mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-11-28 04:31:06 +00:00
375 lines
12 KiB
C++
375 lines
12 KiB
C++
#include "All.h"
|
|
#include "APECompress.h"
|
|
#include "NewPredictor.h"
|
|
|
|
typedef struct {
|
|
int nFilter0Length;
|
|
int nFilter0Shift;
|
|
int nFilter1Length;
|
|
int nFilter1Shift;
|
|
int nFilter2Length;
|
|
int nFilter2Shift;
|
|
} OrderType;
|
|
|
|
// pfk:
|
|
//
|
|
// Where does these numbers come from?
|
|
// What are restrictions so the compressor works for every signal possible?
|
|
// Instead of a magic CompressionLevel these 6 (or 4 numbers) should be stored in the header
|
|
// some MMX code may also increase speed for fast or normal compression, but reduces readability.
|
|
// Also note that shifting is extremly slow on Pentium 4 (why? no idea!)
|
|
// Rotation + Order of the follwoing buffers mirroring: m_rbPredictionA, m_rbPredictionB, m_rbAdaptA, m_rbAdaptB -> allows MMX/Altivec
|
|
|
|
const OrderType OrderTypeArray [] = {
|
|
{ 0, 0, 0, 0, 0, 0 },
|
|
{ 0, 0, 0, 0, 0, 0 },
|
|
{ 16, 11, 0, 0, 0, 0 },
|
|
{ 64, 11, 0, 0, 0, 0 },
|
|
{ 256, 13, 32, 10, 0, 0 },
|
|
{ 1024, 15, 256, 13, 0, 0 },
|
|
{ 1024, 15, 256, 13, 16, 11 },
|
|
};
|
|
|
|
/*****************************************************************************************
|
|
CPredictorCompressNormal
|
|
*****************************************************************************************/
|
|
CPredictorCompressNormal::CPredictorCompressNormal(int nCompressionLevel)
|
|
: IPredictorCompress(nCompressionLevel)
|
|
{
|
|
const OrderType* p = OrderTypeArray + nCompressionLevel / 1000;
|
|
|
|
if ( nCompressionLevel < 1000 || nCompressionLevel > 6000 || nCompressionLevel % 1000 != 0 )
|
|
throw (1);
|
|
|
|
m_pNNFilter0 = p->nFilter0Length ? new CNNFilter (p->nFilter0Length, p->nFilter0Shift) : NULL;
|
|
m_pNNFilter1 = p->nFilter1Length ? new CNNFilter (p->nFilter1Length, p->nFilter1Shift) : NULL;
|
|
m_pNNFilter2 = p->nFilter2Length ? new CNNFilter (p->nFilter2Length, p->nFilter2Shift) : NULL;
|
|
}
|
|
|
|
CPredictorCompressNormal::~CPredictorCompressNormal()
|
|
{
|
|
SAFE_DELETE (m_pNNFilter0)
|
|
SAFE_DELETE (m_pNNFilter1)
|
|
SAFE_DELETE (m_pNNFilter2)
|
|
}
|
|
|
|
int CPredictorCompressNormal::Flush()
|
|
{
|
|
if ( m_pNNFilter0 != NULL ) m_pNNFilter0->Flush();
|
|
if ( m_pNNFilter1 != NULL ) m_pNNFilter1->Flush();
|
|
if ( m_pNNFilter2 != NULL ) m_pNNFilter2->Flush();
|
|
|
|
m_rbPredictionA.Flush(); m_rbPredictionB.Flush();
|
|
m_rbAdaptA .Flush(); m_rbAdaptB .Flush();
|
|
m_Stage1FilterA.Flush(); m_Stage1FilterB.Flush();
|
|
|
|
ZeroMemory(m_aryMA, sizeof(m_aryMA));
