u-jpg.c

// %sys-jpg.h appears to be a mostly unmodified third-party file.  The defines
// are used to configure it.
//
#define JPEG_INTERNALS
#define NO_GETENV
#include "sys-jpg.h"

// setjmp and longjmp are used as the error reporting hook.  It's not clear
// if this choice was made by Rebol or if the original file used it.
//
#include <setjmp.h>

extern jmp_buf jpeg_state;
extern void jpeg_info(char *buffer, int nbytes, int *w, int *h);
extern void jpeg_load(char *buffer, int nbytes, char *output);


#include "pstdint.h" // for uint32_t


/*
 * jdatasrc.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 decompression data source routines for the case of
 * reading JPEG data from a file (or any stdio stream).  While these routines
 * are sufficient for most applications, some will want to use a different
 * source manager.
 * IMPORTANT: we assume that fread() will correctly transcribe an array of
 * JOCTETs from 8-bit-wide elements on external storage.  If char is wider
 * than 8 bits on your machine, you may need to do some tweaking.
 */

/* this is not a core library module, so it doesn't define JPEG_INTERNALS */
//#include "jinclude.h"
//#include "jpeglib.h"
//#include "jerror.h"

typedef uint32_t uinteger32;

/* Expanded data source object for stdio input */

typedef struct {
  struct jpeg_source_mgr pub;   /* public fields */

  JOCTET * buffer;      /* start of buffer */
  size_t   nbytes;
  boolean start_of_file;    /* have we gotten any data yet? */
} my_source_mgr;

typedef my_source_mgr * my_src_ptr;

/*
 * Initialize source --- called by jpeg_read_header
 * before any data is actually read.
 */

METHODDEF(void)
init_source (j_decompress_ptr cinfo)
{
  my_src_ptr src = (my_src_ptr) cinfo->src;

  /* We reset the empty-input-file flag for each image,
   * but we don't clear the input buffer.
   * This is correct behavior for reading a series of images from one source.
   */
  src->start_of_file = TRUE;
}


/*
 * Fill the input buffer --- called whenever buffer is emptied.
 *
 * In typical applications, this should read fresh data into the buffer
 * (ignoring the current state of next_input_byte & bytes_in_buffer),
 * reset the pointer & count to the start of the buffer, and return TRUE
 * indicating that the buffer has been reloaded.  It is not necessary to
 * fill the buffer entirely, only to obtain at least one more byte.
 *
 * There is no such thing as an EOF return.  If the end of the file has been
 * reached, the routine has a choice of ERREXIT() or inserting fake data into
 * the buffer.  In most cases, generating a warning message and inserting a
 * fake EOI marker is the best course of action --- this will allow the
 * decompressor to output however much of the image is there.  However,
 * the resulting error message is misleading if the real problem is an empty
 * input file, so we handle that case specially.
 *
 * In applications that need to be able to suspend compression due to input
 * not being available yet, a FALSE return indicates that no more data can be
 * obtained right now, but more may be forthcoming later.  In this situation,
 * the decompressor will return to its caller (with an indication of the
 * number of scanlines it has read, if any).  The application should resume
 * decompression after it has loaded more data into the input buffer.  Note
 * that there are substantial restrictions on the use of suspension --- see
 * the documentation.
 *
 * When suspending, the decompressor will back up to a convenient restart point
 * (typically the start of the current MCU). next_input_byte & bytes_in_buffer
 * indicate where the restart point will be if the current call returns FALSE.
 * Data beyond this point must be rescanned after resumption, so move it to
 * the front of the buffer rather than discarding it.
 */

METHODDEF(boolean)
fill_input_buffer (j_decompress_ptr cinfo)
{
  my_src_ptr src = (my_src_ptr) cinfo->src;
  static JOCTET buffer[ 2 ];

  if (src->nbytes <= 0) {
    if (src->start_of_file) /* Treat empty input file as fatal error */
      ERREXIT(cinfo, JERR_INPUT_EMPTY);
    WARNMS(cinfo, JWRN_JPEG_EOF);
    /* Insert a fake EOI marker */
    buffer[0] = (JOCTET) 0xFF;
    buffer[1] = (JOCTET) JPEG_EOI;
    src->pub.next_input_byte = buffer;
    src->pub.bytes_in_buffer = 2;
  }
  else {
      src->pub.next_input_byte = src->buffer;
      src->pub.bytes_in_buffer = src->nbytes;
  }

  src->start_of_file = FALSE;

  return TRUE;
}


/*
 * Skip data --- used to skip over a potentially large amount of
 * uninteresting data (such as an APPn marker).
 *
 * Writers of suspendable-input applications must note that skip_input_data
 * is not granted the right to give a suspension return.  If the skip extends
 * beyond the data currently in the buffer, the buffer can be marked empty so
 * that the next read will cause a fill_input_buffer call that can suspend.
 * Arranging for additional bytes to be discarded before reloading the input
 * buffer is the application writer's problem.
 */

METHODDEF(void)
skip_input_data (j_decompress_ptr cinfo, long num_bytes)
{
  my_src_ptr src = (my_src_ptr) cinfo->src;

  /* Just a dumb implementation for now.  Could use fseek() except
   * it doesn't work on pipes.  Not clear that being smart is worth
   * any trouble anyway --- large skips are infrequent.
   */
  if (num_bytes > 0) {
    src->pub.next_input_byte += (size_t) num_bytes;
    src->pub.bytes_in_buffer -= (size_t) num_bytes;
  }
}


/*
 * An additional method that can be provided by data source modules is the
 * resync_to_restart method for error recovery in the presence of RST markers.
 * For the moment, this source module just uses the default resync method
 * provided by the JPEG library.  That method assumes that no backtracking
 * is possible.
 */


/*
 * Terminate source --- called by jpeg_finish_decompress
 * after all data has been read.  Often a no-op.
 *
 * NB: *not* called by jpeg_abort or jpeg_destroy; surrounding
 * application must deal with any cleanup that should happen even
 * for error exit.
 */

METHODDEF(void)
term_source (j_decompress_ptr cinfo)
{
  /* no work necessary here */
}


/*
 * Prepare for input from a stdio stream.
 * The caller must have already opened the stream, and is responsible
 * for closing it after finishing decompression.
 */

GLOBAL(void)
jpeg_series_src (j_decompress_ptr cinfo, JOCTET *buffer, size_t nbytes )
{
  my_src_ptr src;

  /* The source object and input buffer are made permanent so that a series
   * of JPEG images can be read from the same file by calling jpeg_stdio_src
   * only before the first one.  (If we discarded the buffer at the end of
   * one image, we'd likely lose the start of the next one.)
   * This makes it unsafe to use this manager and a different source
   * manager serially with the same JPEG object.  Caveat programmer.
   */
  if (cinfo->src == NULL) { /* first time for this JPEG object? */
    cinfo->src = (struct jpeg_source_mgr *)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
                  SIZEOF(my_source_mgr));
    src = (my_src_ptr) cinfo->src;
  }

  src = (my_src_ptr) cinfo->src;
  src->pub.init_source = init_source;
  src->pub.fill_input_buffer = fill_input_buffer;
  src->pub.skip_input_data = skip_input_data;
  src->pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */
  src->pub.term_source = term_source;
  src->buffer = buffer;
  src->nbytes = nbytes;
  src->pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */
  src->pub.next_input_byte = NULL; /* until buffer loaded */
}

void jpeg_info( char *buffer, int nbytes, int *w, int *h )
{
  struct jpeg_decompress_struct cinfo;
  struct jpeg_error_mgr jerr;

  /* Initialize the JPEG decompression object with default error handling. */
  cinfo.err = jpeg_std_error(&jerr);
  jpeg_create_decompress(&cinfo);

  /* Specify data source for decompression */
  jpeg_series_src(&cinfo, (unsigned char *)buffer, nbytes);

  /* Read file header, set default decompression parameters */
  (void) jpeg_read_header(&cinfo, TRUE);
  *w = cinfo.image_width;
  *h = cinfo.image_height;

  jpeg_destroy_decompress(&cinfo);
}

void jpeg_load( char *buffer, int nbytes, char *output )
{
  struct jpeg_decompress_struct cinfo;
  struct jpeg_error_mgr jerr;
  JSAMPROW  array[ 4 ];
  unsigned int  i, j;

  /* Initialize the JPEG decompression object with default error handling. */
  cinfo.err = jpeg_std_error(&jerr);
  jpeg_create_decompress(&cinfo);

  /* Specify data source for decompression */
  jpeg_series_src(&cinfo, (unsigned char *)buffer, nbytes);

  /* Read file header, set default decompression parameters */
  (void) jpeg_read_header(&cinfo, TRUE);

  /* Start decompressor */
  (void) jpeg_start_decompress(&cinfo);

  /* Process data */
  while (cinfo.output_scanline < cinfo.output_height) {
    array[ 0 ] = (JSAMPROW)(output + cinfo.output_scanline * cinfo.image_width * 4);
    array[ 1 ] = array[ 0 ] + cinfo.image_width * 4;
    array[ 2 ] = array[ 1 ] + cinfo.image_width * 4;
    array[ 3 ] = array[ 2 ] + cinfo.image_width * 4;
    jpeg_read_scanlines(&cinfo, array, 4 );
  }

  if (cinfo.out_color_space != JCS_GRAYSCALE)
  // convert 3 byte values into four byte ones
  for ( i=0; i<cinfo.image_height; i++ ) {
    unsigned char   *cp;
    unsigned char   *dp;

    cp = (unsigned char *)(output + cinfo.image_width * 3);
    dp = (unsigned char *)(output + cinfo.image_width * 4);
    output = ( char * )dp;
    for ( j=0; j<cinfo.image_width; j++ ) {
        cp -= 3;
        *--dp = 0xff; // opaque alpha (going in reverse rgba order...)
        *--dp = cp[2]; // blue
        *--dp = cp[1]; // green
        *--dp = cp[0]; // red
    }
  }
  else
  // convert 1 byte value into four byte ones
    for ( i=0; i<cinfo.image_height; i++ ) {
      unsigned char *cp;
      unsigned char *dp;
      unsigned char c;

      cp = (unsigned char *)(output + cinfo.image_width);
      dp = (unsigned char *)(output + cinfo.image_width * 4);
      output = ( char * )dp;
      for ( j=0; j<cinfo.image_width; j++ ) {
        c = *--cp;
        *--dp = 0xff; // opaque alpha (going in reverse rgba order...)
        *--dp = c; // blue
        *--dp = c; // green
        *--dp = c; // red
      }
    }

  /* Finish decompression and release memory.
   * I must do it in this order because output module has allocated memory
   * of lifespan JPOOL_IMAGE; it needs to finish before releasing memory.
   */
  (void) jpeg_finish_decompress(&cinfo);
  jpeg_destroy_decompress(&cinfo);
}

/*
 * jdapimin.c
 *
 * Copyright (C) 1994-1998, 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 application interface code for the decompression half
 * of the JPEG library.  These are the "minimum" API routines that may be
 * needed in either the normal full-decompression case or the
 * transcoding-only case.
 *
 * Most of the routines intended to be called directly by an application
 * are in this file or in jdapistd.c.  But also see jcomapi.c for routines
 * shared by compression and decompression, and jdtrans.c for the transcoding
 * case.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/*
 * Initialization of a JPEG decompression object.
 * The error manager must already be set up (in case memory manager fails).
 */

GLOBAL(void)
jpeg_CreateDecompress (j_decompress_ptr cinfo, int version, size_t structsize)
{
  int i;

  /* Guard against version mismatches between library and caller. */
  cinfo->mem = NULL;        /* so jpeg_destroy knows mem mgr not called */
  if (version != JPEG_LIB_VERSION)
    ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version);
  if (structsize != SIZEOF(struct jpeg_decompress_struct))
    ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE,
         (int) SIZEOF(struct jpeg_decompress_struct), (int) structsize);

  /* For debugging purposes, we zero the whole master structure.
   * But the application has already set the err pointer, and may have set
   * client_data, so we have to save and restore those fields.
   * Note: if application hasn't set client_data, tools like Purify may
   * complain here.
   */
  {
    struct jpeg_error_mgr * err = cinfo->err;
    void * client_data = cinfo->client_data; /* ignore Purify complaint here */
    MEMZERO(cinfo, SIZEOF(struct jpeg_decompress_struct));
    cinfo->err = err;
    cinfo->client_data = client_data;
  }
  cinfo->is_decompressor = TRUE;

  /* Initialize a memory manager instance for this object */
  jinit_memory_mgr((j_common_ptr) cinfo);

  /* Zero out pointers to permanent structures. */
  cinfo->progress = NULL;
  cinfo->src = NULL;

  for (i = 0; i < NUM_QUANT_TBLS; i++)
    cinfo->quant_tbl_ptrs[i] = NULL;

  for (i = 0; i < NUM_HUFF_TBLS; i++) {
    cinfo->dc_huff_tbl_ptrs[i] = NULL;
    cinfo->ac_huff_tbl_ptrs[i] = NULL;
  }

  /* Initialize marker processor so application can override methods
   * for COM, APPn markers before calling jpeg_read_header.
   */
  cinfo->marker_list = NULL;
  jinit_marker_reader(cinfo);

  /* And initialize the overall input controller. */
  jinit_input_controller(cinfo);

  /* OK, I'm ready */
  cinfo->global_state = DSTATE_START;
}


/*
 * Destruction of a JPEG decompression object
 */

GLOBAL(void)
jpeg_destroy_decompress (j_decompress_ptr cinfo)
{
  jpeg_destroy((j_common_ptr) cinfo); /* use common routine */
}


/*
 * Abort processing of a JPEG decompression operation,
 * but don't destroy the object itself.
 */

GLOBAL(void)
jpeg_abort_decompress (j_decompress_ptr cinfo)
{
  jpeg_abort((j_common_ptr) cinfo); /* use common routine */
}


/*
 * Set default decompression parameters.
 */

LOCAL(void)
default_decompress_parms (j_decompress_ptr cinfo)
{
  /* Guess the input colorspace, and set output colorspace accordingly. */
  /* (Wish JPEG committee had provided a real way to specify this...) */
  /* Note application may override our guesses. */
  switch (cinfo->num_components) {
  case 1:
    cinfo->jpeg_color_space = JCS_GRAYSCALE;
    cinfo->out_color_space = JCS_GRAYSCALE;
    break;

  case 3:
    if (cinfo->saw_JFIF_marker) {
      cinfo->jpeg_color_space = JCS_YCbCr; /* JFIF implies YCbCr */
    } else if (cinfo->saw_Adobe_marker) {
      switch (cinfo->Adobe_transform) {
      case 0:
    cinfo->jpeg_color_space = JCS_RGB;
    break;
      case 1:
    cinfo->jpeg_color_space = JCS_YCbCr;
    break;
      default:
    WARNMS1(cinfo, JWRN_ADOBE_XFORM, cinfo->Adobe_transform);
    cinfo->jpeg_color_space = JCS_YCbCr; /* assume it's YCbCr */
    break;
      }
    } else {
      /* Saw no special markers, try to guess from the component IDs */
      int cid0 = cinfo->comp_info[0].component_id;
      int cid1 = cinfo->comp_info[1].component_id;
      int cid2 = cinfo->comp_info[2].component_id;

      if (cid0 == 1 && cid1 == 2 && cid2 == 3)
    cinfo->jpeg_color_space = JCS_YCbCr; /* assume JFIF w/out marker */
      else if (cid0 == 82 && cid1 == 71 && cid2 == 66)
    cinfo->jpeg_color_space = JCS_RGB; /* ASCII 'R', 'G', 'B' */
      else {
    TRACEMS3(cinfo, 1, JTRC_UNKNOWN_IDS, cid0, cid1, cid2);
    cinfo->jpeg_color_space = JCS_YCbCr; /* assume it's YCbCr */
      }
    }
    /* Always guess RGB is proper output colorspace. */
    cinfo->out_color_space = JCS_RGB;
    break;

  case 4:
    if (cinfo->saw_Adobe_marker) {
      switch (cinfo->Adobe_transform) {
      case 0:
    cinfo->jpeg_color_space = JCS_CMYK;
    break;
      case 2:
    cinfo->jpeg_color_space = JCS_YCCK;
    break;
      default:
    WARNMS1(cinfo, JWRN_ADOBE_XFORM, cinfo->Adobe_transform);
    cinfo->jpeg_color_space = JCS_YCCK; /* assume it's YCCK */
    break;
      }
    } else {
      /* No special markers, assume straight CMYK. */
      cinfo->jpeg_color_space = JCS_CMYK;
    }
    cinfo->out_color_space = JCS_CMYK;
    break;

  default:
    cinfo->jpeg_color_space = JCS_UNKNOWN;
    cinfo->out_color_space = JCS_UNKNOWN;
    break;
  }

  /* Set defaults for other decompression parameters. */
  cinfo->scale_num = 1;     /* 1:1 scaling */
  cinfo->scale_denom = 1;
  cinfo->output_gamma = 1.0;
  cinfo->buffered_image = FALSE;
  cinfo->raw_data_out = FALSE;
  cinfo->dct_method = JDCT_DEFAULT;
  cinfo->do_fancy_upsampling = TRUE;
  cinfo->do_block_smoothing = TRUE;
  cinfo->quantize_colors = FALSE;
  /* We set these in case application only sets quantize_colors. */
  cinfo->dither_mode = JDITHER_FS;
#ifdef QUANT_2PASS_SUPPORTED
  cinfo->two_pass_quantize = TRUE;
#else
  cinfo->two_pass_quantize = FALSE;
#endif
  cinfo->desired_number_of_colors = 256;
  cinfo->colormap = NULL;
  /* Initialize for no mode change in buffered-image mode. */
  cinfo->enable_1pass_quant = FALSE;
  cinfo->enable_external_quant = FALSE;
  cinfo->enable_2pass_quant = FALSE;
}


/*
 * Decompression startup: read start of JPEG datastream to see what's there.
 * Need only initialize JPEG object and supply a data source before calling.
 *
 * This routine will read as far as the first SOS marker (ie, actual start of
 * compressed data), and will save all tables and parameters in the JPEG
 * object.  It will also initialize the decompression parameters to default
 * values, and finally return JPEG_HEADER_OK.  On return, the application may
 * adjust the decompression parameters and then call jpeg_start_decompress.
 * (Or, if the application only wanted to determine the image parameters,
 * the data need not be decompressed.  In that case, call jpeg_abort or
 * jpeg_destroy to release any temporary space.)
 * If an abbreviated (tables only) datastream is presented, the routine will
 * return JPEG_HEADER_TABLES_ONLY upon reaching EOI.  The application may then
 * re-use the JPEG object to read the abbreviated image datastream(s).
 * It is unnecessary (but OK) to call jpeg_abort in this case.
 * The JPEG_SUSPENDED return code only occurs if the data source module
 * requests suspension of the decompressor.  In this case the application
 * should load more source data and then re-call jpeg_read_header to resume
 * processing.
 * If a non-suspending data source is used and require_image is TRUE, then the
 * return code need not be inspected since only JPEG_HEADER_OK is possible.
 *
 * This routine is now just a front end to jpeg_consume_input, with some
 * extra error checking.
 */

GLOBAL(int)
jpeg_read_header (j_decompress_ptr cinfo, boolean require_image)
{
  int retcode;

  if (cinfo->global_state != DSTATE_START &&
      cinfo->global_state != DSTATE_INHEADER)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);

  retcode = jpeg_consume_input(cinfo);

  switch (retcode) {
  case JPEG_REACHED_SOS:
    retcode = JPEG_HEADER_OK;
    break;
  case JPEG_REACHED_EOI:
    if (require_image)      /* Complain if application wanted an image */
      ERREXIT(cinfo, JERR_NO_IMAGE);
    /* Reset to start state; it would be safer to require the application to
     * call jpeg_abort, but we can't change it now for compatibility reasons.
     * A side effect is to free any temporary memory (there shouldn't be any).
     */
    jpeg_abort((j_common_ptr) cinfo); /* sets state = DSTATE_START */
    retcode = JPEG_HEADER_TABLES_ONLY;
    break;
  case JPEG_SUSPENDED:
    /* no work */
    break;
  }

  return retcode;
}


/*
 * Consume data in advance of what the decompressor requires.
 * This can be called at any time once the decompressor object has
 * been created and a data source has been set up.
 *
 * This routine is essentially a state machine that handles a couple
 * of critical state-transition actions, namely initial setup and
 * transition from header scanning to ready-for-start_decompress.
 * All the actual input is done via the input controller's consume_input
 * method.
 */

GLOBAL(int)
jpeg_consume_input (j_decompress_ptr cinfo)
{
  int retcode = JPEG_SUSPENDED;

  /* NB: every possible DSTATE value should be listed in this switch */
  switch (cinfo->global_state) {
  case DSTATE_START:
    /* Start-of-datastream actions: reset appropriate modules */
    (*cinfo->inputctl->reset_input_controller) (cinfo);
    /* Initialize application's data source module */
    (*cinfo->src->init_source) (cinfo);
    cinfo->global_state = DSTATE_INHEADER;
    /*FALLTHROUGH*/
  case DSTATE_INHEADER:
    retcode = (*cinfo->inputctl->consume_input) (cinfo);
    if (retcode == JPEG_REACHED_SOS) { /* Found SOS, prepare to decompress */
      /* Set up default parameters based on header data */
      default_decompress_parms(cinfo);
      /* Set global state: ready for start_decompress */
      cinfo->global_state = DSTATE_READY;
    }
    break;
  case DSTATE_READY:
    /* Can't advance past first SOS until start_decompress is called */
    retcode = JPEG_REACHED_SOS;
    break;
  case DSTATE_PRELOAD:
  case DSTATE_PRESCAN:
  case DSTATE_SCANNING:
  case DSTATE_RAW_OK:
  case DSTATE_BUFIMAGE:
  case DSTATE_BUFPOST:
  case DSTATE_STOPPING:
    retcode = (*cinfo->inputctl->consume_input) (cinfo);
    break;
  default:
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  }
  return retcode;
}


/*
 * Have we finished reading the input file?
 */

GLOBAL(boolean)
jpeg_input_complete (j_decompress_ptr cinfo)
{
  /* Check for valid jpeg object */
  if (cinfo->global_state < DSTATE_START ||
      cinfo->global_state > DSTATE_STOPPING)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  return cinfo->inputctl->eoi_reached;
}


/*
 * Is there more than one scan?
 */

GLOBAL(boolean)
jpeg_has_multiple_scans (j_decompress_ptr cinfo)
{
  /* Only valid after jpeg_read_header completes */
  if (cinfo->global_state < DSTATE_READY ||
      cinfo->global_state > DSTATE_STOPPING)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  return cinfo->inputctl->has_multiple_scans;
}


/*
 * Finish JPEG decompression.
 *
 * This will normally just verify the file trailer and release temp storage.
 *
 * Returns FALSE if suspended.  The return value need be inspected only if
 * a suspending data source is used.
 */

GLOBAL(boolean)
jpeg_finish_decompress (j_decompress_ptr cinfo)
{
  if ((cinfo->global_state == DSTATE_SCANNING ||
       cinfo->global_state == DSTATE_RAW_OK) && ! cinfo->buffered_image) {
    /* Terminate final pass of non-buffered mode */
    if (cinfo->output_scanline < cinfo->output_height)
      ERREXIT(cinfo, JERR_TOO_LITTLE_DATA);
    (*cinfo->master->finish_output_pass) (cinfo);
    cinfo->global_state = DSTATE_STOPPING;
  } else if (cinfo->global_state == DSTATE_BUFIMAGE) {
    /* Finishing after a buffered-image operation */
    cinfo->global_state = DSTATE_STOPPING;
  } else if (cinfo->global_state != DSTATE_STOPPING) {
    /* STOPPING = repeat call after a suspension, anything else is error */
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  }
  /* Read until EOI */
  while (! cinfo->inputctl->eoi_reached) {
    if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED)
      return FALSE;     /* Suspend, come back later */
  }
  /* Do final cleanup */
  (*cinfo->src->term_source) (cinfo);
  /* We can use jpeg_abort to release memory and reset global_state */
  jpeg_abort((j_common_ptr) cinfo);
  return TRUE;
}
/*
 * jdapistd.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 application interface code for the decompression half
 * of the JPEG library.  These are the "standard" API routines that are
 * used in the normal full-decompression case.  They are not used by a
 * transcoding-only application.  Note that if an application links in
 * jpeg_start_decompress, it will end up linking in the entire decompressor.
 * We thus must separate this file from jdapimin.c to avoid linking the
 * whole decompression library into a transcoder.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/* Forward declarations */
LOCAL(boolean) output_pass_setup JPP((j_decompress_ptr cinfo));


/*
 * Decompression initialization.
 * jpeg_read_header must be completed before calling this.
 *
 * If a multipass operating mode was selected, this will do all but the
 * last pass, and thus may take a great deal of time.
 *
 * Returns FALSE if suspended.  The return value need be inspected only if
 * a suspending data source is used.
 */

GLOBAL(boolean)
jpeg_start_decompress (j_decompress_ptr cinfo)
{
  if (cinfo->global_state == DSTATE_READY) {
    /* First call: initialize master control, select active modules */
    jinit_master_decompress(cinfo);
    if (cinfo->buffered_image) {
      /* No more work here; expecting jpeg_start_output next */
      cinfo->global_state = DSTATE_BUFIMAGE;
      return TRUE;
    }
    cinfo->global_state = DSTATE_PRELOAD;
  }
  if (cinfo->global_state == DSTATE_PRELOAD) {
    /* If file has multiple scans, absorb them all into the coef buffer */
    if (cinfo->inputctl->has_multiple_scans) {
#ifdef D_MULTISCAN_FILES_SUPPORTED
      for (;;) {
    int retcode;
    /* Call progress monitor hook if present */
    if (cinfo->progress != NULL)
      (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
    /* Absorb some more input */
    retcode = (*cinfo->inputctl->consume_input) (cinfo);
    if (retcode == JPEG_SUSPENDED)
      return FALSE;
    if (retcode == JPEG_REACHED_EOI)
      break;
    /* Advance progress counter if appropriate */
    if (cinfo->progress != NULL &&
        (retcode == JPEG_ROW_COMPLETED || retcode == JPEG_REACHED_SOS)) {
      if (++cinfo->progress->pass_counter >= cinfo->progress->pass_limit) {
        /* jdmaster underestimated number of scans; ratchet up one scan */
        cinfo->progress->pass_limit += (long) cinfo->total_iMCU_rows;
      }
    }
      }
#else
      ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif /* D_MULTISCAN_FILES_SUPPORTED */
    }
    cinfo->output_scan_number = cinfo->input_scan_number;
  } else if (cinfo->global_state != DSTATE_PRESCAN)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  /* Perform any dummy output passes, and set up for the final pass */
  return output_pass_setup(cinfo);
}


/*
 * Set up for an output pass, and perform any dummy pass(es) needed.
 * Common subroutine for jpeg_start_decompress and jpeg_start_output.
 * Entry: global_state = DSTATE_PRESCAN only if previously suspended.
 * Exit: If done, returns TRUE and sets global_state for proper output mode.
 *       If suspended, returns FALSE and sets global_state = DSTATE_PRESCAN.
 */

LOCAL(boolean)
output_pass_setup (j_decompress_ptr cinfo)
{
  if (cinfo->global_state != DSTATE_PRESCAN) {
    /* First call: do pass setup */
    (*cinfo->master->prepare_for_output_pass) (cinfo);
    cinfo->output_scanline = 0;
    cinfo->global_state = DSTATE_PRESCAN;
  }
  /* Loop over any required dummy passes */
  while (cinfo->master->is_dummy_pass) {
#ifdef QUANT_2PASS_SUPPORTED
    /* Crank through the dummy pass */
    while (cinfo->output_scanline < cinfo->output_height) {
      JDIMENSION last_scanline;
      /* Call progress monitor hook if present */
      if (cinfo->progress != NULL) {
    cinfo->progress->pass_counter = (long) cinfo->output_scanline;
    cinfo->progress->pass_limit = (long) cinfo->output_height;
    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
      }
      /* Process some data */
      last_scanline = cinfo->output_scanline;
      (*cinfo->main_ptr->process_data) (cinfo, (JSAMPARRAY) NULL,
                    &cinfo->output_scanline, (JDIMENSION) 0);
      if (cinfo->output_scanline == last_scanline)
    return FALSE;       /* No progress made, must suspend */
    }
    /* Finish up dummy pass, and set up for another one */
    (*cinfo->master->finish_output_pass) (cinfo);
    (*cinfo->master->prepare_for_output_pass) (cinfo);
    cinfo->output_scanline = 0;
#else
    ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif /* QUANT_2PASS_SUPPORTED */
  }
  /* Ready for application to drive output pass through
   * jpeg_read_scanlines or jpeg_read_raw_data.
   */
  cinfo->global_state = cinfo->raw_data_out ? DSTATE_RAW_OK : DSTATE_SCANNING;
  return TRUE;
}


/*
 * Read some scanlines of data from the JPEG decompressor.
 *
 * The return value will be the number of lines actually read.
 * This may be less than the number requested in several cases,
 * including bottom of image, data source suspension, and operating
 * modes that emit multiple scanlines at a time.
 *
 * Note: we warn about excess calls to jpeg_read_scanlines() since
 * this likely signals an application programmer error.  However,
 * an oversize buffer (max_lines > scanlines remaining) is not an error.
 */

GLOBAL(JDIMENSION)
jpeg_read_scanlines (j_decompress_ptr cinfo, JSAMPARRAY scanlines,
             JDIMENSION max_lines)
{
  JDIMENSION row_ctr;

  if (cinfo->global_state != DSTATE_SCANNING)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  if (cinfo->output_scanline >= cinfo->output_height) {
    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
    return 0;
  }

