entangledstate-isbn 1.4.0 → 1.4.1

data/src/jpeg-7/jddctmgr.c

@@ -0,0 +1,382 @@
+/*
+ * jddctmgr.c
+ *
+ * Copyright (C) 1994-1996, Thomas G. Lane.
+ * 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 the inverse-DCT management logic.
+ * This code selects a particular IDCT implementation to be used,
+ * and it performs related housekeeping chores.  No code in this file
+ * is executed per IDCT step, only during output pass setup.
+ *
+ * Note that the IDCT routines are responsible for performing coefficient
+ * dequantization as well as the IDCT proper.  This module sets up the
+ * dequantization multiplier table needed by the IDCT routine.
+ */
+
+#define JPEG_INTERNALS
+#include "jinclude.h"
+#include "jpeglib.h"
+#include "jdct.h"		/* Private declarations for DCT subsystem */
+
+
+/*
+ * The decompressor input side (jdinput.c) saves away the appropriate
+ * quantization table for each component at the start of the first scan
+ * involving that component.  (This is necessary in order to correctly
+ * decode files that reuse Q-table slots.)
+ * When we are ready to make an output pass, the saved Q-table is converted
+ * to a multiplier table that will actually be used by the IDCT routine.
+ * The multiplier table contents are IDCT-method-dependent.  To support
+ * application changes in IDCT method between scans, we can remake the
+ * multiplier tables if necessary.
+ * In buffered-image mode, the first output pass may occur before any data
+ * has been seen for some components, and thus before their Q-tables have
+ * been saved away.  To handle this case, multiplier tables are preset
+ * to zeroes; the result of the IDCT will be a neutral gray level.
+ */
+
+
+/* Private subobject for this module */
+
+typedef struct {
+  struct jpeg_inverse_dct pub;	/* public fields */
+
+  /* This array contains the IDCT method code that each multiplier table
+   * is currently set up for, or -1 if it's not yet set up.
+   * The actual multiplier tables are pointed to by dct_table in the
+   * per-component comp_info structures.
+   */
+  int cur_method[MAX_COMPONENTS];
+} my_idct_controller;
+
+typedef my_idct_controller * my_idct_ptr;
+
+
+/* Allocated multiplier tables: big enough for any supported variant */
+
+typedef union {
+  ISLOW_MULT_TYPE islow_array[DCTSIZE2];
+#ifdef DCT_IFAST_SUPPORTED
+  IFAST_MULT_TYPE ifast_array[DCTSIZE2];
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+  FLOAT_MULT_TYPE float_array[DCTSIZE2];
+#endif
+} multiplier_table;
+
+
+/* The current scaled-IDCT routines require ISLOW-style multiplier tables,
+ * so be sure to compile that code if either ISLOW or SCALING is requested.
+ */
+#ifdef DCT_ISLOW_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#else
+#ifdef IDCT_SCALING_SUPPORTED
+#define PROVIDE_ISLOW_TABLES
+#endif
+#endif
+
+
+/*
+ * Prepare for an output pass.
+ * Here we select the proper IDCT routine for each component and build
+ * a matching multiplier table.
+ */
+
+METHODDEF(void)
+start_pass (j_decompress_ptr cinfo)
+{
+  my_idct_ptr idct = (my_idct_ptr) cinfo->idct;
+  int ci, i;
+  jpeg_component_info *compptr;
+  int method = 0;
+  inverse_DCT_method_ptr method_ptr = NULL;
+  JQUANT_TBL * qtbl;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Select the proper IDCT routine for this component's scaling */
+    switch ((compptr->DCT_h_scaled_size << 8) + compptr->DCT_v_scaled_size) {
+#ifdef IDCT_SCALING_SUPPORTED
+    case ((1 << 8) + 1):
+      method_ptr = jpeg_idct_1x1;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 2):
+      method_ptr = jpeg_idct_2x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((3 << 8) + 3):
+      method_ptr = jpeg_idct_3x3;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 4):
+      method_ptr = jpeg_idct_4x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((5 << 8) + 5):
+      method_ptr = jpeg_idct_5x5;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 6):
+      method_ptr = jpeg_idct_6x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((7 << 8) + 7):
+      method_ptr = jpeg_idct_7x7;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((9 << 8) + 9):
+      method_ptr = jpeg_idct_9x9;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((10 << 8) + 10):
+      method_ptr = jpeg_idct_10x10;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((11 << 8) + 11):
