/*

 * jdhuff.c

 *

 * Copyright (C) 1991-1997, Thomas G. Lane.

 * Modified 2006-2009 by Guido Vollbeding.

 * This file is part of the Independent JPEG Group's software.

 * For conditions of distribution and use, see the accompanying README file.

 *

 * This file contains Huffman entropy decoding routines.

 * Both sequential and progressive modes are supported in this single module.

 *

 * Much of the complexity here has to do with supporting input suspension.

 * If the data source module demands suspension, we want to be able to back

 * up to the start of the current MCU.  To do this, we copy state variables

 * into local working storage, and update them back to the permanent

 * storage only upon successful completion of an MCU.

 */

#define JPEG_INTERNALS

#include "jinclude.h"

#include "jpeglib.h"

/* Derived data constructed for each Huffman table */

#define HUFF_LOOKAHEAD	8	/* # of bits of lookahead */

typedef struct {

  /* Basic tables: (element [0] of each array is unused) */

  INT32 maxcode[18];		/* largest code of length k (-1 if none) */

  /* (maxcode[17] is a sentinel to ensure jpeg_huff_decode terminates) */

  INT32 valoffset[17];		/* huffval[] offset for codes of length k */

  /* valoffset[k] = huffval[] index of 1st symbol of code length k, less

   * the smallest code of length k; so given a code of length k, the

   * corresponding symbol is huffval[code + valoffset[k]]

   */

  /* Link to public Huffman table (needed only in jpeg_huff_decode) */

  JHUFF_TBL *pub;

  /* Lookahead tables: indexed by the next HUFF_LOOKAHEAD bits of

   * the input data stream.  If the next Huffman code is no more

   * than HUFF_LOOKAHEAD bits long, we can obtain its length and

   * the corresponding symbol directly from these tables.

   */

  int look_nbits[1<<HUFF_LOOKAHEAD]; /* # bits, or 0 if too long */

  UINT8 look_sym[1<<HUFF_LOOKAHEAD]; /* symbol, or unused */

} d_derived_tbl;

/*

 * Fetching the next N bits from the input stream is a time-critical operation

 * for the Huffman decoders.  We implement it with a combination of inline

 * macros and out-of-line subroutines.  Note that N (the number of bits

 * demanded at one time) never exceeds 15 for JPEG use.

 *

 * We read source bytes into get_buffer and dole out bits as needed.

 * If get_buffer already contains enough bits, they are fetched in-line

 * by the macros CHECK_BIT_BUFFER and GET_BITS.  When there aren't enough

 * bits, jpeg_fill_bit_buffer is called; it will attempt to fill get_buffer

 * as full as possible (not just to the number of bits needed; this

 * prefetching reduces the overhead cost of calling jpeg_fill_bit_buffer).

 * Note that jpeg_fill_bit_buffer may return FALSE to indicate suspension.

 * On TRUE return, jpeg_fill_bit_buffer guarantees that get_buffer contains

 * at least the requested number of bits --- dummy zeroes are inserted if

 * necessary.

 */

typedef INT32 bit_buf_type;	/* type of bit-extraction buffer */

#define BIT_BUF_SIZE  32	/* size of buffer in bits */

/* If long is > 32 bits on your machine, and shifting/masking longs is

 * reasonably fast, making bit_buf_type be long and setting BIT_BUF_SIZE

 * appropriately should be a win.  Unfortunately we can't define the size

 * with something like  #define BIT_BUF_SIZE (sizeof(bit_buf_type)*8)

 * because not all machines measure sizeof in 8-bit bytes.

 */

typedef struct {		/* Bitreading state saved across MCUs */

  bit_buf_type get_buffer;	/* current bit-extraction buffer */

  int bits_left;		/* # of unused bits in it */

} bitread_perm_state;

typedef struct {		/* Bitreading working state within an MCU */

  /* Current data source location */

  /* We need a copy, rather than munging the original, in case of suspension */

  const JOCTET * next_input_byte; /* => next byte to read from source */

  size_t bytes_in_buffer;	/* # of bytes remaining in source buffer */

  /* Bit input buffer --- note these values are kept in register variables,

   * not in this struct, inside the inner loops.