  /* Call progress monitor hook if present */
  if (cinfo->progress != NULL) {
    cinfo->progress->pass_counter = (long) cinfo->output_scanline;
    cinfo->progress->pass_limit = (long) cinfo->output_height;
    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
  }

  /* Process some data */
  row_ctr = 0;
  (*cinfo->main_ptr->process_data) (cinfo, scanlines, &row_ctr, max_lines);
  cinfo->output_scanline += row_ctr;
  return row_ctr;
}


/*
 * Alternate entry point to read raw data.
 * Processes exactly one iMCU row per call, unless suspended.
 */

GLOBAL(JDIMENSION)
jpeg_read_raw_data (j_decompress_ptr cinfo, JSAMPIMAGE data,
            JDIMENSION max_lines)
{
  JDIMENSION lines_per_iMCU_row;

  if (cinfo->global_state != DSTATE_RAW_OK)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  if (cinfo->output_scanline >= cinfo->output_height) {
    WARNMS(cinfo, JWRN_TOO_MUCH_DATA);
    return 0;
  }

  /* Call progress monitor hook if present */
  if (cinfo->progress != NULL) {
    cinfo->progress->pass_counter = (long) cinfo->output_scanline;
    cinfo->progress->pass_limit = (long) cinfo->output_height;
    (*cinfo->progress->progress_monitor) ((j_common_ptr) cinfo);
  }

  /* Verify that at least one iMCU row can be returned. */
  lines_per_iMCU_row = cinfo->max_v_samp_factor * cinfo->min_DCT_scaled_size;
  if (max_lines < lines_per_iMCU_row)
    ERREXIT(cinfo, JERR_BUFFER_SIZE);

  /* Decompress directly into user's buffer. */
  if (! (*cinfo->coef->decompress_data) (cinfo, data))
    return 0;           /* suspension forced, can do nothing more */

  /* OK, we processed one iMCU row. */
  cinfo->output_scanline += lines_per_iMCU_row;
  return lines_per_iMCU_row;
}


/* Additional entry points for buffered-image mode. */

#ifdef D_MULTISCAN_FILES_SUPPORTED

/*
 * Initialize for an output pass in buffered-image mode.
 */

GLOBAL(boolean)
jpeg_start_output (j_decompress_ptr cinfo, int scan_number)
{
  if (cinfo->global_state != DSTATE_BUFIMAGE &&
      cinfo->global_state != DSTATE_PRESCAN)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  /* Limit scan number to valid range */
  if (scan_number <= 0)
    scan_number = 1;
  if (cinfo->inputctl->eoi_reached &&
      scan_number > cinfo->input_scan_number)
    scan_number = cinfo->input_scan_number;
  cinfo->output_scan_number = scan_number;
  /* Perform any dummy output passes, and set up for the real pass */
  return output_pass_setup(cinfo);
}


/*
 * Finish up after an output pass in buffered-image mode.
 *
 * Returns FALSE if suspended.  The return value need be inspected only if
 * a suspending data source is used.
 */

GLOBAL(boolean)
jpeg_finish_output (j_decompress_ptr cinfo)
{
  if ((cinfo->global_state == DSTATE_SCANNING ||
       cinfo->global_state == DSTATE_RAW_OK) && cinfo->buffered_image) {
    /* Terminate this pass. */
    /* We do not require the whole pass to have been completed. */
    (*cinfo->master->finish_output_pass) (cinfo);
    cinfo->global_state = DSTATE_BUFPOST;
  } else if (cinfo->global_state != DSTATE_BUFPOST) {
    /* BUFPOST = repeat call after a suspension, anything else is error */
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);
  }
  /* Read markers looking for SOS or EOI */
  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
     ! cinfo->inputctl->eoi_reached) {
    if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED)
      return FALSE;     /* Suspend, come back later */
  }
  cinfo->global_state = DSTATE_BUFIMAGE;
  return TRUE;
}

#endif /* D_MULTISCAN_FILES_SUPPORTED */
/*
 * jdmaster.c
 *
 * Copyright (C) 1991-1997, 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 master control logic for the JPEG decompressor.
 * These routines are concerned with selecting the modules to be executed
 * and with determining the number of passes and the work to be done in each
 * pass.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/* Private state */

typedef struct {
  struct jpeg_decomp_master pub; /* public fields */

  int pass_number;      /* # of passes completed */

  boolean using_merged_upsample; /* TRUE if using merged upsample/cconvert */

  /* Saved references to initialized quantizer modules,
   * in case we need to switch modes.
   */
  struct jpeg_color_quantizer * quantizer_1pass;
  struct jpeg_color_quantizer * quantizer_2pass;
} my_decomp_master;

typedef my_decomp_master * my_master_ptr;


/*
 * Determine whether merged upsample/color conversion should be used.
 * CRUCIAL: this must match the actual capabilities of jdmerge.c!
 */

LOCAL(boolean)
use_merged_upsample (j_decompress_ptr cinfo)
{
#ifdef UPSAMPLE_MERGING_SUPPORTED
  /* Merging is the equivalent of plain box-filter upsampling */
  if (cinfo->do_fancy_upsampling || cinfo->CCIR601_sampling)
    return FALSE;
  /* jdmerge.c only supports YCC=>RGB color conversion */
  if (cinfo->jpeg_color_space != JCS_YCbCr || cinfo->num_components != 3 ||
      cinfo->out_color_space != JCS_RGB ||
      cinfo->out_color_components != RGB_PIXELSIZE)
    return FALSE;
  /* and it only handles 2h1v or 2h2v sampling ratios */
  if (cinfo->comp_info[0].h_samp_factor != 2 ||
      cinfo->comp_info[1].h_samp_factor != 1 ||
      cinfo->comp_info[2].h_samp_factor != 1 ||
      cinfo->comp_info[0].v_samp_factor >  2 ||
      cinfo->comp_info[1].v_samp_factor != 1 ||
      cinfo->comp_info[2].v_samp_factor != 1)
    return FALSE;
  /* furthermore, it doesn't work if we've scaled the IDCTs differently */
  if (cinfo->comp_info[0].DCT_scaled_size != cinfo->min_DCT_scaled_size ||
      cinfo->comp_info[1].DCT_scaled_size != cinfo->min_DCT_scaled_size ||
      cinfo->comp_info[2].DCT_scaled_size != cinfo->min_DCT_scaled_size)
    return FALSE;
  /* ??? also need to test for upsample-time rescaling, when & if supported */
  return TRUE;          /* by golly, it'll work... */
#else
  return FALSE;
#endif
}


/*
 * Compute output image dimensions and related values.
 * NOTE: this is exported for possible use by application.
 * Hence it mustn't do anything that can't be done twice.
 * Also note that it may be called before the master module is initialized!
 */

GLOBAL(void)
jpeg_calc_output_dimensions (j_decompress_ptr cinfo)
/* Do computations that are needed before master selection phase */
{
#ifdef IDCT_SCALING_SUPPORTED
  int ci;
  jpeg_component_info *compptr;
#endif

  /* Prevent application from calling me at wrong times */
  if (cinfo->global_state != DSTATE_READY)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);

#ifdef IDCT_SCALING_SUPPORTED

  /* Compute actual output image dimensions and DCT scaling choices. */
  if (cinfo->scale_num * 8 <= cinfo->scale_denom) {
    /* Provide 1/8 scaling */
    cinfo->output_width = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width, 8L);
    cinfo->output_height = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height, 8L);
    cinfo->min_DCT_scaled_size = 1;
  } else if (cinfo->scale_num * 4 <= cinfo->scale_denom) {
    /* Provide 1/4 scaling */
    cinfo->output_width = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width, 4L);
    cinfo->output_height = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height, 4L);
    cinfo->min_DCT_scaled_size = 2;
  } else if (cinfo->scale_num * 2 <= cinfo->scale_denom) {
    /* Provide 1/2 scaling */
    cinfo->output_width = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width, 2L);
    cinfo->output_height = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height, 2L);
    cinfo->min_DCT_scaled_size = 4;
  } else {
    /* Provide 1/1 scaling */
    cinfo->output_width = cinfo->image_width;
    cinfo->output_height = cinfo->image_height;
    cinfo->min_DCT_scaled_size = DCTSIZE;
  }
  /* In selecting the actual DCT scaling for each component, we try to
   * scale up the chroma components via IDCT scaling rather than upsampling.
   * This saves time if the upsampler gets to use 1:1 scaling.
   * Note this code assumes that the supported DCT scalings are powers of 2.
   */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    int ssize = cinfo->min_DCT_scaled_size;
    while (ssize < DCTSIZE &&
       (compptr->h_samp_factor * ssize * 2 <=
        cinfo->max_h_samp_factor * cinfo->min_DCT_scaled_size) &&
       (compptr->v_samp_factor * ssize * 2 <=
        cinfo->max_v_samp_factor * cinfo->min_DCT_scaled_size)) {
      ssize = ssize * 2;
    }
    compptr->DCT_scaled_size = ssize;
  }

  /* Recompute downsampled dimensions of components;
   * application needs to know these if using raw downsampled data.
   */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    /* Size in samples, after IDCT scaling */
    compptr->downsampled_width = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width *
            (long) (compptr->h_samp_factor * compptr->DCT_scaled_size),
            (long) (cinfo->max_h_samp_factor * DCTSIZE));
    compptr->downsampled_height = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height *
            (long) (compptr->v_samp_factor * compptr->DCT_scaled_size),
            (long) (cinfo->max_v_samp_factor * DCTSIZE));
  }

#else /* !IDCT_SCALING_SUPPORTED */

  /* Hardwire it to "no scaling" */
  cinfo->output_width = cinfo->image_width;
  cinfo->output_height = cinfo->image_height;
  /* jdinput.c has already initialized DCT_scaled_size to DCTSIZE,
   * and has computed unscaled downsampled_width and downsampled_height.
   */

#endif /* IDCT_SCALING_SUPPORTED */

  /* Report number of components in selected colorspace. */
  /* Probably this should be in the color conversion module... */
  switch (cinfo->out_color_space) {
  case JCS_GRAYSCALE:
    cinfo->out_color_components = 1;
    break;
  case JCS_RGB:
#if RGB_PIXELSIZE != 3
    cinfo->out_color_components = RGB_PIXELSIZE;
    break;
#endif /* else share code with YCbCr */
  case JCS_YCbCr:
    cinfo->out_color_components = 3;
    break;
  case JCS_CMYK:
  case JCS_YCCK:
    cinfo->out_color_components = 4;
    break;
  default:          /* else must be same colorspace as in file */
    cinfo->out_color_components = cinfo->num_components;
    break;
  }
  cinfo->output_components = (cinfo->quantize_colors ? 1 :
                  cinfo->out_color_components);

  /* See if upsampler will want to emit more than one row at a time */
  if (use_merged_upsample(cinfo))
    cinfo->rec_outbuf_height = cinfo->max_v_samp_factor;
  else
    cinfo->rec_outbuf_height = 1;
}


/*
 * Several decompression processes need to range-limit values to the range
 * 0..MAXJSAMPLE; the input value may fall somewhat outside this range
 * due to noise introduced by quantization, roundoff error, etc.  These
 * processes are inner loops and need to be as fast as possible.  On most
 * machines, particularly CPUs with pipelines or instruction prefetch,
 * a (subscript-check-less) C table lookup
 *      x = sample_range_limit[x];
 * is faster than explicit tests
 *      if (x < 0)  x = 0;
 *      else if (x > MAXJSAMPLE)  x = MAXJSAMPLE;
 * These processes all use a common table prepared by the routine below.
 *
 * For most steps we can mathematically guarantee that the initial value
 * of x is within MAXJSAMPLE+1 of the legal range, so a table running from
 * -(MAXJSAMPLE+1) to 2*MAXJSAMPLE+1 is sufficient.  But for the initial
 * limiting step (just after the IDCT), a wildly out-of-range value is
 * possible if the input data is corrupt.  To avoid any chance of indexing
 * off the end of memory and getting a bad-pointer trap, we perform the
 * post-IDCT limiting thus:
 *      x = range_limit[x & MASK];
 * where MASK is 2 bits wider than legal sample data, ie 10 bits for 8-bit
 * samples.  Under normal circumstances this is more than enough range and
 * a correct output will be generated; with bogus input data the mask will
 * cause wraparound, and we will safely generate a bogus-but-in-range output.
 * For the post-IDCT step, we want to convert the data from signed to unsigned
 * representation by adding CENTERJSAMPLE at the same time that we limit it.
 * So the post-IDCT limiting table ends up looking like this:
 *   CENTERJSAMPLE,CENTERJSAMPLE+1,...,MAXJSAMPLE,
 *   MAXJSAMPLE (repeat 2*(MAXJSAMPLE+1)-CENTERJSAMPLE times),
 *   0          (repeat 2*(MAXJSAMPLE+1)-CENTERJSAMPLE times),
 *   0,1,...,CENTERJSAMPLE-1
 * Negative inputs select values from the upper half of the table after
 * masking.
 *
 * We can save some space by overlapping the start of the post-IDCT table
 * with the simpler range limiting table.  The post-IDCT table begins at
 * sample_range_limit + CENTERJSAMPLE.
 *
 * Note that the table is allocated in near data space on PCs; it's small
 * enough and used often enough to justify this.
 */

LOCAL(void)
prepare_range_limit_table (j_decompress_ptr cinfo)
/* Allocate and fill in the sample_range_limit table */
{
  JSAMPLE * table;
  int i;

  table = (JSAMPLE *)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
        (5 * (MAXJSAMPLE+1) + CENTERJSAMPLE) * SIZEOF(JSAMPLE));
  table += (MAXJSAMPLE+1);  /* allow negative subscripts of simple table */
  cinfo->sample_range_limit = table;
  /* First segment of "simple" table: limit[x] = 0 for x < 0 */
  MEMZERO(table - (MAXJSAMPLE+1), (MAXJSAMPLE+1) * SIZEOF(JSAMPLE));
  /* Main part of "simple" table: limit[x] = x */
  for (i = 0; i <= MAXJSAMPLE; i++)
    table[i] = (JSAMPLE) i;
  table += CENTERJSAMPLE;   /* Point to where post-IDCT table starts */
  /* End of simple table, rest of first half of post-IDCT table */
  for (i = CENTERJSAMPLE; i < 2*(MAXJSAMPLE+1); i++)
    table[i] = MAXJSAMPLE;
  /* Second half of post-IDCT table */
  MEMZERO(table + (2 * (MAXJSAMPLE+1)),
      (2 * (MAXJSAMPLE+1) - CENTERJSAMPLE) * SIZEOF(JSAMPLE));
  MEMCOPY(table + (4 * (MAXJSAMPLE+1) - CENTERJSAMPLE),
      cinfo->sample_range_limit, CENTERJSAMPLE * SIZEOF(JSAMPLE));
}


/*
 * Master selection of decompression modules.
 * This is done once at jpeg_start_decompress time.  We determine
 * which modules will be used and give them appropriate initialization calls.
 * We also initialize the decompressor input side to begin consuming data.
 *
 * Since jpeg_read_header has finished, we know what is in the SOF
 * and (first) SOS markers.  We also have all the application parameter
 * settings.
 */

LOCAL(void)
master_selection (j_decompress_ptr cinfo)
{
  my_master_ptr master = (my_master_ptr) cinfo->master;
  boolean use_c_buffer;
  long samplesperrow;
  JDIMENSION jd_samplesperrow;

  /* Initialize dimensions and other stuff */
  jpeg_calc_output_dimensions(cinfo);
  prepare_range_limit_table(cinfo);

  /* Width of an output scanline must be representable as JDIMENSION. */
  samplesperrow = (long) cinfo->output_width * (long) cinfo->out_color_components;
  jd_samplesperrow = (JDIMENSION) samplesperrow;
  if ((long) jd_samplesperrow != samplesperrow)
    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);

  /* Initialize my private state */
  master->pass_number = 0;
  master->using_merged_upsample = use_merged_upsample(cinfo);

  /* Color quantizer selection */
  master->quantizer_1pass = NULL;
  master->quantizer_2pass = NULL;
  /* No mode changes if not using buffered-image mode. */
  if (! cinfo->quantize_colors || ! cinfo->buffered_image) {
    cinfo->enable_1pass_quant = FALSE;
    cinfo->enable_external_quant = FALSE;
    cinfo->enable_2pass_quant = FALSE;
  }
  if (cinfo->quantize_colors) {
    if (cinfo->raw_data_out)
      ERREXIT(cinfo, JERR_NOTIMPL);
    /* 2-pass quantizer only works in 3-component color space. */
    if (cinfo->out_color_components != 3) {
      cinfo->enable_1pass_quant = TRUE;
      cinfo->enable_external_quant = FALSE;
      cinfo->enable_2pass_quant = FALSE;
      cinfo->colormap = NULL;
    } else if (cinfo->colormap != NULL) {
      cinfo->enable_external_quant = TRUE;
    } else if (cinfo->two_pass_quantize) {
      cinfo->enable_2pass_quant = TRUE;
    } else {
      cinfo->enable_1pass_quant = TRUE;
    }

    if (cinfo->enable_1pass_quant) {
#ifdef QUANT_1PASS_SUPPORTED
      jinit_1pass_quantizer(cinfo);
      master->quantizer_1pass = cinfo->cquantize;
#else
      ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif
    }

    /* We use the 2-pass code to map to external colormaps. */
    if (cinfo->enable_2pass_quant || cinfo->enable_external_quant) {
#ifdef QUANT_2PASS_SUPPORTED
      jinit_2pass_quantizer(cinfo);
      master->quantizer_2pass = cinfo->cquantize;
#else
      ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif
    }
    /* If both quantizers are initialized, the 2-pass one is left active;
     * this is necessary for starting with quantization to an external map.
     */
  }

  /* Post-processing: in particular, color conversion first */
  if (! cinfo->raw_data_out) {
    if (master->using_merged_upsample) {
#ifdef UPSAMPLE_MERGING_SUPPORTED
      jinit_merged_upsampler(cinfo); /* does color conversion too */
#else
      ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif
    } else {
      jinit_color_deconverter(cinfo);
      jinit_upsampler(cinfo);
    }
    jinit_d_post_controller(cinfo, cinfo->enable_2pass_quant);
  }
  /* Inverse DCT */
  jinit_inverse_dct(cinfo);
  /* Entropy decoding: either Huffman or arithmetic coding. */
  if (cinfo->arith_code) {
    ERREXIT(cinfo, JERR_ARITH_NOTIMPL);
  } else {
    if (cinfo->progressive_mode) {
#ifdef D_PROGRESSIVE_SUPPORTED
      jinit_phuff_decoder(cinfo);
#else
      ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif
    } else
      jinit_huff_decoder(cinfo);
  }

  /* Initialize principal buffer controllers. */
  use_c_buffer = cinfo->inputctl->has_multiple_scans || cinfo->buffered_image;
  jinit_d_coef_controller(cinfo, use_c_buffer);

  if (! cinfo->raw_data_out)
    jinit_d_main_controller(cinfo, FALSE /* never need full buffer here */);

  /* We can now tell the memory manager to allocate virtual arrays. */
  (*cinfo->mem->realize_virt_arrays) ((j_common_ptr) cinfo);

  /* Initialize input side of decompressor to consume first scan. */
  (*cinfo->inputctl->start_input_pass) (cinfo);

#ifdef D_MULTISCAN_FILES_SUPPORTED
  /* If jpeg_start_decompress will read the whole file, initialize
   * progress monitoring appropriately.  The input step is counted
   * as one pass.
   */
  if (cinfo->progress != NULL && ! cinfo->buffered_image &&
      cinfo->inputctl->has_multiple_scans) {
    int nscans;
    /* Estimate number of scans to set pass_limit. */
    if (cinfo->progressive_mode) {
      /* Arbitrarily estimate 2 interleaved DC scans + 3 AC scans/component. */
      nscans = 2 + 3 * cinfo->num_components;
    } else {
      /* For a nonprogressive multiscan file, estimate 1 scan per component. */
      nscans = cinfo->num_components;
    }
    cinfo->progress->pass_counter = 0L;
    cinfo->progress->pass_limit = (long) cinfo->total_iMCU_rows * nscans;
    cinfo->progress->completed_passes = 0;
    cinfo->progress->total_passes = (cinfo->enable_2pass_quant ? 3 : 2);
    /* Count the input pass as done */
    master->pass_number++;
  }
#endif /* D_MULTISCAN_FILES_SUPPORTED */
}


/*
 * Per-pass setup.
 * This is called at the beginning of each output pass.  We determine which
 * modules will be active during this pass and give them appropriate
 * start_pass calls.  We also set is_dummy_pass to indicate whether this
 * is a "real" output pass or a dummy pass for color quantization.
 * (In the latter case, jdapistd.c will crank the pass to completion.)
 */

METHODDEF(void)
prepare_for_output_pass (j_decompress_ptr cinfo)
{
  my_master_ptr master = (my_master_ptr) cinfo->master;

  if (master->pub.is_dummy_pass) {
#ifdef QUANT_2PASS_SUPPORTED
    /* Final pass of 2-pass quantization */
    master->pub.is_dummy_pass = FALSE;
    (*cinfo->cquantize->start_pass) (cinfo, FALSE);
    (*cinfo->post->start_pass) (cinfo, JBUF_CRANK_DEST);
    (*cinfo->main_ptr->start_pass) (cinfo, JBUF_CRANK_DEST);
#else
    ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif /* QUANT_2PASS_SUPPORTED */
  } else {
    if (cinfo->quantize_colors && cinfo->colormap == NULL) {
      /* Select new quantization method */
      if (cinfo->two_pass_quantize && cinfo->enable_2pass_quant) {
    cinfo->cquantize = master->quantizer_2pass;
    master->pub.is_dummy_pass = TRUE;
      } else if (cinfo->enable_1pass_quant) {
    cinfo->cquantize = master->quantizer_1pass;
      } else {
    ERREXIT(cinfo, JERR_MODE_CHANGE);
      }
    }
    (*cinfo->idct->start_pass) (cinfo);
    (*cinfo->coef->start_output_pass) (cinfo);
    if (! cinfo->raw_data_out) {
      if (! master->using_merged_upsample)
    (*cinfo->cconvert->start_pass) (cinfo);
      (*cinfo->upsample->start_pass) (cinfo);
      if (cinfo->quantize_colors)
    (*cinfo->cquantize->start_pass) (cinfo, master->pub.is_dummy_pass);
      (*cinfo->post->start_pass) (cinfo,
        (master->pub.is_dummy_pass ? JBUF_SAVE_AND_PASS : JBUF_PASS_THRU));
      (*cinfo->main_ptr->start_pass) (cinfo, JBUF_PASS_THRU);
    }
  }

  /* Set up progress monitor's pass info if present */
  if (cinfo->progress != NULL) {
    cinfo->progress->completed_passes = master->pass_number;
    cinfo->progress->total_passes = master->pass_number +
                    (master->pub.is_dummy_pass ? 2 : 1);
    /* In buffered-image mode, we assume one more output pass if EOI not
     * yet reached, but no more passes if EOI has been reached.
     */
    if (cinfo->buffered_image && ! cinfo->inputctl->eoi_reached) {
      cinfo->progress->total_passes += (cinfo->enable_2pass_quant ? 2 : 1);
    }
  }
}


/*
 * Finish up at end of an output pass.
 */

METHODDEF(void)
finish_output_pass (j_decompress_ptr cinfo)
{
  my_master_ptr master = (my_master_ptr) cinfo->master;

  if (cinfo->quantize_colors)
    (*cinfo->cquantize->finish_pass) (cinfo);
  master->pass_number++;
}


#ifdef D_MULTISCAN_FILES_SUPPORTED

/*
 * Switch to a new external colormap between output passes.
 */

GLOBAL(void)
jpeg_new_colormap (j_decompress_ptr cinfo)
{
  my_master_ptr master = (my_master_ptr) cinfo->master;

  /* Prevent application from calling me at wrong times */
  if (cinfo->global_state != DSTATE_BUFIMAGE)
    ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state);

  if (cinfo->quantize_colors && cinfo->enable_external_quant &&
      cinfo->colormap != NULL) {
    /* Select 2-pass quantizer for external colormap use */
    cinfo->cquantize = master->quantizer_2pass;
    /* Notify quantizer of colormap change */
    (*cinfo->cquantize->new_color_map) (cinfo);
    master->pub.is_dummy_pass = FALSE; /* just in case */
  } else
    ERREXIT(cinfo, JERR_MODE_CHANGE);
}

#endif /* D_MULTISCAN_FILES_SUPPORTED */


/*
 * Initialize master decompression control and select active modules.
 * This is performed at the start of jpeg_start_decompress.
 */

GLOBAL(void)
jinit_master_decompress (j_decompress_ptr cinfo)
{
  my_master_ptr master;

  master = (my_master_ptr)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                  SIZEOF(my_decomp_master));
  cinfo->master = (struct jpeg_decomp_master *) master;
  master->pub.prepare_for_output_pass = prepare_for_output_pass;
  master->pub.finish_output_pass = finish_output_pass;

  master->pub.is_dummy_pass = FALSE;

  master_selection(cinfo);
}
/*
 * jdinput.c
 *
 * Copyright (C) 1991-1997, 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 input control logic for the JPEG decompressor.
 * These routines are concerned with controlling the decompressor's input
 * processing (marker reading and coefficient decoding).  The actual input
 * reading is done in jdmarker.c, jdhuff.c, and jdphuff.c.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/* Private state */

typedef struct {
  struct jpeg_input_controller pub; /* public fields */

  boolean inheaders;        /* TRUE until first SOS is reached */
} my_input_controller;

typedef my_input_controller * my_inputctl_ptr;


/* Forward declarations */
METHODDEF(int) consume_markers JPP((j_decompress_ptr cinfo));


/*
 * Routines to calculate various quantities related to the size of the image.
 */

LOCAL(void)
initial_setup (j_decompress_ptr cinfo)
/* Called once, when first SOS marker is reached */
{
  int ci;
  jpeg_component_info *compptr;

  /* Make sure image isn't bigger than I can handle */
  if ((long) cinfo->image_height > (long) JPEG_MAX_DIMENSION ||
      (long) cinfo->image_width > (long) JPEG_MAX_DIMENSION)
    ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int) JPEG_MAX_DIMENSION);

  /* For now, precision must match compiled-in value... */
  if (cinfo->data_precision != BITS_IN_JSAMPLE)
    ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);

  /* Check that number of components won't exceed internal array sizes */
  if (cinfo->num_components > MAX_COMPONENTS)
    ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components,
         MAX_COMPONENTS);

  /* Compute maximum sampling factors; check factor validity */
  cinfo->max_h_samp_factor = 1;
  cinfo->max_v_samp_factor = 1;
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    if (compptr->h_samp_factor<=0 || compptr->h_samp_factor>MAX_SAMP_FACTOR ||
    compptr->v_samp_factor<=0 || compptr->v_samp_factor>MAX_SAMP_FACTOR)
      ERREXIT(cinfo, JERR_BAD_SAMPLING);
    cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor,
                   compptr->h_samp_factor);
    cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor,
                   compptr->v_samp_factor);
  }

  /* We initialize DCT_scaled_size and min_DCT_scaled_size to DCTSIZE.
   * In the full decompressor, this will be overridden by jdmaster.c;
   * but in the transcoder, jdmaster.c is not used, so we must do it here.
   */
  cinfo->min_DCT_scaled_size = DCTSIZE;

  /* Compute dimensions of components */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    compptr->DCT_scaled_size = DCTSIZE;
    /* Size in DCT blocks */
    compptr->width_in_blocks = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
            (long) (cinfo->max_h_samp_factor * DCTSIZE));
    compptr->height_in_blocks = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
            (long) (cinfo->max_v_samp_factor * DCTSIZE));
    /* downsampled_width and downsampled_height will also be overridden by
     * jdmaster.c if we are doing full decompression.  The transcoder library
     * doesn't use these values, but the calling application might.
     */
    /* Size in samples */
    compptr->downsampled_width = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width * (long) compptr->h_samp_factor,
            (long) cinfo->max_h_samp_factor);
    compptr->downsampled_height = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height * (long) compptr->v_samp_factor,
            (long) cinfo->max_v_samp_factor);
    /* Mark component needed, until color conversion says otherwise */
    compptr->component_needed = TRUE;
    /* Mark no quantization table yet saved for component */
    compptr->quant_table = NULL;
  }

  /* Compute number of fully interleaved MCU rows. */
  cinfo->total_iMCU_rows = (JDIMENSION)
    jdiv_round_up((long) cinfo->image_height,
          (long) (cinfo->max_v_samp_factor*DCTSIZE));

  /* Decide whether file contains multiple scans */
  if (cinfo->comps_in_scan < cinfo->num_components || cinfo->progressive_mode)
    cinfo->inputctl->has_multiple_scans = TRUE;
  else
    cinfo->inputctl->has_multiple_scans = FALSE;
}


LOCAL(void)
per_scan_setup (j_decompress_ptr cinfo)
/* Do computations that are needed before processing a JPEG scan */
/* cinfo->comps_in_scan and cinfo->cur_comp_info[] were set from SOS marker */
{
  int ci, mcublks, tmp;
  jpeg_component_info *compptr;

  if (cinfo->comps_in_scan == 1) {

    /* Noninterleaved (single-component) scan */
    compptr = cinfo->cur_comp_info[0];

    /* Overall image size in MCUs */
    cinfo->MCUs_per_row = compptr->width_in_blocks;
    cinfo->MCU_rows_in_scan = compptr->height_in_blocks;