+      method_ptr = jpeg_idct_11x11;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((12 << 8) + 12):
+      method_ptr = jpeg_idct_12x12;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((13 << 8) + 13):
+      method_ptr = jpeg_idct_13x13;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((14 << 8) + 14):
+      method_ptr = jpeg_idct_14x14;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((15 << 8) + 15):
+      method_ptr = jpeg_idct_15x15;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((16 << 8) + 16):
+      method_ptr = jpeg_idct_16x16;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((16 << 8) + 8):
+      method_ptr = jpeg_idct_16x8;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((14 << 8) + 7):
+      method_ptr = jpeg_idct_14x7;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((12 << 8) + 6):
+      method_ptr = jpeg_idct_12x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((10 << 8) + 5):
+      method_ptr = jpeg_idct_10x5;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((8 << 8) + 4):
+      method_ptr = jpeg_idct_8x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 3):
+      method_ptr = jpeg_idct_6x3;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 2):
+      method_ptr = jpeg_idct_4x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 1):
+      method_ptr = jpeg_idct_2x1;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((8 << 8) + 16):
+      method_ptr = jpeg_idct_8x16;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((7 << 8) + 14):
+      method_ptr = jpeg_idct_7x14;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((6 << 8) + 12):
+      method_ptr = jpeg_idct_6x12;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((5 << 8) + 10):
+      method_ptr = jpeg_idct_5x10;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((4 << 8) + 8):
+      method_ptr = jpeg_idct_4x8;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((3 << 8) + 6):
+      method_ptr = jpeg_idct_3x6;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((2 << 8) + 4):
+      method_ptr = jpeg_idct_2x4;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+    case ((1 << 8) + 2):
+      method_ptr = jpeg_idct_1x2;
+      method = JDCT_ISLOW;	/* jidctint uses islow-style table */
+      break;
+#endif
+    case ((DCTSIZE << 8) + DCTSIZE):
+      switch (cinfo->dct_method) {
+#ifdef DCT_ISLOW_SUPPORTED
+      case JDCT_ISLOW:
+	method_ptr = jpeg_idct_islow;
+	method = JDCT_ISLOW;
+	break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+      case JDCT_IFAST:
+	method_ptr = jpeg_idct_ifast;
+	method = JDCT_IFAST;
+	break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+      case JDCT_FLOAT:
+	method_ptr = jpeg_idct_float;
+	method = JDCT_FLOAT;
+	break;
+#endif
+      default:
+	ERREXIT(cinfo, JERR_NOT_COMPILED);
+	break;
+      }
+      break;
+    default:
+      ERREXIT2(cinfo, JERR_BAD_DCTSIZE,
+	       compptr->DCT_h_scaled_size, compptr->DCT_v_scaled_size);
+      break;
+    }
+    idct->pub.inverse_DCT[ci] = method_ptr;
+    /* Create multiplier table from quant table.
+     * However, we can skip this if the component is uninteresting
+     * or if we already built the table.  Also, if no quant table
+     * has yet been saved for the component, we leave the
+     * multiplier table all-zero; we'll be reading zeroes from the
+     * coefficient controller's buffer anyway.
+     */
+    if (! compptr->component_needed || idct->cur_method[ci] == method)
+      continue;
+    qtbl = compptr->quant_table;
+    if (qtbl == NULL)		/* happens if no data yet for component */
+      continue;
+    idct->cur_method[ci] = method;
+    switch (method) {
+#ifdef PROVIDE_ISLOW_TABLES
+    case JDCT_ISLOW:
+      {
+	/* For LL&M IDCT method, multipliers are equal to raw quantization
+	 * coefficients, but are stored as ints to ensure access efficiency.
+	 */
+	ISLOW_MULT_TYPE * ismtbl = (ISLOW_MULT_TYPE *) compptr->dct_table;
+	for (i = 0; i < DCTSIZE2; i++) {
+	  ismtbl[i] = (ISLOW_MULT_TYPE) qtbl->quantval[i];
+	}
+      }
+      break;
+#endif
+#ifdef DCT_IFAST_SUPPORTED
+    case JDCT_IFAST:
+      {
+	/* For AA&N IDCT method, multipliers are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 * For integer operation, the multiplier table is to be scaled by
+	 * IFAST_SCALE_BITS.