|
|
ZeroMemory(m_aryMB, sizeof(m_aryMB));
|
|
|
|
m_aryMA[0] = 360;
|
|
m_aryMA[1] = 317;
|
|
m_aryMA[2] = -109;
|
|
m_aryMA[3] = 98;
|
|
|
|
m_nLastValueA = 0;
|
|
|
|
m_nCurrentIndex = 0;
|
|
|
|
return 0;
|
|
}
|
|
|
|
|
|
int CPredictorCompressNormal::CompressValue(int nA, int nB)
|
|
{
|
|
if (m_nCurrentIndex == WINDOW_BLOCKS)
|
|
{
|
|
m_rbPredictionA.Roll(); m_rbPredictionB.Roll();
|
|
m_rbAdaptA .Roll(); m_rbAdaptB .Roll();
|
|
|
|
m_nCurrentIndex = 0;
|
|
}
|
|
|
|
// stage 1: simple, non-adaptive order 1 prediction
|
|
int nCurrentA = m_Stage1FilterA.Compress(nA);
|
|
int nCurrentB = m_Stage1FilterB.Compress(nB);
|
|
|
|
// stage 2: adaptive offset filter(s)
|
|
m_rbPredictionA[0] = m_nLastValueA;
|
|
m_rbPredictionA[-1] = m_rbPredictionA[0] - m_rbPredictionA[-1];
|
|
|
|
m_rbPredictionB[0] = nCurrentB;
|
|
m_rbPredictionB[-1] = m_rbPredictionB[0] - m_rbPredictionB[-1];
|
|
|
|
int nPredictionA = (m_rbPredictionA[0] * m_aryMA[0]) + (m_rbPredictionA[-1] * m_aryMA[1]) + (m_rbPredictionA[-2] * m_aryMA[2]) + (m_rbPredictionA[-3] * m_aryMA[3]);
|
|
int nPredictionB = (m_rbPredictionB[0] * m_aryMB[0]) + (m_rbPredictionB[-1] * m_aryMB[1]) + (m_rbPredictionB[-2] * m_aryMB[2]) + (m_rbPredictionB[-3] * m_aryMB[3]) + (m_rbPredictionB[-4] * m_aryMB[4]);
|
|
|
|
int nOutput = nCurrentA - ((nPredictionA + (nPredictionB >> 1)) >> 10);
|
|
|
|
m_nLastValueA = nCurrentA;
|
|
|
|
m_rbAdaptA[0] = (m_rbPredictionA[0]) ? ((m_rbPredictionA[0] >> 30) & 2) - 1 : 0;
|
|
m_rbAdaptA[-1] = (m_rbPredictionA[-1]) ? ((m_rbPredictionA[-1] >> 30) & 2) - 1 : 0;
|
|
|
|
m_rbAdaptB[0] = (m_rbPredictionB[0]) ? ((m_rbPredictionB[0] >> 30) & 2) - 1 : 0;
|
|
m_rbAdaptB[-1] = (m_rbPredictionB[-1]) ? ((m_rbPredictionB[-1] >> 30) & 2) - 1 : 0;
|
|
|
|
if (nOutput > 0)
|
|
{
|
|
m_aryMA[0] -= m_rbAdaptA[0];
|
|
m_aryMA[1] -= m_rbAdaptA[-1];
|
|
m_aryMA[2] -= m_rbAdaptA[-2];
|
|
m_aryMA[3] -= m_rbAdaptA[-3];
|
|
|
|
m_aryMB[0] -= m_rbAdaptB[0];
|
|
m_aryMB[1] -= m_rbAdaptB[-1];
|
|
m_aryMB[2] -= m_rbAdaptB[-2];
|
|
m_aryMB[3] -= m_rbAdaptB[-3];
|
|
m_aryMB[4] -= m_rbAdaptB[-4];
|
|
}
|
|
else if (nOutput < 0)
|
|
{
|
|
m_aryMA[0] += m_rbAdaptA[0];
|
|
m_aryMA[1] += m_rbAdaptA[-1];
|
|
m_aryMA[2] += m_rbAdaptA[-2];
|
|
m_aryMA[3] += m_rbAdaptA[-3];
|
|
|
|
m_aryMB[0] += m_rbAdaptB[0];
|
|
m_aryMB[1] += m_rbAdaptB[-1];
|
|
m_aryMB[2] += m_rbAdaptB[-2];
|
|
m_aryMB[3] += m_rbAdaptB[-3];
|
|
m_aryMB[4] += m_rbAdaptB[-4];
|
|
}
|
|
|
|
// stage 3: NNFilters
|
|
|
|
if ( m_pNNFilter0 != NULL ) nOutput = m_pNNFilter0->Compress(nOutput);
|
|
if ( m_pNNFilter1 != NULL ) nOutput = m_pNNFilter1->Compress(nOutput);
