heretic2-sdk/Toolkit/Programming/Tools/qdata/old.c
1998-11-24 00:00:00 +00:00

559 lines
No EOL
9.6 KiB
C

// Some redundant stuff I found in the video creatin code....
// Could be handy in the future tho...
// JJS
// from qdata.c
#if 0
if (g_compress_pak && len < 4096*1024 )
{
cblock_t in, out;
cblock_t Huffman (cblock_t in);
in.count = len;
in.data = buf;
out = Huffman (in);
if (out.count < in.count)
{
printf (" compressed from %i to %i\n", in.count, out.count);
free (in.data);
buf = out.data;
len = out.count;
}
else
free (out.data);
}
#endif
#if 0
int SmallestNode()
{
int i;
int best, bestnode;
best = 99999999;
bestnode = -1;
for(i = 0; i < numhnodes; i++)
{
if(hnodes[i].used)
continue;
if(!hnodes[i].count)
continue;
if(hnodes[i].count < best)
{
best = hnodes[i].count;
bestnode = i;
}
}
if(bestnode == -1)
return(-1);
hnodes[bestnode].used = true;
return(bestnode);
}
void BuildChars(int nodenum, unsigned bits, int bitcount)
{
hnode_t *node;
if(nodenum < 256)
{
if(bitcount > 32)
Error ("bitcount > 32");
charbits[nodenum] = bits;
charbitscount[nodenum] = bitcount;
return;
}
node = &hnodes[nodenum];
bits <<= 1;
BuildChars (node->children[0], bits, bitcount + 1);
bits |= 1;
BuildChars (node->children[1], bits, bitcount + 1);
}
/*
==================
Huffman
==================
*/
cblock_t Huffman (cblock_t in)
{
int i;
hnode_t *node;
int outbits, c;
unsigned bits;
byte *out_p;
cblock_t out;
int max, maxchar;
// count
memset(hnodes, 0, sizeof(hnodes));
for(i = 0; i < in.count; i++)
hnodes[in.data[i]].count++;
// normalize counts
max = 0;
maxchar = 0;
for(i = 0; i < 256; i++)
{
if (hnodes[i].count > max)
{
max = hnodes[i].count;
maxchar = i;
}
}
if(max == 0)
Error ("Huffman: max == 0");
for(i = 0; i < 256; i++)
hnodes[i].count = (hnodes[i].count * 255 + max - 1) / max;
// build the nodes
numhnodes = 256;
while(numhnodes != 511)
{
node = &hnodes[numhnodes];
// pick two lowest counts
node->children[0] = SmallestNode();
if(node->children[0] == -1)
break; // no more
node->children[1] = SmallestNode ();
if(node->children[1] == -1)
{
if (node->children[0] != numhnodes - 1)
Error ("Bad smallestnode");
break;
}
node->count = hnodes[node->children[0]].count + hnodes[node->children[1]].count;
numhnodes++;
}
BuildChars(numhnodes-1, 0, 0);
out_p = out.data = malloc(in.count * 2 + 1024);
memset (out_p, 0, in.count * 2 + 1024);
// write count
*out_p++ = in.count&255;
*out_p++ = (in.count>>8)&255;
*out_p++ = (in.count>>16)&255;
*out_p++ = (in.count>>24)&255;
// save out the 256 normalized counts so the tree can be recreated
for(i = 0; i < 256; i++)
*out_p++ = hnodes[i].count;
// write bits
outbits = 0;
for (i = 0; i < in.count; i++)
{
c = charbitscount[in.data[i]];
bits = charbits[in.data[i]];
while (c)
{
c--;
if(bits & (1 << c))
out_p[outbits >> 3] |= 1<< (outbits & 7);
outbits++;
}
}
out_p += (outbits + 7) >> 3;
out.count = out_p - out.data;
return(out);
}
#endif
#if 0
int bwt_size;
byte *bwt_data;
int bwtCompare (const void *elem1, const void *elem2)
{
int i;
int i1, i2;
int b1, b2;
i1 = *(int *)elem1;
i2 = *(int *)elem2;
for (i=0 ; i<bwt_size ; i++)
{
b1 = bwt_data[i1];
b2 = bwt_data[i2];
if (b1 < b2)
return -1;
if (b1 > b2)
return 1;
if (++i1 == bwt_size)
i1 = 0;
if (++i2 == bwt_size)
i2 = 0;
}
return 0;
}
/*
==================
BWT
==================
*/
cblock_t BWT (cblock_t in)
{
int *sorted;
int i;
byte *out_p;
cblock_t out;
bwt_size = in.count;
bwt_data = in.data;
sorted = malloc(in.count*sizeof(*sorted));
for (i=0 ; i<in.count ; i++)
sorted[i] = i;
qsort (sorted, in.count, sizeof(*sorted), bwtCompare);
out_p = out.data = malloc(in.count + 8);
// write count
*out_p++ = in.count&255;
*out_p++ = (in.count>>8)&255;
*out_p++ = (in.count>>16)&255;
*out_p++ = (in.count>>24)&255;
// write head index
for (i=0 ; i<in.count ; i++)
if (sorted[i] == 0)
break;
*out_p++ = i&255;
*out_p++ = (i>>8)&255;
*out_p++ = (i>>16)&255;
*out_p++ = (i>>24)&255;
// write the L column
for (i=0 ; i<in.count ; i++)
*out_p++ = in.data[(sorted[i]+in.count-1)%in.count];
free (sorted);
out.count = out_p - out.data;
return out;
}
#endif
#if 0
/*
==================
MTF
==================
*/
cblock_t MTF (cblock_t in)
{
int i, j, b, code;
byte *out_p;
int index[256];
cblock_t out;
out_p = out.data = malloc(in.count + 4);