+	 */
+	IFAST_MULT_TYPE * ifmtbl = (IFAST_MULT_TYPE *) compptr->dct_table;
+#define CONST_BITS 14
+	static const INT16 aanscales[DCTSIZE2] = {
+	  /* precomputed values scaled up by 14 bits */
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
+	  21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
+	  19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
+	  16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
+	  12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
+	   8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
+	   4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
+	};
+	SHIFT_TEMPS
+
+	for (i = 0; i < DCTSIZE2; i++) {
+	  ifmtbl[i] = (IFAST_MULT_TYPE)
+	    DESCALE(MULTIPLY16V16((INT32) qtbl->quantval[i],
+				  (INT32) aanscales[i]),
+		    CONST_BITS-IFAST_SCALE_BITS);
+	}
+      }
+      break;
+#endif
+#ifdef DCT_FLOAT_SUPPORTED
+    case JDCT_FLOAT:
+      {
+	/* For float AA&N IDCT method, multipliers are equal to quantization
+	 * coefficients scaled by scalefactor[row]*scalefactor[col], where
+	 *   scalefactor[0] = 1
+	 *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
+	 */
+	FLOAT_MULT_TYPE * fmtbl = (FLOAT_MULT_TYPE *) compptr->dct_table;
+	int row, col;
+	static const double aanscalefactor[DCTSIZE] = {
+	  1.0, 1.387039845, 1.306562965, 1.175875602,
+	  1.0, 0.785694958, 0.541196100, 0.275899379
+	};
+
+	i = 0;
+	for (row = 0; row < DCTSIZE; row++) {
+	  for (col = 0; col < DCTSIZE; col++) {
+	    fmtbl[i] = (FLOAT_MULT_TYPE)
+	      ((double) qtbl->quantval[i] *
+	       aanscalefactor[row] * aanscalefactor[col]);
+	    i++;
+	  }
+	}
+      }
+      break;
+#endif
+    default:
+      ERREXIT(cinfo, JERR_NOT_COMPILED);
+      break;
+    }
+  }
+}
+
+
+/*
+ * Initialize IDCT manager.
+ */
+
+GLOBAL(void)
+jinit_inverse_dct (j_decompress_ptr cinfo)
+{
+  my_idct_ptr idct;
+  int ci;
+  jpeg_component_info *compptr;
+
+  idct = (my_idct_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(my_idct_controller));
+  cinfo->idct = (struct jpeg_inverse_dct *) idct;
+  idct->pub.start_pass = start_pass;
+
+  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
+       ci++, compptr++) {
+    /* Allocate and pre-zero a multiplier table for each component */
+    compptr->dct_table =
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  SIZEOF(multiplier_table));
+    MEMZERO(compptr->dct_table, SIZEOF(multiplier_table));
+    /* Mark multiplier table not yet set up for any method */
+    idct->cur_method[ci] = -1;
+  }
+}

data/src/jpeg-7/jdhuff.c

@@ -0,0 +1,1309 @@
+/*
+ * 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. */
+  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 }
+};
+
+
+/*
+ * 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);
+
+	if (c == 0) {
+	  /* Found FF/00, which represents an FF data byte */
+	  c = 0xFF;
+	} else {
+	  /* Oops, it's actually a marker indicating end of compressed data.
+	   * Save the marker code for later use.
+	   * Fine point: it might appear that we should save the marker into
+	   * bitread working state, not straight into permanent state.  But
+	   * once we have hit a marker, we cannot need to suspend within the
+	   * current MCU, because we will read no more bytes from the data
+	   * source.  So it is OK to update permanent state right away.
+	   */
+	  cinfo->unread_marker = c;
+	  /* See if we need to insert some fake zero bits. */
+	  goto no_more_bytes;
+	}
+      }
+
+      /* OK, load c into get_buffer */
+      get_buffer = (get_buffer << 8) | c;
+      bits_left += 8;
+    } /* end while */
+  } else {
+  no_more_bytes:
+    /* We get here if we've read the marker that terminates the compressed
+     * data segment.  There should be enough bits in the buffer register
+     * to satisfy the request; if so, no problem.
+     */
+    if (nbits > bits_left) {
+      /* Uh-oh.  Report corrupted data to user and stuff zeroes into
+       * the data stream, so that we can produce some kind of image.
+       * We use a nonvolatile flag to ensure that only one warning message
+       * appears per data segment.
+       */
+      if (! cinfo->entropy->insufficient_data) {
+	WARNMS(cinfo, JWRN_HIT_MARKER);
+	cinfo->entropy->insufficient_data = TRUE;
+      }
+      /* Fill the buffer with zero bits */
+      get_buffer <<= MIN_GET_BITS - bits_left;
+      bits_left = MIN_GET_BITS;
+    }
+  }
+
+  /* Unload the local registers */
+  state->next_input_byte = next_input_byte;
+  state->bytes_in_buffer = bytes_in_buffer;
+  state->get_buffer = get_buffer;
+  state->bits_left = bits_left;
+
+  return TRUE;
+}
+
+
+/*
+ * Figure F.12: extend sign bit.
+ * On some machines, a shift and sub will be faster than a table lookup.
+ */
+
+#ifdef AVOID_TABLES
+
+#define BIT_MASK(nbits)   ((1<<(nbits))-1)
+#define HUFF_EXTEND(x,s)  ((x) < (1<<((s)-1)) ? (x) - ((1<<(s))-1) : (x))
+
+#else
+
+#define BIT_MASK(nbits)   bmask[nbits]
+#define HUFF_EXTEND(x,s)  ((x) <= bmask[(s) - 1] ? (x) - bmask[s] : (x))
+
+static const int bmask[16] =	/* bmask[n] is mask for n rightmost bits */
+  { 0, 0x0001, 0x0003, 0x0007, 0x000F, 0x001F, 0x003F, 0x007F, 0x00FF,
+    0x01FF, 0x03FF, 0x07FF, 0x0FFF, 0x1FFF, 0x3FFF, 0x7FFF };
+
+#endif /* AVOID_TABLES */
+
+
+/*
+ * Out-of-line code for Huffman code decoding.