    /* For noninterleaved scan, always one block per MCU */
    compptr->MCU_width = 1;
    compptr->MCU_height = 1;
    compptr->MCU_blocks = 1;
    compptr->MCU_sample_width = compptr->DCT_scaled_size;
    compptr->last_col_width = 1;
    /* For noninterleaved scans, it is convenient to define last_row_height
     * as the number of block rows present in the last iMCU row.
     */
    tmp = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
    if (tmp == 0) tmp = compptr->v_samp_factor;
    compptr->last_row_height = tmp;

    /* Prepare array describing MCU composition */
    cinfo->blocks_in_MCU = 1;
    cinfo->MCU_membership[0] = 0;

  } else {

    /* Interleaved (multi-component) scan */
    if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN)
      ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan,
           MAX_COMPS_IN_SCAN);

    /* Overall image size in MCUs */
    cinfo->MCUs_per_row = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_width,
            (long) (cinfo->max_h_samp_factor*DCTSIZE));
    cinfo->MCU_rows_in_scan = (JDIMENSION)
      jdiv_round_up((long) cinfo->image_height,
            (long) (cinfo->max_v_samp_factor*DCTSIZE));

    cinfo->blocks_in_MCU = 0;

    for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
      compptr = cinfo->cur_comp_info[ci];
      /* Sampling factors give # of blocks of component in each MCU */
      compptr->MCU_width = compptr->h_samp_factor;
      compptr->MCU_height = compptr->v_samp_factor;
      compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height;
      compptr->MCU_sample_width = compptr->MCU_width * compptr->DCT_scaled_size;
      /* Figure number of non-dummy blocks in last MCU column & row */
      tmp = (int) (compptr->width_in_blocks % compptr->MCU_width);
      if (tmp == 0) tmp = compptr->MCU_width;
      compptr->last_col_width = tmp;
      tmp = (int) (compptr->height_in_blocks % compptr->MCU_height);
      if (tmp == 0) tmp = compptr->MCU_height;
      compptr->last_row_height = tmp;
      /* Prepare array describing MCU composition */
      mcublks = compptr->MCU_blocks;
      if (cinfo->blocks_in_MCU + mcublks > D_MAX_BLOCKS_IN_MCU)
    ERREXIT(cinfo, JERR_BAD_MCU_SIZE);
      while (mcublks-- > 0) {
    cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci;
      }
    }

  }
}


/*
 * Save away a copy of the Q-table referenced by each component present
 * in the current scan, unless already saved during a prior scan.
 *
 * In a multiple-scan JPEG file, the encoder could assign different components
 * the same Q-table slot number, but change table definitions between scans
 * so that each component uses a different Q-table.  (The IJG encoder is not
 * currently capable of doing this, but other encoders might.)  Since we want
 * to be able to dequantize all the components at the end of the file, this
 * means that we have to save away the table actually used for each component.
 * We do this by copying the table at the start of the first scan containing
 * the component.
 * The JPEG spec prohibits the encoder from changing the contents of a Q-table
 * slot between scans of a component using that slot.  If the encoder does so
 * anyway, this decoder will simply use the Q-table values that were current
 * at the start of the first scan for the component.
 *
 * The decompressor output side looks only at the saved quant tables,
 * not at the current Q-table slots.
 */

LOCAL(void)
latch_quant_tables (j_decompress_ptr cinfo)
{
  int ci, qtblno;
  jpeg_component_info *compptr;
  JQUANT_TBL * qtbl;

  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
    compptr = cinfo->cur_comp_info[ci];
    /* No work if we already saved Q-table for this component */
    if (compptr->quant_table != NULL)
      continue;
    /* Make sure specified quantization table is present */
    qtblno = compptr->quant_tbl_no;
    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
    cinfo->quant_tbl_ptrs[qtblno] == NULL)
      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
    /* OK, save away the quantization table */
    qtbl = (JQUANT_TBL *)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                  SIZEOF(JQUANT_TBL));
    MEMCOPY(qtbl, cinfo->quant_tbl_ptrs[qtblno], SIZEOF(JQUANT_TBL));
    compptr->quant_table = qtbl;
  }
}


/*
 * Initialize the input modules to read a scan of compressed data.
 * The first call to this is done by jdmaster.c after initializing
 * the entire decompressor (during jpeg_start_decompress).
 * Subsequent calls come from consume_markers, below.
 */

METHODDEF(void)
start_input_pass (j_decompress_ptr cinfo)
{
  per_scan_setup(cinfo);
  latch_quant_tables(cinfo);
  (*cinfo->entropy->start_pass) (cinfo);
  (*cinfo->coef->start_input_pass) (cinfo);
  cinfo->inputctl->consume_input = cinfo->coef->consume_data;
}


/*
 * Finish up after inputting a compressed-data scan.
 * This is called by the coefficient controller after it's read all
 * the expected data of the scan.
 */

METHODDEF(void)
finish_input_pass (j_decompress_ptr cinfo)
{
  cinfo->inputctl->consume_input = consume_markers;
}


/*
 * Read JPEG markers before, between, or after compressed-data scans.
 * Change state as necessary when a new scan is reached.
 * Return value is JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI.
 *
 * The consume_input method pointer points either here or to the
 * coefficient controller's consume_data routine, depending on whether
 * we are reading a compressed data segment or inter-segment markers.
 */

METHODDEF(int)
consume_markers (j_decompress_ptr cinfo)
{
  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;
  int val;

  if (inputctl->pub.eoi_reached) /* After hitting EOI, read no further */
    return JPEG_REACHED_EOI;

  val = (*cinfo->marker->read_markers) (cinfo);

  switch (val) {
  case JPEG_REACHED_SOS:    /* Found SOS */
    if (inputctl->inheaders) {  /* 1st SOS */
      initial_setup(cinfo);
      inputctl->inheaders = FALSE;
      /* Note: start_input_pass must be called by jdmaster.c
       * before any more input can be consumed.  jdapimin.c is
       * responsible for enforcing this sequencing.
       */
    } else {            /* 2nd or later SOS marker */
      if (! inputctl->pub.has_multiple_scans)
    ERREXIT(cinfo, JERR_EOI_EXPECTED); /* Oops, I wasn't expecting this! */
      start_input_pass(cinfo);
    }
    break;
  case JPEG_REACHED_EOI:    /* Found EOI */
    inputctl->pub.eoi_reached = TRUE;
    if (inputctl->inheaders) {  /* Tables-only datastream, apparently */
      if (cinfo->marker->saw_SOF)
    ERREXIT(cinfo, JERR_SOF_NO_SOS);
    } else {
      /* Prevent infinite loop in coef ctlr's decompress_data routine
       * if user set output_scan_number larger than number of scans.
       */
      if (cinfo->output_scan_number > cinfo->input_scan_number)
    cinfo->output_scan_number = cinfo->input_scan_number;
    }
    break;
  case JPEG_SUSPENDED:
    break;
  }

  return val;
}


/*
 * Reset state to begin a fresh datastream.
 */

METHODDEF(void)
reset_input_controller (j_decompress_ptr cinfo)
{
  my_inputctl_ptr inputctl = (my_inputctl_ptr) cinfo->inputctl;

  inputctl->pub.consume_input = consume_markers;
  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
  inputctl->pub.eoi_reached = FALSE;
  inputctl->inheaders = TRUE;
  /* Reset other modules */
  (*cinfo->err->reset_error_mgr) ((j_common_ptr) cinfo);
  (*cinfo->marker->reset_marker_reader) (cinfo);
  /* Reset progression state -- would be cleaner if entropy decoder did this */
  cinfo->coef_bits = NULL;
}


/*
 * Initialize the input controller module.
 * This is called only once, when the decompression object is created.
 */

GLOBAL(void)
jinit_input_controller (j_decompress_ptr cinfo)
{
  my_inputctl_ptr inputctl;

  /* Create subobject in permanent pool */
  inputctl = (my_inputctl_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
                SIZEOF(my_input_controller));
  cinfo->inputctl = (struct jpeg_input_controller *) inputctl;
  /* Initialize method pointers */
  inputctl->pub.consume_input = consume_markers;
  inputctl->pub.reset_input_controller = reset_input_controller;
  inputctl->pub.start_input_pass = start_input_pass;
  inputctl->pub.finish_input_pass = finish_input_pass;
  /* Initialize state: can't use reset_input_controller since we don't
   * want to try to reset other modules yet.
   */
  inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */
  inputctl->pub.eoi_reached = FALSE;
  inputctl->inheaders = TRUE;
}
/*
 * jdmarker.c
 *
 * Copyright (C) 1991-1998, 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 routines to decode JPEG datastream markers.
 * Most of the complexity arises from our desire to support input
 * suspension: if not all of the data for a marker is available,
 * we must exit back to the application.  On resumption, we reprocess
 * the marker.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


typedef enum {          /* JPEG marker codes */
  M_SOF0  = 0xc0,
  M_SOF1  = 0xc1,
  M_SOF2  = 0xc2,
  M_SOF3  = 0xc3,

  M_SOF5  = 0xc5,
  M_SOF6  = 0xc6,
  M_SOF7  = 0xc7,

  M_JPG   = 0xc8,
  M_SOF9  = 0xc9,
  M_SOF10 = 0xca,
  M_SOF11 = 0xcb,

  M_SOF13 = 0xcd,
  M_SOF14 = 0xce,
  M_SOF15 = 0xcf,

  M_DHT   = 0xc4,

  M_DAC   = 0xcc,

  M_RST0  = 0xd0,
  M_RST1  = 0xd1,
  M_RST2  = 0xd2,
  M_RST3  = 0xd3,
  M_RST4  = 0xd4,
  M_RST5  = 0xd5,
  M_RST6  = 0xd6,
  M_RST7  = 0xd7,

  M_SOI   = 0xd8,
  M_EOI   = 0xd9,
  M_SOS   = 0xda,
  M_DQT   = 0xdb,
  M_DNL   = 0xdc,
  M_DRI   = 0xdd,
  M_DHP   = 0xde,
  M_EXP   = 0xdf,

  M_APP0  = 0xe0,
  M_APP1  = 0xe1,
  M_APP2  = 0xe2,
  M_APP3  = 0xe3,
  M_APP4  = 0xe4,
  M_APP5  = 0xe5,
  M_APP6  = 0xe6,
  M_APP7  = 0xe7,
  M_APP8  = 0xe8,
  M_APP9  = 0xe9,
  M_APP10 = 0xea,
  M_APP11 = 0xeb,
  M_APP12 = 0xec,
  M_APP13 = 0xed,
  M_APP14 = 0xee,
  M_APP15 = 0xef,

  M_JPG0  = 0xf0,
  M_JPG13 = 0xfd,
  M_COM   = 0xfe,

  M_TEM   = 0x01,

  M_ERROR = 0x100
} JPEG_MARKER;


/* Private state */

typedef struct {
  struct jpeg_marker_reader pub; /* public fields */

  /* Application-overridable marker processing methods */
  jpeg_marker_parser_method process_COM;
  jpeg_marker_parser_method process_APPn[16];

  /* Limit on marker data length to save for each marker type */
  unsigned int length_limit_COM;
  unsigned int length_limit_APPn[16];

  /* Status of COM/APPn marker saving */
  jpeg_saved_marker_ptr cur_marker; /* NULL if not processing a marker */
  unsigned int bytes_read;      /* data bytes read so far in marker */
  /* Note: cur_marker is not linked into marker_list until it's all read. */
} my_marker_reader;

typedef my_marker_reader * my_marker_ptr;


/*
 * Macros for fetching data from the data source module.
 *
 * At all times, cinfo->src->next_input_byte and ->bytes_in_buffer reflect
 * the current restart point; we update them only when we have reached a
 * suitable place to restart if a suspension occurs.
 */

/* Declare and initialize local copies of input pointer/count */
#define jdar_INPUT_VARS(cinfo)  \
    struct jpeg_source_mgr * datasrc = (cinfo)->src;  \
    const JOCTET * next_input_byte = datasrc->next_input_byte;  \
    size_t bytes_in_buffer = datasrc->bytes_in_buffer

/* Unload the local copies --- do this only at a restart boundary */
#define jdar_INPUT_SYNC(cinfo)  \
    ( datasrc->next_input_byte = next_input_byte,  \
      datasrc->bytes_in_buffer = bytes_in_buffer )

/* Reload the local copies --- used only in MAKE_BYTE_AVAIL */
#define jdar_INPUT_RELOAD(cinfo)  \
    ( next_input_byte = datasrc->next_input_byte,  \
      bytes_in_buffer = datasrc->bytes_in_buffer )

/* Internal macro for INPUT_BYTE and INPUT_2BYTES: make a byte available.
 * Note we do *not* do INPUT_SYNC before calling fill_input_buffer,
 * but we must reload the local copies after a successful fill.
 */
#define jdar_MAKE_BYTE_AVAIL(cinfo,action)  \
    if (bytes_in_buffer == 0) {  \
      if (! (*datasrc->fill_input_buffer) (cinfo))  \
        { action; }  \
      jdar_INPUT_RELOAD(cinfo);  \
    }

/* Read a byte into variable V.
 * If must suspend, take the specified action (typically "return FALSE").
 */
#define jdar_INPUT_BYTE(cinfo,V,action)  \
    MAKESTMT( jdar_MAKE_BYTE_AVAIL(cinfo,action); \
          bytes_in_buffer--; \
          V = GETJOCTET(*next_input_byte++); )

/* As above, but read two bytes interpreted as an unsigned 16-bit integer.
 * V should be declared unsigned int or perhaps INT32.
 */
#define jdar_INPUT_2BYTES(cinfo,V,action)  \
    MAKESTMT( jdar_MAKE_BYTE_AVAIL(cinfo,action); \
          bytes_in_buffer--; \
          V = ((unsigned int) GETJOCTET(*next_input_byte++)) << 8; \
          jdar_MAKE_BYTE_AVAIL(cinfo,action); \
          bytes_in_buffer--; \
          V += GETJOCTET(*next_input_byte++); )


/*
 * Routines to process JPEG markers.
 *
 * Entry condition: JPEG marker itself has been read and its code saved
 *   in cinfo->unread_marker; input restart point is just after the marker.
 *
 * Exit: if return TRUE, have read and processed any parameters, and have
 *   updated the restart point to point after the parameters.
 *   If return FALSE, was forced to suspend before reaching end of
 *   marker parameters; restart point has not been moved.  Same routine
 *   will be called again after application supplies more input data.
 *
 * This approach to suspension assumes that all of a marker's parameters
 * can fit into a single input bufferload.  This should hold for "normal"
 * markers.  Some COM/APPn markers might have large parameter segments
 * that might not fit.  If we are simply dropping such a marker, we use
 * skip_input_data to get past it, and thereby put the problem on the
 * source manager's shoulders.  If we are saving the marker's contents
 * into memory, we use a slightly different convention: when forced to
 * suspend, the marker processor updates the restart point to the end of
 * what it's consumed (ie, the end of the buffer) before returning FALSE.
 * On resumption, cinfo->unread_marker still contains the marker code,
 * but the data source will point to the next chunk of marker data.
 * The marker processor must retain internal state to deal with this.
 *
 * Note that we don't bother to avoid duplicate trace messages if a
 * suspension occurs within marker parameters.  Other side effects
 * require more care.
 */


LOCAL(boolean)
get_soi (j_decompress_ptr cinfo)
/* Process an SOI marker */
{
  int i;

  TRACEMS(cinfo, 1, JTRC_SOI);

  if (cinfo->marker->saw_SOI)
    ERREXIT(cinfo, JERR_SOI_DUPLICATE);

  /* Reset all parameters that are defined to be reset by SOI */

  for (i = 0; i < NUM_ARITH_TBLS; i++) {
    cinfo->arith_dc_L[i] = 0;
    cinfo->arith_dc_U[i] = 1;
    cinfo->arith_ac_K[i] = 5;
  }
  cinfo->restart_interval = 0;

  /* Set initial assumptions for colorspace etc */

  cinfo->jpeg_color_space = JCS_UNKNOWN;
  cinfo->CCIR601_sampling = FALSE; /* Assume non-CCIR sampling??? */

  cinfo->saw_JFIF_marker = FALSE;
  cinfo->JFIF_major_version = 1; /* set default JFIF APP0 values */
  cinfo->JFIF_minor_version = 1;
  cinfo->density_unit = 0;
  cinfo->X_density = 1;
  cinfo->Y_density = 1;
  cinfo->saw_Adobe_marker = FALSE;
  cinfo->Adobe_transform = 0;

  cinfo->marker->saw_SOI = TRUE;

  return TRUE;
}


LOCAL(boolean)
get_sof (j_decompress_ptr cinfo, boolean is_prog, boolean is_arith)
/* Process a SOFn marker */
{
  INT32 length;
  int c, ci;
  jpeg_component_info * compptr;
  jdar_INPUT_VARS(cinfo);

  cinfo->progressive_mode = is_prog;
  cinfo->arith_code = is_arith;

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);

  jdar_INPUT_BYTE(cinfo, cinfo->data_precision, return FALSE);
  jdar_INPUT_2BYTES(cinfo, cinfo->image_height, return FALSE);
  jdar_INPUT_2BYTES(cinfo, cinfo->image_width, return FALSE);
  jdar_INPUT_BYTE(cinfo, cinfo->num_components, return FALSE);

  length -= 8;

  TRACEMS4(cinfo, 1, JTRC_SOF, cinfo->unread_marker,
       (int) cinfo->image_width, (int) cinfo->image_height,
       cinfo->num_components);

  if (cinfo->marker->saw_SOF)
    ERREXIT(cinfo, JERR_SOF_DUPLICATE);

  /* We don't support files in which the image height is initially specified */
  /* as 0 and is later redefined by DNL.  As long as we have to check that,  */
  /* might as well have a general sanity check. */
  if (cinfo->image_height <= 0 || cinfo->image_width <= 0
      || cinfo->num_components <= 0)
    ERREXIT(cinfo, JERR_EMPTY_IMAGE);

  if (length != (cinfo->num_components * 3))
    ERREXIT(cinfo, JERR_BAD_LENGTH);

  if (cinfo->comp_info == NULL) /* do only once, even if suspend */
    cinfo->comp_info = (jpeg_component_info *) (*cinfo->mem->alloc_small)
            ((j_common_ptr) cinfo, JPOOL_IMAGE,
             cinfo->num_components * SIZEOF(jpeg_component_info));

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    compptr->component_index = ci;
    jdar_INPUT_BYTE(cinfo, compptr->component_id, return FALSE);
    jdar_INPUT_BYTE(cinfo, c, return FALSE);
    compptr->h_samp_factor = (c >> 4) & 15;
    compptr->v_samp_factor = (c     ) & 15;
    jdar_INPUT_BYTE(cinfo, compptr->quant_tbl_no, return FALSE);

    TRACEMS4(cinfo, 1, JTRC_SOF_COMPONENT,
         compptr->component_id, compptr->h_samp_factor,
         compptr->v_samp_factor, compptr->quant_tbl_no);
  }

  cinfo->marker->saw_SOF = TRUE;

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


LOCAL(boolean)
get_sos (j_decompress_ptr cinfo)
/* Process a SOS marker */
{
  INT32 length;
  int i, ci, n, c, cc;
  jpeg_component_info * compptr;
  jdar_INPUT_VARS(cinfo);

  if (! cinfo->marker->saw_SOF)
    ERREXIT(cinfo, JERR_SOS_NO_SOF);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);

  jdar_INPUT_BYTE(cinfo, n, return FALSE); /* Number of components */

  TRACEMS1(cinfo, 1, JTRC_SOS, n);

  if (length != (n * 2 + 6) || n < 1 || n > MAX_COMPS_IN_SCAN)
    ERREXIT(cinfo, JERR_BAD_LENGTH);

  cinfo->comps_in_scan = n;

  /* Collect the component-spec parameters */

  for (i = 0; i < n; i++) {
    jdar_INPUT_BYTE(cinfo, cc, return FALSE);
    jdar_INPUT_BYTE(cinfo, c, return FALSE);

    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
     ci++, compptr++) {
      if (cc == compptr->component_id)
    goto id_found;
    }

    ERREXIT1(cinfo, JERR_BAD_COMPONENT_ID, cc);

  id_found:

    cinfo->cur_comp_info[i] = compptr;
    compptr->dc_tbl_no = (c >> 4) & 15;
    compptr->ac_tbl_no = (c     ) & 15;

    TRACEMS3(cinfo, 1, JTRC_SOS_COMPONENT, cc,
         compptr->dc_tbl_no, compptr->ac_tbl_no);
  }

  /* Collect the additional scan parameters Ss, Se, Ah/Al. */
  jdar_INPUT_BYTE(cinfo, c, return FALSE);
  cinfo->Ss = c;
  jdar_INPUT_BYTE(cinfo, c, return FALSE);
  cinfo->Se = c;
  jdar_INPUT_BYTE(cinfo, c, return FALSE);
  cinfo->Ah = (c >> 4) & 15;
  cinfo->Al = (c     ) & 15;

  TRACEMS4(cinfo, 1, JTRC_SOS_PARAMS, cinfo->Ss, cinfo->Se,
       cinfo->Ah, cinfo->Al);

  /* Prepare to scan data & restart markers */
  cinfo->marker->next_restart_num = 0;

  /* Count another SOS marker */
  cinfo->input_scan_number++;

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


#ifdef D_ARITH_CODING_SUPPORTED

LOCAL(boolean)
get_dac (j_decompress_ptr cinfo)
/* Process a DAC marker */
{
  INT32 length;
  int index, val;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);
  length -= 2;

  while (length > 0) {
    jdar_INPUT_BYTE(cinfo, index, return FALSE);
    jdar_INPUT_BYTE(cinfo, val, return FALSE);

    length -= 2;

    TRACEMS2(cinfo, 1, JTRC_DAC, index, val);

    if (index < 0 || index >= (2*NUM_ARITH_TBLS))
      ERREXIT1(cinfo, JERR_DAC_INDEX, index);

    if (index >= NUM_ARITH_TBLS) { /* define AC table */
      cinfo->arith_ac_K[index-NUM_ARITH_TBLS] = (UINT8) val;
    } else {            /* define DC table */
      cinfo->arith_dc_L[index] = (UINT8) (val & 0x0F);
      cinfo->arith_dc_U[index] = (UINT8) (val >> 4);
      if (cinfo->arith_dc_L[index] > cinfo->arith_dc_U[index])
    ERREXIT1(cinfo, JERR_DAC_VALUE, val);
    }
  }

  if (length != 0)
    ERREXIT(cinfo, JERR_BAD_LENGTH);

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}

#else /* ! D_ARITH_CODING_SUPPORTED */

#define jdar_get_dac(cinfo)  skip_variable(cinfo)

#endif /* D_ARITH_CODING_SUPPORTED */


LOCAL(boolean)
get_dht (j_decompress_ptr cinfo)
/* Process a DHT marker */
{
  INT32 length;
  UINT8 bits[17];
  UINT8 huffval[256];
  int i, index, count;
  JHUFF_TBL **htblptr;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);
  length -= 2;

  while (length > 16) {
    jdar_INPUT_BYTE(cinfo, index, return FALSE);

    TRACEMS1(cinfo, 1, JTRC_DHT, index);

    bits[0] = 0;
    count = 0;
    for (i = 1; i <= 16; i++) {
      jdar_INPUT_BYTE(cinfo, bits[i], return FALSE);
      count += bits[i];
    }

    length -= 1 + 16;

    TRACEMS8(cinfo, 2, JTRC_HUFFBITS,
         bits[1], bits[2], bits[3], bits[4],
         bits[5], bits[6], bits[7], bits[8]);
    TRACEMS8(cinfo, 2, JTRC_HUFFBITS,
         bits[9], bits[10], bits[11], bits[12],
         bits[13], bits[14], bits[15], bits[16]);

    /* Here we just do minimal validation of the counts to avoid walking
     * off the end of our table space.  jdhuff.c will check more carefully.
     */
    if (count > 256 || ((INT32) count) > length)
      ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);

    for (i = 0; i < count; i++)
      jdar_INPUT_BYTE(cinfo, huffval[i], return FALSE);

    length -= count;

    if (index & 0x10) {     /* AC table definition */
      index -= 0x10;
      htblptr = &cinfo->ac_huff_tbl_ptrs[index];
    } else {            /* DC table definition */
      if (index < 0 || index >= NUM_HUFF_TBLS)
        ERREXIT1(cinfo, JERR_DHT_INDEX, index);
      htblptr = &cinfo->dc_huff_tbl_ptrs[index];
    }

    if (*htblptr == NULL)
      *htblptr = jpeg_alloc_huff_table((j_common_ptr) cinfo);

    MEMCOPY((*htblptr)->bits, bits, SIZEOF((*htblptr)->bits));
    MEMCOPY((*htblptr)->huffval, huffval, SIZEOF((*htblptr)->huffval));
  }

  if (length != 0)
    ERREXIT(cinfo, JERR_BAD_LENGTH);

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


LOCAL(boolean)
get_dqt (j_decompress_ptr cinfo)
/* Process a DQT marker */
{
  INT32 length;
  int n, i, prec;
  unsigned int tmp;
  JQUANT_TBL *quant_ptr;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);
  length -= 2;

  while (length > 0) {
    jdar_INPUT_BYTE(cinfo, n, return FALSE);
    prec = n >> 4;
    n &= 0x0F;

    TRACEMS2(cinfo, 1, JTRC_DQT, n, prec);

    if (n >= NUM_QUANT_TBLS)
      ERREXIT1(cinfo, JERR_DQT_INDEX, n);

    if (cinfo->quant_tbl_ptrs[n] == NULL)
      cinfo->quant_tbl_ptrs[n] = jpeg_alloc_quant_table((j_common_ptr) cinfo);
    quant_ptr = cinfo->quant_tbl_ptrs[n];

    for (i = 0; i < DCTSIZE2; i++) {
      if (prec)
    jdar_INPUT_2BYTES(cinfo, tmp, return FALSE);
      else
    jdar_INPUT_BYTE(cinfo, tmp, return FALSE);
      /* We convert the zigzag-order table to natural array order. */
      quant_ptr->quantval[jpeg_natural_order[i]] = (UINT16) tmp;
    }

    if (cinfo->err->trace_level >= 2) {
      for (i = 0; i < DCTSIZE2; i += 8) {
    TRACEMS8(cinfo, 2, JTRC_QUANTVALS,
         quant_ptr->quantval[i],   quant_ptr->quantval[i+1],
         quant_ptr->quantval[i+2], quant_ptr->quantval[i+3],
         quant_ptr->quantval[i+4], quant_ptr->quantval[i+5],
         quant_ptr->quantval[i+6], quant_ptr->quantval[i+7]);
      }
    }

    length -= DCTSIZE2+1;
    if (prec) length -= DCTSIZE2;
  }

  if (length != 0)
    ERREXIT(cinfo, JERR_BAD_LENGTH);

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


LOCAL(boolean)
get_dri (j_decompress_ptr cinfo)
/* Process a DRI marker */
{
  INT32 length;
  unsigned int tmp;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);

  if (length != 4)
    ERREXIT(cinfo, JERR_BAD_LENGTH);

  jdar_INPUT_2BYTES(cinfo, tmp, return FALSE);

  TRACEMS1(cinfo, 1, JTRC_DRI, tmp);

  cinfo->restart_interval = tmp;

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


/*
 * Routines for processing APPn and COM markers.
 * These are either saved in memory or discarded, per application request.
 * APP0 and APP14 are specially checked to see if they are
 * JFIF and Adobe markers, respectively.
 */

#define APP0_DATA_LEN   14  /* Length of interesting data in APP0 */
#define APP14_DATA_LEN  12  /* Length of interesting data in APP14 */
#define APPN_DATA_LEN   14  /* Must be the largest of the above!! */


