Codeword*
Huffman::obtainCodewords()
{
// *************************************
// This implementation is optimized for managing codewords up to 32 bits.
// *************************************
// Codewords are right-aligned to be compliant with the decoding table
Codeword* codewords = new Codeword[huff_table.max+1];
// The Huffman tree is rebuilt for the new code assignment
tree = new BinaryNode();
uint *stream = new uint[256];
for (uint i=0; i<256; i++) stream[i] = 0;
//for (uint i=0; i<=huff_table.max; i++)
for (uint i=0; i<=255; i++)
{
BinaryNode *current = tree;
codewords[i].bits = encodeHuff(huff_table, i, stream, (size_t)0);
codewords[i].codeword = 0;
for (uint j=0; j<codewords[i].bits; j++)
{
bool bit = (stream[0] >> j) & 1;
codewords[i].codeword = codewords[i].codeword << 1;
codewords[i].codeword = (codewords[i].codeword | bit);
if (bit == 0)
{
if (current->leftChild == NULL) current->leftChild = new BinaryNode();
current = current->leftChild;
}
else
{
if (current->rightChild == NULL) current->rightChild = new BinaryNode();
current = current->rightChild;
}
}
current->position = i;
stream[0] = 0;
}
delete [] stream;
return codewords;
}
GonzaloCompressedPsi gonzaloCompressPsi(uint *Psi, uint psiSize, uint T, uint HUFF) {
GonzaloCompressedPsi compressedPsi;
register uint i;
uint oi,j;
int ok,k;
register uint _cptr;
uint *_cPsi;
uint *_bposS;
uint links = psiSize;
uint samplen = T;
uint _bplen;
uint pslen;
uint totexc;
uint *acc,*lacc;
THuff Hacc, Hlen;
uint totalSize;
// Construe os arboles de huffman, o dos valores directos
// e o das lonxitudes dos runs. Usa como vectores auxiliares de frecuencias
// a acc e lacc, que finalmente libera.
acc = (uint *)malloc (HUFF*sizeof(uint));
lacc = (uint *)malloc ((samplen-1)*sizeof(uint));
for (k=0;k<HUFF;k++) acc[k]=0;
for (k=0;k<samplen-1;k++) lacc[k]=0;
ok = 0;
k = Psi[0];
for (i=0;i<=links;i++) {
if ((k == 1) && (i % samplen)) { if (ok != 1) oi = i; }
else {
if (ok == 1) {
acc[1]++;
lacc[i-oi-1]++;
}
if (i % samplen)
if ((k < 1) || (k >= HUFF)) acc[0]++;
else acc[k]++;
}
ok = (i % samplen) ? k : 0;
k = Psi[i+1]-Psi[i];
}
if (ok == 1) {
acc[1]++;
lacc[i-oi-1]++;
}
Hacc = createHuff (acc,HUFF-1, UNSORTED);
Hlen = createHuff (lacc,samplen-2, UNSORTED);
totexc = acc[0];
pslen = bits(psiSize+1);
_bplen = bits(Hacc.total+Hlen.total+(1+links/samplen+totexc)*pslen);
_bposS = (uint *)malloc ((((1+links/samplen)*_bplen+W-1)/W)*sizeof(uint));
_cPsi = (uint *)malloc (((Hacc.total+Hlen.total+(1+links/samplen+totexc)*pslen+W-1)/W)*sizeof(uint));
_cptr = 0;
ok = 0;
k = Psi[0];
for (i=0;i<=links;i++) {
if ((k == 1) && (i % samplen)) { if (ok != 1) oi = i; }
else {
if (ok == 1) {
_cptr = encodeHuff (Hacc,1,_cPsi,_cptr);
_cptr = encodeHuff(Hlen,i-oi-1,_cPsi,_cptr);
}
if (i % samplen) {
if ((k > 1) && (k < HUFF)) _cptr = encodeHuff (Hacc,k,_cPsi,_cptr);
else {
_cptr = encodeHuff (Hacc,0,_cPsi,_cptr);
bitwrite (_cPsi,_cptr,pslen,Psi[i]);
_cptr += pslen;
}
}
else {
bitwrite (_bposS,(i/samplen)*_bplen,_bplen,_cptr);
bitwrite (_cPsi,_cptr,pslen,Psi[i]);
_cptr += pslen;
}
}
ok = (i % samplen) ? k : 0;
k = Psi[i+1]-Psi[i];
}
if (ok == 1) {
_cptr = encodeHuff (Hacc,1,_cPsi,_cptr);
_cptr = encodeHuff(Hlen,i-oi-1,_cPsi,_cptr);
}
// Calculamos o espacio total