+ */
+
+LOCAL(int)
+jpeg_huff_decode (bitread_working_state * state,
+		  register bit_buf_type get_buffer, register int bits_left,
+		  d_derived_tbl * htbl, int min_bits)
+{
+  register int l = min_bits;
+  register INT32 code;
+
+  /* HUFF_DECODE has determined that the code is at least min_bits */
+  /* bits long, so fetch that many bits in one swoop. */
+
+  CHECK_BIT_BUFFER(*state, l, return -1);
+  code = GET_BITS(l);
+
+  /* Collect the rest of the Huffman code one bit at a time. */
+  /* This is per Figure F.16 in the JPEG spec. */
+
+  while (code > htbl->maxcode[l]) {
+    code <<= 1;
+    CHECK_BIT_BUFFER(*state, 1, return -1);
+    code |= GET_BITS(1);
+    l++;
+  }
+
+  /* Unload the local registers */
+  state->get_buffer = get_buffer;
+  state->bits_left = bits_left;
+
+  /* With garbage input we may reach the sentinel value l = 17. */
+
+  if (l > 16) {
+    WARNMS(state->cinfo, JWRN_HUFF_BAD_CODE);
+    return 0;			/* fake a zero as the safest result */
+  }
+
+  return htbl->pub->huffval[ (int) (code + htbl->valoffset[l]) ];
+}
+
+
+/*
+ * Check for a restart marker & resynchronize decoder.
+ * Returns FALSE if must suspend.
+ */
+
+LOCAL(boolean)
+process_restart (j_decompress_ptr cinfo)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci;
+
+  /* Throw away any unused bits remaining in bit buffer; */
+  /* include any full bytes in next_marker's count of discarded bytes */
+  cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8;
+  entropy->bitstate.bits_left = 0;
+
+  /* Advance past the RSTn marker */
+  if (! (*cinfo->marker->read_restart_marker) (cinfo))
+    return FALSE;
+
+  /* Re-initialize DC predictions to 0 */
+  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
+    entropy->saved.last_dc_val[ci] = 0;
+  /* Re-init EOB run count, too */
+  entropy->saved.EOBRUN = 0;
+
+  /* Reset restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+
+  /* Reset out-of-data flag, unless read_restart_marker left us smack up
+   * against a marker.  In that case we will end up treating the next data
+   * segment as empty, and we can avoid producing bogus output pixels by
+   * leaving the flag set.
+   */
+  if (cinfo->unread_marker == 0)
+    entropy->pub.insufficient_data = FALSE;
+
+  return TRUE;
+}
+
+
+/*
+ * Huffman MCU decoding.
+ * Each of these routines decodes and returns one MCU's worth of
+ * Huffman-compressed coefficients. 
+ * The coefficients are reordered from zigzag order into natural array order,
+ * but are not dequantized.
+ *
+ * The i'th block of the MCU is stored into the block pointed to by
+ * MCU_data[i].  WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER.
+ * (Wholesale zeroing is usually a little faster than retail...)
+ *
+ * We return FALSE if data source requested suspension.  In that case no
+ * changes have been made to permanent state.  (Exception: some output
+ * coefficients may already have been assigned.  This is harmless for
+ * spectral selection, since we'll just re-assign them on the next call.
+ * Successive approximation AC refinement has to be more careful, however.)
+ */
+
+/*
+ * MCU decoding for DC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int Al = cinfo->Al;
+  register int s, r;
+  int blkn, ci;
+  JBLOCKROW block;
+  BITREAD_STATE_VARS;
+  savable_state state;
+  d_derived_tbl * tbl;
+  jpeg_component_info * compptr;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->pub.insufficient_data) {
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(state, entropy->saved);
+
+    /* Outer loop handles each block in the MCU */
+
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      block = MCU_data[blkn];
+      ci = cinfo->MCU_membership[blkn];
+      compptr = cinfo->cur_comp_info[ci];
+      tbl = entropy->derived_tbls[compptr->dc_tbl_no];
+
+      /* Decode a single block's worth of coefficients */
+
+      /* Section F.2.2.1: decode the DC coefficient difference */
+      HUFF_DECODE(s, br_state, tbl, return FALSE, label1);
+      if (s) {
+	CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	r = GET_BITS(s);
+	s = HUFF_EXTEND(r, s);
+      }
+
+      /* Convert DC difference to actual value, update last_dc_val */
+      s += state.last_dc_val[ci];
+      state.last_dc_val[ci] = s;
+      /* Scale and output the coefficient (assumes jpeg_natural_order[0]=0) */
+      (*block)[0] = (JCOEF) (s << Al);
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(entropy->saved, state);
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC initial scan (either spectral selection,
+ * or first pass of successive approximation).