   */

  bit_buf_type get_buffer;	/* current bit-extraction buffer */

  int bits_left;		/* # of unused bits in it */

  /* Pointer needed by jpeg_fill_bit_buffer. */

  j_decompress_ptr cinfo;	/* back link to decompress master record */

} bitread_working_state;

/* Macros to declare and load/save bitread local variables. */

#define BITREAD_STATE_VARS  \

	register bit_buf_type get_buffer;  \

	register int bits_left;  \

	bitread_working_state br_state

#define BITREAD_LOAD_STATE(cinfop,permstate)  \

	br_state.cinfo = cinfop; \

	br_state.next_input_byte = cinfop->src->next_input_byte; \

	br_state.bytes_in_buffer = cinfop->src->bytes_in_buffer; \

	get_buffer = permstate.get_buffer; \

	bits_left = permstate.bits_left;

#define BITREAD_SAVE_STATE(cinfop,permstate)  \

	cinfop->src->next_input_byte = br_state.next_input_byte; \

	cinfop->src->bytes_in_buffer = br_state.bytes_in_buffer; \

	permstate.get_buffer = get_buffer; \

	permstate.bits_left = bits_left

/*

 * These macros provide the in-line portion of bit fetching.

 * Use CHECK_BIT_BUFFER to ensure there are N bits in get_buffer

 * before using GET_BITS, PEEK_BITS, or DROP_BITS.

 * The variables get_buffer and bits_left are assumed to be locals,

 * but the state struct might not be (jpeg_huff_decode needs this).

 *	CHECK_BIT_BUFFER(state,n,action);

 *		Ensure there are N bits in get_buffer; if suspend, take action.

 *      val = GET_BITS(n);

 *		Fetch next N bits.

 *      val = PEEK_BITS(n);

 *		Fetch next N bits without removing them from the buffer.

 *	DROP_BITS(n);

 *		Discard next N bits.

 * The value N should be a simple variable, not an expression, because it

 * is evaluated multiple times.

 */

#define CHECK_BIT_BUFFER(state,nbits,action) \

	{ if (bits_left < (nbits)) {  \

	    if (! jpeg_fill_bit_buffer(&(state),get_buffer,bits_left,nbits))  \

	      { action; }  \

	    get_buffer = (state).get_buffer; bits_left = (state).bits_left; } }

#define GET_BITS(nbits) \

	(((int) (get_buffer >> (bits_left -= (nbits)))) & BIT_MASK(nbits))

#define PEEK_BITS(nbits) \

	(((int) (get_buffer >> (bits_left -  (nbits)))) & BIT_MASK(nbits))

#define DROP_BITS(nbits) \

	(bits_left -= (nbits))

/*

 * Code for extracting next Huffman-coded symbol from input bit stream.

 * Again, this is time-critical and we make the main paths be macros.

 *

 * We use a lookahead table to process codes of up to HUFF_LOOKAHEAD bits

 * without looping.  Usually, more than 95% of the Huffman codes will be 8

 * or fewer bits long.  The few overlength codes are handled with a loop,

 * which need not be inline code.

 *

 * Notes about the HUFF_DECODE macro:

 * 1. Near the end of the data segment, we may fail to get enough bits

 *    for a lookahead.  In that case, we do it the hard way.

 * 2. If the lookahead table contains no entry, the next code must be

 *    more than HUFF_LOOKAHEAD bits long.

 * 3. jpeg_huff_decode returns -1 if forced to suspend.

 */

#define HUFF_DECODE(result,state,htbl,failaction,slowlabel) \

{ register int nb, look; \

  if (bits_left < HUFF_LOOKAHEAD) { \

    if (! jpeg_fill_bit_buffer(&state,get_buffer,bits_left, 0)) {failaction;} \

    get_buffer = state.get_buffer; bits_left = state.bits_left; \

    if (bits_left < HUFF_LOOKAHEAD) { \

      nb = 1; goto slowlabel; \

    } \

  } \

  look = PEEK_BITS(HUFF_LOOKAHEAD); \

  if ((nb = htbl->look_nbits[look]) != 0) { \

    DROP_BITS(nb); \

    result = htbl->look_sym[look]; \

  } else { \

    nb = HUFF_LOOKAHEAD+1; \

slowlabel: \

    if ((result=jpeg_huff_decode(&state,get_buffer,bits_left,htbl,nb)) < 0) \

	{ failaction; } \

    get_buffer = state.get_buffer; bits_left = state.bits_left; \

  } \

}

/*

 * Expanded entropy decoder object for Huffman decoding.