totalSize = (((1+links/samplen)*_bplen+W-1)/W)*sizeof(uint) +
((Hacc.total+Hlen.total+(1+links/samplen+totexc)*pslen+W-1)/W)*sizeof(uint) +
5*sizeof(int) + sizeHuff(Hacc) + sizeHuff(Hlen);
printf("\n\tCompressed Psi size = %d bytes\n", totalSize);
// Necesario antes de decodificar
prepareToDecode(&Hacc);
prepareToDecode(&Hlen);
// Asignamos os valores e devolvemos psi comprimido
compressedPsi.links = psiSize;
compressedPsi.totexc = totexc;
compressedPsi.cPsi = _cPsi;
compressedPsi.samplen = samplen;
compressedPsi.bposS = _bposS;
compressedPsi.bplen = _bplen;
compressedPsi.pslen = pslen;
compressedPsi.Hacc = Hacc;
compressedPsi.Hlen = Hlen;
compressedPsi.totalMem = totalSize;
free(acc);
free(lacc);
return compressedPsi;
}
uint64_t Huffman::encode(uint symb, uint * stream, uint64_t pos){
return encodeHuff(huff_table, symb, stream, pos);
}
HuffmanCompressedPsi huffmanCompressPsi(unsigned int *Psi, size_t psiSize, unsigned int T, unsigned int nS) {
HuffmanCompressedPsi cPsi;
uint absolute_value;
register size_t index;
register size_t ptr, samplesPtr, samplePointersPtr;
unsigned int runLenght, binaryLenght;
ssize_t *diffs;
unsigned int *huffmanDst;
// Estructuras da funcion comprimida (para logo asignar)
// Tam�n se podian almacenar directamente
THuff diffsHT;
size_t numberOfSamples;
unsigned int *samples;
unsigned int sampleSize;
size_t *samplePointers;
unsigned int pointerSize;
unsigned int *stream;
size_t streamSize;
// Variables que marcan os intervalos dentro do vector de frecuencias
unsigned int runLenghtStart = nS - 64 - T; // Inicio das Runs
unsigned int negStart = nS - 64; // Inicio dos Negativos
unsigned int bigStart = nS - 32; // Inicio dos Grandes (>runLenghtStart)
// Para estadistica
size_t totalSize;
// Reservamos espacio para a distribuci�n de valores de Psi
huffmanDst = (unsigned int *)malloc(sizeof(int)*nS);
for(index=0;index<nS;index++) huffmanDst[index]=0;
// Inicializamos diferencias
diffs = (ssize_t *)malloc(sizeof(ssize_t)*psiSize);
diffs[0] = 0;
for(index=1; index<psiSize; index++)
diffs[index] = ((ssize_t)Psi[index]) - ((ssize_t)Psi[index-1]);
// Calculamos a distribucion de frecuencias
runLenght = 0;
for(index=0; index<psiSize; index++) {
if(index%T) {
if(diffs[index]== ((ssize_t) 1) ) {
runLenght++;
} else { // Non estamos nun run
if(runLenght) {
huffmanDst[runLenght+runLenghtStart]++;
runLenght = 0;
}
if(diffs[index]>((ssize_t)1) && diffs[index]<runLenghtStart)
huffmanDst[diffs[index]]++;
else
if(diffs[index]< ((ssize_t) 0) ) { // Valor negativo
absolute_value = (uint) (-diffs[index]);
binaryLenght = bits(absolute_value);
huffmanDst[binaryLenght+negStart-1]++;
} else { // Valor grande >= 128
absolute_value = (uint)(diffs[index]);
binaryLenght = bits(absolute_value);
huffmanDst[binaryLenght+bigStart-1]++;
}
}
} else { // Rompemos o run porque atopamos unha mostra
if(runLenght) {
huffmanDst[runLenght+runLenghtStart]++;
runLenght = 0;
}
}
}
if(runLenght) huffmanDst[runLenght+runLenghtStart]++;
// Creamos o arbol de Huffman
diffsHT = createHuff(huffmanDst,nS-1,UNSORTED);
// Calculamos o espacio total ocupado pola secuencia Huffman + RLE
streamSize = diffsHT.total;
for(index=negStart;index<bigStart;index++)
streamSize += ((size_t)huffmanDst[index])*(index-negStart+1); // Negativos