|
|
if ( m_pNNFilter2 != NULL ) nOutput = m_pNNFilter2->Compress(nOutput);
|
|
|
|
m_rbPredictionA.IncrementFast(); m_rbPredictionB.IncrementFast();
|
|
m_rbAdaptA.IncrementFast(); m_rbAdaptB.IncrementFast();
|
|
|
|
m_nCurrentIndex++;
|
|
|
|
return nOutput;
|
|
}
|
|
|
|
/*****************************************************************************************
|
|
CPredictorDecompressNormal3930to3950
|
|
*****************************************************************************************/
|
|
CPredictorDecompressNormal3930to3950::CPredictorDecompressNormal3930to3950(int nCompressionLevel)
|
|
: IPredictorDecompress(nCompressionLevel)
|
|
{
|
|
const OrderType* p = OrderTypeArray + nCompressionLevel / 1000;
|
|
m_pBuffer[0] = new int [HISTORY_ELEMENTS + WINDOW_BLOCKS];
|
|
|
|
if ( nCompressionLevel < 1000 || nCompressionLevel > 6000 || nCompressionLevel % 1000 != 0 )
|
|
throw (1);
|
|
|
|
m_pNNFilter0 = p->nFilter0Length ? new CNNFilter (p->nFilter0Length, p->nFilter0Shift) : NULL;
|
|
m_pNNFilter1 = p->nFilter1Length ? new CNNFilter (p->nFilter1Length, p->nFilter1Shift) : NULL;
|
|
m_pNNFilter2 = p->nFilter2Length ? new CNNFilter (p->nFilter2Length, p->nFilter2Shift) : NULL;
|
|
}
|
|
|
|
CPredictorDecompressNormal3930to3950::~CPredictorDecompressNormal3930to3950()
|
|
{
|
|
SAFE_DELETE (m_pNNFilter0)
|
|
SAFE_DELETE (m_pNNFilter1)
|
|
SAFE_DELETE (m_pNNFilter2)
|
|
SAFE_ARRAY_DELETE(m_pBuffer[0])
|
|
}
|
|
|
|
int CPredictorDecompressNormal3930to3950::Flush()
|
|
{
|
|
if (m_pNNFilter0) m_pNNFilter0->Flush();
|
|
if (m_pNNFilter1) m_pNNFilter1->Flush();
|
|
if (m_pNNFilter2) m_pNNFilter2->Flush();
|
|
|
|
ZeroMemory(m_pBuffer[0], (HISTORY_ELEMENTS + 1) * sizeof(int));
|
|
ZeroMemory(&m_aryM[0], M_COUNT * sizeof(int));
|
|
|
|
m_aryM[0] = 360;
|
|
m_aryM[1] = 317;
|
|
m_aryM[2] = -109;
|
|
m_aryM[3] = 98;
|
|
|
|
m_pInputBuffer = &m_pBuffer[0][HISTORY_ELEMENTS];
|
|
|
|
m_nLastValue = 0;
|
|
m_nCurrentIndex = 0;
|
|
|
|
return 0;
|
|
}
|
|
|
|
int CPredictorDecompressNormal3930to3950::DecompressValue(int nInput, int)
|
|
{
|
|
if (m_nCurrentIndex == WINDOW_BLOCKS)
|
|
{
|
|
// copy forward and adjust pointers
|
|
memcpy(&m_pBuffer[0][0], &m_pBuffer[0][WINDOW_BLOCKS], HISTORY_ELEMENTS * sizeof(int));
|
|
m_pInputBuffer = &m_pBuffer[0][HISTORY_ELEMENTS];
|
|
|
|
m_nCurrentIndex = 0;
|
|
}
|
|
|
|
// stage 2: NNFilter
|
|
if (m_pNNFilter2 != NULL) nInput = m_pNNFilter2->Decompress(nInput);
|
|
if (m_pNNFilter1 != NULL) nInput = m_pNNFilter1->Decompress(nInput);
|
|
if (m_pNNFilter0 != NULL) nInput = m_pNNFilter0->Decompress(nInput);
|
|
|
|