 *

 * The savable_state subrecord contains fields that change within an MCU,

 * but must not be updated permanently until we complete the MCU.

 */

typedef struct {

  unsigned int EOBRUN;			/* remaining EOBs in EOBRUN */

  int last_dc_val[MAX_COMPS_IN_SCAN];	/* last DC coef for each component */

} savable_state;

/* This macro is to work around compilers with missing or broken

 * structure assignment.  You'll need to fix this code if you have

 * such a compiler and you change MAX_COMPS_IN_SCAN.

 */

#ifndef NO_STRUCT_ASSIGN

#define ASSIGN_STATE(dest,src)  ((dest) = (src))

#else

#if MAX_COMPS_IN_SCAN == 4

#define ASSIGN_STATE(dest,src)  \

	((dest).EOBRUN = (src).EOBRUN, \

	 (dest).last_dc_val[0] = (src).last_dc_val[0], \

	 (dest).last_dc_val[1] = (src).last_dc_val[1], \

	 (dest).last_dc_val[2] = (src).last_dc_val[2], \

	 (dest).last_dc_val[3] = (src).last_dc_val[3])

#endif

#endif

typedef struct {

  struct jpeg_entropy_decoder pub; /* public fields */

  /* These fields are loaded into local variables at start of each MCU.

   * In case of suspension, we exit WITHOUT updating them.

   */

  bitread_perm_state bitstate;	/* Bit buffer at start of MCU */

  savable_state saved;		/* Other state at start of MCU */

  /* These fields are NOT loaded into local working state. */

  boolean insufficient_data;	/* set TRUE after emitting warning */

  unsigned int restarts_to_go;	/* MCUs left in this restart interval */

  /* Following two fields used only in progressive mode */

  /* Pointers to derived tables (these workspaces have image lifespan) */

  d_derived_tbl * derived_tbls[NUM_HUFF_TBLS];

  d_derived_tbl * ac_derived_tbl; /* active table during an AC scan */

  /* Following fields used only in sequential mode */

  /* Pointers to derived tables (these workspaces have image lifespan) */

  d_derived_tbl * dc_derived_tbls[NUM_HUFF_TBLS];

  d_derived_tbl * ac_derived_tbls[NUM_HUFF_TBLS];

  /* Precalculated info set up by start_pass for use in decode_mcu: */

  /* Pointers to derived tables to be used for each block within an MCU */

  d_derived_tbl * dc_cur_tbls[D_MAX_BLOCKS_IN_MCU];

  d_derived_tbl * ac_cur_tbls[D_MAX_BLOCKS_IN_MCU];

  /* Whether we care about the DC and AC coefficient values for each block */

  int coef_limit[D_MAX_BLOCKS_IN_MCU];

} huff_entropy_decoder;

typedef huff_entropy_decoder * huff_entropy_ptr;

static const int jpeg_zigzag_order[8][8] = {

  {  0,  1,  5,  6, 14, 15, 27, 28 },

  {  2,  4,  7, 13, 16, 26, 29, 42 },

  {  3,  8, 12, 17, 25, 30, 41, 43 },

  {  9, 11, 18, 24, 31, 40, 44, 53 },

  { 10, 19, 23, 32, 39, 45, 52, 54 },

  { 20, 22, 33, 38, 46, 51, 55, 60 },

  { 21, 34, 37, 47, 50, 56, 59, 61 },

  { 35, 36, 48, 49, 57, 58, 62, 63 }

};

static const int jpeg_zigzag_order7[7][7] = {

  {  0,  1,  5,  6, 14, 15, 27 },

  {  2,  4,  7, 13, 16, 26, 28 },

  {  3,  8, 12, 17, 25, 29, 38 },

  {  9, 11, 18, 24, 30, 37, 39 },

  { 10, 19, 23, 31, 36, 40, 45 },

  { 20, 22, 32, 35, 41, 44, 46 },

  { 21, 33, 34, 42, 43, 47, 48 }

};