for(index=bigStart;index<nS;index++)
streamSize += ((size_t)huffmanDst[index])*(index-bigStart+1); // Grandes
// Calculamos o numero de mostras e o espacio ocupado por cada mostra e por cada punteiro
numberOfSamples = (psiSize+T-1)/T;
sampleSize = bits(psiSize);
pointerSize = bits(streamSize);
// Reservamos espacio para a secuencia e para as mostras e punteiros
samples = (unsigned int *)malloc(sizeof(uint)*((numberOfSamples*sampleSize+W-1)/W));
samples[((numberOfSamples*sampleSize+W-1)/W)-1] =0000; //initialized only to avoid valgrind warnings
samplePointers = (size_t *)malloc(sizeof(size_t)* (ulong_len(pointerSize,numberOfSamples)) );
samplePointers[ (ulong_len(pointerSize,numberOfSamples)) -1] = 00000000; //initialized only to avoid valgrind warnings
stream = (unsigned int *)malloc(sizeof(int)*((streamSize+W-1)/W));
stream[((streamSize+W-1)/W)-1]=0000;//initialized only to avoid valgrind warnings
// Comprimimos secuencialmente (haber� que levar un punteiro desde o inicio)
ptr = 0;
samplesPtr = 0;
samplePointersPtr = 0;
runLenght = 0;
for(index=0; index<psiSize; index++) {
if(index%T) {
if(diffs[index]==((ssize_t)1)) {
runLenght++;
} else { // Non estamos nun run
if(runLenght) {
ptr = encodeHuff(diffsHT,runLenght+runLenghtStart,stream,ptr);
runLenght = 0;
}
if(diffs[index]>((ssize_t)1) && diffs[index]<runLenghtStart) {
ptr = encodeHuff(diffsHT,(uint)diffs[index],stream,ptr);
}
else
if(diffs[index]< ((ssize_t)0) ) { // Valor negativo
absolute_value = (uint) (-diffs[index]);
binaryLenght = bits(absolute_value);
ptr = encodeHuff(diffsHT,binaryLenght+negStart-1,stream,ptr);
bitwrite(stream,ptr,binaryLenght,absolute_value);
ptr += binaryLenght;
} else { // Valor grande >= 128
absolute_value = (uint) diffs[index];
binaryLenght = bits(absolute_value);
ptr = encodeHuff(diffsHT,binaryLenght+bigStart-1,stream,ptr);
bitwrite(stream,ptr,binaryLenght,absolute_value);
ptr += binaryLenght;
}
}
} else { // Rompemos o run porque atopamos unha mostra
if(runLenght) {
ptr = encodeHuff(diffsHT,runLenght+runLenghtStart,stream,ptr);
runLenght = 0;
}
bitwrite(samples,samplesPtr,sampleSize, Psi[index]);
samplesPtr += sampleSize;
bitwrite64(samplePointers,samplePointersPtr,pointerSize,ptr);
samplePointersPtr += pointerSize;
}
}
if(runLenght) {
ptr = encodeHuff(diffsHT,runLenght+runLenghtStart,stream,ptr);
}
// Amosamos o espacio ocupado
totalSize = sizeof(HuffmanCompressedPsi) +
sizeof(int)*((numberOfSamples*sampleSize+W-1)/W) +
sizeof(size_t)*((numberOfSamples*pointerSize+WW-1)/WW) +
sizeof(int)*((streamSize+W-1)/W) + sizeHuff(diffsHT);
printf("\n\t Compressed Psi size = %zu bytes, with %d different symbols.", totalSize, nS);
// Necesario antes de decodificar
prepareToDecode(&diffsHT);
// Asignamos os valores a cPsi e devolvemolo
cPsi.T = T;
cPsi.diffsHT = diffsHT;
cPsi.nS = nS;
cPsi.numberOfSamples = numberOfSamples;
cPsi.samples = samples;
cPsi.sampleSize = sampleSize;
cPsi.samplePointers = samplePointers;
cPsi.pointerSize = pointerSize;
cPsi.stream = stream;
cPsi.streamSize = streamSize;
cPsi.totalMem = totalSize;
//frees resources not needed in advance
free(diffs);
free(huffmanDst);
//returns the data structure that holds the compressed psi.
return cPsi;
}