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_first (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int Se = cinfo->Se;
+  int Al = cinfo->Al;
+  register int s, k, r;
+  unsigned int EOBRUN;
+  JBLOCKROW block;
+  BITREAD_STATE_VARS;
+  d_derived_tbl * tbl;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->pub.insufficient_data) {
+
+    /* Load up working state.
+     * We can avoid loading/saving bitread state if in an EOB run.
+     */
+    EOBRUN = entropy->saved.EOBRUN;	/* only part of saved state we need */
+
+    /* There is always only one block per MCU */
+
+    if (EOBRUN > 0)		/* if it's a band of zeroes... */
+      EOBRUN--;			/* ...process it now (we do nothing) */
+    else {
+      BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+      block = MCU_data[0];
+      tbl = entropy->ac_derived_tbl;
+
+      for (k = cinfo->Ss; k <= Se; k++) {
+	HUFF_DECODE(s, br_state, tbl, return FALSE, label2);
+	r = s >> 4;
+	s &= 15;
+	if (s) {
+	  k += r;
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  r = GET_BITS(s);
+	  s = HUFF_EXTEND(r, s);
+	  /* Scale and output coefficient in natural (dezigzagged) order */
+	  (*block)[jpeg_natural_order[k]] = (JCOEF) (s << Al);
+	} else {
+	  if (r == 15) {	/* ZRL */
+	    k += 15;		/* skip 15 zeroes in band */
+	  } else {		/* EOBr, run length is 2^r + appended bits */
+	    EOBRUN = 1 << r;
+	    if (r) {		/* EOBr, r > 0 */
+	      CHECK_BIT_BUFFER(br_state, r, return FALSE);
+	      r = GET_BITS(r);
+	      EOBRUN += r;
+	    }
+	    EOBRUN--;		/* this band is processed at this moment */
+	    break;		/* force end-of-band */
+	  }
+	}
+      }
+
+      BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    }
+
+    /* Completed MCU, so update state */
+    entropy->saved.EOBRUN = EOBRUN;	/* only part of saved state we need */
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for DC successive approximation refinement scan.
+ * Note: we assume such scans can be multi-component, although the spec
+ * is not very clear on the point.
+ */
+
+METHODDEF(boolean)
+decode_mcu_DC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
+  int blkn;
+  JBLOCKROW block;
+  BITREAD_STATE_VARS;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* Not worth the cycles to check insufficient_data here,
+   * since we will not change the data anyway if we read zeroes.
+   */
+
+  /* Load up working state */
+  BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+
+  /* Outer loop handles each block in the MCU */
+
+  for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+    block = MCU_data[blkn];
+
+    /* Encoded data is simply the next bit of the two's-complement DC value */
+    CHECK_BIT_BUFFER(br_state, 1, return FALSE);
+    if (GET_BITS(1))
+      (*block)[0] |= p1;
+    /* Note: since we use |=, repeating the assignment later is safe */
+  }
+
+  /* Completed MCU, so update state */
+  BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * MCU decoding for AC successive approximation refinement scan.
+ */
+
+METHODDEF(boolean)
+decode_mcu_AC_refine (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{   
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int Se = cinfo->Se;
+  int p1 = 1 << cinfo->Al;	/* 1 in the bit position being coded */
+  int m1 = (-1) << cinfo->Al;	/* -1 in the bit position being coded */
+  register int s, k, r;
+  unsigned int EOBRUN;
+  JBLOCKROW block;
+  JCOEFPTR thiscoef;
+  BITREAD_STATE_VARS;
+  d_derived_tbl * tbl;
+  int num_newnz;
+  int newnz_pos[DCTSIZE2];
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, don't modify the MCU.
+   */
+  if (! entropy->pub.insufficient_data) {
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    EOBRUN = entropy->saved.EOBRUN; /* only part of saved state we need */
+
+    /* There is always only one block per MCU */
+    block = MCU_data[0];
+    tbl = entropy->ac_derived_tbl;
+
+    /* If we are forced to suspend, we must undo the assignments to any newly
+     * nonzero coefficients in the block, because otherwise we'd get confused
+     * next time about which coefficients were already nonzero.
+     * But we need not undo addition of bits to already-nonzero coefficients;
+     * instead, we can test the current bit to see if we already did it.