static const int jpeg_zigzag_order6[6][6] = {

  {  0,  1,  5,  6, 14, 15 },

  {  2,  4,  7, 13, 16, 25 },

  {  3,  8, 12, 17, 24, 26 },

  {  9, 11, 18, 23, 27, 32 },

  { 10, 19, 22, 28, 31, 33 },

  { 20, 21, 29, 30, 34, 35 }

};

static const int jpeg_zigzag_order5[5][5] = {

  {  0,  1,  5,  6, 14 },

  {  2,  4,  7, 13, 15 },

  {  3,  8, 12, 16, 21 },

  {  9, 11, 17, 20, 22 },

  { 10, 18, 19, 23, 24 }

};

static const int jpeg_zigzag_order4[4][4] = {

  { 0,  1,  5,  6 },

  { 2,  4,  7, 12 },

  { 3,  8, 11, 13 },

  { 9, 10, 14, 15 }

};

static const int jpeg_zigzag_order3[3][3] = {

  { 0, 1, 5 },

  { 2, 4, 6 },

  { 3, 7, 8 }

};

static const int jpeg_zigzag_order2[2][2] = {

  { 0, 1 },

  { 2, 3 }

};

/*

 * Compute the derived values for a Huffman table.

 * This routine also performs some validation checks on the table.

 */

LOCAL(void)

jpeg_make_d_derived_tbl (j_decompress_ptr cinfo, boolean isDC, int tblno,

			 d_derived_tbl ** pdtbl)

{

  JHUFF_TBL *htbl;

  d_derived_tbl *dtbl;

  int p, i, l, si, numsymbols;

  int lookbits, ctr;

  char huffsize[257];

  unsigned int huffcode[257];

  unsigned int code;

  /* Note that huffsize[] and huffcode[] are filled in code-length order,

   * paralleling the order of the symbols themselves in htbl->huffval[].

   */

  /* Find the input Huffman table */

  if (tblno < 0 || tblno >= NUM_HUFF_TBLS)

    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);

  htbl =

    isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno];

  if (htbl == NULL)

    ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno);

  /* Allocate a workspace if we haven't already done so. */

  if (*pdtbl == NULL)

    *pdtbl = (d_derived_tbl *)

      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,

				  SIZEOF(d_derived_tbl));

  dtbl = *pdtbl;

  dtbl->pub = htbl;		/* fill in back link */

  /* Figure C.1: make table of Huffman code length for each symbol */

  p = 0;

  for (l = 1; l <= 16; l++) {

    i = (int) htbl->bits[l];

    if (i < 0 || p + i > 256)	/* protect against table overrun */

      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);

    while (i--)

      huffsize[p++] = (char) l;

  }

  huffsize[p] = 0;

  numsymbols = p;

  /* Figure C.2: generate the codes themselves */

  /* We also validate that the counts represent a legal Huffman code tree. */

  code = 0;

  si = huffsize[0];

  p = 0;

  while (huffsize[p]) {

    while (((int) huffsize[p]) == si) {

      huffcode[p++] = code;

      code++;

    }

    /* code is now 1 more than the last code used for codelength si; but

     * it must still fit in si bits, since no code is allowed to be all ones.

     */

    if (((INT32) code) >= (((INT32) 1) << si))

      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);

    code <<= 1;

    si++;

  }

  /* Figure F.15: generate decoding tables for bit-sequential decoding */

  p = 0;

  for (l = 1; l <= 16; l++) {

    if (htbl->bits[l]) {

      /* valoffset[l] = huffval[] index of 1st symbol of code length l,

       * minus the minimum code of length l

       */

      dtbl->valoffset[l] = (INT32) p - (INT32) huffcode[p];

      p += htbl->bits[l];

      dtbl->maxcode[l] = huffcode[p-1]; /* maximum code of length l */

    } else {

      dtbl->maxcode[l] = -1;	/* -1 if no codes of this length */

    }

  }

  dtbl->maxcode[17] = 0xFFFFFL; /* ensures jpeg_huff_decode terminates */

  /* Compute lookahead tables to speed up decoding.

   * First we set all the table entries to 0, indicating "too long";

   * then we iterate through the Huffman codes that are short enough and

   * fill in all the entries that correspond to bit sequences starting

   * with that code.