// stage 1: multiple predictors (order 2 and offset 1)
|
|
|
|
int p1 = m_pInputBuffer[-1];
|
|
int p2 = m_pInputBuffer[-1] - m_pInputBuffer[-2];
|
|
int p3 = m_pInputBuffer[-2] - m_pInputBuffer[-3];
|
|
int p4 = m_pInputBuffer[-3] - m_pInputBuffer[-4];
|
|
|
|
m_pInputBuffer[0] = nInput + ((p1 * m_aryM[0] + p2 * m_aryM[1] + p3 * m_aryM[2] + p4 * m_aryM[3]) >> 9);
|
|
|
|
if (nInput > 0)
|
|
{
|
|
m_aryM[0] -= ((p1 >> 30) & 2) - 1;
|
|
m_aryM[1] -= ((p2 >> 30) & 2) - 1;
|
|
m_aryM[2] -= ((p3 >> 30) & 2) - 1;
|
|
m_aryM[3] -= ((p4 >> 30) & 2) - 1;
|
|
}
|
|
else if (nInput < 0)
|
|
{
|
|
m_aryM[0] += ((p1 >> 30) & 2) - 1;
|
|
m_aryM[1] += ((p2 >> 30) & 2) - 1;
|
|
m_aryM[2] += ((p3 >> 30) & 2) - 1;
|
|
m_aryM[3] += ((p4 >> 30) & 2) - 1;
|
|
}
|
|
|
|
int nRetVal = m_pInputBuffer[0] + ((m_nLastValue * 31) >> 5);
|
|
m_nLastValue = nRetVal;
|
|
|
|
m_nCurrentIndex++;
|
|
m_pInputBuffer++;
|
|
|
|
return nRetVal;
|
|
}
|
|
|
|
/*****************************************************************************************
|
|
CPredictorDecompress3950toCurrent
|
|
*****************************************************************************************/
|
|
CPredictorDecompress3950toCurrent::CPredictorDecompress3950toCurrent(int nCompressionLevel)
|
|
: IPredictorDecompress(nCompressionLevel)
|
|
{
|
|
const OrderType* p = OrderTypeArray + nCompressionLevel / 1000;
|
|
|
|
if ( nCompressionLevel < 1000 || nCompressionLevel > 6000 || nCompressionLevel % 1000 != 0 )
|
|
throw (1);
|
|
|
|
m_pNNFilter0 = p->nFilter0Length ? new CNNFilter (p->nFilter0Length, p->nFilter0Shift) : NULL;
|
|
m_pNNFilter1 = p->nFilter1Length ? new CNNFilter (p->nFilter1Length, p->nFilter1Shift) : NULL;
|
|
m_pNNFilter2 = p->nFilter2Length ? new CNNFilter (p->nFilter2Length, p->nFilter2Shift) : NULL;
|
|
}
|
|
|
|
CPredictorDecompress3950toCurrent::~CPredictorDecompress3950toCurrent()
|
|
{
|
|
SAFE_DELETE(m_pNNFilter0)
|
|
SAFE_DELETE(m_pNNFilter1)
|
|
SAFE_DELETE(m_pNNFilter2)
|
|
}
|
|
|
|
int CPredictorDecompress3950toCurrent::Flush()
|
|
{
|
|
if (m_pNNFilter0 != NULL) m_pNNFilter0->Flush();
|
|
if (m_pNNFilter1 != NULL) m_pNNFilter1->Flush();
|
|
if (m_pNNFilter2 != NULL) m_pNNFilter2->Flush();
|
|
|
|
ZeroMemory(m_aryMA, sizeof(m_aryMA));
|
|
ZeroMemory(m_aryMB, sizeof(m_aryMB));
|
|
|
|
m_rbPredictionA.Flush();
|
|
m_rbPredictionB.Flush();
|
|
m_rbAdaptA.Flush();
|
|
m_rbAdaptB.Flush();
|
|
|
|
m_aryMA[0] = 360;
|
|
m_aryMA[1] = 317;
|
|
m_aryMA[2] = -109;
|
|
m_aryMA[3] = 98;
|
|
|
|
m_Stage1FilterA.Flush();
|
|
m_Stage1FilterB.Flush();
|
|
|
|
m_nLastValueA = 0;
|
|
|
|
m_nCurrentIndex = 0;
|
|
|
|