// write count
*out_p++ = in.count&255;
*out_p++ = (in.count>>8)&255;
*out_p++ = (in.count>>16)&255;
*out_p++ = (in.count>>24)&255;
for (i=0 ; i<256 ; i++)
index[i] = i;
for (i=0 ; i<in.count ; i++)
{
b = in.data[i];
code = index[b];
*out_p++ = code;
// shuffle b indexes to 0
for (j=0 ; j<256 ; j++)
if (index[j] < code)
index[j]++;
index[b] = 0;
}
out.count = out_p - out.data;
return out;
}
/*
==================
RLE
==================
*/
#define RLE_CODE 0xe8
#define RLE_TRIPPLE 0xe9
int rle_counts[256];
int rle_bytes[256];
cblock_t RLE (cblock_t in)
{
int i;
byte *out_p;
int val;
int repeat;
cblock_t out;
out_p = out.data = malloc (in.count*2);
// write count
*out_p++ = in.count&255;
*out_p++ = (in.count>>8)&255;
*out_p++ = (in.count>>16)&255;
*out_p++ = (in.count>>24)&255;
for (i=0 ; i<in.count ; )
{
val = in.data[i];
rle_bytes[val]++;
repeat = 1;
i++;
while (i<in.count && repeat < 255 && in.data[i] == val)
{
repeat++;
i++;
}
if (repeat < 256)
rle_counts[repeat]++;
if (repeat > 3 || val == RLE_CODE)
{
*out_p++ = RLE_CODE;
*out_p++ = val;
*out_p++ = repeat;
}
else
{
while (repeat--)
*out_p++ = val;
}
}
out.count = out_p - out.data;
return out;
}
#endif
//==========================================================================
#if 0
unsigned lzss_head[256];
unsigned lzss_next[0x20000];
/*
==================
LZSS
==================
*/
#define BACK_WINDOW 0x10000
#define BACK_BITS 16
#define FRONT_WINDOW 16
#define FRONT_BITS 4
cblock_t LZSS (cblock_t in)
{
int i;
byte *out_p;
cblock_t out;
int val;
int j, start, max;
int bestlength, beststart;
int outbits;
if (in.count >= sizeof(lzss_next)/4)
Error ("LZSS: too big");
memset (lzss_head, -1, sizeof(lzss_head));
out_p = out.data = malloc (in.count*2);
memset (out.data, 0, in.count*2);
// write count
*out_p++ = in.count&255;
*out_p++ = (in.count>>8)&255;
*out_p++ = (in.count>>16)&255;
*out_p++ = (in.count>>24)&255;
outbits = 0;
for (i=0 ; i<in.count ; )
{
val = in.data[i];
#if 1
// chained search
bestlength = 0;
beststart = 0;
max = FRONT_WINDOW;
if (i + max > in.count)
max = in.count - i;
start = lzss_head[val];
while (start != -1 && start >= i-BACK_WINDOW)
{
// count match length
for (j=0 ; j<max ; j++)
if (in.data[start+j] != in.data[i+j])
break;
if (j > bestlength)
{
bestlength = j;
beststart = start;
}
start = lzss_next[start];
}
#else
// slow simple search
// search for a match
max = FRONT_WINDOW;
if (i + max > in.count)
max = in.count - i;
start = i - BACK_WINDOW;
if (start < 0)
start = 0;
bestlength = 0;
beststart = 0;
for ( ; start < i ; start++)
{
if (in.data[start] != val)
continue;
// count match length
for (j=0 ; j<max ; j++)
if (in.data[start+j] != in.data[i+j])
break;
if (j > bestlength)
{
bestlength = j;
beststart = start;
}
}
#endif
beststart = BACK_WINDOW - (i-beststart);
if (bestlength < 3)
{ // output a single char
bestlength = 1;
out_p[outbits>>3] |= 1<<(outbits&7); // set bit to mark char
outbits++;
for (j=0 ; j<8 ; j++, outbits++)
if (val & (1<<j) )
out_p[outbits>>3] |= 1<<(outbits&7);
}
else
{ // output a phrase
outbits++; // leave a 0 bit to mark phrase
for (j=0 ; j<BACK_BITS ; j++, outbits++)
if (beststart & (1<<j) )
out_p[outbits>>3] |= 1<<(outbits&7);
for (j=0 ; j<FRONT_BITS ; j++, outbits++)
if (bestlength & (1<<j) )
out_p[outbits>>3] |= 1<<(outbits&7);
}
while (bestlength--)
{
val = in.data[i];
lzss_next[i] = lzss_head[val];
lzss_head[val] = i;
i++;
}
}
out_p += (outbits+7)>>3;
out.count = out_p - out.data;
return out;
}
#endif
#if 0
char *RLE(char *in, char *out)
{
char *end;
char *srun;
char count;
end = in + 64; // size of DCT block
while(in < end)
{
srun = in; // Start of run
while(in < end)
{
if(in[0] == in[1])
break;
in++;
}
count = in - srun; // count of unique bytes
if(!count)
{
while(in < end)
{
if(in[0] != in[1])
break;
in++;
}
count = in - srun + 1; // count of repeated bytes
*out++ = -count;
*out++ = *in;
in++;
}
else
{
*out++ = count;
while(count--)
*out++ = *srun++;
}
}
return(out);
}
char *IRLE(char *in, char *out)
{
char count;
char data;
char *end;
end = out + 64;
while(out < end)
{
count = *in++;
if(count < 0)
{
data = *in++;
while(count++)
*out++ = data;
}
else
{
while(count--)
*out++ = *in++;
}
}
if(out != end)
printf("Rle decompression error\n");
return(in);
}
#endif