LOCAL(void)
examine_app0 (j_decompress_ptr cinfo, JOCTET FAR * data,
          unsigned int datalen, INT32 remaining)
/* Examine first few bytes from an APP0.
 * Take appropriate action if it is a JFIF marker.
 * datalen is # of bytes at data[], remaining is length of rest of marker data.
 */
{
  INT32 totallen = (INT32) datalen + remaining;

  if (datalen >= APP0_DATA_LEN &&
      GETJOCTET(data[0]) == 0x4A &&
      GETJOCTET(data[1]) == 0x46 &&
      GETJOCTET(data[2]) == 0x49 &&
      GETJOCTET(data[3]) == 0x46 &&
      GETJOCTET(data[4]) == 0) {
    /* Found JFIF APP0 marker: save info */
    cinfo->saw_JFIF_marker = TRUE;
    cinfo->JFIF_major_version = GETJOCTET(data[5]);
    cinfo->JFIF_minor_version = GETJOCTET(data[6]);
    cinfo->density_unit = GETJOCTET(data[7]);
    cinfo->X_density = (GETJOCTET(data[8]) << 8) + GETJOCTET(data[9]);
    cinfo->Y_density = (GETJOCTET(data[10]) << 8) + GETJOCTET(data[11]);
    /* Check version.
     * Major version must be 1, anything else signals an incompatible change.
     * (We used to treat this as an error, but now it's a nonfatal warning,
     * because some bozo at Hijaak couldn't read the spec.)
     * Minor version should be 0..2, but process anyway if newer.
     */
    if (cinfo->JFIF_major_version != 1)
      WARNMS2(cinfo, JWRN_JFIF_MAJOR,
          cinfo->JFIF_major_version, cinfo->JFIF_minor_version);
    /* Generate trace messages */
    TRACEMS5(cinfo, 1, JTRC_JFIF,
         cinfo->JFIF_major_version, cinfo->JFIF_minor_version,
         cinfo->X_density, cinfo->Y_density, cinfo->density_unit);
    /* Validate thumbnail dimensions and issue appropriate messages */
    if (GETJOCTET(data[12]) | GETJOCTET(data[13]))
      TRACEMS2(cinfo, 1, JTRC_JFIF_THUMBNAIL,
           GETJOCTET(data[12]), GETJOCTET(data[13]));
    totallen -= APP0_DATA_LEN;
    if (totallen !=
    ((INT32)GETJOCTET(data[12]) * (INT32)GETJOCTET(data[13]) * (INT32) 3))
      TRACEMS1(cinfo, 1, JTRC_JFIF_BADTHUMBNAILSIZE, (int) totallen);
  } else if (datalen >= 6 &&
      GETJOCTET(data[0]) == 0x4A &&
      GETJOCTET(data[1]) == 0x46 &&
      GETJOCTET(data[2]) == 0x58 &&
      GETJOCTET(data[3]) == 0x58 &&
      GETJOCTET(data[4]) == 0) {
    /* Found JFIF "JFXX" extension APP0 marker */
    /* The library doesn't actually do anything with these,
     * but we try to produce a helpful trace message.
     */
    switch (GETJOCTET(data[5])) {
    case 0x10:
      TRACEMS1(cinfo, 1, JTRC_THUMB_JPEG, (int) totallen);
      break;
    case 0x11:
      TRACEMS1(cinfo, 1, JTRC_THUMB_PALETTE, (int) totallen);
      break;
    case 0x13:
      TRACEMS1(cinfo, 1, JTRC_THUMB_RGB, (int) totallen);
      break;
    default:
      TRACEMS2(cinfo, 1, JTRC_JFIF_EXTENSION,
           GETJOCTET(data[5]), (int) totallen);
      break;
    }
  } else {
    /* Start of APP0 does not match "JFIF" or "JFXX", or too short */
    TRACEMS1(cinfo, 1, JTRC_APP0, (int) totallen);
  }
}


LOCAL(void)
examine_app14 (j_decompress_ptr cinfo, JOCTET FAR * data,
           unsigned int datalen, INT32 remaining)
/* Examine first few bytes from an APP14.
 * Take appropriate action if it is an Adobe marker.
 * datalen is # of bytes at data[], remaining is length of rest of marker data.
 */
{
  unsigned int version, flags0, flags1, transform;

  if (datalen >= APP14_DATA_LEN &&
      GETJOCTET(data[0]) == 0x41 &&
      GETJOCTET(data[1]) == 0x64 &&
      GETJOCTET(data[2]) == 0x6F &&
      GETJOCTET(data[3]) == 0x62 &&
      GETJOCTET(data[4]) == 0x65) {
    /* Found Adobe APP14 marker */
    version = (GETJOCTET(data[5]) << 8) + GETJOCTET(data[6]);
    flags0 = (GETJOCTET(data[7]) << 8) + GETJOCTET(data[8]);
    flags1 = (GETJOCTET(data[9]) << 8) + GETJOCTET(data[10]);
    transform = GETJOCTET(data[11]);
    TRACEMS4(cinfo, 1, JTRC_ADOBE, version, flags0, flags1, transform);
    cinfo->saw_Adobe_marker = TRUE;
    cinfo->Adobe_transform = (UINT8) transform;
  } else {
    /* Start of APP14 does not match "Adobe", or too short */
    TRACEMS1(cinfo, 1, JTRC_APP14, (int) (datalen + remaining));
  }
}


METHODDEF(boolean)
get_interesting_appn (j_decompress_ptr cinfo)
/* Process an APP0 or APP14 marker without saving it */
{
  INT32 length;
  JOCTET b[APPN_DATA_LEN];
  unsigned int i, numtoread;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);
  length -= 2;

  /* get the interesting part of the marker data */
  if (length >= APPN_DATA_LEN)
    numtoread = APPN_DATA_LEN;
  else if (length > 0)
    numtoread = (unsigned int) length;
  else
    numtoread = 0;
  for (i = 0; i < numtoread; i++)
    jdar_INPUT_BYTE(cinfo, b[i], return FALSE);
  length -= numtoread;

  /* process it */
  switch (cinfo->unread_marker) {
  case M_APP0:
    examine_app0(cinfo, (JOCTET FAR *) b, numtoread, length);
    break;
  case M_APP14:
    examine_app14(cinfo, (JOCTET FAR *) b, numtoread, length);
    break;
  default:
    /* can't get here unless jpeg_save_markers chooses wrong processor */
    ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, cinfo->unread_marker);
    break;
  }

  /* skip any remaining data -- could be lots */
  jdar_INPUT_SYNC(cinfo);
  if (length > 0)
    (*cinfo->src->skip_input_data) (cinfo, (long) length);

  return TRUE;
}


#ifdef SAVE_MARKERS_SUPPORTED

METHODDEF(boolean)
save_marker (j_decompress_ptr cinfo)
/* Save an APPn or COM marker into the marker list */
{
  my_marker_ptr marker = (my_marker_ptr) cinfo->marker;
  jpeg_saved_marker_ptr cur_marker = marker->cur_marker;
  unsigned int bytes_read, data_length;
  JOCTET FAR * data;
  INT32 length = 0;
  jdar_INPUT_VARS(cinfo);

  if (cur_marker == NULL) {
    /* begin reading a marker */
    jdar_INPUT_2BYTES(cinfo, length, return FALSE);
    length -= 2;
    if (length >= 0) {      /* watch out for bogus length word */
      /* figure out how much we want to save */
      unsigned int limit;
      if (cinfo->unread_marker == (int) M_COM)
    limit = marker->length_limit_COM;
      else
    limit = marker->length_limit_APPn[cinfo->unread_marker - (int) M_APP0];
      if ((unsigned int) length < limit)
    limit = (unsigned int) length;
      /* allocate and initialize the marker item */
      cur_marker = (jpeg_saved_marker_ptr)
    (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                    SIZEOF(struct jpeg_marker_struct) + limit);
      cur_marker->next = NULL;
      cur_marker->marker = (UINT8) cinfo->unread_marker;
      cur_marker->original_length = (unsigned int) length;
      cur_marker->data_length = limit;
      /* data area is just beyond the jpeg_marker_struct */
      data = cur_marker->data = (JOCTET FAR *) (cur_marker + 1);
      marker->cur_marker = cur_marker;
      marker->bytes_read = 0;
      bytes_read = 0;
      data_length = limit;
    } else {
      /* deal with bogus length word */
      bytes_read = data_length = 0;
      data = NULL;
    }
  } else {
    /* resume reading a marker */
    bytes_read = marker->bytes_read;
    data_length = cur_marker->data_length;
    data = cur_marker->data + bytes_read;
  }

  while (bytes_read < data_length) {
    jdar_INPUT_SYNC(cinfo);     /* move the restart point to here */
    marker->bytes_read = bytes_read;
    /* If there's not at least one byte in buffer, suspend */
    jdar_MAKE_BYTE_AVAIL(cinfo, return FALSE);
    /* Copy bytes with reasonable rapidity */
    while (bytes_read < data_length && bytes_in_buffer > 0) {
      *data++ = *next_input_byte++;
      bytes_in_buffer--;
      bytes_read++;
    }
  }

  /* Done reading what we want to read */
  if (cur_marker != NULL) { /* will be NULL if bogus length word */
    /* Add new marker to end of list */
    if (cinfo->marker_list == NULL) {
      cinfo->marker_list = cur_marker;
    } else {
      jpeg_saved_marker_ptr prev = cinfo->marker_list;
      while (prev->next != NULL)
    prev = prev->next;
      prev->next = cur_marker;
    }
    /* Reset pointer & calc remaining data length */
    data = cur_marker->data;
    length = cur_marker->original_length - data_length;
  }
  /* Reset to initial state for next marker */
  marker->cur_marker = NULL;

  /* Process the marker if interesting; else just make a generic trace msg */
  switch (cinfo->unread_marker) {
  case M_APP0:
    examine_app0(cinfo, data, data_length, length);
    break;
  case M_APP14:
    examine_app14(cinfo, data, data_length, length);
    break;
  default:
    TRACEMS2(cinfo, 1, JTRC_MISC_MARKER, cinfo->unread_marker,
         (int) (data_length + length));
    break;
  }

  /* skip any remaining data -- could be lots */
  jdar_INPUT_SYNC(cinfo);       /* do before skip_input_data */
  if (length > 0)
    (*cinfo->src->skip_input_data) (cinfo, (long) length);

  return TRUE;
}

#endif /* SAVE_MARKERS_SUPPORTED */


METHODDEF(boolean)
skip_variable (j_decompress_ptr cinfo)
/* Skip over an unknown or uninteresting variable-length marker */
{
  INT32 length;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_2BYTES(cinfo, length, return FALSE);
  length -= 2;

  TRACEMS2(cinfo, 1, JTRC_MISC_MARKER, cinfo->unread_marker, (int) length);

  jdar_INPUT_SYNC(cinfo);       /* do before skip_input_data */
  if (length > 0)
    (*cinfo->src->skip_input_data) (cinfo, (long) length);

  return TRUE;
}


/*
 * Find the next JPEG marker, save it in cinfo->unread_marker.
 * Returns FALSE if had to suspend before reaching a marker;
 * in that case cinfo->unread_marker is unchanged.
 *
 * Note that the result might not be a valid marker code,
 * but it will never be 0 or FF.
 */

LOCAL(boolean)
next_marker (j_decompress_ptr cinfo)
{
  int c;
  jdar_INPUT_VARS(cinfo);

  for (;;) {
    jdar_INPUT_BYTE(cinfo, c, return FALSE);
    /* Skip any non-FF bytes.
     * This may look a bit inefficient, but it will not occur in a valid file.
     * We sync after each discarded byte so that a suspending data source
     * can discard the byte from its buffer.
     */
    while (c != 0xFF) {
      cinfo->marker->discarded_bytes++;
      jdar_INPUT_SYNC(cinfo);
      jdar_INPUT_BYTE(cinfo, c, return FALSE);
    }
    /* This loop swallows any duplicate FF bytes.  Extra FFs are legal as
     * pad bytes, so don't count them in discarded_bytes.  We assume there
     * will not be so many consecutive FF bytes as to overflow a suspending
     * data source's input buffer.
     */
    do {
      jdar_INPUT_BYTE(cinfo, c, return FALSE);
    } while (c == 0xFF);
    if (c != 0)
      break;            /* found a valid marker, exit loop */
    /* Reach here if we found a stuffed-zero data sequence (FF/00).
     * Discard it and loop back to try again.
     */
    cinfo->marker->discarded_bytes += 2;
    jdar_INPUT_SYNC(cinfo);
  }

  if (cinfo->marker->discarded_bytes != 0) {
    WARNMS2(cinfo, JWRN_EXTRANEOUS_DATA, cinfo->marker->discarded_bytes, c);
    cinfo->marker->discarded_bytes = 0;
  }

  cinfo->unread_marker = c;

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


LOCAL(boolean)
first_marker (j_decompress_ptr cinfo)
/* Like next_marker, but used to obtain the initial SOI marker. */
/* For this marker, we do not allow preceding garbage or fill; otherwise,
 * we might well scan an entire input file before realizing it ain't JPEG.
 * If an application wants to process non-JFIF files, it must seek to the
 * SOI before calling the JPEG library.
 */
{
  int c, c2;
  jdar_INPUT_VARS(cinfo);

  jdar_INPUT_BYTE(cinfo, c, return FALSE);
  jdar_INPUT_BYTE(cinfo, c2, return FALSE);
  if (c != 0xFF || c2 != (int) M_SOI)
    ERREXIT2(cinfo, JERR_NO_SOI, c, c2);

  cinfo->unread_marker = c2;

  jdar_INPUT_SYNC(cinfo);
  return TRUE;
}


/*
 * Read markers until SOS or EOI.
 *
 * Returns same codes as are defined for jpeg_consume_input:
 * JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI.
 */

METHODDEF(int)
read_markers (j_decompress_ptr cinfo)
{
  /* Outer loop repeats once for each marker. */
  for (;;) {
    /* Collect the marker proper, unless we already did. */
    /* NB: first_marker() enforces the requirement that SOI appear first. */
    if (cinfo->unread_marker == 0) {
      if (! cinfo->marker->saw_SOI) {
    if (! first_marker(cinfo))
      return JPEG_SUSPENDED;
      } else {
    if (! next_marker(cinfo))
      return JPEG_SUSPENDED;
      }
    }
    /* At this point cinfo->unread_marker contains the marker code and the
     * input point is just past the marker proper, but before any parameters.
     * A suspension will cause us to return with this state still true.
     */
    switch (cinfo->unread_marker) {
    case M_SOI:
      if (! get_soi(cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_SOF0:        /* Baseline */
    case M_SOF1:        /* Extended sequential, Huffman */
      if (! get_sof(cinfo, FALSE, FALSE))
    return JPEG_SUSPENDED;
      break;

    case M_SOF2:        /* Progressive, Huffman */
      if (! get_sof(cinfo, TRUE, FALSE))
    return JPEG_SUSPENDED;
      break;

    case M_SOF9:        /* Extended sequential, arithmetic */
      if (! get_sof(cinfo, FALSE, TRUE))
    return JPEG_SUSPENDED;
      break;

    case M_SOF10:       /* Progressive, arithmetic */
      if (! get_sof(cinfo, TRUE, TRUE))
    return JPEG_SUSPENDED;
      break;

    /* Currently unsupported SOFn types */
    case M_SOF3:        /* Lossless, Huffman */
    case M_SOF5:        /* Differential sequential, Huffman */
    case M_SOF6:        /* Differential progressive, Huffman */
    case M_SOF7:        /* Differential lossless, Huffman */
    case M_JPG:         /* Reserved for JPEG extensions */
    case M_SOF11:       /* Lossless, arithmetic */
    case M_SOF13:       /* Differential sequential, arithmetic */
    case M_SOF14:       /* Differential progressive, arithmetic */
    case M_SOF15:       /* Differential lossless, arithmetic */
      ERREXIT1(cinfo, JERR_SOF_UNSUPPORTED, cinfo->unread_marker);
      break;

    case M_SOS:
      if (! get_sos(cinfo))
    return JPEG_SUSPENDED;
      cinfo->unread_marker = 0; /* processed the marker */
      return JPEG_REACHED_SOS;

    case M_EOI:
      TRACEMS(cinfo, 1, JTRC_EOI);
      cinfo->unread_marker = 0; /* processed the marker */
      return JPEG_REACHED_EOI;

    case M_DAC:
      if (! jdar_get_dac(cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_DHT:
      if (! get_dht(cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_DQT:
      if (! get_dqt(cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_DRI:
      if (! get_dri(cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_APP0:
    case M_APP1:
    case M_APP2:
    case M_APP3:
    case M_APP4:
    case M_APP5:
    case M_APP6:
    case M_APP7:
    case M_APP8:
    case M_APP9:
    case M_APP10:
    case M_APP11:
    case M_APP12:
    case M_APP13:
    case M_APP14:
    case M_APP15:
      if (! (*((my_marker_ptr) cinfo->marker)->process_APPn[
        cinfo->unread_marker - (int) M_APP0]) (cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_COM:
      if (! (*((my_marker_ptr) cinfo->marker)->process_COM) (cinfo))
    return JPEG_SUSPENDED;
      break;

    case M_RST0:        /* these are all parameterless */
    case M_RST1:
    case M_RST2:
    case M_RST3:
    case M_RST4:
    case M_RST5:
    case M_RST6:
    case M_RST7:
    case M_TEM:
      TRACEMS1(cinfo, 1, JTRC_PARMLESS_MARKER, cinfo->unread_marker);
      break;

    case M_DNL:         /* Ignore DNL ... perhaps the wrong thing */
      if (! skip_variable(cinfo))
    return JPEG_SUSPENDED;
      break;

    default:            /* must be DHP, EXP, JPGn, or RESn */
      /* For now, we treat the reserved markers as fatal errors since they are
       * likely to be used to signal incompatible JPEG Part 3 extensions.
       * Once the JPEG 3 version-number marker is well defined, this code
       * ought to change!
       */
      ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, cinfo->unread_marker);
      break;
    }
    /* Successfully processed marker, so reset state variable */
    cinfo->unread_marker = 0;
  } /* end loop */
}


/*
 * Read a restart marker, which is expected to appear next in the datastream;
 * if the marker is not there, take appropriate recovery action.
 * Returns FALSE if suspension is required.
 *
 * This is called by the entropy decoder after it has read an appropriate
 * number of MCUs.  cinfo->unread_marker may be nonzero if the entropy decoder
 * has already read a marker from the data source.  Under normal conditions
 * cinfo->unread_marker will be reset to 0 before returning; if not reset,
 * it holds a marker which the decoder will be unable to read past.
 */

METHODDEF(boolean)
read_restart_marker (j_decompress_ptr cinfo)
{
  /* Obtain a marker unless we already did. */
  /* Note that next_marker will complain if it skips any data. */
  if (cinfo->unread_marker == 0) {
    if (! next_marker(cinfo))
      return FALSE;
  }

  if (cinfo->unread_marker ==
      ((int) M_RST0 + cinfo->marker->next_restart_num)) {
    /* Normal case --- swallow the marker and let entropy decoder continue */
    TRACEMS1(cinfo, 3, JTRC_RST, cinfo->marker->next_restart_num);
    cinfo->unread_marker = 0;
  } else {
    /* Uh-oh, the restart markers have been messed up. */
    /* Let the data source manager determine how to resync. */
    if (! (*cinfo->src->resync_to_restart) (cinfo,
                        cinfo->marker->next_restart_num))
      return FALSE;
  }

  /* Update next-restart state */
  cinfo->marker->next_restart_num = (cinfo->marker->next_restart_num + 1) & 7;

  return TRUE;
}


/*
 * This is the default resync_to_restart method for data source managers
 * to use if they don't have any better approach.  Some data source managers
 * may be able to back up, or may have additional knowledge about the data
 * which permits a more intelligent recovery strategy; such managers would
 * presumably supply their own resync method.
 *
 * read_restart_marker calls resync_to_restart if it finds a marker other than
 * the restart marker it was expecting.  (This code is *not* used unless
 * a nonzero restart interval has been declared.)  cinfo->unread_marker is
 * the marker code actually found (might be anything, except 0 or FF).
 * The desired restart marker number (0..7) is passed as a parameter.
 * This routine is supposed to apply whatever error recovery strategy seems
 * appropriate in order to position the input stream to the next data segment.
 * Note that cinfo->unread_marker is treated as a marker appearing before
 * the current data-source input point; usually it should be reset to zero
 * before returning.
 * Returns FALSE if suspension is required.
 *
 * This implementation is substantially constrained by wanting to treat the
 * input as a data stream; this means we can't back up.  Therefore, we have
 * only the following actions to work with:
 *   1. Simply discard the marker and let the entropy decoder resume at next
 *      byte of file.
 *   2. Read forward until we find another marker, discarding intervening
 *      data.  (In theory we could look ahead within the current bufferload,
 *      without having to discard data if we don't find the desired marker.
 *      This idea is not implemented here, in part because it makes behavior
 *      dependent on buffer size and chance buffer-boundary positions.)
 *   3. Leave the marker unread (by failing to zero cinfo->unread_marker).
 *      This will cause the entropy decoder to process an empty data segment,
 *      inserting dummy zeroes, and then we will reprocess the marker.
 *
 * #2 is appropriate if we think the desired marker lies ahead, while #3 is
 * appropriate if the found marker is a future restart marker (indicating
 * that we have missed the desired restart marker, probably because it got
 * corrupted).
 * We apply #2 or #3 if the found marker is a restart marker no more than
 * two counts behind or ahead of the expected one.  We also apply #2 if the
 * found marker is not a legal JPEG marker code (it's certainly bogus data).
 * If the found marker is a restart marker more than 2 counts away, we do #1
 * (too much risk that the marker is erroneous; with luck we will be able to
 * resync at some future point).
 * For any valid non-restart JPEG marker, we apply #3.  This keeps us from
 * overrunning the end of a scan.  An implementation limited to single-scan
 * files might find it better to apply #2 for markers other than EOI, since
 * any other marker would have to be bogus data in that case.
 */

GLOBAL(boolean)
jpeg_resync_to_restart (j_decompress_ptr cinfo, int desired)
{
  int marker = cinfo->unread_marker;
  int action = 1;

  /* Always put up a warning. */
  WARNMS2(cinfo, JWRN_MUST_RESYNC, marker, desired);

  /* Outer loop handles repeated decision after scanning forward. */
  for (;;) {
    if (marker < (int) M_SOF0)
      action = 2;       /* invalid marker */
    else if (marker < (int) M_RST0 || marker > (int) M_RST7)
      action = 3;       /* valid non-restart marker */
    else {
      if (marker == ((int) M_RST0 + ((desired+1) & 7)) ||
      marker == ((int) M_RST0 + ((desired+2) & 7)))
    action = 3;     /* one of the next two expected restarts */
      else if (marker == ((int) M_RST0 + ((desired-1) & 7)) ||
           marker == ((int) M_RST0 + ((desired-2) & 7)))
    action = 2;     /* a prior restart, so advance */
      else
    action = 1;     /* desired restart or too far away */
    }
    TRACEMS2(cinfo, 4, JTRC_RECOVERY_ACTION, marker, action);
    switch (action) {
    case 1:
      /* Discard marker and let entropy decoder resume processing. */
      cinfo->unread_marker = 0;
      return TRUE;
    case 2:
      /* Scan to the next marker, and repeat the decision loop. */
      if (! next_marker(cinfo))
    return FALSE;
      marker = cinfo->unread_marker;
      break;
    case 3:
      /* Return without advancing past this marker. */
      /* Entropy decoder will be forced to process an empty segment. */
      return TRUE;
    }
  } /* end loop */
}


/*
 * Reset marker processing state to begin a fresh datastream.
 */

METHODDEF(void)
reset_marker_reader (j_decompress_ptr cinfo)
{
  my_marker_ptr marker = (my_marker_ptr) cinfo->marker;

  cinfo->comp_info = NULL;      /* until allocated by get_sof */
  cinfo->input_scan_number = 0;     /* no SOS seen yet */
  cinfo->unread_marker = 0;     /* no pending marker */
  marker->pub.saw_SOI = FALSE;      /* set internal state too */
  marker->pub.saw_SOF = FALSE;
  marker->pub.discarded_bytes = 0;
  marker->cur_marker = NULL;
}


/*
 * Initialize the marker reader module.
 * This is called only once, when the decompression object is created.
 */

GLOBAL(void)
jinit_marker_reader (j_decompress_ptr cinfo)
{
  my_marker_ptr marker;
  int i;

  /* Create subobject in permanent pool */
  marker = (my_marker_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
                SIZEOF(my_marker_reader));
  cinfo->marker = (struct jpeg_marker_reader *) marker;
  /* Initialize public method pointers */
  marker->pub.reset_marker_reader = reset_marker_reader;
  marker->pub.read_markers = read_markers;
  marker->pub.read_restart_marker = read_restart_marker;
  /* Initialize COM/APPn processing.
   * By default, we examine and then discard APP0 and APP14,
   * but simply discard COM and all other APPn.
   */
  marker->process_COM = skip_variable;
  marker->length_limit_COM = 0;
  for (i = 0; i < 16; i++) {
    marker->process_APPn[i] = skip_variable;
    marker->length_limit_APPn[i] = 0;
  }
  marker->process_APPn[0] = get_interesting_appn;
  marker->process_APPn[14] = get_interesting_appn;
  /* Reset marker processing state */
  reset_marker_reader(cinfo);
}


/*
 * Control saving of COM and APPn markers into marker_list.
 */

#ifdef SAVE_MARKERS_SUPPORTED

GLOBAL(void)
jpeg_save_markers (j_decompress_ptr cinfo, int marker_code,
           unsigned int length_limit)
{
  my_marker_ptr marker = (my_marker_ptr) cinfo->marker;
  long maxlength;
  jpeg_marker_parser_method processor;

  /* Length limit mustn't be larger than what we can allocate
   * (should only be a concern in a 16-bit environment).
   */
  maxlength = cinfo->mem->max_alloc_chunk - SIZEOF(struct jpeg_marker_struct);
  if (((long) length_limit) > maxlength)
    length_limit = (unsigned int) maxlength;

  /* Choose processor routine to use.
   * APP0/APP14 have special requirements.
   */
  if (length_limit) {
    processor = save_marker;
    /* If saving APP0/APP14, save at least enough for our internal use. */
    if (marker_code == (int) M_APP0 && length_limit < APP0_DATA_LEN)
      length_limit = APP0_DATA_LEN;
    else if (marker_code == (int) M_APP14 && length_limit < APP14_DATA_LEN)
      length_limit = APP14_DATA_LEN;
  } else {
    processor = skip_variable;
    /* If discarding APP0/APP14, use our regular on-the-fly processor. */
    if (marker_code == (int) M_APP0 || marker_code == (int) M_APP14)
      processor = get_interesting_appn;
  }

  if (marker_code == (int) M_COM) {
    marker->process_COM = processor;
    marker->length_limit_COM = length_limit;
  } else if (marker_code >= (int) M_APP0 && marker_code <= (int) M_APP15) {
    marker->process_APPn[marker_code - (int) M_APP0] = processor;
    marker->length_limit_APPn[marker_code - (int) M_APP0] = length_limit;
  } else
    ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, marker_code);
}

#endif /* SAVE_MARKERS_SUPPORTED */


/*
 * Install a special processing method for COM or APPn markers.
 */

GLOBAL(void)
jpeg_set_marker_processor (j_decompress_ptr cinfo, int marker_code,
               jpeg_marker_parser_method routine)
{
  my_marker_ptr marker = (my_marker_ptr) cinfo->marker;

  if (marker_code == (int) M_COM)
    marker->process_COM = routine;
  else if (marker_code >= (int) M_APP0 && marker_code <= (int) M_APP15)
    marker->process_APPn[marker_code - (int) M_APP0] = routine;
  else
    ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, marker_code);
}
/*
 * jdhuff.c
 *
 * Copyright (C) 1991-1997, 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 Huffman entropy decoding routines.
 *
 * 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"
//#include "jdhuff.h"       /* Declarations shared with jdphuff.c */


/*
 * 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 {
  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 jdu_jdu_ASSIGN_STATE(dest,src)  ((dest) = (src))
#else
#if MAX_COMPS_IN_SCAN == 4
#define jdu_jdu_ASSIGN_STATE(dest,src)  \
    ((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 */

  /* 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 */
  boolean dc_needed[D_MAX_BLOCKS_IN_MCU];
  boolean ac_needed[D_MAX_BLOCKS_IN_MCU];
} huff_entropy_decoder;

typedef huff_entropy_decoder * huff_entropy_ptr;


/*
 * 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;
  jpeg_component_info * compptr;

  /* 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);

  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) {
      entropy->dc_needed[blkn] = TRUE;
      /* we don't need the ACs if producing a 1/8th-size image */
      entropy->ac_needed[blkn] = (compptr->DCT_scaled_size > 1);
    } else {
      entropy->dc_needed[blkn] = entropy->ac_needed[blkn] = FALSE;
    }
  }

  /* 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;
}


/*
 * Compute the derived values for a Huffman table.
 * This routine also performs some validation checks on the table.
 *
 * Note this is also used by jdphuff.c.
 */

GLOBAL(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 (shared with jdphuff.c).
 * See jdhuff.h for info about usage.
 * 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


GLOBAL(boolean)
jpeg_fill_bit_buffer (bitread_working_state * state,
              bit_buf_type get_buffer, 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) */
  const JOCTET * next_input_byte = state->next_input_byte;
  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) {
      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;
}


/*
 * Out-of-line code for Huffman code decoding.
 * See jdhuff.h for info about usage.
 */

GLOBAL(int)
jpeg_huff_decode (bitread_working_state * state,
          bit_buf_type get_buffer, int bits_left,
          d_derived_tbl * htbl, int min_bits)
{
  int l = min_bits;
  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]) ];
}


/*
 * Figure F.12: extend sign bit.
 * On some machines, a shift and add will be faster than a table lookup.
 */

#ifdef AVOID_TABLES

#define jdu_jdu_HUFF_EXTEND(x,s)  ((x) < (1<<((s)-1)) ? (x) + (((-1)<<(s)) + 1) : (x))

#else

#define jdu_jdu_HUFF_EXTEND(x,s)  ((x) < extend_test[s] ? (x) + extend_offset[s] : (x))

static const int extend_test[16] =   /* entry n is 2**(n-1) */
  { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080,
    0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };

static const int extend_offset[16] = /* entry n is (-1 << n) + 1 */
  { 0, -(1<<1) + 1, -(1<<2) + 1, -(1<<3) + 1, -(1<<4) + 1,
    -(1<<5) + 1, -(1<<6) + 1, -(1<<7) + 1, -(1<<8) + 1,
    -(1<<9) + 1, -(1<<10) + 1, -(1<<11) + 1, -(1<<12) + 1,
    -(1<<13) + 1, -(1<<14) + 1, -(1<<15) + 1 };

#endif /* AVOID_TABLES */


/*
 * 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;

  /* 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;
}


/*
 * Decode and return 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 HAS BEEN ZEROED BY THE CALLER.
 * (Wholesale zeroing is usually a little faster than retail...)
 *
 * Returns 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
 * this module, since we'll just re-assign them on the next call.)
 */

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);
    jdu_jdu_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 * dctbl = entropy->dc_cur_tbls[blkn];
      d_derived_tbl * actbl = entropy->ac_cur_tbls[blkn];
      int s, k, r;