return 0;
|
|
}
|
|
|
|
int CPredictorDecompress3950toCurrent::DecompressValue(int nA, int nB)
|
|
{
|
|
if (m_nCurrentIndex == WINDOW_BLOCKS)
|
|
{
|
|
// copy forward and adjust pointers
|
|
m_rbPredictionA.Roll(); m_rbPredictionB.Roll();
|
|
m_rbAdaptA.Roll(); m_rbAdaptB.Roll();
|
|
|
|
m_nCurrentIndex = 0;
|
|
}
|
|
|
|
// stage 2: NNFilter
|
|
if (m_pNNFilter2 != NULL) nA = m_pNNFilter2->Decompress(nA);
|
|
if (m_pNNFilter1 != NULL) nA = m_pNNFilter1->Decompress(nA);
|
|
if (m_pNNFilter0 != NULL) nA = m_pNNFilter0->Decompress(nA);
|
|
|
|
// stage 1: multiple predictors (order 2 and offset 1)
|
|
m_rbPredictionA[0] = m_nLastValueA;
|
|
m_rbPredictionA[-1] = m_rbPredictionA[0] - m_rbPredictionA[-1];
|
|
|
|
m_rbPredictionB[0] = m_Stage1FilterB.Compress(nB);
|
|
m_rbPredictionB[-1] = m_rbPredictionB[0] - m_rbPredictionB[-1];
|
|
|
|
int nPredictionA = (m_rbPredictionA[0] * m_aryMA[0]) + (m_rbPredictionA[-1] * m_aryMA[1]) + (m_rbPredictionA[-2] * m_aryMA[2]) + (m_rbPredictionA[-3] * m_aryMA[3]);
|
|
int nPredictionB = (m_rbPredictionB[0] * m_aryMB[0]) + (m_rbPredictionB[-1] * m_aryMB[1]) + (m_rbPredictionB[-2] * m_aryMB[2]) + (m_rbPredictionB[-3] * m_aryMB[3]) + (m_rbPredictionB[-4] * m_aryMB[4]);
|
|
|
|
int nCurrentA = nA + ((nPredictionA + (nPredictionB >> 1)) >> 10);
|
|
|
|
m_rbAdaptA[0] = (m_rbPredictionA[0]) ? ((m_rbPredictionA[0] >> 30) & 2) - 1 : 0;
|
|
m_rbAdaptA[-1] = (m_rbPredictionA[-1]) ? ((m_rbPredictionA[-1] >> 30) & 2) - 1 : 0;
|
|
|
|
m_rbAdaptB[0] = (m_rbPredictionB[0]) ? ((m_rbPredictionB[0] >> 30) & 2) - 1 : 0;
|
|
m_rbAdaptB[-1] = (m_rbPredictionB[-1]) ? ((m_rbPredictionB[-1] >> 30) & 2) - 1 : 0;
|
|
|
|
if (nA > 0)
|
|
{
|
|
m_aryMA[0] -= m_rbAdaptA[0];
|
|
m_aryMA[1] -= m_rbAdaptA[-1];
|
|
m_aryMA[2] -= m_rbAdaptA[-2];
|
|
m_aryMA[3] -= m_rbAdaptA[-3];
|
|
|
|
m_aryMB[0] -= m_rbAdaptB[0];
|
|
m_aryMB[1] -= m_rbAdaptB[-1];
|
|
m_aryMB[2] -= m_rbAdaptB[-2];
|
|
m_aryMB[3] -= m_rbAdaptB[-3];
|
|
m_aryMB[4] -= m_rbAdaptB[-4];
|
|
}
|
|
else if (nA < 0)
|
|
{
|
|
m_aryMA[0] += m_rbAdaptA[0];
|
|
m_aryMA[1] += m_rbAdaptA[-1];
|
|
m_aryMA[2] += m_rbAdaptA[-2];
|
|
m_aryMA[3] += m_rbAdaptA[-3];
|
|
|
|
m_aryMB[0] += m_rbAdaptB[0];
|
|
m_aryMB[1] += m_rbAdaptB[-1];
|
|
m_aryMB[2] += m_rbAdaptB[-2];
|
|
m_aryMB[3] += m_rbAdaptB[-3];
|
|
m_aryMB[4] += m_rbAdaptB[-4];
|
|
}
|
|
|
|
int nRetVal = m_Stage1FilterA.Decompress(nCurrentA);
|
|
m_nLastValueA = nCurrentA;
|
|
|
|
m_rbPredictionA.IncrementFast(); m_rbPredictionB.IncrementFast();
|
|
m_rbAdaptA.IncrementFast(); m_rbAdaptB.IncrementFast();
|
|
|
|
m_nCurrentIndex++;
|
|
|
|
return nRetVal;
|
|
}
|