+     */
+    num_newnz = 0;
+
+    /* initialize coefficient loop counter to start of band */
+    k = cinfo->Ss;
+
+    if (EOBRUN == 0) {
+      for (; k <= Se; k++) {
+	HUFF_DECODE(s, br_state, tbl, goto undoit, label3);
+	r = s >> 4;
+	s &= 15;
+	if (s) {
+	  if (s != 1)		/* size of new coef should always be 1 */
+	    WARNMS(cinfo, JWRN_HUFF_BAD_CODE);
+	  CHECK_BIT_BUFFER(br_state, 1, goto undoit);
+	  if (GET_BITS(1))
+	    s = p1;		/* newly nonzero coef is positive */
+	  else
+	    s = m1;		/* newly nonzero coef is negative */
+	} else {
+	  if (r != 15) {
+	    EOBRUN = 1 << r;	/* EOBr, run length is 2^r + appended bits */
+	    if (r) {
+	      CHECK_BIT_BUFFER(br_state, r, goto undoit);
+	      r = GET_BITS(r);
+	      EOBRUN += r;
+	    }
+	    break;		/* rest of block is handled by EOB logic */
+	  }
+	  /* note s = 0 for processing ZRL */
+	}
+	/* Advance over already-nonzero coefs and r still-zero coefs,
+	 * appending correction bits to the nonzeroes.  A correction bit is 1
+	 * if the absolute value of the coefficient must be increased.
+	 */
+	do {
+	  thiscoef = *block + jpeg_natural_order[k];
+	  if (*thiscoef != 0) {
+	    CHECK_BIT_BUFFER(br_state, 1, goto undoit);
+	    if (GET_BITS(1)) {
+	      if ((*thiscoef & p1) == 0) { /* do nothing if already set it */
+		if (*thiscoef >= 0)
+		  *thiscoef += p1;
+		else
+		  *thiscoef += m1;
+	      }
+	    }
+	  } else {
+	    if (--r < 0)
+	      break;		/* reached target zero coefficient */
+	  }
+	  k++;
+	} while (k <= Se);
+	if (s) {
+	  int pos = jpeg_natural_order[k];
+	  /* Output newly nonzero coefficient */
+	  (*block)[pos] = (JCOEF) s;
+	  /* Remember its position in case we have to suspend */
+	  newnz_pos[num_newnz++] = pos;
+	}
+      }
+    }
+
+    if (EOBRUN > 0) {
+      /* Scan any remaining coefficient positions after the end-of-band
+       * (the last newly nonzero coefficient, if any).  Append a correction
+       * bit to each already-nonzero coefficient.  A correction bit is 1
+       * if the absolute value of the coefficient must be increased.
+       */
+      for (; k <= Se; k++) {
+	thiscoef = *block + jpeg_natural_order[k];
+	if (*thiscoef != 0) {
+	  CHECK_BIT_BUFFER(br_state, 1, goto undoit);
+	  if (GET_BITS(1)) {
+	    if ((*thiscoef & p1) == 0) { /* do nothing if already changed it */
+	      if (*thiscoef >= 0)
+		*thiscoef += p1;
+	      else
+		*thiscoef += m1;
+	    }
+	  }
+	}
+      }
+      /* Count one block completed in EOB run */
+      EOBRUN--;
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    entropy->saved.EOBRUN = EOBRUN; /* only part of saved state we need */
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+
+undoit:
+  /* Re-zero any output coefficients that we made newly nonzero */
+  while (num_newnz > 0)
+    (*block)[newnz_pos[--num_newnz]] = 0;
+
+  return FALSE;
+}
+
+
+/*
+ * Decode one MCU's worth of Huffman-compressed coefficients.
+ */
+
+METHODDEF(boolean)
+decode_mcu (j_decompress_ptr cinfo, JBLOCKROW *MCU_data)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int blkn;
+  BITREAD_STATE_VARS;
+  savable_state state;
+
+  /* Process restart marker if needed; may have to suspend */
+  if (cinfo->restart_interval) {
+    if (entropy->restarts_to_go == 0)
+      if (! process_restart(cinfo))
+	return FALSE;
+  }
+
+  /* If we've run out of data, just leave the MCU set to zeroes.
+   * This way, we return uniform gray for the remainder of the segment.
+   */
+  if (! entropy->pub.insufficient_data) {
+
+    /* Load up working state */
+    BITREAD_LOAD_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(state, entropy->saved);
+
+    /* Outer loop handles each block in the MCU */
+
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      JBLOCKROW block = MCU_data[blkn];
+      d_derived_tbl * htbl;
+      register int s, k, r;
+      int coef_limit, ci;
+
+      /* Decode a single block's worth of coefficients */
+
+      /* Section F.2.2.1: decode the DC coefficient difference */
+      htbl = entropy->dc_cur_tbls[blkn];
+      HUFF_DECODE(s, br_state, htbl, return FALSE, label1);
+
+      htbl = entropy->ac_cur_tbls[blkn];
+      k = 1;
+      coef_limit = entropy->coef_limit[blkn];
+      if (coef_limit) {
+	/* Convert DC difference to actual value, update last_dc_val */
+	if (s) {
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  r = GET_BITS(s);
+	  s = HUFF_EXTEND(r, s);
+	}
+	ci = cinfo->MCU_membership[blkn];
+	s += state.last_dc_val[ci];
+	state.last_dc_val[ci] = s;
+	/* Output the DC coefficient */
+	(*block)[0] = (JCOEF) s;
+
+	/* Section F.2.2.2: decode the AC coefficients */
+	/* Since zeroes are skipped, output area must be cleared beforehand */
+	for (; k < coef_limit; k++) {
+	  HUFF_DECODE(s, br_state, htbl, return FALSE, label2);
+
+	  r = s >> 4;
+	  s &= 15;
+
+	  if (s) {
+	    k += r;
+	    CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	    r = GET_BITS(s);
+	    s = HUFF_EXTEND(r, s);
+	    /* Output coefficient in natural (dezigzagged) order.