      /* Decode a single block's worth of coefficients */

      /* Section F.2.2.1: decode the DC coefficient difference */
      HUFF_DECODE(s, br_state, dctbl, return FALSE, label1);
      if (s) {
    CHECK_BIT_BUFFER(br_state, s, return FALSE);
    r = GET_BITS(s);
    s = jdu_jdu_HUFF_EXTEND(r, s);
      }

      if (entropy->dc_needed[blkn]) {
    /* Convert DC difference to actual value, update last_dc_val */
    int ci = cinfo->MCU_membership[blkn];
    s += state.last_dc_val[ci];
    state.last_dc_val[ci] = s;
    /* Output the DC coefficient (assumes jpeg_natural_order[0] = 0) */
    (*block)[0] = (JCOEF) s;
      }

      if (entropy->ac_needed[blkn]) {

    /* Section F.2.2.2: decode the AC coefficients */
    /* Since zeroes are skipped, output area must be cleared beforehand */
    for (k = 1; k < DCTSIZE2; k++) {
      HUFF_DECODE(s, br_state, actbl, 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 = jdu_jdu_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)
          break;
        k += 15;
      }
    }

      } else {

    /* Section F.2.2.2: decode the AC coefficients */
    /* In this path we just discard the values */
    for (k = 1; k < DCTSIZE2; k++) {
      HUFF_DECODE(s, br_state, actbl, 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;
      }
    }

      }
    }

    /* Completed MCU, so update state */
    BITREAD_SAVE_STATE(cinfo,entropy->bitstate);
    jdu_jdu_ASSIGN_STATE(entropy->saved, state);
  }

  /* Account for restart interval (no-op if not using restarts) */
  entropy->restarts_to_go--;

  return TRUE;
}


/*
 * 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;
  entropy->pub.decode_mcu = decode_mcu;

  /* Mark tables unallocated */
  for (i = 0; i < NUM_HUFF_TBLS; i++) {
    entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL;
  }
}
/*
 * jdmainct.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 main buffer controller for decompression.
 * The main buffer lies between the JPEG decompressor proper and the
 * post-processor; it holds downsampled data in the JPEG colorspace.
 *
 * Note that this code is bypassed in raw-data mode, since the application
 * supplies the equivalent of the main buffer in that case.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/*
 * In the current system design, the main buffer need never be a full-image
 * buffer; any full-height buffers will be found inside the coefficient or
 * postprocessing controllers.  Nonetheless, the main controller is not
 * trivial.  Its responsibility is to provide context rows for upsampling/
 * rescaling, and doing this in an efficient fashion is a bit tricky.
 *
 * Postprocessor input data is counted in "row groups".  A row group
 * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
 * sample rows of each component.  (We require DCT_scaled_size values to be
 * chosen such that these numbers are integers.  In practice DCT_scaled_size
 * values will likely be powers of two, so we actually have the stronger
 * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.)
 * Upsampling will typically produce max_v_samp_factor pixel rows from each
 * row group (times any additional scale factor that the upsampler is
 * applying).
 *
 * The coefficient controller will deliver data to us one iMCU row at a time;
 * each iMCU row contains v_samp_factor * DCT_scaled_size sample rows, or
 * exactly min_DCT_scaled_size row groups.  (This amount of data corresponds
 * to one row of MCUs when the image is fully interleaved.)  Note that the
 * number of sample rows varies across components, but the number of row
 * groups does not.  Some garbage sample rows may be included in the last iMCU
 * row at the bottom of the image.
 *
 * Depending on the vertical scaling algorithm used, the upsampler may need
 * access to the sample row(s) above and below its current input row group.
 * The upsampler is required to set need_context_rows TRUE at global selection
 * time if so.  When need_context_rows is FALSE, this controller can simply
 * obtain one iMCU row at a time from the coefficient controller and dole it
 * out as row groups to the postprocessor.
 *
 * When need_context_rows is TRUE, this controller guarantees that the buffer
 * passed to postprocessing contains at least one row group's worth of samples
 * above and below the row group(s) being processed.  Note that the context
 * rows "above" the first passed row group appear at negative row offsets in
 * the passed buffer.  At the top and bottom of the image, the required
 * context rows are manufactured by duplicating the first or last real sample
 * row; this avoids having special cases in the upsampling inner loops.
 *
 * The amount of context is fixed at one row group just because that's a
 * convenient number for this controller to work with.  The existing
 * upsamplers really only need one sample row of context.  An upsampler
 * supporting arbitrary output rescaling might wish for more than one row
 * group of context when shrinking the image; tough, we don't handle that.
 * (This is justified by the assumption that downsizing will be handled mostly
 * by adjusting the DCT_scaled_size values, so that the actual scale factor at
 * the upsample step needn't be much less than one.)
 *
 * To provide the desired context, we have to retain the last two row groups
 * of one iMCU row while reading in the next iMCU row.  (The last row group
 * can't be processed until we have another row group for its below-context,
 * and so we have to save the next-to-last group too for its above-context.)
 * We could do this most simply by copying data around in our buffer, but
 * that'd be very slow.  We can avoid copying any data by creating a rather
 * strange pointer structure.  Here's how it works.  We allocate a workspace
 * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number
 * of row groups per iMCU row).  We create two sets of redundant pointers to
 * the workspace.  Labeling the physical row groups 0 to M+1, the synthesized
 * pointer lists look like this:
 *                   M+1                          M-1
 * master pointer --> 0         master pointer --> 0
 *                    1                            1
 *                   ...                          ...
 *                   M-3                          M-3
 *                   M-2                           M
 *                   M-1                          M+1
 *                    M                           M-2
 *                   M+1                          M-1
 *                    0                            0
 * We read alternate iMCU rows using each master pointer; thus the last two
 * row groups of the previous iMCU row remain un-overwritten in the workspace.
 * The pointer lists are set up so that the required context rows appear to
 * be adjacent to the proper places when we pass the pointer lists to the
 * upsampler.
 *
 * The above pictures describe the normal state of the pointer lists.
 * At top and bottom of the image, we diddle the pointer lists to duplicate
 * the first or last sample row as necessary (this is cheaper than copying
 * sample rows around).
 *
 * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1.  In that
 * situation each iMCU row provides only one row group so the buffering logic
 * must be different (eg, we must read two iMCU rows before we can emit the
 * first row group).  For now, we simply do not support providing context
 * rows when min_DCT_scaled_size is 1.  That combination seems unlikely to
 * be worth providing --- if someone wants a 1/8th-size preview, they probably
 * want it quick and dirty, so a context-free upsampler is sufficient.
 */


/* Private buffer controller object */

typedef struct {
  struct jpeg_d_main_controller pub; /* public fields */

  /* Pointer to allocated workspace (M or M+2 row groups). */
  JSAMPARRAY buffer[MAX_COMPONENTS];

  boolean buffer_full;      /* Have we gotten an iMCU row from decoder? */
  JDIMENSION rowgroup_ctr;  /* counts row groups output to postprocessor */

  /* Remaining fields are only used in the context case. */

  /* These are the master pointers to the funny-order pointer lists. */
  JSAMPIMAGE xbuffer[2];    /* pointers to weird pointer lists */

  int whichptr;         /* indicates which pointer set is now in use */
  int context_state;        /* process_data state machine status */
  JDIMENSION rowgroups_avail;   /* row groups available to postprocessor */
  JDIMENSION iMCU_row_ctr;  /* counts iMCU rows to detect image top/bot */
} my_main_controller;

typedef my_main_controller * my_main_ptr;

/* context_state values: */
#define CTX_PREPARE_FOR_IMCU    0   /* need to prepare for MCU row */
#define CTX_PROCESS_IMCU    1   /* feeding iMCU to postprocessor */
#define CTX_POSTPONED_ROW   2   /* feeding postponed row group */


/* Forward declarations */
METHODDEF(void) process_data_simple_main
    JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
         JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
METHODDEF(void) process_data_context_main
    JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
         JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
#ifdef QUANT_2PASS_SUPPORTED
METHODDEF(void) process_data_crank_post
    JPP((j_decompress_ptr cinfo, JSAMPARRAY output_buf,
         JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail));
#endif


LOCAL(void)
alloc_funny_pointers (j_decompress_ptr cinfo)
/* Allocate space for the funny pointer lists.
 * This is done only once, not once per pass.
 */
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;
  int ci, rgroup;
  int M = cinfo->min_DCT_scaled_size;
  jpeg_component_info *compptr;
  JSAMPARRAY xbuf;

  /* Get top-level space for component array pointers.
   * We alloc both arrays with one call to save a few cycles.
   */
  main_ptr->xbuffer[0] = (JSAMPIMAGE)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                cinfo->num_components * 2 * SIZEOF(JSAMPARRAY));
  main_ptr->xbuffer[1] = main_ptr->xbuffer[0] + cinfo->num_components;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
      cinfo->min_DCT_scaled_size; /* height of a row group of component */
    /* Get space for pointer lists --- M+4 row groups in each list.
     * We alloc both pointer lists with one call to save a few cycles.
     */
    xbuf = (JSAMPARRAY)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                  2 * (rgroup * (M + 4)) * SIZEOF(JSAMPROW));
    xbuf += rgroup;     /* want one row group at negative offsets */
    main_ptr->xbuffer[0][ci] = xbuf;
    xbuf += rgroup * (M + 4);
    main_ptr->xbuffer[1][ci] = xbuf;
  }
}


LOCAL(void)
make_funny_pointers (j_decompress_ptr cinfo)
/* Create the funny pointer lists discussed in the comments above.
 * The actual workspace is already allocated (in main_ptr->buffer),
 * and the space for the pointer lists is allocated too.
 * This routine just fills in the curiously ordered lists.
 * This will be repeated at the beginning of each pass.
 */
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;
  int ci, i, rgroup;
  int M = cinfo->min_DCT_scaled_size;
  jpeg_component_info *compptr;
  JSAMPARRAY buf, xbuf0, xbuf1;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
      cinfo->min_DCT_scaled_size; /* height of a row group of component */
    xbuf0 = main_ptr->xbuffer[0][ci];
    xbuf1 = main_ptr->xbuffer[1][ci];
    /* First copy the workspace pointers as-is */
    buf = main_ptr->buffer[ci];
    for (i = 0; i < rgroup * (M + 2); i++) {
      xbuf0[i] = xbuf1[i] = buf[i];
    }
    /* In the second list, put the last four row groups in swapped order */
    for (i = 0; i < rgroup * 2; i++) {
      xbuf1[rgroup*(M-2) + i] = buf[rgroup*M + i];
      xbuf1[rgroup*M + i] = buf[rgroup*(M-2) + i];
    }
    /* The wraparound pointers at top and bottom will be filled later
     * (see set_wraparound_pointers, below).  Initially we want the "above"
     * pointers to duplicate the first actual data line.  This only needs
     * to happen in xbuffer[0].
     */
    for (i = 0; i < rgroup; i++) {
      xbuf0[i - rgroup] = xbuf0[0];
    }
  }
}


LOCAL(void)
set_wraparound_pointers (j_decompress_ptr cinfo)
/* Set up the "wraparound" pointers at top and bottom of the pointer lists.
 * This changes the pointer list state from top-of-image to the normal state.
 */
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;
  int ci, i, rgroup;
  int M = cinfo->min_DCT_scaled_size;
  jpeg_component_info *compptr;
  JSAMPARRAY xbuf0, xbuf1;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
      cinfo->min_DCT_scaled_size; /* height of a row group of component */
    xbuf0 = main_ptr->xbuffer[0][ci];
    xbuf1 = main_ptr->xbuffer[1][ci];
    for (i = 0; i < rgroup; i++) {
      xbuf0[i - rgroup] = xbuf0[rgroup*(M+1) + i];
      xbuf1[i - rgroup] = xbuf1[rgroup*(M+1) + i];
      xbuf0[rgroup*(M+2) + i] = xbuf0[i];
      xbuf1[rgroup*(M+2) + i] = xbuf1[i];
    }
  }
}


LOCAL(void)
set_bottom_pointers (j_decompress_ptr cinfo)
/* Change the pointer lists to duplicate the last sample row at the bottom
 * of the image.  whichptr indicates which xbuffer holds the final iMCU row.
 * Also sets rowgroups_avail to indicate number of nondummy row groups in row.
 */
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;
  int ci, i, rgroup, iMCUheight, rows_left;
  jpeg_component_info *compptr;
  JSAMPARRAY xbuf;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    /* Count sample rows in one iMCU row and in one row group */
    iMCUheight = compptr->v_samp_factor * compptr->DCT_scaled_size;
    rgroup = iMCUheight / cinfo->min_DCT_scaled_size;
    /* Count nondummy sample rows remaining for this component */
    rows_left = (int) (compptr->downsampled_height % (JDIMENSION) iMCUheight);
    if (rows_left == 0) rows_left = iMCUheight;
    /* Count nondummy row groups.  Should get same answer for each component,
     * so we need only do it once.
     */
    if (ci == 0) {
      main_ptr->rowgroups_avail = (JDIMENSION) ((rows_left-1) / rgroup + 1);
    }
    /* Duplicate the last real sample row rgroup*2 times; this pads out the
     * last partial rowgroup and ensures at least one full rowgroup of context.
     */
    xbuf = main_ptr->xbuffer[main_ptr->whichptr][ci];
    for (i = 0; i < rgroup * 2; i++) {
      xbuf[rows_left + i] = xbuf[rows_left-1];
    }
  }
}


/*
 * Initialize for a processing pass.
 */

METHODDEF(void)
start_pass_main (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;

  switch (pass_mode) {
  case JBUF_PASS_THRU:
    if (cinfo->upsample->need_context_rows) {
      main_ptr->pub.process_data = process_data_context_main;
      make_funny_pointers(cinfo); /* Create the xbuffer[] lists */
      main_ptr->whichptr = 0;   /* Read first iMCU row into xbuffer[0] */
      main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
      main_ptr->iMCU_row_ctr = 0;
    } else {
      /* Simple case with no context needed */
      main_ptr->pub.process_data = process_data_simple_main;
    }
    main_ptr->buffer_full = FALSE;  /* Mark buffer empty */
    main_ptr->rowgroup_ctr = 0;
    break;
#ifdef QUANT_2PASS_SUPPORTED
  case JBUF_CRANK_DEST:
    /* For last pass of 2-pass quantization, just crank the postprocessor */
    main_ptr->pub.process_data = process_data_crank_post;
    break;
#endif
  default:
    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
    break;
  }
}


/*
 * Process some data.
 * This handles the simple case where no context is required.
 */

METHODDEF(void)
process_data_simple_main (j_decompress_ptr cinfo,
              JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
              JDIMENSION out_rows_avail)
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;
  JDIMENSION rowgroups_avail;

  /* Read input data if we haven't filled the main buffer yet */
  if (! main_ptr->buffer_full) {
    if (! (*cinfo->coef->decompress_data) (cinfo, main_ptr->buffer))
      return;           /* suspension forced, can do nothing more */
    main_ptr->buffer_full = TRUE;   /* OK, we have an iMCU row to work with */
  }

  /* There are always min_DCT_scaled_size row groups in an iMCU row. */
  rowgroups_avail = (JDIMENSION) cinfo->min_DCT_scaled_size;
  /* Note: at the bottom of the image, we may pass extra garbage row groups
   * to the postprocessor.  The postprocessor has to check for bottom
   * of image anyway (at row resolution), so no point in us doing it too.
   */

  /* Feed the postprocessor */
  (*cinfo->post->post_process_data) (cinfo, main_ptr->buffer,
                     &main_ptr->rowgroup_ctr, rowgroups_avail,
                     output_buf, out_row_ctr, out_rows_avail);

  /* Has postprocessor consumed all the data yet? If so, mark buffer empty */
  if (main_ptr->rowgroup_ctr >= rowgroups_avail) {
    main_ptr->buffer_full = FALSE;
    main_ptr->rowgroup_ctr = 0;
  }
}


/*
 * Process some data.
 * This handles the case where context rows must be provided.
 */

METHODDEF(void)
process_data_context_main (j_decompress_ptr cinfo,
               JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
               JDIMENSION out_rows_avail)
{
  my_main_ptr main_ptr = (my_main_ptr) cinfo->main_ptr;

  /* Read input data if we haven't filled the main buffer yet */
  if (! main_ptr->buffer_full) {
    if (! (*cinfo->coef->decompress_data) (cinfo,
                       main_ptr->xbuffer[main_ptr->whichptr]))
      return;           /* suspension forced, can do nothing more */
    main_ptr->buffer_full = TRUE;   /* OK, we have an iMCU row to work with */
    main_ptr->iMCU_row_ctr++;   /* count rows received */
  }

  /* Postprocessor typically will not swallow all the input data it is handed
   * in one call (due to filling the output buffer first).  Must be prepared
   * to exit and restart.  This switch lets us keep track of how far we got.
   * Note that each case falls through to the next on successful completion.
   */
  switch (main_ptr->context_state) {
  case CTX_POSTPONED_ROW:
    /* Call postprocessor using previously set pointers for postponed row */
    (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr],
            &main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail,
            output_buf, out_row_ctr, out_rows_avail);
    if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
      return;           /* Need to suspend */
    main_ptr->context_state = CTX_PREPARE_FOR_IMCU;
    if (*out_row_ctr >= out_rows_avail)
      return;           /* Postprocessor exactly filled output buf */
    /*FALLTHROUGH*/
  case CTX_PREPARE_FOR_IMCU:
    /* Prepare to process first M-1 row groups of this iMCU row */
    main_ptr->rowgroup_ctr = 0;
    main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size - 1);
    /* Check for bottom of image: if so, tweak pointers to "duplicate"
     * the last sample row, and adjust rowgroups_avail to ignore padding rows.
     */
    if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows)
      set_bottom_pointers(cinfo);
    main_ptr->context_state = CTX_PROCESS_IMCU;
    /*FALLTHROUGH*/
  case CTX_PROCESS_IMCU:
    /* Call postprocessor using previously set pointers */
    (*cinfo->post->post_process_data) (cinfo, main_ptr->xbuffer[main_ptr->whichptr],
            &main_ptr->rowgroup_ctr, main_ptr->rowgroups_avail,
            output_buf, out_row_ctr, out_rows_avail);
    if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail)
      return;           /* Need to suspend */
    /* After the first iMCU, change wraparound pointers to normal state */
    if (main_ptr->iMCU_row_ctr == 1)
      set_wraparound_pointers(cinfo);
    /* Prepare to load new iMCU row using other xbuffer list */
    main_ptr->whichptr ^= 1;    /* 0=>1 or 1=>0 */
    main_ptr->buffer_full = FALSE;
    /* Still need to process last row group of this iMCU row, */
    /* which is saved at index M+1 of the other xbuffer */
    main_ptr->rowgroup_ctr = (JDIMENSION) (cinfo->min_DCT_scaled_size + 1);
    main_ptr->rowgroups_avail = (JDIMENSION) (cinfo->min_DCT_scaled_size + 2);
    main_ptr->context_state = CTX_POSTPONED_ROW;
  }
}


/*
 * Process some data.
 * Final pass of two-pass quantization: just call the postprocessor.
 * Source data will be the postprocessor controller's internal buffer.
 */

#ifdef QUANT_2PASS_SUPPORTED

METHODDEF(void)
process_data_crank_post (j_decompress_ptr cinfo,
             JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
             JDIMENSION out_rows_avail)
{
  (*cinfo->post->post_process_data) (cinfo, (JSAMPIMAGE) NULL,
                     (JDIMENSION *) NULL, (JDIMENSION) 0,
                     output_buf, out_row_ctr, out_rows_avail);
}

#endif /* QUANT_2PASS_SUPPORTED */


/*
 * Initialize main buffer controller.
 */

GLOBAL(void)
jinit_d_main_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
{
  my_main_ptr main_ptr;
  int ci, rgroup, ngroups;
  jpeg_component_info *compptr;

  main_ptr = (my_main_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(my_main_controller));
  cinfo->main_ptr = (struct jpeg_d_main_controller *) main_ptr;
  main_ptr->pub.start_pass = start_pass_main;

  if (need_full_buffer)     /* shouldn't happen */
    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);

  /* Allocate the workspace.
   * ngroups is the number of row groups we need.
   */
  if (cinfo->upsample->need_context_rows) {
    if (cinfo->min_DCT_scaled_size < 2) /* unsupported, see comments above */
      ERREXIT(cinfo, JERR_NOTIMPL);
    alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */
    ngroups = cinfo->min_DCT_scaled_size + 2;
  } else {
    ngroups = cinfo->min_DCT_scaled_size;
  }

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    rgroup = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
      cinfo->min_DCT_scaled_size; /* height of a row group of component */
    main_ptr->buffer[ci] = (*cinfo->mem->alloc_sarray)
            ((j_common_ptr) cinfo, JPOOL_IMAGE,
             compptr->width_in_blocks * compptr->DCT_scaled_size,
             (JDIMENSION) (rgroup * ngroups));
  }
}
/*
 * jdcoefct.c
 *
 * Copyright (C) 1994-1997, 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 coefficient buffer controller for decompression.
 * This controller is the top level of the JPEG decompressor proper.
 * The coefficient buffer lies between entropy decoding and inverse-DCT steps.
 *
 * In buffered-image mode, this controller is the interface between
 * input-oriented processing and output-oriented processing.
 * Also, the input side (only) is used when reading a file for transcoding.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"

/* Block smoothing is only applicable for progressive JPEG, so: */
#ifndef D_PROGRESSIVE_SUPPORTED
#undef BLOCK_SMOOTHING_SUPPORTED
#endif

/* Private buffer controller object */

typedef struct {
  struct jpeg_d_coef_controller pub; /* public fields */

  /* These variables keep track of the current location of the input side. */
  /* cinfo->input_iMCU_row is also used for this. */
  JDIMENSION MCU_ctr;       /* counts MCUs processed in current row */
  int MCU_vert_offset;      /* counts MCU rows within iMCU row */
  int MCU_rows_per_iMCU_row;    /* number of such rows needed */

  /* The output side's location is represented by cinfo->output_iMCU_row. */

  /* In single-pass modes, it's sufficient to buffer just one MCU.
   * We allocate a workspace of D_MAX_BLOCKS_IN_MCU coefficient blocks,
   * and let the entropy decoder write into that workspace each time.
   * (On 80x86, the workspace is FAR even though it's not really very big;
   * this is to keep the module interfaces unchanged when a large coefficient
   * buffer is necessary.)
   * In multi-pass modes, this array points to the current MCU's blocks
   * within the virtual arrays; it is used only by the input side.
   */
  JBLOCKROW MCU_buffer[D_MAX_BLOCKS_IN_MCU];

#ifdef D_MULTISCAN_FILES_SUPPORTED
  /* In multi-pass modes, we need a virtual block array for each component. */
  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
#endif

#ifdef BLOCK_SMOOTHING_SUPPORTED
  /* When doing block smoothing, we latch coefficient Al values here */
  int * coef_bits_latch;
#define SAVED_COEFS  6      /* we save coef_bits[0..5] */
#endif
} my_coef_controller;

typedef my_coef_controller * my_coef_ptr;

/* Forward declarations */
METHODDEF(int) decompress_onepass
    JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
#ifdef D_MULTISCAN_FILES_SUPPORTED
METHODDEF(int) decompress_data
    JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
#endif
#ifdef BLOCK_SMOOTHING_SUPPORTED
LOCAL(boolean) smoothing_ok JPP((j_decompress_ptr cinfo));
METHODDEF(int) decompress_smooth_data
    JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
#endif


LOCAL(void)
start_iMCU_row (j_decompress_ptr cinfo)
/* Reset within-iMCU-row counters for a new row (input side) */
{
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;

  /* In an interleaved scan, an MCU row is the same as an iMCU row.
   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
   * But at the bottom of the image, process only what's left.
   */
  if (cinfo->comps_in_scan > 1) {
    coef->MCU_rows_per_iMCU_row = 1;
  } else {
    if (cinfo->input_iMCU_row < (cinfo->total_iMCU_rows-1))
      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
    else
      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
  }

  coef->MCU_ctr = 0;
  coef->MCU_vert_offset = 0;
}


/*
 * Initialize for an input processing pass.
 */

METHODDEF(void)
co_start_input_pass (j_decompress_ptr cinfo)
{
  cinfo->input_iMCU_row = 0;
  start_iMCU_row(cinfo);
}


/*
 * Initialize for an output processing pass.
 */

METHODDEF(void)
start_output_pass (j_decompress_ptr cinfo)
{
#ifdef BLOCK_SMOOTHING_SUPPORTED
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;

  /* If multipass, check to see whether to use block smoothing on this pass */
  if (coef->pub.coef_arrays != NULL) {
    if (cinfo->do_block_smoothing && smoothing_ok(cinfo))
      coef->pub.decompress_data = decompress_smooth_data;
    else
      coef->pub.decompress_data = decompress_data;
  }
#endif
  cinfo->output_iMCU_row = 0;
}


/*
 * Decompress and return some data in the single-pass case.
 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
 * Input and output must run in lockstep since we have only a one-MCU buffer.
 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
 *
 * NB: output_buf contains a plane for each component in image,
 * which we index according to the component's SOF position.
 */

METHODDEF(int)
decompress_onepass (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
{
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
  JDIMENSION MCU_col_num;   /* index of current MCU within row */
  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
  int blkn, ci, xindex, yindex, yoffset, useful_width;
  JSAMPARRAY output_ptr;
  JDIMENSION start_col, output_col;
  jpeg_component_info *compptr;
  inverse_DCT_method_ptr inverse_DCT;

  /* Loop to process as much as one whole iMCU row */
  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
       yoffset++) {
    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
     MCU_col_num++) {
      /* Try to fetch an MCU.  Entropy decoder expects buffer to be zeroed. */
      jzero_far((void FAR *) coef->MCU_buffer[0],
        (size_t) (cinfo->blocks_in_MCU * SIZEOF(JBLOCK)));
      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
    /* Suspension forced; update state counters and exit */
    coef->MCU_vert_offset = yoffset;
    coef->MCU_ctr = MCU_col_num;
    return JPEG_SUSPENDED;
      }
      /* Determine where data should go in output_buf and do the IDCT thing.
       * We skip dummy blocks at the right and bottom edges (but blkn gets
       * incremented past them!).  Note the inner loop relies on having
       * allocated the MCU_buffer[] blocks sequentially.
       */
      blkn = 0;         /* index of current DCT block within MCU */
      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
    compptr = cinfo->cur_comp_info[ci];
    /* Don't bother to IDCT an uninteresting component. */
    if (! compptr->component_needed) {
      blkn += compptr->MCU_blocks;
      continue;
    }
    inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index];
    useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
                            : compptr->last_col_width;
    output_ptr = output_buf[compptr->component_index] +
      yoffset * compptr->DCT_scaled_size;
    start_col = MCU_col_num * compptr->MCU_sample_width;
    for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
      if (cinfo->input_iMCU_row < last_iMCU_row ||
          yoffset+yindex < compptr->last_row_height) {
        output_col = start_col;
        for (xindex = 0; xindex < useful_width; xindex++) {
          (*inverse_DCT) (cinfo, compptr,
                  (JCOEFPTR) coef->MCU_buffer[blkn+xindex],
                  output_ptr, output_col);
          output_col += compptr->DCT_scaled_size;
        }
      }
      blkn += compptr->MCU_width;
      output_ptr += compptr->DCT_scaled_size;
    }
      }
    }
    /* Completed an MCU row, but perhaps not an iMCU row */
    coef->MCU_ctr = 0;
  }
  /* Completed the iMCU row, advance counters for next one */
  cinfo->output_iMCU_row++;
  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
    start_iMCU_row(cinfo);
    return JPEG_ROW_COMPLETED;
  }
  /* Completed the scan */
  (*cinfo->inputctl->finish_input_pass) (cinfo);
  return JPEG_SCAN_COMPLETED;
}


/*
 * Dummy consume-input routine for single-pass operation.
 */

METHODDEF(int)
dummy_consume_data (j_decompress_ptr cinfo)
{
  return JPEG_SUSPENDED;    /* Always indicate nothing was done */
}


#ifdef D_MULTISCAN_FILES_SUPPORTED

/*
 * Consume input data and store it in the full-image coefficient buffer.
 * We read as much as one fully interleaved MCU row ("iMCU" row) per call,
 * ie, v_samp_factor block rows for each component in the scan.
 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
 */

METHODDEF(int)
consume_data (j_decompress_ptr cinfo)
{
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
  JDIMENSION MCU_col_num;   /* index of current MCU within row */
  int blkn, ci, xindex, yindex, yoffset;
  JDIMENSION start_col;
  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
  JBLOCKROW buffer_ptr;
  jpeg_component_info *compptr;

  /* Align the virtual buffers for the components used in this scan. */
  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
    compptr = cinfo->cur_comp_info[ci];
    buffer[ci] = (*cinfo->mem->access_virt_barray)
      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
       cinfo->input_iMCU_row * compptr->v_samp_factor,
       (JDIMENSION) compptr->v_samp_factor, TRUE);
    /* Note: entropy decoder expects buffer to be zeroed,
     * but this is handled automatically by the memory manager
     * because we requested a pre-zeroed array.
     */
  }

