDownFactor ) const
{
float fCutoffFrequency = GetCutoffFrequency( iDownFactor );
return PolyphaseFilterCache::FindNearestPolyphaseFilter( m_iUpFactor, fCutoffFrequency );
}
const PolyphaseFilter *m_pPolyphase;
PolyphaseFilter::State *m_pState;
int m_iUpFactor;
int m_iDownFactor;
};
int RageSoundReader_Resample_Good::GetNextSourceFrame() const
{
int64_t iPosition = m_pSource->GetNextSourceFrame();
iPosition -= m_apResamplers[0]->GetFilled();
iPosition *= m_iSampleRate;
iPosition /= m_pSource->GetSampleRate();
return (int) iPosition;
}
bool RageSoundReader_Resample_Good::SetProperty( const RString &sProperty, float fValue )
{
if( sProperty == "Rate" )
{
SetRate( fValue );
return true;
}
return m_pSource->SetProperty( sProperty, fValue );
}
float RageSoundReader_Resample_Good::GetStreamToSourceRatio() const
{
float fRatio = m_pSource->GetStreamToSourceRatio();
if( m_fRate != -1 )
fRatio *= m_fRate;
return fRatio;
}
RageSoundReader_Resample_Good::RageSoundReader_Resample_Good( RageSoundReader *pSource, int iSampleRate ):
RageSoundReader_Filter( pSource )
{
m_iSampleRate = iSampleRate;
m_fRate = -1;
ReopenResampler();
}
/* Call this if the input position is changed or reset. */
void RageSoundReader_Resample_Good::Reset()
{
for( size_t iChannel = 0; iChannel < m_pSource->GetNumChannels(); ++iChannel )
m_apResamplers[iChannel]->Reset();
}
void RageSoundReader_Resample_Good::GetFactors( int &iDownFactor, int &iUpFactor ) const
{
iDownFactor = m_pSource->GetSampleRate();
iUpFactor = m_iSampleRate;
{
int iGCD = GCD( iUpFactor, iDownFactor );
iUpFactor /= iGCD;
iDownFactor /= iGCD;
}
bool bRateChangingEnabled = m_fRate != -1;
if( bRateChangingEnabled )
{
iUpFactor *= 100;
iDownFactor *= 100;
}
}
/* Call this if the sample factor changes. */
void RageSoundReader_Resample_Good::ReopenResampler()
{
for( size_t iChannel = 0; iChannel < m_apResamplers.size(); ++iChannel )
delete m_apResamplers[iChannel];
m_apResamplers.clear();
int iDownFactor, iUpFactor;
GetFactors( iDownFactor, iUpFactor );
for( size_t iChannel = 0; iChannel < m_pSource->GetNumChannels(); ++iChannel )
{
int iMinDownFactor = iDownFactor;
int iMaxDownFactor = iDownFactor;
if( m_fRate != -1 )
iMaxDownFactor *= 5;
RageSoundResampler_Polyphase *p = new RageSoundResampler_Polyphase( iUpFactor, iMinDownFactor, iMaxDownFactor );
m_apResamplers.push_back( p );
}
if( m_fRate != -1 )
iDownFactor = lrintf( m_fRate * iDownFactor );
for( size_t iChannel = 0; iChannel < m_apResamplers.size(); ++iChannel )
m_apResamplers[iChannel]->SetDownFactor( iDownFactor );
}
RageSoundReader_Resample_Good::~RageSoundReader_Resample_Good()
{
for( size_t iChannel = 0; iChannel < m_apResamplers.size(); ++iChannel )
delete m_apResamplers[iChannel];
}
/* iFrame is in the destination rate. Seek the source in its own sample rate. */
int RageSoundReader_Resample_Good::SetPosition( int iFrame )
{
Reset();
iFrame = (int) SCALE( iFrame, 0, (int64_t) m_iSampleRate, 0, (int64_t) m_pSource->GetSampleRate() );
return m_pSource->SetPosition( iFrame );
}
int RageSoundReader_Resample_Good::Read( float *pBuf, int iFrames )
{
int iChannels = m_apResamplers.size();
int iFramesRead = 0;
/* If the ratio is 1:1, then we're effectively disabled, and we can read
* directly into the buffer. */
int iDownFactor, iUpFactor;