   */

  MEMZERO(dtbl->look_nbits, SIZEOF(dtbl->look_nbits));

  p = 0;

  for (l = 1; l <= HUFF_LOOKAHEAD; l++) {

    for (i = 1; i <= (int) htbl->bits[l]; i++, p++) {

      /* l = current code's length, p = its index in huffcode[] & huffval[]. */

      /* Generate left-justified code followed by all possible bit sequences */

      lookbits = huffcode[p] << (HUFF_LOOKAHEAD-l);

      for (ctr = 1 << (HUFF_LOOKAHEAD-l); ctr > 0; ctr--) {

	dtbl->look_nbits[lookbits] = l;

	dtbl->look_sym[lookbits] = htbl->huffval[p];

	lookbits++;

      }

    }

  }

  /* Validate symbols as being reasonable.

   * For AC tables, we make no check, but accept all byte values 0..255.

   * For DC tables, we require the symbols to be in range 0..15.

   * (Tighter bounds could be applied depending on the data depth and mode,

   * but this is sufficient to ensure safe decoding.)

   */

  if (isDC) {

    for (i = 0; i < numsymbols; i++) {

      int sym = htbl->huffval[i];

      if (sym < 0 || sym > 15)

	ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);

    }

  }

}

/*

 * Out-of-line code for bit fetching.

 * Note: current values of get_buffer and bits_left are passed as parameters,

 * but are returned in the corresponding fields of the state struct.

 *

 * On most machines MIN_GET_BITS should be 25 to allow the full 32-bit width

 * of get_buffer to be used.  (On machines with wider words, an even larger

 * buffer could be used.)  However, on some machines 32-bit shifts are

 * quite slow and take time proportional to the number of places shifted.

 * (This is true with most PC compilers, for instance.)  In this case it may

 * be a win to set MIN_GET_BITS to the minimum value of 15.  This reduces the

 * average shift distance at the cost of more calls to jpeg_fill_bit_buffer.

 */

#ifdef SLOW_SHIFT_32

#define MIN_GET_BITS  15	/* minimum allowable value */

#else

#define MIN_GET_BITS  (BIT_BUF_SIZE-7)

#endif

LOCAL(boolean)

jpeg_fill_bit_buffer (bitread_working_state * state,

		      register bit_buf_type get_buffer, register int bits_left,

		      int nbits)

/* Load up the bit buffer to a depth of at least nbits */

{

  /* Copy heavily used state fields into locals (hopefully registers) */

  register const JOCTET * next_input_byte = state->next_input_byte;

  register size_t bytes_in_buffer = state->bytes_in_buffer;

  j_decompress_ptr cinfo = state->cinfo;

  /* Attempt to load at least MIN_GET_BITS bits into get_buffer. */

  /* (It is assumed that no request will be for more than that many bits.) */

  /* We fail to do so only if we hit a marker or are forced to suspend. */

  if (cinfo->unread_marker == 0) {	/* cannot advance past a marker */

    while (bits_left < MIN_GET_BITS) {

      register int c;

      /* Attempt to read a byte */

      if (bytes_in_buffer == 0) {

	if (! (*cinfo->src->fill_input_buffer) (cinfo))

	  return FALSE;

	next_input_byte = cinfo->src->next_input_byte;

	bytes_in_buffer = cinfo->src->bytes_in_buffer;

      }

      bytes_in_buffer--;

      c = GETJOCTET(*next_input_byte++);

      /* If it's 0xFF, check and discard stuffed zero byte */

      if (c == 0xFF) {

	/* Loop here to discard any padding FF's on terminating marker,

	 * so that we can save a valid unread_marker value.  NOTE: we will

	 * accept multiple FF's followed by a 0 as meaning a single FF data

	 * byte.  This data pattern is not valid according to the standard.

	 */

	do {

	  if (bytes_in_buffer == 0) {

	    if (! (*cinfo->src->fill_input_buffer) (cinfo))

	      return FALSE;

	    next_input_byte = cinfo->src->next_input_byte;

	    bytes_in_buffer = cinfo->src->bytes_in_buffer;

	  }

	  bytes_in_buffer--;

	  c = GETJOCTET(*next_input_byte++);

	} while (c == 0xFF);