  /* Loop to process one whole iMCU row */
  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
       yoffset++) {
    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
     MCU_col_num++) {
      /* Construct list of pointers to DCT blocks belonging to this MCU */
      blkn = 0;         /* index of current DCT block within MCU */
      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
    compptr = cinfo->cur_comp_info[ci];
    start_col = MCU_col_num * compptr->MCU_width;
    for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
      buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
      for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
        coef->MCU_buffer[blkn++] = buffer_ptr++;
      }
    }
      }
      /* Try to fetch the MCU. */
      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
    /* Suspension forced; update state counters and exit */
    coef->MCU_vert_offset = yoffset;
    coef->MCU_ctr = MCU_col_num;
    return JPEG_SUSPENDED;
      }
    }
    /* Completed an MCU row, but perhaps not an iMCU row */
    coef->MCU_ctr = 0;
  }
  /* Completed the iMCU row, advance counters for next one */
  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
    start_iMCU_row(cinfo);
    return JPEG_ROW_COMPLETED;
  }
  /* Completed the scan */
  (*cinfo->inputctl->finish_input_pass) (cinfo);
  return JPEG_SCAN_COMPLETED;
}


/*
 * Decompress and return some data in the multi-pass case.
 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
 *
 * NB: output_buf contains a plane for each component in image.
 */

METHODDEF(int)
decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
{
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
  JDIMENSION block_num;
  int ci, block_row, block_rows;
  JBLOCKARRAY buffer;
  JBLOCKROW buffer_ptr;
  JSAMPARRAY output_ptr;
  JDIMENSION output_col;
  jpeg_component_info *compptr;
  inverse_DCT_method_ptr inverse_DCT;

  /* Force some input to be done if we are getting ahead of the input. */
  while (cinfo->input_scan_number < cinfo->output_scan_number ||
     (cinfo->input_scan_number == cinfo->output_scan_number &&
      cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
      return JPEG_SUSPENDED;
  }

  /* OK, output from the virtual arrays. */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    /* Don't bother to IDCT an uninteresting component. */
    if (! compptr->component_needed)
      continue;
    /* Align the virtual buffer for this component. */
    buffer = (*cinfo->mem->access_virt_barray)
      ((j_common_ptr) cinfo, coef->whole_image[ci],
       cinfo->output_iMCU_row * compptr->v_samp_factor,
       (JDIMENSION) compptr->v_samp_factor, FALSE);
    /* Count non-dummy DCT block rows in this iMCU row. */
    if (cinfo->output_iMCU_row < last_iMCU_row)
      block_rows = compptr->v_samp_factor;
    else {
      /* NB: can't use last_row_height here; it is input-side-dependent! */
      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
      if (block_rows == 0) block_rows = compptr->v_samp_factor;
    }
    inverse_DCT = cinfo->idct->inverse_DCT[ci];
    output_ptr = output_buf[ci];
    /* Loop over all DCT blocks to be processed. */
    for (block_row = 0; block_row < block_rows; block_row++) {
      buffer_ptr = buffer[block_row];
      output_col = 0;
      for (block_num = 0; block_num < compptr->width_in_blocks; block_num++) {
    (*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
            output_ptr, output_col);
    buffer_ptr++;
    output_col += compptr->DCT_scaled_size;
      }
      output_ptr += compptr->DCT_scaled_size;
    }
  }

  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
    return JPEG_ROW_COMPLETED;
  return JPEG_SCAN_COMPLETED;
}

#endif /* D_MULTISCAN_FILES_SUPPORTED */


#ifdef BLOCK_SMOOTHING_SUPPORTED

/*
 * This code applies interblock smoothing as described by section K.8
 * of the JPEG standard: the first 5 AC coefficients are estimated from
 * the DC values of a DCT block and its 8 neighboring blocks.
 * We apply smoothing only for progressive JPEG decoding, and only if
 * the coefficients it can estimate are not yet known to full precision.
 */

/* Natural-order array positions of the first 5 zigzag-order coefficients */
#define Q01_POS  1
#define Q10_POS  8
#define Q20_POS  16
#define Q11_POS  9
#define Q02_POS  2

/*
 * Determine whether block smoothing is applicable and safe.
 * We also latch the current states of the coef_bits[] entries for the
 * AC coefficients; otherwise, if the input side of the decompressor
 * advances into a new scan, we might think the coefficients are known
 * more accurately than they really are.
 */

LOCAL(boolean)
smoothing_ok (j_decompress_ptr cinfo)
{
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
  boolean smoothing_useful = FALSE;
  int ci, coefi;
  jpeg_component_info *compptr;
  JQUANT_TBL * qtable;
  int * coef_bits;
  int * coef_bits_latch;

  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
    return FALSE;

  /* Allocate latch area if not already done */
  if (coef->coef_bits_latch == NULL)
    coef->coef_bits_latch = (int *)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                  cinfo->num_components *
                  (SAVED_COEFS * SIZEOF(int)));
  coef_bits_latch = coef->coef_bits_latch;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    /* All components' quantization values must already be latched. */
    if ((qtable = compptr->quant_table) == NULL)
      return FALSE;
    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
    if (qtable->quantval[0] == 0 ||
    qtable->quantval[Q01_POS] == 0 ||
    qtable->quantval[Q10_POS] == 0 ||
    qtable->quantval[Q20_POS] == 0 ||
    qtable->quantval[Q11_POS] == 0 ||
    qtable->quantval[Q02_POS] == 0)
      return FALSE;
    /* DC values must be at least partly known for all components. */
    coef_bits = cinfo->coef_bits[ci];
    if (coef_bits[0] < 0)
      return FALSE;
    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
    for (coefi = 1; coefi <= 5; coefi++) {
      coef_bits_latch[coefi] = coef_bits[coefi];
      if (coef_bits[coefi] != 0)
    smoothing_useful = TRUE;
    }
    coef_bits_latch += SAVED_COEFS;
  }

  return smoothing_useful;
}


/*
 * Variant of decompress_data for use when doing block smoothing.
 */

METHODDEF(int)
decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
{
  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
  JDIMENSION block_num, last_block_column;
  int ci, block_row, block_rows, access_rows;
  JBLOCKARRAY buffer;
  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
  JSAMPARRAY output_ptr;
  JDIMENSION output_col;
  jpeg_component_info *compptr;
  inverse_DCT_method_ptr inverse_DCT;
  boolean first_row, last_row;
  JBLOCK workspace;
  int *coef_bits;
  JQUANT_TBL *quanttbl;
  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
  int Al, pred;

  /* Force some input to be done if we are getting ahead of the input. */
  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
     ! cinfo->inputctl->eoi_reached) {
    if (cinfo->input_scan_number == cinfo->output_scan_number) {
      /* If input is working on current scan, we ordinarily want it to
       * have completed the current row.  But if input scan is DC,
       * we want it to keep one row ahead so that next block row's DC
       * values are up to date.
       */
      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
    break;
    }
    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
      return JPEG_SUSPENDED;
  }

  /* OK, output from the virtual arrays. */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    /* Don't bother to IDCT an uninteresting component. */
    if (! compptr->component_needed)
      continue;
    /* Count non-dummy DCT block rows in this iMCU row. */
    if (cinfo->output_iMCU_row < last_iMCU_row) {
      block_rows = compptr->v_samp_factor;
      access_rows = block_rows * 2; /* this and next iMCU row */
      last_row = FALSE;
    } else {
      /* NB: can't use last_row_height here; it is input-side-dependent! */
      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
      if (block_rows == 0) block_rows = compptr->v_samp_factor;
      access_rows = block_rows; /* this iMCU row only */
      last_row = TRUE;
    }
    /* Align the virtual buffer for this component. */
    if (cinfo->output_iMCU_row > 0) {
      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
      buffer = (*cinfo->mem->access_virt_barray)
    ((j_common_ptr) cinfo, coef->whole_image[ci],
     (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
     (JDIMENSION) access_rows, FALSE);
      buffer += compptr->v_samp_factor; /* point to current iMCU row */
      first_row = FALSE;
    } else {
      buffer = (*cinfo->mem->access_virt_barray)
    ((j_common_ptr) cinfo, coef->whole_image[ci],
     (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
      first_row = TRUE;
    }
    /* Fetch component-dependent info */
    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
    quanttbl = compptr->quant_table;
    Q00 = quanttbl->quantval[0];
    Q01 = quanttbl->quantval[Q01_POS];
    Q10 = quanttbl->quantval[Q10_POS];
    Q20 = quanttbl->quantval[Q20_POS];
    Q11 = quanttbl->quantval[Q11_POS];
    Q02 = quanttbl->quantval[Q02_POS];
    inverse_DCT = cinfo->idct->inverse_DCT[ci];
    output_ptr = output_buf[ci];
    /* Loop over all DCT blocks to be processed. */
    for (block_row = 0; block_row < block_rows; block_row++) {
      buffer_ptr = buffer[block_row];
      if (first_row && block_row == 0)
    prev_block_row = buffer_ptr;
      else
    prev_block_row = buffer[block_row-1];
      if (last_row && block_row == block_rows-1)
    next_block_row = buffer_ptr;
      else
    next_block_row = buffer[block_row+1];
      /* We fetch the surrounding DC values using a sliding-register approach.
       * Initialize all nine here so as to do the right thing on narrow pics.
       */
      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
      output_col = 0;
      last_block_column = compptr->width_in_blocks - 1;
      for (block_num = 0; block_num <= last_block_column; block_num++) {
    /* Fetch current DCT block into workspace so we can modify it. */
    jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
    /* Update DC values */
    if (block_num < last_block_column) {
      DC3 = (int) prev_block_row[1][0];
      DC6 = (int) buffer_ptr[1][0];
      DC9 = (int) next_block_row[1][0];
    }
    /* Compute coefficient estimates per K.8.
     * An estimate is applied only if coefficient is still zero,
     * and is not known to be fully accurate.
     */
    /* AC01 */
    if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
      num = 36 * Q00 * (DC4 - DC6);
      if (num >= 0) {
        pred = (int) (((Q01<<7) + num) / (Q01<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
      } else {
        pred = (int) (((Q01<<7) - num) / (Q01<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
        pred = -pred;
      }
      workspace[1] = (JCOEF) pred;
    }
    /* AC10 */
    if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
      num = 36 * Q00 * (DC2 - DC8);
      if (num >= 0) {
        pred = (int) (((Q10<<7) + num) / (Q10<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
      } else {
        pred = (int) (((Q10<<7) - num) / (Q10<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
        pred = -pred;
      }
      workspace[8] = (JCOEF) pred;
    }
    /* AC20 */
    if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
      num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
      if (num >= 0) {
        pred = (int) (((Q20<<7) + num) / (Q20<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
      } else {
        pred = (int) (((Q20<<7) - num) / (Q20<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
        pred = -pred;
      }
      workspace[16] = (JCOEF) pred;
    }
    /* AC11 */
    if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
      num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
      if (num >= 0) {
        pred = (int) (((Q11<<7) + num) / (Q11<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
      } else {
        pred = (int) (((Q11<<7) - num) / (Q11<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
        pred = -pred;
      }
      workspace[9] = (JCOEF) pred;
    }
    /* AC02 */
    if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
      num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
      if (num >= 0) {
        pred = (int) (((Q02<<7) + num) / (Q02<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
      } else {
        pred = (int) (((Q02<<7) - num) / (Q02<<8));
        if (Al > 0 && pred >= (1<<Al))
          pred = (1<<Al)-1;
        pred = -pred;
      }
      workspace[2] = (JCOEF) pred;
    }
    /* OK, do the IDCT */
    (*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
            output_ptr, output_col);
    /* Advance for next column */
    DC1 = DC2; DC2 = DC3;
    DC4 = DC5; DC5 = DC6;
    DC7 = DC8; DC8 = DC9;
    buffer_ptr++, prev_block_row++, next_block_row++;
    output_col += compptr->DCT_scaled_size;
      }
      output_ptr += compptr->DCT_scaled_size;
    }
  }

  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
    return JPEG_ROW_COMPLETED;
  return JPEG_SCAN_COMPLETED;
}

#endif /* BLOCK_SMOOTHING_SUPPORTED */


/*
 * Initialize coefficient buffer controller.
 */

GLOBAL(void)
jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
{
  my_coef_ptr coef;

  coef = (my_coef_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(my_coef_controller));
  cinfo->coef = (struct jpeg_d_coef_controller *) coef;
  coef->pub.start_input_pass = co_start_input_pass;
  coef->pub.start_output_pass = start_output_pass;
#ifdef BLOCK_SMOOTHING_SUPPORTED
  coef->coef_bits_latch = NULL;
#endif

  /* Create the coefficient buffer. */
  if (need_full_buffer) {
#ifdef D_MULTISCAN_FILES_SUPPORTED
    /* Allocate a full-image virtual array for each component, */
    /* padded to a multiple of samp_factor DCT blocks in each direction. */
    /* Note we ask for a pre-zeroed array. */
    int ci, access_rows;
    jpeg_component_info *compptr;

    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
     ci++, compptr++) {
      access_rows = compptr->v_samp_factor;
#ifdef BLOCK_SMOOTHING_SUPPORTED
      /* If block smoothing could be used, need a bigger window */
      if (cinfo->progressive_mode)
    access_rows *= 3;
#endif
      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
    ((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
     (JDIMENSION) jround_up((long) compptr->width_in_blocks,
                (long) compptr->h_samp_factor),
     (JDIMENSION) jround_up((long) compptr->height_in_blocks,
                (long) compptr->v_samp_factor),
     (JDIMENSION) access_rows);
    }
    coef->pub.consume_data = consume_data;
    coef->pub.decompress_data = decompress_data;
    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
#else
    ERREXIT(cinfo, JERR_NOT_COMPILED);
#endif
  } else {
    /* We only need a single-MCU buffer. */
    JBLOCKROW buffer;
    int i;

    buffer = (JBLOCKROW)
      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
    for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
      coef->MCU_buffer[i] = buffer + i;
    }
    coef->pub.consume_data = dummy_consume_data;
    coef->pub.decompress_data = decompress_onepass;
    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
  }
}
/*
 * jdpostct.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 decompression postprocessing controller.
 * This controller manages the upsampling, color conversion, and color
 * quantization/reduction steps; specifically, it controls the buffering
 * between upsample/color conversion and color quantization/reduction.
 *
 * If no color quantization/reduction is required, then this module has no
 * work to do, and it just hands off to the upsample/color conversion code.
 * An integrated upsample/convert/quantize process would replace this module
 * entirely.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/* Private buffer controller object */

typedef struct {
  struct jpeg_d_post_controller pub; /* public fields */

  /* Color quantization source buffer: this holds output data from
   * the upsample/color conversion step to be passed to the quantizer.
   * For two-pass color quantization, we need a full-image buffer;
   * for one-pass operation, a strip buffer is sufficient.
   */
  jvirt_sarray_ptr whole_image; /* virtual array, or NULL if one-pass */
  JSAMPARRAY buffer;        /* strip buffer, or current strip of virtual */
  JDIMENSION strip_height;  /* buffer size in rows */
  /* for two-pass mode only: */
  JDIMENSION starting_row;  /* row # of first row in current strip */
  JDIMENSION next_row;      /* index of next row to fill/empty in strip */
} my_post_controller;

typedef my_post_controller * my_post_ptr;


/* Forward declarations */
METHODDEF(void) post_process_1pass
    JPP((j_decompress_ptr cinfo,
         JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
         JDIMENSION in_row_groups_avail,
         JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
         JDIMENSION out_rows_avail));
#ifdef QUANT_2PASS_SUPPORTED
METHODDEF(void) post_process_prepass
    JPP((j_decompress_ptr cinfo,
         JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
         JDIMENSION in_row_groups_avail,
         JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
         JDIMENSION out_rows_avail));
METHODDEF(void) post_process_2pass
    JPP((j_decompress_ptr cinfo,
         JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
         JDIMENSION in_row_groups_avail,
         JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
         JDIMENSION out_rows_avail));
#endif


/*
 * Initialize for a processing pass.
 */

METHODDEF(void)
start_pass_dpost (j_decompress_ptr cinfo, J_BUF_MODE pass_mode)
{
  my_post_ptr post = (my_post_ptr) cinfo->post;

  switch (pass_mode) {
  case JBUF_PASS_THRU:
    if (cinfo->quantize_colors) {
      /* Single-pass processing with color quantization. */
      post->pub.post_process_data = post_process_1pass;
      /* We could be doing buffered-image output before starting a 2-pass
       * color quantization; in that case, jinit_d_post_controller did not
       * allocate a strip buffer.  Use the virtual-array buffer as workspace.
       */
      if (post->buffer == NULL) {
    post->buffer = (*cinfo->mem->access_virt_sarray)
      ((j_common_ptr) cinfo, post->whole_image,
       (JDIMENSION) 0, post->strip_height, TRUE);
      }
    } else {
      /* For single-pass processing without color quantization,
       * I have no work to do; just call the upsampler directly.
       */
      post->pub.post_process_data = cinfo->upsample->upsample;
    }
    break;
#ifdef QUANT_2PASS_SUPPORTED
  case JBUF_SAVE_AND_PASS:
    /* First pass of 2-pass quantization */
    if (post->whole_image == NULL)
      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
    post->pub.post_process_data = post_process_prepass;
    break;
  case JBUF_CRANK_DEST:
    /* Second pass of 2-pass quantization */
    if (post->whole_image == NULL)
      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
    post->pub.post_process_data = post_process_2pass;
    break;
#endif /* QUANT_2PASS_SUPPORTED */
  default:
    ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
    break;
  }
  post->starting_row = post->next_row = 0;
}


/*
 * Process some data in the one-pass (strip buffer) case.
 * This is used for color precision reduction as well as one-pass quantization.
 */

METHODDEF(void)
post_process_1pass (j_decompress_ptr cinfo,
            JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
            JDIMENSION in_row_groups_avail,
            JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
            JDIMENSION out_rows_avail)
{
  my_post_ptr post = (my_post_ptr) cinfo->post;
  JDIMENSION num_rows, max_rows;

  /* Fill the buffer, but not more than what we can dump out in one go. */
  /* Note we rely on the upsampler to detect bottom of image. */
  max_rows = out_rows_avail - *out_row_ctr;
  if (max_rows > post->strip_height)
    max_rows = post->strip_height;
  num_rows = 0;
  (*cinfo->upsample->upsample) (cinfo,
        input_buf, in_row_group_ctr, in_row_groups_avail,
        post->buffer, &num_rows, max_rows);
  /* Quantize and emit data. */
  (*cinfo->cquantize->color_quantize) (cinfo,
        post->buffer, output_buf + *out_row_ctr, (int) num_rows);
  *out_row_ctr += num_rows;
}


#ifdef QUANT_2PASS_SUPPORTED

/*
 * Process some data in the first pass of 2-pass quantization.
 */

METHODDEF(void)
post_process_prepass (j_decompress_ptr cinfo,
              JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
              JDIMENSION in_row_groups_avail,
              JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
              JDIMENSION out_rows_avail)
{
  my_post_ptr post = (my_post_ptr) cinfo->post;
  JDIMENSION old_next_row, num_rows;

  /* Reposition virtual buffer if at start of strip. */
  if (post->next_row == 0) {
    post->buffer = (*cinfo->mem->access_virt_sarray)
    ((j_common_ptr) cinfo, post->whole_image,
     post->starting_row, post->strip_height, TRUE);
  }

  /* Upsample some data (up to a strip height's worth). */
  old_next_row = post->next_row;
  (*cinfo->upsample->upsample) (cinfo,
        input_buf, in_row_group_ctr, in_row_groups_avail,
        post->buffer, &post->next_row, post->strip_height);

  /* Allow quantizer to scan new data.  No data is emitted, */
  /* but we advance out_row_ctr so outer loop can tell when we're done. */
  if (post->next_row > old_next_row) {
    num_rows = post->next_row - old_next_row;
    (*cinfo->cquantize->color_quantize) (cinfo, post->buffer + old_next_row,
                     (JSAMPARRAY) NULL, (int) num_rows);
    *out_row_ctr += num_rows;
  }

  /* Advance if we filled the strip. */
  if (post->next_row >= post->strip_height) {
    post->starting_row += post->strip_height;
    post->next_row = 0;
  }
}


/*
 * Process some data in the second pass of 2-pass quantization.
 */

METHODDEF(void)
post_process_2pass (j_decompress_ptr cinfo,
            JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
            JDIMENSION in_row_groups_avail,
            JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
            JDIMENSION out_rows_avail)
{
  my_post_ptr post = (my_post_ptr) cinfo->post;
  JDIMENSION num_rows, max_rows;

  /* Reposition virtual buffer if at start of strip. */
  if (post->next_row == 0) {
    post->buffer = (*cinfo->mem->access_virt_sarray)
    ((j_common_ptr) cinfo, post->whole_image,
     post->starting_row, post->strip_height, FALSE);
  }

  /* Determine number of rows to emit. */
  num_rows = post->strip_height - post->next_row; /* available in strip */
  max_rows = out_rows_avail - *out_row_ctr; /* available in output area */
  if (num_rows > max_rows)
    num_rows = max_rows;
  /* We have to check bottom of image here, can't depend on upsampler. */
  max_rows = cinfo->output_height - post->starting_row;
  if (num_rows > max_rows)
    num_rows = max_rows;

  /* Quantize and emit data. */
  (*cinfo->cquantize->color_quantize) (cinfo,
        post->buffer + post->next_row, output_buf + *out_row_ctr,
        (int) num_rows);
  *out_row_ctr += num_rows;

  /* Advance if we filled the strip. */
  post->next_row += num_rows;
  if (post->next_row >= post->strip_height) {
    post->starting_row += post->strip_height;
    post->next_row = 0;
  }
}

#endif /* QUANT_2PASS_SUPPORTED */


/*
 * Initialize postprocessing controller.
 */

GLOBAL(void)
jinit_d_post_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
{
  my_post_ptr post;

  post = (my_post_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(my_post_controller));
  cinfo->post = (struct jpeg_d_post_controller *) post;
  post->pub.start_pass = start_pass_dpost;
  post->whole_image = NULL; /* flag for no virtual arrays */
  post->buffer = NULL;      /* flag for no strip buffer */

  /* Create the quantization buffer, if needed */
  if (cinfo->quantize_colors) {
    /* The buffer strip height is max_v_samp_factor, which is typically
     * an efficient number of rows for upsampling to return.
     * (In the presence of output rescaling, we might want to be smarter?)
     */
    post->strip_height = (JDIMENSION) cinfo->max_v_samp_factor;
    if (need_full_buffer) {
      /* Two-pass color quantization: need full-image storage. */
      /* We round up the number of rows to a multiple of the strip height. */
#ifdef QUANT_2PASS_SUPPORTED
      post->whole_image = (*cinfo->mem->request_virt_sarray)
    ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE,
     cinfo->output_width * cinfo->out_color_components,
     (JDIMENSION) jround_up((long) cinfo->output_height,
                (long) post->strip_height),
     post->strip_height);
#else
      ERREXIT(cinfo, JERR_BAD_BUFFER_MODE);
#endif /* QUANT_2PASS_SUPPORTED */
    } else {
      /* One-pass color quantization: just make a strip buffer. */
      post->buffer = (*cinfo->mem->alloc_sarray)
    ((j_common_ptr) cinfo, JPOOL_IMAGE,
     cinfo->output_width * cinfo->out_color_components,
     post->strip_height);
    }
  }
}
/*
 * 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_scaled_size) {
#ifdef IDCT_SCALING_SUPPORTED
    case 1:
      method_ptr = jpeg_idct_1x1;
      method = JDCT_ISLOW;  /* jidctred uses islow-style table */
      break;
    case 2:
      method_ptr = jpeg_idct_2x2;
      method = JDCT_ISLOW;  /* jidctred uses islow-style table */
      break;
    case 4:
      method_ptr = jpeg_idct_4x4;
      method = JDCT_ISLOW;  /* jidctred uses islow-style table */
      break;
#endif
    case 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:
      ERREXIT1(cinfo, JERR_BAD_DCTSIZE, compptr->DCT_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;
  }
}
/*
 * jidctfst.c
 *
 * Copyright (C) 1994-1998, 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 a fast, not so accurate integer implementation of the
 * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
 * must also perform dequantization of the input coefficients.
 *
 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
 * on each row (or vice versa, but it's more convenient to emit a row at
 * a time).  Direct algorithms are also available, but they are much more
 * complex and seem not to be any faster when reduced to code.
 *
 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
 * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
 * JPEG textbook (see REFERENCES section in file README).  The following code
 * is based directly on figure 4-8 in P&M.
 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
 * possible to arrange the computation so that many of the multiplies are
 * simple scalings of the final outputs.  These multiplies can then be
 * folded into the multiplications or divisions by the JPEG quantization
 * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
 * to be done in the DCT itself.
 * The primary disadvantage of this method is that with fixed-point math,
 * accuracy is lost due to imprecise representation of the scaled
 * quantization values.  The smaller the quantization table entry, the less
 * precise the scaled value, so this implementation does worse with high-
 * quality-setting files than with low-quality ones.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"
//#include "jdct.h"     /* Private declarations for DCT subsystem */

#ifdef DCT_IFAST_SUPPORTED


/*
 * This module is specialized to the case DCTSIZE = 8.
 */

#if DCTSIZE != 8
  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
#endif


/* Scaling decisions are generally the same as in the LL&M algorithm;
 * see jidctint.c for more details.  However, we choose to descale
 * (right shift) multiplication products as soon as they are formed,
 * rather than carrying additional fractional bits into subsequent additions.
 * This compromises accuracy slightly, but it lets us save a few shifts.
 * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples)
 * everywhere except in the multiplications proper; this saves a good deal
 * of work on 16-bit-int machines.
 *
 * The dequantized coefficients are not integers because the AA&N scaling
 * factors have been incorporated.  We represent them scaled up by PASS1_BITS,
 * so that the first and second IDCT rounds have the same input scaling.
 * For 8-bit JSAMPLEs, we choose IFAST_SCALE_BITS = PASS1_BITS so as to
 * avoid a descaling shift; this compromises accuracy rather drastically
 * for small quantization table entries, but it saves a lot of shifts.
 * For 12-bit JSAMPLEs, there's no hope of using 16x16 multiplies anyway,
 * so we use a much larger scaling factor to preserve accuracy.
 *
 * A final compromise is to represent the multiplicative constants to only
 * 8 fractional bits, rather than 13.  This saves some shifting work on some
 * machines, and may also reduce the cost of multiplication (since there
 * are fewer one-bits in the constants).
 */

#undef CONST_BITS
#if BITS_IN_JSAMPLE == 8
#define CONST_BITS  8
#define PASS1_BITS  2
#else
#define CONST_BITS  8
#define PASS1_BITS  1       /* lose a little precision to avoid overflow */
#endif

/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
 * causing a lot of useless floating-point operations at run time.
 * To get around this we use the following pre-calculated constants.
 * If you change CONST_BITS you may want to add appropriate values.
 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
 */

#if CONST_BITS == 8
#define FIX_1_082392200  ((INT32)  277)     /* FIX(1.082392200) */
#define FIX_1_414213562  ((INT32)  362)     /* FIX(1.414213562) */
#define FIX_1_847759065  ((INT32)  473)     /* FIX(1.847759065) */
#define FIX_2_613125930  ((INT32)  669)     /* FIX(2.613125930) */
#else
#define FIX_1_082392200  FIX(1.082392200)
#define FIX_1_414213562  FIX(1.414213562)
#define FIX_1_847759065  FIX(1.847759065)
#define FIX_2_613125930  FIX(2.613125930)
#endif


/* We can gain a little more speed, with a further compromise in accuracy,
 * by omitting the addition in a descaling shift.  This yields an incorrectly
 * rounded result half the time...
 */

#ifndef USE_ACCURATE_ROUNDING
//#undef DESCALE
#define jic_DESCALE(x,n)  RIGHT_SHIFT(x, n)
#endif


/* Multiply a DCTELEM variable by an INT32 constant, and immediately
 * descale to yield a DCTELEM result.
 */

#define jic_MULTIPLY(var,const)  ((DCTELEM) jic_DESCALE((var) * (const), CONST_BITS))


/* Dequantize a coefficient by multiplying it by the multiplier-table
 * entry; produce a DCTELEM result.  For 8-bit data a 16x16->16
 * multiplication will do.  For 12-bit data, the multiplier table is
 * declared INT32, so a 32-bit multiply will be used.
 */

#if BITS_IN_JSAMPLE == 8
#define jic_jic_DEQUANTIZE(coef,quantval)  (((IFAST_MULT_TYPE) (coef)) * (quantval))
#else
#define jic_jic_DEQUANTIZE(coef,quantval)  \
    jic_DESCALE((coef)*(quantval), IFAST_SCALE_BITS-PASS1_BITS)
#endif


/* Like DESCALE, but applies to a DCTELEM and produces an int.
 * We assume that int right shift is unsigned if INT32 right shift is.
 */

#ifdef RIGHT_SHIFT_IS_UNSIGNED
#define ISHIFT_TEMPS    DCTELEM ishift_temp;
#if BITS_IN_JSAMPLE == 8
#define DCTELEMBITS  16     /* DCTELEM may be 16 or 32 bits */
#else
#define DCTELEMBITS  32     /* DCTELEM must be 32 bits */
#endif
#define jic_jic_IRIGHT_SHIFT(x,shft)  \
    ((ishift_temp = (x)) < 0 ? \
     (ishift_temp >> (shft)) | ((~((DCTELEM) 0)) << (DCTELEMBITS-(shft))) : \
     (ishift_temp >> (shft)))
#else
#define ISHIFT_TEMPS
#define jic_jic_IRIGHT_SHIFT(x,shft)    ((x) >> (shft))
#endif

#ifdef USE_ACCURATE_ROUNDING
#define Ijic_DESCALE(x,n)  ((int) jic_jic_IRIGHT_SHIFT((x) + (1 << ((n)-1)), n))
#else
#define Ijic_DESCALE(x,n)  ((int) jic_jic_IRIGHT_SHIFT(x, n))
#endif