+	     * Note: the extra entries in jpeg_natural_order[] will save us
+	     * if k >= DCTSIZE2, which could happen if the data is corrupted.
+	     */
+	    (*block)[jpeg_natural_order[k]] = (JCOEF) s;
+	  } else {
+	    if (r != 15)
+	      goto EndOfBlock;
+	    k += 15;
+	  }
+	}
+      } else {
+	if (s) {
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  DROP_BITS(s);
+	}
+      }
+
+      /* Section F.2.2.2: decode the AC coefficients */
+      /* In this path we just discard the values */
+      for (; k < DCTSIZE2; k++) {
+	HUFF_DECODE(s, br_state, htbl, return FALSE, label3);
+
+	r = s >> 4;
+	s &= 15;
+
+	if (s) {
+	  k += r;
+	  CHECK_BIT_BUFFER(br_state, s, return FALSE);
+	  DROP_BITS(s);
+	} else {
+	  if (r != 15)
+	    break;
+	  k += 15;
+	}
+      }
+
+      EndOfBlock: ;
+    }
+
+    /* Completed MCU, so update state */
+    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
+    ASSIGN_STATE(entropy->saved, state);
+  }
+
+  /* Account for restart interval (no-op if not using restarts) */
+  entropy->restarts_to_go--;
+
+  return TRUE;
+}
+
+
+/*
+ * Initialize for a Huffman-compressed scan.
+ */
+
+METHODDEF(void)
+start_pass_huff_decoder (j_decompress_ptr cinfo)
+{
+  huff_entropy_ptr entropy = (huff_entropy_ptr) cinfo->entropy;
+  int ci, blkn, dctbl, actbl, i;
+  jpeg_component_info * compptr;
+
+  if (cinfo->progressive_mode) {
+    /* Validate progressive scan parameters */
+    if (cinfo->Ss == 0) {
+      if (cinfo->Se != 0)
+	goto bad;
+    } else {
+      /* need not check Ss/Se < 0 since they came from unsigned bytes */
+      if (cinfo->Se < cinfo->Ss || cinfo->Se >= DCTSIZE2)
+	goto bad;
+      /* AC scans may have only one component */
+      if (cinfo->comps_in_scan != 1)
+	goto bad;
+    }
+    if (cinfo->Ah != 0) {
+      /* Successive approximation refinement scan: must have Al = Ah-1. */
+      if (cinfo->Ah-1 != cinfo->Al)
+	goto bad;
+    }
+    if (cinfo->Al > 13) {	/* need not check for < 0 */
+      /* Arguably the maximum Al value should be less than 13 for 8-bit precision,
+       * but the spec doesn't say so, and we try to be liberal about what we
+       * accept.  Note: large Al values could result in out-of-range DC
+       * coefficients during early scans, leading to bizarre displays due to
+       * overflows in the IDCT math.  But we won't crash.
+       */
+      bad:
+      ERREXIT4(cinfo, JERR_BAD_PROGRESSION,
+	       cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al);
+    }
+    /* Update progression status, and verify that scan order is legal.
+     * Note that inter-scan inconsistencies are treated as warnings
+     * not fatal errors ... not clear if this is right way to behave.
+     */
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      int coefi, cindex = cinfo->cur_comp_info[ci]->component_index;
+      int *coef_bit_ptr = & cinfo->coef_bits[cindex][0];
+      if (cinfo->Ss && coef_bit_ptr[0] < 0) /* AC without prior DC scan */
+	WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0);
+      for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) {
+	int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi];
+	if (cinfo->Ah != expected)
+	  WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi);
+	coef_bit_ptr[coefi] = cinfo->Al;
+      }
+    }
+
+    /* Select MCU decoding routine */
+    if (cinfo->Ah == 0) {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_first;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_first;
+    } else {
+      if (cinfo->Ss == 0)
+	entropy->pub.decode_mcu = decode_mcu_DC_refine;
+      else
+	entropy->pub.decode_mcu = decode_mcu_AC_refine;
+    }
+
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      /* Make sure requested tables are present, and compute derived tables.