GetFactors( iDownFactor, iUpFactor );
if( m_apResamplers[0]->GetFilled() == 0 && iDownFactor == iUpFactor && GetRate() == 1.0f )
return m_pSource->Read( pBuf, iFrames );
{
int iFramesNeeded = m_apResamplers[0]->NumInputsForOutputSamples(iFrames);
float *pTmpBuf = (float *) alloca( iFramesNeeded * sizeof(float) * iChannels );
ASSERT( pTmpBuf != NULL );
int iFramesIn = m_pSource->Read( pTmpBuf, iFramesNeeded );
if( iFramesIn < 0 )
return iFramesIn;
for( int iChannel = 0; iChannel < iChannels; ++iChannel )
{
int iGotFrames = m_apResamplers[iChannel]->Run( pTmpBuf + iChannel, iFramesIn, pBuf + iChannel, iFrames, iChannels );
ASSERT( iGotFrames <= iFrames );
if( iChannel == 0 )
iFramesRead += iGotFrames;
}
}
return iFramesRead;
}
/*
* A resampler is commonly used for two things: to change the sample rate of audio,
* in order to give an audio driver what it wants (SetSampleRate), and to change the
* sound of audio, changing its speed and pitch (SetRate). These are the same
* operation, and we do both in the same pass; the only difference is that SetSampleRate
* causes GetSampleRate() to change, while SetRate() causes GetStreamToSourceRatio() to change.
*
* Changing these values will take effect immediately, with a buffering latency of L/4
* frames.
*/
void RageSoundReader_Resample_Good::SetRate( float fRatio )
{
ASSERT( fRatio > 0 );
bool bRateChangingWasEnabled = m_fRate != -1;
m_fRate = fRatio;
if( !bRateChangingWasEnabled )
ReopenResampler();
int iDownFactor, iUpFactor;
GetFactors( iDownFactor, iUpFactor );
if( m_fRate != -1 )
iDownFactor = lrintf( m_fRate * iDownFactor );
/* Set m_fRate to the actual rate, after quantization by iUpFactor. */
m_fRate = float(iDownFactor) / iUpFactor;
for( size_t iChannel = 0; iChannel < m_apResamplers.size(); ++iChannel )
m_apResamplers[iChannel]->SetDownFactor( iDownFactor );
}
float RageSoundReader_Resample_Good::GetRate() const
{
if( m_fRate == -1 )
return 1.0f;
else
return m_fRate;
}
RageSoundReader_Resample_Good::RageSoundReader_Resample_Good( const RageSoundReader_Resample_Good &cpy ):
RageSoundReader_Filter(cpy)
{
for( size_t i = 0; i < cpy.m_apResamplers.size(); ++i )
this->m_apResamplers.push_back( new RageSoundResampler_Polyphase(*cpy.m_apResamplers[i]) );
this->m_iSampleRate = cpy.m_iSampleRate;
this->m_fRate = cpy.m_fRate;
}
RageSoundReader_Resample_Good *RageSoundReader_Resample_Good::Copy() const
{
return new RageSoundReader_Resample_Good( *this );
}
/*
* (c) 2006 Glenn Maynard
* All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, and/or sell copies of the Software, and to permit persons to
* whom the Software is furnished to do so, provided that the above
* copyright notice(s) and this permission notice appear in all copies of
* the Software and that both the above copyright notice(s) and this
* permission notice appear in supporting documentation.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT OF
* THIRD PARTY RIGHTS. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR HOLDERS
* INCLUDED IN THIS NOTICE BE LIABLE FOR ANY CLAIM, OR ANY SPECIAL INDIRECT
* OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS
* OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
* OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
* PERFORMANCE OF THIS SOFTWARE.
*/