/*
 * Perform dequantization and inverse DCT on one block of coefficients.
 */

GLOBAL(void)
jpeg_idct_ifast (j_decompress_ptr cinfo, jpeg_component_info * compptr,
         JCOEFPTR coef_block,
         JSAMPARRAY output_buf, JDIMENSION output_col)
{
  DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  DCTELEM tmp10, tmp11, tmp12, tmp13;
  DCTELEM z5, z10, z11, z12, z13;
  JCOEFPTR inptr;
  IFAST_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[DCTSIZE2];  /* buffers data between passes */
  SHIFT_TEMPS           /* for DESCALE */
  ISHIFT_TEMPS          /* for IDESCALE */

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (IFAST_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = DCTSIZE; ctr > 0; ctr--) {
    /* Due to quantization, we will usually find that many of the input
     * coefficients are zero, especially the AC terms.  We can exploit this
     * by short-circuiting the IDCT calculation for any column in which all
     * the AC terms are zero.  In that case each output is equal to the
     * DC coefficient (with scale factor as needed).
     * With typical images and quantization tables, half or more of the
     * column DCT calculations can be simplified this way.
     */

    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
    inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
    inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
    inptr[DCTSIZE*7] == 0) {
      /* AC terms all zero */
      int dcval = (int) jic_jic_DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);

      wsptr[DCTSIZE*0] = dcval;
      wsptr[DCTSIZE*1] = dcval;
      wsptr[DCTSIZE*2] = dcval;
      wsptr[DCTSIZE*3] = dcval;
      wsptr[DCTSIZE*4] = dcval;
      wsptr[DCTSIZE*5] = dcval;
      wsptr[DCTSIZE*6] = dcval;
      wsptr[DCTSIZE*7] = dcval;

      inptr++;          /* advance pointers to next column */
      quantptr++;
      wsptr++;
      continue;
    }

    /* Even part */

    tmp0 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp1 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp2 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp3 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp10 = tmp0 + tmp2;    /* phase 3 */
    tmp11 = tmp0 - tmp2;

    tmp13 = tmp1 + tmp3;    /* phases 5-3 */
    tmp12 = jic_MULTIPLY(tmp1 - tmp3, FIX_1_414213562) - tmp13; /* 2*c4 */

    tmp0 = tmp10 + tmp13;   /* phase 2 */
    tmp3 = tmp10 - tmp13;
    tmp1 = tmp11 + tmp12;
    tmp2 = tmp11 - tmp12;

    /* Odd part */

    tmp4 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    tmp5 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    tmp6 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp7 = jic_jic_DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    z13 = tmp6 + tmp5;      /* phase 6 */
    z10 = tmp6 - tmp5;
    z11 = tmp4 + tmp7;
    z12 = tmp4 - tmp7;

    tmp7 = z11 + z13;       /* phase 5 */
    tmp11 = jic_MULTIPLY(z11 - z13, FIX_1_414213562); /* 2*c4 */

    z5 = jic_MULTIPLY(z10 + z12, FIX_1_847759065); /* 2*c2 */
    tmp10 = jic_MULTIPLY(z12, FIX_1_082392200) - z5; /* 2*(c2-c6) */
    tmp12 = jic_MULTIPLY(z10, - FIX_2_613125930) + z5; /* -2*(c2+c6) */

    tmp6 = tmp12 - tmp7;    /* phase 2 */
    tmp5 = tmp11 - tmp6;
    tmp4 = tmp10 + tmp5;

    wsptr[DCTSIZE*0] = (int) (tmp0 + tmp7);
    wsptr[DCTSIZE*7] = (int) (tmp0 - tmp7);
    wsptr[DCTSIZE*1] = (int) (tmp1 + tmp6);
    wsptr[DCTSIZE*6] = (int) (tmp1 - tmp6);
    wsptr[DCTSIZE*2] = (int) (tmp2 + tmp5);
    wsptr[DCTSIZE*5] = (int) (tmp2 - tmp5);
    wsptr[DCTSIZE*4] = (int) (tmp3 + tmp4);
    wsptr[DCTSIZE*3] = (int) (tmp3 - tmp4);

    inptr++;            /* advance pointers to next column */
    quantptr++;
    wsptr++;
  }

  /* Pass 2: process rows from work array, store into output array. */
  /* Note that we must descale the results by a factor of 8 == 2**3, */
  /* and also undo the PASS1_BITS scaling. */

  wsptr = workspace;
  for (ctr = 0; ctr < DCTSIZE; ctr++) {
    outptr = output_buf[ctr] + output_col;
    /* Rows of zeroes can be exploited in the same way as we did with columns.
     * However, the column calculation has created many nonzero AC terms, so
     * the simplification applies less often (typically 5% to 10% of the time).
     * On machines with very fast multiplication, it's possible that the
     * test takes more time than it's worth.  In that case this section
     * may be commented out.
     */

#ifndef NO_ZERO_ROW_TEST
    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
    wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
      /* AC terms all zero */
      JSAMPLE dcval = range_limit[Ijic_DESCALE(wsptr[0], PASS1_BITS+3)
                  & RANGE_MASK];

      outptr[0] = dcval;
      outptr[1] = dcval;
      outptr[2] = dcval;
      outptr[3] = dcval;
      outptr[4] = dcval;
      outptr[5] = dcval;
      outptr[6] = dcval;
      outptr[7] = dcval;

      wsptr += DCTSIZE;     /* advance pointer to next row */
      continue;
    }
#endif

    /* Even part */

    tmp10 = ((DCTELEM) wsptr[0] + (DCTELEM) wsptr[4]);
    tmp11 = ((DCTELEM) wsptr[0] - (DCTELEM) wsptr[4]);

    tmp13 = ((DCTELEM) wsptr[2] + (DCTELEM) wsptr[6]);
    tmp12 = jic_MULTIPLY((DCTELEM) wsptr[2] - (DCTELEM) wsptr[6], FIX_1_414213562)
        - tmp13;

    tmp0 = tmp10 + tmp13;
    tmp3 = tmp10 - tmp13;
    tmp1 = tmp11 + tmp12;
    tmp2 = tmp11 - tmp12;

    /* Odd part */

    z13 = (DCTELEM) wsptr[5] + (DCTELEM) wsptr[3];
    z10 = (DCTELEM) wsptr[5] - (DCTELEM) wsptr[3];
    z11 = (DCTELEM) wsptr[1] + (DCTELEM) wsptr[7];
    z12 = (DCTELEM) wsptr[1] - (DCTELEM) wsptr[7];

    tmp7 = z11 + z13;       /* phase 5 */
    tmp11 = jic_MULTIPLY(z11 - z13, FIX_1_414213562); /* 2*c4 */

    z5 = jic_MULTIPLY(z10 + z12, FIX_1_847759065); /* 2*c2 */
    tmp10 = jic_MULTIPLY(z12, FIX_1_082392200) - z5; /* 2*(c2-c6) */
    tmp12 = jic_MULTIPLY(z10, - FIX_2_613125930) + z5; /* -2*(c2+c6) */

    tmp6 = tmp12 - tmp7;    /* phase 2 */
    tmp5 = tmp11 - tmp6;
    tmp4 = tmp10 + tmp5;

    /* Final output stage: scale down by a factor of 8 and range-limit */

    outptr[0] = range_limit[Ijic_DESCALE(tmp0 + tmp7, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[7] = range_limit[Ijic_DESCALE(tmp0 - tmp7, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[1] = range_limit[Ijic_DESCALE(tmp1 + tmp6, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[6] = range_limit[Ijic_DESCALE(tmp1 - tmp6, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[2] = range_limit[Ijic_DESCALE(tmp2 + tmp5, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[5] = range_limit[Ijic_DESCALE(tmp2 - tmp5, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[4] = range_limit[Ijic_DESCALE(tmp3 + tmp4, PASS1_BITS+3)
                & RANGE_MASK];
    outptr[3] = range_limit[Ijic_DESCALE(tmp3 - tmp4, PASS1_BITS+3)
                & RANGE_MASK];

    wsptr += DCTSIZE;       /* advance pointer to next row */
  }
}

#endif /* DCT_IFAST_SUPPORTED */
/*
 * jidctflt.c
 *
 * Copyright (C) 1994-1998, 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 a floating-point implementation of the
 * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
 * must also perform dequantization of the input coefficients.
 *
 * This implementation should be more accurate than either of the integer
 * IDCT implementations.  However, it may not give the same results on all
 * machines because of differences in roundoff behavior.  Speed will depend
 * on the hardware's floating point capacity.
 *
 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
 * on each row (or vice versa, but it's more convenient to emit a row at
 * a time).  Direct algorithms are also available, but they are much more
 * complex and seem not to be any faster when reduced to code.
 *
 * This implementation is based on Arai, Agui, and Nakajima's algorithm for
 * scaled DCT.  Their original paper (Trans. IEICE E-71(11):1095) is in
 * Japanese, but the algorithm is described in the Pennebaker & Mitchell
 * JPEG textbook (see REFERENCES section in file README).  The following code
 * is based directly on figure 4-8 in P&M.
 * While an 8-point DCT cannot be done in less than 11 multiplies, it is
 * possible to arrange the computation so that many of the multiplies are
 * simple scalings of the final outputs.  These multiplies can then be
 * folded into the multiplications or divisions by the JPEG quantization
 * table entries.  The AA&N method leaves only 5 multiplies and 29 adds
 * to be done in the DCT itself.
 * The primary disadvantage of this method is that with a fixed-point
 * implementation, accuracy is lost due to imprecise representation of the
 * scaled quantization values.  However, that problem does not arise if
 * we use floating point arithmetic.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"
//#include "jdct.h"     /* Private declarations for DCT subsystem */

#ifdef DCT_FLOAT_SUPPORTED


/*
 * This module is specialized to the case DCTSIZE = 8.
 */

#if DCTSIZE != 8
  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
#endif


/* Dequantize a coefficient by multiplying it by the multiplier-table
 * entry; produce a float result.
 */

#define jict_DEQUANTIZE(coef,quantval)  (((FAST_FLOAT) (coef)) * (quantval))


/*
 * Perform dequantization and inverse DCT on one block of coefficients.
 */

GLOBAL(void)
jpeg_idct_float (j_decompress_ptr cinfo, jpeg_component_info * compptr,
         JCOEFPTR coef_block,
         JSAMPARRAY output_buf, JDIMENSION output_col)
{
  FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
  FAST_FLOAT tmp10, tmp11, tmp12, tmp13;
  FAST_FLOAT z5, z10, z11, z12, z13;
  JCOEFPTR inptr;
  FLOAT_MULT_TYPE * quantptr;
  FAST_FLOAT * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (FLOAT_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = DCTSIZE; ctr > 0; ctr--) {
    /* Due to quantization, we will usually find that many of the input
     * coefficients are zero, especially the AC terms.  We can exploit this
     * by short-circuiting the IDCT calculation for any column in which all
     * the AC terms are zero.  In that case each output is equal to the
     * DC coefficient (with scale factor as needed).
     * With typical images and quantization tables, half or more of the
     * column DCT calculations can be simplified this way.
     */

    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
    inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
    inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
    inptr[DCTSIZE*7] == 0) {
      /* AC terms all zero */
      FAST_FLOAT dcval = jict_DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);

      wsptr[DCTSIZE*0] = dcval;
      wsptr[DCTSIZE*1] = dcval;
      wsptr[DCTSIZE*2] = dcval;
      wsptr[DCTSIZE*3] = dcval;
      wsptr[DCTSIZE*4] = dcval;
      wsptr[DCTSIZE*5] = dcval;
      wsptr[DCTSIZE*6] = dcval;
      wsptr[DCTSIZE*7] = dcval;

      inptr++;          /* advance pointers to next column */
      quantptr++;
      wsptr++;
      continue;
    }

    /* Even part */

    tmp0 = jict_DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp1 = jict_DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp2 = jict_DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp3 = jict_DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp10 = tmp0 + tmp2;    /* phase 3 */
    tmp11 = tmp0 - tmp2;

    tmp13 = tmp1 + tmp3;    /* phases 5-3 */
    tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT) 1.414213562) - tmp13; /* 2*c4 */

    tmp0 = tmp10 + tmp13;   /* phase 2 */
    tmp3 = tmp10 - tmp13;
    tmp1 = tmp11 + tmp12;
    tmp2 = tmp11 - tmp12;

    /* Odd part */

    tmp4 = jict_DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    tmp5 = jict_DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    tmp6 = jict_DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp7 = jict_DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    z13 = tmp6 + tmp5;      /* phase 6 */
    z10 = tmp6 - tmp5;
    z11 = tmp4 + tmp7;
    z12 = tmp4 - tmp7;

    tmp7 = z11 + z13;       /* phase 5 */
    tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562); /* 2*c4 */

    z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
    tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
    tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */

    tmp6 = tmp12 - tmp7;    /* phase 2 */
    tmp5 = tmp11 - tmp6;
    tmp4 = tmp10 + tmp5;

    wsptr[DCTSIZE*0] = tmp0 + tmp7;
    wsptr[DCTSIZE*7] = tmp0 - tmp7;
    wsptr[DCTSIZE*1] = tmp1 + tmp6;
    wsptr[DCTSIZE*6] = tmp1 - tmp6;
    wsptr[DCTSIZE*2] = tmp2 + tmp5;
    wsptr[DCTSIZE*5] = tmp2 - tmp5;
    wsptr[DCTSIZE*4] = tmp3 + tmp4;
    wsptr[DCTSIZE*3] = tmp3 - tmp4;

    inptr++;            /* advance pointers to next column */
    quantptr++;
    wsptr++;
  }

  /* Pass 2: process rows from work array, store into output array. */
  /* Note that we must descale the results by a factor of 8 == 2**3. */

  wsptr = workspace;
  for (ctr = 0; ctr < DCTSIZE; ctr++) {
    outptr = output_buf[ctr] + output_col;
    /* Rows of zeroes can be exploited in the same way as we did with columns.
     * However, the column calculation has created many nonzero AC terms, so
     * the simplification applies less often (typically 5% to 10% of the time).
     * And testing floats for zero is relatively expensive, so we don't bother.
     */

    /* Even part */

    tmp10 = wsptr[0] + wsptr[4];
    tmp11 = wsptr[0] - wsptr[4];

    tmp13 = wsptr[2] + wsptr[6];
    tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT) 1.414213562) - tmp13;

    tmp0 = tmp10 + tmp13;
    tmp3 = tmp10 - tmp13;
    tmp1 = tmp11 + tmp12;
    tmp2 = tmp11 - tmp12;

    /* Odd part */

    z13 = wsptr[5] + wsptr[3];
    z10 = wsptr[5] - wsptr[3];
    z11 = wsptr[1] + wsptr[7];
    z12 = wsptr[1] - wsptr[7];

    tmp7 = z11 + z13;
    tmp11 = (z11 - z13) * ((FAST_FLOAT) 1.414213562);

    z5 = (z10 + z12) * ((FAST_FLOAT) 1.847759065); /* 2*c2 */
    tmp10 = ((FAST_FLOAT) 1.082392200) * z12 - z5; /* 2*(c2-c6) */
    tmp12 = ((FAST_FLOAT) -2.613125930) * z10 + z5; /* -2*(c2+c6) */

    tmp6 = tmp12 - tmp7;
    tmp5 = tmp11 - tmp6;
    tmp4 = tmp10 + tmp5;

    /* Final output stage: scale down by a factor of 8 and range-limit */

    outptr[0] = range_limit[(int) DESCALE((INT32) (tmp0 + tmp7), 3)
                & RANGE_MASK];
    outptr[7] = range_limit[(int) DESCALE((INT32) (tmp0 - tmp7), 3)
                & RANGE_MASK];
    outptr[1] = range_limit[(int) DESCALE((INT32) (tmp1 + tmp6), 3)
                & RANGE_MASK];
    outptr[6] = range_limit[(int) DESCALE((INT32) (tmp1 - tmp6), 3)
                & RANGE_MASK];
    outptr[2] = range_limit[(int) DESCALE((INT32) (tmp2 + tmp5), 3)
                & RANGE_MASK];
    outptr[5] = range_limit[(int) DESCALE((INT32) (tmp2 - tmp5), 3)
                & RANGE_MASK];
    outptr[4] = range_limit[(int) DESCALE((INT32) (tmp3 + tmp4), 3)
                & RANGE_MASK];
    outptr[3] = range_limit[(int) DESCALE((INT32) (tmp3 - tmp4), 3)
                & RANGE_MASK];

    wsptr += DCTSIZE;       /* advance pointer to next row */
  }
}

#endif /* DCT_FLOAT_SUPPORTED */
/*
 * jidctint.c
 *
 * Copyright (C) 1991-1998, 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 a slow-but-accurate integer implementation of the
 * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
 * must also perform dequantization of the input coefficients.
 *
 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
 * on each row (or vice versa, but it's more convenient to emit a row at
 * a time).  Direct algorithms are also available, but they are much more
 * complex and seem not to be any faster when reduced to code.
 *
 * This implementation is based on an algorithm described in
 *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
 *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
 *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
 * The primary algorithm described there uses 11 multiplies and 29 adds.
 * We use their alternate method with 12 multiplies and 32 adds.
 * The advantage of this method is that no data path contains more than one
 * multiplication; this allows a very simple and accurate implementation in
 * scaled fixed-point arithmetic, with a minimal number of shifts.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"
//#include "jdct.h"     /* Private declarations for DCT subsystem */

#ifdef DCT_ISLOW_SUPPORTED


/*
 * This module is specialized to the case DCTSIZE = 8.
 */

#if DCTSIZE != 8
  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
#endif


/*
 * The poop on this scaling stuff is as follows:
 *
 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
 * larger than the true IDCT outputs.  The final outputs are therefore
 * a factor of N larger than desired; since N=8 this can be cured by
 * a simple right shift at the end of the algorithm.  The advantage of
 * this arrangement is that we save two multiplications per 1-D IDCT,
 * because the y0 and y4 inputs need not be divided by sqrt(N).
 *
 * We have to do addition and subtraction of the integer inputs, which
 * is no problem, and multiplication by fractional constants, which is
 * a problem to do in integer arithmetic.  We multiply all the constants
 * by CONST_SCALE and convert them to integer constants (thus retaining
 * CONST_BITS bits of precision in the constants).  After doing a
 * multiplication we have to divide the product by CONST_SCALE, with proper
 * rounding, to produce the correct output.  This division can be done
 * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
 * as long as possible so that partial sums can be added together with
 * full fractional precision.
 *
 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
 * they are represented to better-than-integral precision.  These outputs
 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
 * with the recommended scaling.  (To scale up 12-bit sample data further, an
 * intermediate INT32 array would be needed.)
 *
 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
 * shows that the values given below are the most effective.
 */

#undef CONST_BITS
#if BITS_IN_JSAMPLE == 8
#define CONST_BITS  13
#define PASS1_BITS  2
#else
#define CONST_BITS  13
#define PASS1_BITS  1       /* lose a little precision to avoid overflow */
#endif

/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
 * causing a lot of useless floating-point operations at run time.
 * To get around this we use the following pre-calculated constants.
 * If you change CONST_BITS you may want to add appropriate values.
 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
 */

#undef FIX_1_847759065
#if CONST_BITS == 13
#define FIX_0_298631336  ((INT32)  2446)    /* FIX(0.298631336) */
#define FIX_0_390180644  ((INT32)  3196)    /* FIX(0.390180644) */
#define FIX_0_541196100  ((INT32)  4433)    /* FIX(0.541196100) */
#define FIX_0_765366865  ((INT32)  6270)    /* FIX(0.765366865) */
#define FIX_0_899976223  ((INT32)  7373)    /* FIX(0.899976223) */
#define FIX_1_175875602  ((INT32)  9633)    /* FIX(1.175875602) */
#define FIX_1_501321110  ((INT32)  12299)   /* FIX(1.501321110) */
#define FIX_1_847759065  ((INT32)  15137)   /* FIX(1.847759065) */
#define FIX_1_961570560  ((INT32)  16069)   /* FIX(1.961570560) */
#define FIX_2_053119869  ((INT32)  16819)   /* FIX(2.053119869) */
#define FIX_2_562915447  ((INT32)  20995)   /* FIX(2.562915447) */
#define FIX_3_072711026  ((INT32)  25172)   /* FIX(3.072711026) */
#else
#define FIX_0_298631336  FIX(0.298631336)
#define FIX_0_390180644  FIX(0.390180644)
#define FIX_0_541196100  FIX(0.541196100)
#define FIX_0_765366865  FIX(0.765366865)
#define FIX_0_899976223  FIX(0.899976223)
#define FIX_1_175875602  FIX(1.175875602)
#define FIX_1_501321110  FIX(1.501321110)
#define FIX_1_847759065  FIX(1.847759065)
#define FIX_1_961570560  FIX(1.961570560)
#define FIX_2_053119869  FIX(2.053119869)
#define FIX_2_562915447  FIX(2.562915447)
#define FIX_3_072711026  FIX(3.072711026)
#endif


/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
 * For 8-bit samples with the recommended scaling, all the variable
 * and constant values involved are no more than 16 bits wide, so a
 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
 * For 12-bit samples, a full 32-bit multiplication will be needed.
 */

#if BITS_IN_JSAMPLE == 8
#define jicti_jicti_MULTIPLY(var,const)  MULTIPLY16C16(var,const)
#else
#define jicti_jicti_MULTIPLY(var,const)  ((var) * (const))
#endif


/* Dequantize a coefficient by multiplying it by the multiplier-table
 * entry; produce an int result.  In this module, both inputs and result
 * are 16 bits or less, so either int or short multiply will work.
 */

#define jicti_DEQUANTIZE(coef,quantval)  (((ISLOW_MULT_TYPE) (coef)) * (quantval))


/*
 * Perform dequantization and inverse DCT on one block of coefficients.
 */

GLOBAL(void)
jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
         JCOEFPTR coef_block,
         JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3;
  INT32 tmp10, tmp11, tmp12, tmp13;
  INT32 z1, z2, z3, z4, z5;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[DCTSIZE2];  /* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */
  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
  /* furthermore, we scale the results by 2**PASS1_BITS. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = DCTSIZE; ctr > 0; ctr--) {
    /* Due to quantization, we will usually find that many of the input
     * coefficients are zero, especially the AC terms.  We can exploit this
     * by short-circuiting the IDCT calculation for any column in which all
     * the AC terms are zero.  In that case each output is equal to the
     * DC coefficient (with scale factor as needed).
     * With typical images and quantization tables, half or more of the
     * column DCT calculations can be simplified this way.
     */

    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
    inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
    inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
    inptr[DCTSIZE*7] == 0) {
      /* AC terms all zero */
      int dcval = jicti_DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;

      wsptr[DCTSIZE*0] = dcval;
      wsptr[DCTSIZE*1] = dcval;
      wsptr[DCTSIZE*2] = dcval;
      wsptr[DCTSIZE*3] = dcval;
      wsptr[DCTSIZE*4] = dcval;
      wsptr[DCTSIZE*5] = dcval;
      wsptr[DCTSIZE*6] = dcval;
      wsptr[DCTSIZE*7] = dcval;

      inptr++;          /* advance pointers to next column */
      quantptr++;
      wsptr++;
      continue;
    }

    /* Even part: reverse the even part of the forward DCT. */
    /* The rotator is sqrt(2)*c(-6). */

    z2 = jicti_DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = jicti_DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z1 = jicti_jicti_MULTIPLY(z2 + z3, FIX_0_541196100);
    tmp2 = z1 + jicti_jicti_MULTIPLY(z3, - FIX_1_847759065);
    tmp3 = z1 + jicti_jicti_MULTIPLY(z2, FIX_0_765366865);

    z2 = jicti_DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 = jicti_DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);

    tmp0 = (z2 + z3) << CONST_BITS;
    tmp1 = (z2 - z3) << CONST_BITS;

    tmp10 = tmp0 + tmp3;
    tmp13 = tmp0 - tmp3;
    tmp11 = tmp1 + tmp2;
    tmp12 = tmp1 - tmp2;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = jicti_DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
    tmp1 = jicti_DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp2 = jicti_DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    tmp3 = jicti_DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);

    z1 = tmp0 + tmp3;
    z2 = tmp1 + tmp2;
    z3 = tmp0 + tmp2;
    z4 = tmp1 + tmp3;
    z5 = jicti_jicti_MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */

    tmp0 = jicti_jicti_MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
    tmp1 = jicti_jicti_MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
    tmp2 = jicti_jicti_MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
    tmp3 = jicti_jicti_MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
    z1 = jicti_jicti_MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
    z2 = jicti_jicti_MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
    z3 = jicti_jicti_MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
    z4 = jicti_jicti_MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */

    z3 += z5;
    z4 += z5;

    tmp0 += z1 + z3;
    tmp1 += z2 + z4;
    tmp2 += z2 + z3;
    tmp3 += z1 + z4;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    wsptr[DCTSIZE*0] = (int) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*7] = (int) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*1] = (int) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*6] = (int) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*2] = (int) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*5] = (int) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*3] = (int) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*4] = (int) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);

    inptr++;            /* advance pointers to next column */
    quantptr++;
    wsptr++;
  }

  /* Pass 2: process rows from work array, store into output array. */
  /* Note that we must descale the results by a factor of 8 == 2**3, */
  /* and also undo the PASS1_BITS scaling. */

  wsptr = workspace;
  for (ctr = 0; ctr < DCTSIZE; ctr++) {
    outptr = output_buf[ctr] + output_col;
    /* Rows of zeroes can be exploited in the same way as we did with columns.
     * However, the column calculation has created many nonzero AC terms, so
     * the simplification applies less often (typically 5% to 10% of the time).
     * On machines with very fast multiplication, it's possible that the
     * test takes more time than it's worth.  In that case this section
     * may be commented out.
     */

#ifndef NO_ZERO_ROW_TEST
    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
    wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
      /* AC terms all zero */
      JSAMPLE dcval = range_limit[(int) DESCALE((INT32) wsptr[0], PASS1_BITS+3)
                  & RANGE_MASK];

      outptr[0] = dcval;
      outptr[1] = dcval;
      outptr[2] = dcval;
      outptr[3] = dcval;
      outptr[4] = dcval;
      outptr[5] = dcval;
      outptr[6] = dcval;
      outptr[7] = dcval;

      wsptr += DCTSIZE;     /* advance pointer to next row */
      continue;
    }
#endif

    /* Even part: reverse the even part of the forward DCT. */
    /* The rotator is sqrt(2)*c(-6). */

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[6];

    z1 = jicti_jicti_MULTIPLY(z2 + z3, FIX_0_541196100);
    tmp2 = z1 + jicti_jicti_MULTIPLY(z3, - FIX_1_847759065);
    tmp3 = z1 + jicti_jicti_MULTIPLY(z2, FIX_0_765366865);

    tmp0 = ((INT32) wsptr[0] + (INT32) wsptr[4]) << CONST_BITS;
    tmp1 = ((INT32) wsptr[0] - (INT32) wsptr[4]) << CONST_BITS;

    tmp10 = tmp0 + tmp3;
    tmp13 = tmp0 - tmp3;
    tmp11 = tmp1 + tmp2;
    tmp12 = tmp1 - tmp2;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = (INT32) wsptr[7];
    tmp1 = (INT32) wsptr[5];
    tmp2 = (INT32) wsptr[3];
    tmp3 = (INT32) wsptr[1];

    z1 = tmp0 + tmp3;
    z2 = tmp1 + tmp2;
    z3 = tmp0 + tmp2;
    z4 = tmp1 + tmp3;
    z5 = jicti_jicti_MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */

    tmp0 = jicti_jicti_MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */
    tmp1 = jicti_jicti_MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */
    tmp2 = jicti_jicti_MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */
    tmp3 = jicti_jicti_MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */
    z1 = jicti_jicti_MULTIPLY(z1, - FIX_0_899976223); /* sqrt(2) * (c7-c3) */
    z2 = jicti_jicti_MULTIPLY(z2, - FIX_2_562915447); /* sqrt(2) * (-c1-c3) */
    z3 = jicti_jicti_MULTIPLY(z3, - FIX_1_961570560); /* sqrt(2) * (-c3-c5) */
    z4 = jicti_jicti_MULTIPLY(z4, - FIX_0_390180644); /* sqrt(2) * (c5-c3) */

    z3 += z5;
    z4 += z5;

    tmp0 += z1 + z3;
    tmp1 += z2 + z4;
    tmp2 += z2 + z3;
    tmp3 += z1 + z4;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    outptr[0] = range_limit[(int) DESCALE(tmp10 + tmp3,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[7] = range_limit[(int) DESCALE(tmp10 - tmp3,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[1] = range_limit[(int) DESCALE(tmp11 + tmp2,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[6] = range_limit[(int) DESCALE(tmp11 - tmp2,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[2] = range_limit[(int) DESCALE(tmp12 + tmp1,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[5] = range_limit[(int) DESCALE(tmp12 - tmp1,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[3] = range_limit[(int) DESCALE(tmp13 + tmp0,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];
    outptr[4] = range_limit[(int) DESCALE(tmp13 - tmp0,
                      CONST_BITS+PASS1_BITS+3)
                & RANGE_MASK];

    wsptr += DCTSIZE;       /* advance pointer to next row */
  }
}

#endif /* DCT_ISLOW_SUPPORTED */
/*
 * jdsample.c
 *
 * Copyright (C) 1991-1996, 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 upsampling routines.
 *
 * Upsampling input data is counted in "row groups".  A row group
 * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size)
 * sample rows of each component.  Upsampling will normally produce
 * max_v_samp_factor pixel rows from each row group (but this could vary
 * if the upsampler is applying a scale factor of its own).
 *
 * An excellent reference for image resampling is
 *   Digital Image Warping, George Wolberg, 1990.
 *   Pub. by IEEE Computer Society Press, Los Alamitos, CA. ISBN 0-8186-8944-7.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/* Pointer to routine to upsample a single component */
typedef JMETHOD(void, upsample1_ptr,
        (j_decompress_ptr cinfo, jpeg_component_info * compptr,
         JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr));