+       * We may build same derived table more than once, but it's not expensive.
+       */
+      if (cinfo->Ss == 0) {
+	if (cinfo->Ah == 0) {	/* DC refinement needs no table */
+	  i = compptr->dc_tbl_no;
+	  jpeg_make_d_derived_tbl(cinfo, TRUE, i,
+				  & entropy->derived_tbls[i]);
+	}
+      } else {
+	i = compptr->ac_tbl_no;
+	jpeg_make_d_derived_tbl(cinfo, FALSE, i,
+				& entropy->derived_tbls[i]);
+	/* remember the single active table */
+	entropy->ac_derived_tbl = entropy->derived_tbls[i];
+      }
+      /* Initialize DC predictions to 0 */
+      entropy->saved.last_dc_val[ci] = 0;
+    }
+
+    /* Initialize private state variables */
+    entropy->saved.EOBRUN = 0;
+  } else {
+    /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG.
+     * This ought to be an error condition, but we make it a warning because
+     * there are some baseline files out there with all zeroes in these bytes.
+     */
+    if (cinfo->Ss != 0 || cinfo->Se != DCTSIZE2-1 ||
+	cinfo->Ah != 0 || cinfo->Al != 0)
+      WARNMS(cinfo, JWRN_NOT_SEQUENTIAL);
+
+    /* Select MCU decoding routine */
+    entropy->pub.decode_mcu = decode_mcu;
+
+    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
+      compptr = cinfo->cur_comp_info[ci];
+      dctbl = compptr->dc_tbl_no;
+      actbl = compptr->ac_tbl_no;
+      /* Compute derived values for Huffman tables */
+      /* We may do this more than once for a table, but it's not expensive */
+      jpeg_make_d_derived_tbl(cinfo, TRUE, dctbl,
+			      & entropy->dc_derived_tbls[dctbl]);
+      jpeg_make_d_derived_tbl(cinfo, FALSE, actbl,
+			      & entropy->ac_derived_tbls[actbl]);
+      /* Initialize DC predictions to 0 */
+      entropy->saved.last_dc_val[ci] = 0;
+    }
+
+    /* Precalculate decoding info for each block in an MCU of this scan */
+    for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) {
+      ci = cinfo->MCU_membership[blkn];
+      compptr = cinfo->cur_comp_info[ci];
+      /* Precalculate which table to use for each block */
+      entropy->dc_cur_tbls[blkn] = entropy->dc_derived_tbls[compptr->dc_tbl_no];
+      entropy->ac_cur_tbls[blkn] = entropy->ac_derived_tbls[compptr->ac_tbl_no];
+      /* Decide whether we really care about the coefficient values */
+      if (compptr->component_needed) {
+	ci = compptr->DCT_v_scaled_size;
+	if (ci <= 0 || ci > 8) ci = 8;
+	i = compptr->DCT_h_scaled_size;
+	if (i <= 0 || i > 8) i = 8;
+	entropy->coef_limit[blkn] = 1 + jpeg_zigzag_order[ci - 1][i - 1];
+      } else {
+	entropy->coef_limit[blkn] = 0;
+      }
+    }
+  }
+
+  /* Initialize bitread state variables */
+  entropy->bitstate.bits_left = 0;
+  entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */
+  entropy->pub.insufficient_data = FALSE;
+
+  /* Initialize restart counter */
+  entropy->restarts_to_go = cinfo->restart_interval;
+}
+
+
+/*
+ * Module initialization routine for Huffman entropy decoding.
+ */
+
+GLOBAL(void)
+jinit_huff_decoder (j_decompress_ptr cinfo)
+{
+  huff_entropy_ptr entropy;
+  int i;
+
+  entropy = (huff_entropy_ptr)
+    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				SIZEOF(huff_entropy_decoder));
+  cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
+  entropy->pub.start_pass = start_pass_huff_decoder;
+
+  if (cinfo->progressive_mode) {
+    /* Create progression status table */
+    int *coef_bit_ptr, ci;
+    cinfo->coef_bits = (int (*)[DCTSIZE2])
+      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
+				  cinfo->num_components*DCTSIZE2*SIZEOF(int));
+    coef_bit_ptr = & cinfo->coef_bits[0][0];
+    for (ci = 0; ci < cinfo->num_components; ci++)
+      for (i = 0; i < DCTSIZE2; i++)
+	*coef_bit_ptr++ = -1;
+
+    /* Mark derived tables unallocated */
+    for (i = 0; i < NUM_HUFF_TBLS; i++) {
+      entropy->derived_tbls[i] = NULL;
+    }
+  } else {
+    /* Mark tables unallocated */
+    for (i = 0; i < NUM_HUFF_TBLS; i++) {
+      entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
+    }
+  }
+}