huffman.c File Reference

#include "bzlib_private.h"

Go to the source code of this file.

Macros
#define	WEIGHTOF(zz0)   ((zz0) & 0xffffff00)
#define	DEPTHOF(zz1)   ((zz1) & 0x000000ff)
#define	MYMAX(zz2, zz3)   ((zz2) > (zz3) ? (zz2) : (zz3))
#define	ADDWEIGHTS(zw1, zw2)
#define	UPHEAP(z)
#define	DOWNHEAP(z)

Functions
void	BZ2_hbMakeCodeLengths (UChar *len, Int32 *freq, Int32 alphaSize, Int32 maxLen)
void	BZ2_hbAssignCodes (Int32 *code, UChar *length, Int32 minLen, Int32 maxLen, Int32 alphaSize)
void	BZ2_hbCreateDecodeTables (Int32 *limit, Int32 *base, Int32 *perm, UChar *length, Int32 minLen, Int32 maxLen, Int32 alphaSize)

Macro Definition Documentation

#define ADDWEIGHTS	(		zw1,
			zw2
	)

Value:

(WEIGHTOF(zw1)+WEIGHTOF(zw2)) |                    \
   (1 + MYMAX(DEPTHOF(zw1),DEPTHOF(zw2)))

Definition at line 32 of file huffman.c.

#define DEPTHOF

(

zz1

)

((zz1) & 0x000000ff)

Definition at line 29 of file huffman.c.

#define DOWNHEAP

(

z

)

Value:

{                                                     \
   Int32 zz, yy, tmp;                                 \
   zz = z; tmp = heap[zz];                            \
   while (True) {                                     \
      yy = zz << 1;                                   \
      if (yy > nHeap) break;                          \
      if (yy < nHeap &&                               \
          weight[heap[yy+1]] < weight[heap[yy]])      \
         yy++;                                        \
      if (weight[tmp] < weight[heap[yy]]) break;      \
      heap[zz] = heap[yy];                            \
      zz = yy;                                        \
   }                                                  \
   heap[zz] = tmp;                                    \
}

Definition at line 47 of file huffman.c.

#define MYMAX	(		zz2,
			zz3
	)		((zz2) > (zz3) ? (zz2) : (zz3))

Definition at line 30 of file huffman.c.

#define UPHEAP

(

z

)

Value:

{                                                     \
   Int32 zz, tmp;                                     \
   zz = z; tmp = heap[zz];                            \
   while (weight[tmp] < weight[heap[zz >> 1]]) {      \
      heap[zz] = heap[zz >> 1];                       \
      zz >>= 1;                                       \
   }                                                  \
   heap[zz] = tmp;                                    \
}

Definition at line 36 of file huffman.c.

#define WEIGHTOF

(

zz0

)

((zz0) & 0xffffff00)

Definition at line 28 of file huffman.c.

Function Documentation

void BZ2_hbAssignCodes	(	Int32 *	code,
		UChar *	length,
		Int32	minLen,
		Int32	maxLen,
		Int32	alphaSize
	)

Definition at line 155 of file huffman.c.

{
   Int32 n, vec, i;

   vec = 0;
   for (n = minLen; n <= maxLen; n++) {
      for (i = 0; i < alphaSize; i++)
         if (length[i] == n) { code[i] = vec; vec++; };
      vec <<= 1;
   }
}

void BZ2_hbCreateDecodeTables	(	Int32 *	limit,
		Int32 *	base,
		Int32 *	perm,
		UChar *	length,
		Int32	minLen,
		Int32	maxLen,
		Int32	alphaSize
	)

Definition at line 173 of file huffman.c.

{
   Int32 pp, i, j, vec;

   pp = 0;
   for (i = minLen; i <= maxLen; i++)
      for (j = 0; j < alphaSize; j++)
         if (length[j] == i) { perm[pp] = j; pp++; };

   for (i = 0; i < BZ_MAX_CODE_LEN; i++) base[i] = 0;
   for (i = 0; i < alphaSize; i++) base[length[i]+1]++;

   for (i = 1; i < BZ_MAX_CODE_LEN; i++) base[i] += base[i-1];

   for (i = 0; i < BZ_MAX_CODE_LEN; i++) limit[i] = 0;
   vec = 0;

   for (i = minLen; i <= maxLen; i++) {
      vec += (base[i+1] - base[i]);
      limit[i] = vec-1;
      vec <<= 1;
   }
   for (i = minLen + 1; i <= maxLen; i++)
      base[i] = ((limit[i-1] + 1) << 1) - base[i];
}

void BZ2_hbMakeCodeLengths	(	UChar *	len,
		Int32 *	freq,
		Int32	alphaSize,
		Int32	maxLen
	)

Definition at line 66 of file huffman.c.

{
   /*--
      Nodes and heap entries run from 1.  Entry 0
      for both the heap and nodes is a sentinel.
   --*/
   Int32 nNodes, nHeap, n1, n2, i, j, k;
   Bool  tooLong;

   Int32 heap   [ BZ_MAX_ALPHA_SIZE + 2 ];
   Int32 weight [ BZ_MAX_ALPHA_SIZE * 2 ];
   Int32 parent [ BZ_MAX_ALPHA_SIZE * 2 ]; 

   for (i = 0; i < alphaSize; i++)
      weight[i+1] = (freq[i] == 0 ? 1 : freq[i]) << 8;

   while (True) {

      nNodes = alphaSize;
      nHeap = 0;

      heap[0] = 0;
      weight[0] = 0;
      parent[0] = -2;

      for (i = 1; i <= alphaSize; i++) {
         parent[i] = -1;
         nHeap++;
         heap[nHeap] = i;
         UPHEAP(nHeap);
      }

      AssertH( nHeap < (BZ_MAX_ALPHA_SIZE+2), 2001 );
   
      while (nHeap > 1) {
         n1 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1);
         n2 = heap[1]; heap[1] = heap[nHeap]; nHeap--; DOWNHEAP(1);
         nNodes++;
         parent[n1] = parent[n2] = nNodes;
         weight[nNodes] = ADDWEIGHTS(weight[n1], weight[n2]);
         parent[nNodes] = -1;
         nHeap++;
         heap[nHeap] = nNodes;
         UPHEAP(nHeap);
      }

      AssertH( nNodes < (BZ_MAX_ALPHA_SIZE * 2), 2002 );

      tooLong = False;
      for (i = 1; i <= alphaSize; i++) {
         j = 0;
         k = i;
         while (parent[k] >= 0) { k = parent[k]; j++; }
         len[i-1] = j;
         if (j > maxLen) tooLong = True;
      }
      
      if (! tooLong) break;

      /* 17 Oct 04: keep-going condition for the following loop used
         to be 'i < alphaSize', which missed the last element,
         theoretically leading to the possibility of the compressor
         looping.  However, this count-scaling step is only needed if
         one of the generated Huffman code words is longer than
         maxLen, which up to and including version 1.0.2 was 20 bits,
         which is extremely unlikely.  In version 1.0.3 maxLen was
         changed to 17 bits, which has minimal effect on compression
         ratio, but does mean this scaling step is used from time to
         time, enough to verify that it works.

         This means that bzip2-1.0.3 and later will only produce
         Huffman codes with a maximum length of 17 bits.  However, in
         order to preserve backwards compatibility with bitstreams
         produced by versions pre-1.0.3, the decompressor must still
         handle lengths of up to 20. */

      for (i = 1; i <= alphaSize; i++) {
         j = weight[i] >> 8;
         j = 1 + (j / 2);
         weight[i] = j << 8;
      }
   }
}