/* Private subobject */

typedef struct {
  struct jpeg_upsampler pub;    /* public fields */

  /* Color conversion buffer.  When using separate upsampling and color
   * conversion steps, this buffer holds one upsampled row group until it
   * has been color converted and output.
   * Note: we do not allocate any storage for component(s) which are full-size,
   * ie do not need rescaling.  The corresponding entry of color_buf[] is
   * simply set to point to the input data array, thereby avoiding copying.
   */
  JSAMPARRAY color_buf[MAX_COMPONENTS];

  /* Per-component upsampling method pointers */
  upsample1_ptr methods[MAX_COMPONENTS];

  int next_row_out;     /* counts rows emitted from color_buf */
  JDIMENSION rows_to_go;    /* counts rows remaining in image */

  /* Height of an input row group for each component. */
  int rowgroup_height[MAX_COMPONENTS];

  /* These arrays save pixel expansion factors so that int_expand need not
   * recompute them each time.  They are unused for other upsampling methods.
   */
  UINT8 h_expand[MAX_COMPONENTS];
  UINT8 v_expand[MAX_COMPONENTS];
} my_upsampler;

typedef my_upsampler * my_upsample_ptr;


/*
 * Initialize for an upsampling pass.
 */

METHODDEF(void)
start_pass_upsample (j_decompress_ptr cinfo)
{
  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;

  /* Mark the conversion buffer empty */
  upsample->next_row_out = cinfo->max_v_samp_factor;
  /* Initialize total-height counter for detecting bottom of image */
  upsample->rows_to_go = cinfo->output_height;
}


/*
 * Control routine to do upsampling (and color conversion).
 *
 * In this version we upsample each component independently.
 * We upsample one row group into the conversion buffer, then apply
 * color conversion a row at a time.
 */

METHODDEF(void)
sep_upsample (j_decompress_ptr cinfo,
          JSAMPIMAGE input_buf, JDIMENSION *in_row_group_ctr,
          JDIMENSION in_row_groups_avail,
          JSAMPARRAY output_buf, JDIMENSION *out_row_ctr,
          JDIMENSION out_rows_avail)
{
  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
  int ci;
  jpeg_component_info * compptr;
  JDIMENSION num_rows;

  /* Fill the conversion buffer, if it's empty */
  if (upsample->next_row_out >= cinfo->max_v_samp_factor) {
    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
     ci++, compptr++) {
      /* Invoke per-component upsample method.  Notice we pass a POINTER
       * to color_buf[ci], so that fullsize_upsample can change it.
       */
      (*upsample->methods[ci]) (cinfo, compptr,
    input_buf[ci] + (*in_row_group_ctr * upsample->rowgroup_height[ci]),
    upsample->color_buf + ci);
    }
    upsample->next_row_out = 0;
  }

  /* Color-convert and emit rows */

  /* How many we have in the buffer: */
  num_rows = (JDIMENSION) (cinfo->max_v_samp_factor - upsample->next_row_out);
  /* Not more than the distance to the end of the image.  Need this test
   * in case the image height is not a multiple of max_v_samp_factor:
   */
  if (num_rows > upsample->rows_to_go)
    num_rows = upsample->rows_to_go;
  /* And not more than what the client can accept: */
  out_rows_avail -= *out_row_ctr;
  if (num_rows > out_rows_avail)
    num_rows = out_rows_avail;

  (*cinfo->cconvert->color_convert) (cinfo, upsample->color_buf,
                     (JDIMENSION) upsample->next_row_out,
                     output_buf + *out_row_ctr,
                     (int) num_rows);

  /* Adjust counts */
  *out_row_ctr += num_rows;
  upsample->rows_to_go -= num_rows;
  upsample->next_row_out += num_rows;
  /* When the buffer is emptied, declare this input row group consumed */
  if (upsample->next_row_out >= cinfo->max_v_samp_factor)
    (*in_row_group_ctr)++;
}


/*
 * These are the routines invoked by sep_upsample to upsample pixel values
 * of a single component.  One row group is processed per call.
 */


/*
 * For full-size components, we just make color_buf[ci] point at the
 * input buffer, and thus avoid copying any data.  Note that this is
 * safe only because sep_upsample doesn't declare the input row group
 * "consumed" until we are done color converting and emitting it.
 */

METHODDEF(void)
fullsize_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
           JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  *output_data_ptr = input_data;
}


/*
 * This is a no-op version used for "uninteresting" components.
 * These components will not be referenced by color conversion.
 */

METHODDEF(void)
noop_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
           JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  *output_data_ptr = NULL;  /* safety check */
}


/*
 * This version handles any integral sampling ratios.
 * This is not used for typical JPEG files, so it need not be fast.
 * Nor, for that matter, is it particularly accurate: the algorithm is
 * simple replication of the input pixel onto the corresponding output
 * pixels.  The hi-falutin sampling literature refers to this as a
 * "box filter".  A box filter tends to introduce visible artifacts,
 * so if you are actually going to use 3:1 or 4:1 sampling ratios
 * you would be well advised to improve this code.
 */

METHODDEF(void)
int_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
          JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  my_upsample_ptr upsample = (my_upsample_ptr) cinfo->upsample;
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr, outptr;
  JSAMPLE invalue;
  int h;
  JSAMPROW outend;
  int h_expand, v_expand;
  int inrow, outrow;

  h_expand = upsample->h_expand[compptr->component_index];
  v_expand = upsample->v_expand[compptr->component_index];

  inrow = outrow = 0;
  while (outrow < cinfo->max_v_samp_factor) {
    /* Generate one output row with proper horizontal expansion */
    inptr = input_data[inrow];
    outptr = output_data[outrow];
    outend = outptr + cinfo->output_width;
    while (outptr < outend) {
      invalue = *inptr++;   /* don't need GETJSAMPLE() here */
      for (h = h_expand; h > 0; h--) {
    *outptr++ = invalue;
      }
    }
    /* Generate any additional output rows by duplicating the first one */
    if (v_expand > 1) {
      jcopy_sample_rows(output_data, outrow, output_data, outrow+1,
            v_expand-1, cinfo->output_width);
    }
    inrow++;
    outrow += v_expand;
  }
}


/*
 * Fast processing for the common case of 2:1 horizontal and 1:1 vertical.
 * It's still a box filter.
 */

METHODDEF(void)
h2v1_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
           JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr, outptr;
  JSAMPLE invalue;
  JSAMPROW outend;
  int inrow;

  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
    inptr = input_data[inrow];
    outptr = output_data[inrow];
    outend = outptr + cinfo->output_width;
    while (outptr < outend) {
      invalue = *inptr++;   /* don't need GETJSAMPLE() here */
      *outptr++ = invalue;
      *outptr++ = invalue;
    }
  }
}


/*
 * Fast processing for the common case of 2:1 horizontal and 2:1 vertical.
 * It's still a box filter.
 */

METHODDEF(void)
h2v2_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
           JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr, outptr;
  JSAMPLE invalue;
  JSAMPROW outend;
  int inrow, outrow;

  inrow = outrow = 0;
  while (outrow < cinfo->max_v_samp_factor) {
    inptr = input_data[inrow];
    outptr = output_data[outrow];
    outend = outptr + cinfo->output_width;
    while (outptr < outend) {
      invalue = *inptr++;   /* don't need GETJSAMPLE() here */
      *outptr++ = invalue;
      *outptr++ = invalue;
    }
    jcopy_sample_rows(output_data, outrow, output_data, outrow+1,
              1, cinfo->output_width);
    inrow++;
    outrow += 2;
  }
}


/*
 * Fancy processing for the common case of 2:1 horizontal and 1:1 vertical.
 *
 * The upsampling algorithm is linear interpolation between pixel centers,
 * also known as a "triangle filter".  This is a good compromise between
 * speed and visual quality.  The centers of the output pixels are 1/4 and 3/4
 * of the way between input pixel centers.
 *
 * A note about the "bias" calculations: when rounding fractional values to
 * integer, we do not want to always round 0.5 up to the next integer.
 * If we did that, we'd introduce a noticeable bias towards larger values.
 * Instead, this code is arranged so that 0.5 will be rounded up or down at
 * alternate pixel locations (a simple ordered dither pattern).
 */

METHODDEF(void)
h2v1_fancy_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
             JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr, outptr;
  int invalue;
  JDIMENSION colctr;
  int inrow;

  for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) {
    inptr = input_data[inrow];
    outptr = output_data[inrow];
    /* Special case for first column */
    invalue = GETJSAMPLE(*inptr++);
    *outptr++ = (JSAMPLE) invalue;
    *outptr++ = (JSAMPLE) ((invalue * 3 + GETJSAMPLE(*inptr) + 2) >> 2);

    for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) {
      /* General case: 3/4 * nearer pixel + 1/4 * further pixel */
      invalue = GETJSAMPLE(*inptr++) * 3;
      *outptr++ = (JSAMPLE) ((invalue + GETJSAMPLE(inptr[-2]) + 1) >> 2);
      *outptr++ = (JSAMPLE) ((invalue + GETJSAMPLE(*inptr) + 2) >> 2);
    }

    /* Special case for last column */
    invalue = GETJSAMPLE(*inptr);
    *outptr++ = (JSAMPLE) ((invalue * 3 + GETJSAMPLE(inptr[-1]) + 1) >> 2);
    *outptr++ = (JSAMPLE) invalue;
  }
}


/*
 * Fancy processing for the common case of 2:1 horizontal and 2:1 vertical.
 * Again a triangle filter; see comments for h2v1 case, above.
 *
 * It is OK for us to reference the adjacent input rows because we demanded
 * context from the main buffer controller (see initialization code).
 */

METHODDEF(void)
h2v2_fancy_upsample (j_decompress_ptr cinfo, jpeg_component_info * compptr,
             JSAMPARRAY input_data, JSAMPARRAY * output_data_ptr)
{
  JSAMPARRAY output_data = *output_data_ptr;
  JSAMPROW inptr0, inptr1, outptr;
#if BITS_IN_JSAMPLE == 8
  int thiscolsum, lastcolsum, nextcolsum;
#else
  INT32 thiscolsum, lastcolsum, nextcolsum;
#endif
  JDIMENSION colctr;
  int inrow, outrow, v;

  inrow = outrow = 0;
  while (outrow < cinfo->max_v_samp_factor) {
    for (v = 0; v < 2; v++) {
      /* inptr0 points to nearest input row, inptr1 points to next nearest */
      inptr0 = input_data[inrow];
      if (v == 0)       /* next nearest is row above */
    inptr1 = input_data[inrow-1];
      else          /* next nearest is row below */
    inptr1 = input_data[inrow+1];
      outptr = output_data[outrow++];

      /* Special case for first column */
      thiscolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
      nextcolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
      *outptr++ = (JSAMPLE) ((thiscolsum * 4 + 8) >> 4);
      *outptr++ = (JSAMPLE) ((thiscolsum * 3 + nextcolsum + 7) >> 4);
      lastcolsum = thiscolsum; thiscolsum = nextcolsum;

      for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) {
    /* General case: 3/4 * nearer pixel + 1/4 * further pixel in each */
    /* dimension, thus 9/16, 3/16, 3/16, 1/16 overall */
    nextcolsum = GETJSAMPLE(*inptr0++) * 3 + GETJSAMPLE(*inptr1++);
    *outptr++ = (JSAMPLE) ((thiscolsum * 3 + lastcolsum + 8) >> 4);
    *outptr++ = (JSAMPLE) ((thiscolsum * 3 + nextcolsum + 7) >> 4);
    lastcolsum = thiscolsum; thiscolsum = nextcolsum;
      }

      /* Special case for last column */
      *outptr++ = (JSAMPLE) ((thiscolsum * 3 + lastcolsum + 8) >> 4);
      *outptr++ = (JSAMPLE) ((thiscolsum * 4 + 7) >> 4);
    }
    inrow++;
  }
}


/*
 * Module initialization routine for upsampling.
 */

GLOBAL(void)
jinit_upsampler (j_decompress_ptr cinfo)
{
  my_upsample_ptr upsample;
  int ci;
  jpeg_component_info * compptr;
  boolean need_buffer, do_fancy;
  int h_in_group, v_in_group, h_out_group, v_out_group;

  upsample = (my_upsample_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(my_upsampler));
  cinfo->upsample = (struct jpeg_upsampler *) upsample;
  upsample->pub.start_pass = start_pass_upsample;
  upsample->pub.upsample = sep_upsample;
  upsample->pub.need_context_rows = FALSE; /* until we find out differently */

  if (cinfo->CCIR601_sampling)  /* this isn't supported */
    ERREXIT(cinfo, JERR_CCIR601_NOTIMPL);

  /* jdmainct.c doesn't support context rows when min_DCT_scaled_size = 1,
   * so don't ask for it.
   */
  do_fancy = cinfo->do_fancy_upsampling && cinfo->min_DCT_scaled_size > 1;

  /* Verify we can handle the sampling factors, select per-component methods,
   * and create storage as needed.
   */
  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    /* Compute size of an "input group" after IDCT scaling.  This many samples
     * are to be converted to max_h_samp_factor * max_v_samp_factor pixels.
     */
    h_in_group = (compptr->h_samp_factor * compptr->DCT_scaled_size) /
         cinfo->min_DCT_scaled_size;
    v_in_group = (compptr->v_samp_factor * compptr->DCT_scaled_size) /
         cinfo->min_DCT_scaled_size;
    h_out_group = cinfo->max_h_samp_factor;
    v_out_group = cinfo->max_v_samp_factor;
    upsample->rowgroup_height[ci] = v_in_group; /* save for use later */
    need_buffer = TRUE;
    if (! compptr->component_needed) {
      /* Don't bother to upsample an uninteresting component. */
      upsample->methods[ci] = noop_upsample;
      need_buffer = FALSE;
    } else if (h_in_group == h_out_group && v_in_group == v_out_group) {
      /* Fullsize components can be processed without any work. */
      upsample->methods[ci] = fullsize_upsample;
      need_buffer = FALSE;
    } else if (h_in_group * 2 == h_out_group &&
           v_in_group == v_out_group) {
      /* Special cases for 2h1v upsampling */
      if (do_fancy && compptr->downsampled_width > 2)
    upsample->methods[ci] = h2v1_fancy_upsample;
      else
    upsample->methods[ci] = h2v1_upsample;
    } else if (h_in_group * 2 == h_out_group &&
           v_in_group * 2 == v_out_group) {
      /* Special cases for 2h2v upsampling */
      if (do_fancy && compptr->downsampled_width > 2) {
    upsample->methods[ci] = h2v2_fancy_upsample;
    upsample->pub.need_context_rows = TRUE;
      } else
    upsample->methods[ci] = h2v2_upsample;
    } else if ((h_out_group % h_in_group) == 0 &&
           (v_out_group % v_in_group) == 0) {
      /* Generic integral-factors upsampling method */
      upsample->methods[ci] = int_upsample;
      upsample->h_expand[ci] = (UINT8) (h_out_group / h_in_group);
      upsample->v_expand[ci] = (UINT8) (v_out_group / v_in_group);
    } else
      ERREXIT(cinfo, JERR_FRACT_SAMPLE_NOTIMPL);
    if (need_buffer) {
      upsample->color_buf[ci] = (*cinfo->mem->alloc_sarray)
    ((j_common_ptr) cinfo, JPOOL_IMAGE,
     (JDIMENSION) jround_up((long) cinfo->output_width,
                (long) cinfo->max_h_samp_factor),
     (JDIMENSION) cinfo->max_v_samp_factor);
    }
  }
}
/*
 * jdcolor.c
 *
 * Copyright (C) 1991-1997, 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 output colorspace conversion routines.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/* Private subobject */

typedef struct {
  struct jpeg_color_deconverter pub; /* public fields */

  /* Private state for YCC->RGB conversion */
  int * Cr_r_tab;       /* => table for Cr to R conversion */
  int * Cb_b_tab;       /* => table for Cb to B conversion */
  INT32 * Cr_g_tab;     /* => table for Cr to G conversion */
  INT32 * Cb_g_tab;     /* => table for Cb to G conversion */
} my_color_deconverter;

typedef my_color_deconverter * my_cconvert_ptr;


/**************** YCbCr -> RGB conversion: most common case **************/

/*
 * YCbCr is defined per CCIR 601-1, except that Cb and Cr are
 * normalized to the range 0..MAXJSAMPLE rather than -0.5 .. 0.5.
 * The conversion equations to be implemented are therefore
 *  R = Y                + 1.40200 * Cr
 *  G = Y - 0.34414 * Cb - 0.71414 * Cr
 *  B = Y + 1.77200 * Cb
 * where Cb and Cr represent the incoming values less CENTERJSAMPLE.
 * (These numbers are derived from TIFF 6.0 section 21, dated 3-June-92.)
 *
 * To avoid floating-point arithmetic, we represent the fractional constants
 * as integers scaled up by 2^16 (about 4 digits precision); we have to divide
 * the products by 2^16, with appropriate rounding, to get the correct answer.
 * Notice that Y, being an integral input, does not contribute any fraction
 * so it need not participate in the rounding.
 *
 * For even more speed, we avoid doing any multiplications in the inner loop
 * by precalculating the constants times Cb and Cr for all possible values.
 * For 8-bit JSAMPLEs this is very reasonable (only 256 entries per table);
 * for 12-bit samples it is still acceptable.  It's not very reasonable for
 * 16-bit samples, but if you want lossless storage you shouldn't be changing
 * colorspace anyway.
 * The Cr=>R and Cb=>B values can be rounded to integers in advance; the
 * values for the G calculation are left scaled up, since we must add them
 * together before rounding.
 */

#define SCALEBITS   16  /* speediest right-shift on some machines */
#define ONE_HALF    ((INT32) 1 << (SCALEBITS-1))
#define jdol_FIX(x)     ((INT32) ((x) * (1L<<SCALEBITS) + 0.5))


/*
 * Initialize tables for YCC->RGB colorspace conversion.
 */

LOCAL(void)
build_ycc_rgb_table (j_decompress_ptr cinfo)
{
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
  int i;
  INT32 x;
  SHIFT_TEMPS

  cconvert->Cr_r_tab = (int *)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                (MAXJSAMPLE+1) * SIZEOF(int));
  cconvert->Cb_b_tab = (int *)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                (MAXJSAMPLE+1) * SIZEOF(int));
  cconvert->Cr_g_tab = (INT32 *)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                (MAXJSAMPLE+1) * SIZEOF(INT32));
  cconvert->Cb_g_tab = (INT32 *)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                (MAXJSAMPLE+1) * SIZEOF(INT32));

  for (i = 0, x = -CENTERJSAMPLE; i <= MAXJSAMPLE; i++, x++) {
    /* i is the actual input pixel value, in the range 0..MAXJSAMPLE */
    /* The Cb or Cr value we are thinking of is x = i - CENTERJSAMPLE */
    /* Cr=>R value is nearest int to 1.40200 * x */
    cconvert->Cr_r_tab[i] = (int)
            RIGHT_SHIFT(jdol_FIX(1.40200) * x + ONE_HALF, SCALEBITS);
    /* Cb=>B value is nearest int to 1.77200 * x */
    cconvert->Cb_b_tab[i] = (int)
            RIGHT_SHIFT(jdol_FIX(1.77200) * x + ONE_HALF, SCALEBITS);
    /* Cr=>G value is scaled-up -0.71414 * x */
    cconvert->Cr_g_tab[i] = (- jdol_FIX(0.71414)) * x;
    /* Cb=>G value is scaled-up -0.34414 * x */
    /* We also add in ONE_HALF so that need not do it in inner loop */
    cconvert->Cb_g_tab[i] = (- jdol_FIX(0.34414)) * x + ONE_HALF;
  }
}


/*
 * Convert some rows of samples to the output colorspace.
 *
 * Note that we change from noninterleaved, one-plane-per-component format
 * to interleaved-pixel format.  The output buffer is therefore three times
 * as wide as the input buffer.
 * A starting row offset is provided only for the input buffer.  The caller
 * can easily adjust the passed output_buf value to accommodate any row
 * offset required on that side.
 */

METHODDEF(void)
ycc_rgb_convert (j_decompress_ptr cinfo,
         JSAMPIMAGE input_buf, JDIMENSION input_row,
         JSAMPARRAY output_buf, int num_rows)
{
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
  int y, cb, cr;
  JSAMPROW outptr;
  JSAMPROW inptr0, inptr1, inptr2;
  JDIMENSION col;
  JDIMENSION num_cols = cinfo->output_width;
  /* copy these pointers into registers if possible */
  JSAMPLE * range_limit = cinfo->sample_range_limit;
  int * Crrtab = cconvert->Cr_r_tab;
  int * Cbbtab = cconvert->Cb_b_tab;
  INT32 * Crgtab = cconvert->Cr_g_tab;
  INT32 * Cbgtab = cconvert->Cb_g_tab;
  SHIFT_TEMPS

  while (--num_rows >= 0) {
    inptr0 = input_buf[0][input_row];
    inptr1 = input_buf[1][input_row];
    inptr2 = input_buf[2][input_row];
    input_row++;
    outptr = *output_buf++;
    for (col = 0; col < num_cols; col++) {
      y  = GETJSAMPLE(inptr0[col]);
      cb = GETJSAMPLE(inptr1[col]);
      cr = GETJSAMPLE(inptr2[col]);
      /* Range-limiting is essential due to noise introduced by DCT losses. */
      outptr[RGB_RED] =   range_limit[y + Crrtab[cr]];
      outptr[RGB_GREEN] = range_limit[y +
                  ((int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
                         SCALEBITS))];
      outptr[RGB_BLUE] =  range_limit[y + Cbbtab[cb]];
      outptr += RGB_PIXELSIZE;
    }
  }
}


/**************** Cases other than YCbCr -> RGB **************/


/*
 * Color conversion for no colorspace change: just copy the data,
 * converting from separate-planes to interleaved representation.
 */

METHODDEF(void)
null_convert (j_decompress_ptr cinfo,
          JSAMPIMAGE input_buf, JDIMENSION input_row,
          JSAMPARRAY output_buf, int num_rows)
{
  JSAMPROW inptr, outptr;
  JDIMENSION count;
  int num_components = cinfo->num_components;
  JDIMENSION num_cols = cinfo->output_width;
  int ci;

  while (--num_rows >= 0) {
    for (ci = 0; ci < num_components; ci++) {
      inptr = input_buf[ci][input_row];
      outptr = output_buf[0] + ci;
      for (count = num_cols; count > 0; count--) {
    *outptr = *inptr++; /* needn't bother with GETJSAMPLE() here */
    outptr += num_components;
      }
    }
    input_row++;
    output_buf++;
  }
}


/*
 * Color conversion for grayscale: just copy the data.
 * This also works for YCbCr -> grayscale conversion, in which
 * we just copy the Y (luminance) component and ignore chrominance.
 */

METHODDEF(void)
grayscale_convert (j_decompress_ptr cinfo,
           JSAMPIMAGE input_buf, JDIMENSION input_row,
           JSAMPARRAY output_buf, int num_rows)
{
  jcopy_sample_rows(input_buf[0], (int) input_row, output_buf, 0,
            num_rows, cinfo->output_width);
}


/*
 * Convert grayscale to RGB: just duplicate the graylevel three times.
 * This is provided to support applications that don't want to cope
 * with grayscale as a separate case.
 */

METHODDEF(void)
gray_rgb_convert (j_decompress_ptr cinfo,
          JSAMPIMAGE input_buf, JDIMENSION input_row,
          JSAMPARRAY output_buf, int num_rows)
{
  JSAMPROW inptr, outptr;
  JDIMENSION col;
  JDIMENSION num_cols = cinfo->output_width;

  while (--num_rows >= 0) {
    inptr = input_buf[0][input_row++];
    outptr = *output_buf++;
    for (col = 0; col < num_cols; col++) {
      /* We can dispense with GETJSAMPLE() here */
      outptr[RGB_RED] = outptr[RGB_GREEN] = outptr[RGB_BLUE] = inptr[col];
      outptr += RGB_PIXELSIZE;
    }
  }
}


/*
 * Adobe-style YCCK->CMYK conversion.
 * We convert YCbCr to R=1-C, G=1-M, and B=1-Y using the same
 * conversion as above, while passing K (black) unchanged.
 * We assume build_ycc_rgb_table has been called.
 */

METHODDEF(void)
ycck_cmyk_convert (j_decompress_ptr cinfo,
           JSAMPIMAGE input_buf, JDIMENSION input_row,
           JSAMPARRAY output_buf, int num_rows)
{
  my_cconvert_ptr cconvert = (my_cconvert_ptr) cinfo->cconvert;
  int y, cb, cr;
  JSAMPROW outptr;
  JSAMPROW inptr0, inptr1, inptr2, inptr3;
  JDIMENSION col;
  JDIMENSION num_cols = cinfo->output_width;
  /* copy these pointers into registers if possible */
  JSAMPLE * range_limit = cinfo->sample_range_limit;
  int * Crrtab = cconvert->Cr_r_tab;
  int * Cbbtab = cconvert->Cb_b_tab;
  INT32 * Crgtab = cconvert->Cr_g_tab;
  INT32 * Cbgtab = cconvert->Cb_g_tab;
  SHIFT_TEMPS

  while (--num_rows >= 0) {
    inptr0 = input_buf[0][input_row];
    inptr1 = input_buf[1][input_row];
    inptr2 = input_buf[2][input_row];
    inptr3 = input_buf[3][input_row];
    input_row++;
    outptr = *output_buf++;
    for (col = 0; col < num_cols; col++) {
      y  = GETJSAMPLE(inptr0[col]);
      cb = GETJSAMPLE(inptr1[col]);
      cr = GETJSAMPLE(inptr2[col]);
      /* Range-limiting is essential due to noise introduced by DCT losses. */
      outptr[0] = range_limit[MAXJSAMPLE - (y + Crrtab[cr])];   /* red */
      outptr[1] = range_limit[MAXJSAMPLE - (y +         /* green */
                  ((int) RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr],
                         SCALEBITS)))];
      outptr[2] = range_limit[MAXJSAMPLE - (y + Cbbtab[cb])];   /* blue */
      /* K passes through unchanged */
      outptr[3] = inptr3[col];  /* don't need GETJSAMPLE here */
      outptr += 4;
    }
  }
}


/*
 * Empty method for start_pass.
 */

METHODDEF(void)
start_pass_dcolor (j_decompress_ptr cinfo)
{
  /* no work needed */
}


/*
 * Module initialization routine for output colorspace conversion.
 */

GLOBAL(void)
jinit_color_deconverter (j_decompress_ptr cinfo)
{
  my_cconvert_ptr cconvert;
  int ci;

  cconvert = (my_cconvert_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(my_color_deconverter));
  cinfo->cconvert = (struct jpeg_color_deconverter *) cconvert;
  cconvert->pub.start_pass = start_pass_dcolor;

  /* Make sure num_components agrees with jpeg_color_space */
  switch (cinfo->jpeg_color_space) {
  case JCS_GRAYSCALE:
    if (cinfo->num_components != 1)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
    break;

  case JCS_RGB:
  case JCS_YCbCr:
    if (cinfo->num_components != 3)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
    break;

  case JCS_CMYK:
  case JCS_YCCK:
    if (cinfo->num_components != 4)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
    break;

  default:          /* JCS_UNKNOWN can be anything */
    if (cinfo->num_components < 1)
      ERREXIT(cinfo, JERR_BAD_J_COLORSPACE);
    break;
  }

  /* Set out_color_components and conversion method based on requested space.
   * Also clear the component_needed flags for any unused components,
   * so that earlier pipeline stages can avoid useless computation.
   */

  switch (cinfo->out_color_space) {
  case JCS_GRAYSCALE:
    cinfo->out_color_components = 1;
    if (cinfo->jpeg_color_space == JCS_GRAYSCALE ||
    cinfo->jpeg_color_space == JCS_YCbCr) {
      cconvert->pub.color_convert = grayscale_convert;
      /* For color->grayscale conversion, only the Y (0) component is needed */
      for (ci = 1; ci < cinfo->num_components; ci++)
    cinfo->comp_info[ci].component_needed = FALSE;
    } else
      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
    break;

  case JCS_RGB:
    cinfo->out_color_components = RGB_PIXELSIZE;
    if (cinfo->jpeg_color_space == JCS_YCbCr) {
      cconvert->pub.color_convert = ycc_rgb_convert;
      build_ycc_rgb_table(cinfo);
    } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) {
      cconvert->pub.color_convert = gray_rgb_convert;
    } else if (cinfo->jpeg_color_space == JCS_RGB && RGB_PIXELSIZE == 3) {
      cconvert->pub.color_convert = null_convert;
    } else
      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
    break;

  case JCS_CMYK:
    cinfo->out_color_components = 4;
    if (cinfo->jpeg_color_space == JCS_YCCK) {
      cconvert->pub.color_convert = ycck_cmyk_convert;
      build_ycc_rgb_table(cinfo);
    } else if (cinfo->jpeg_color_space == JCS_CMYK) {
      cconvert->pub.color_convert = null_convert;
    } else
      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
    break;

  default:
    /* Permit null conversion to same output space */
    if (cinfo->out_color_space == cinfo->jpeg_color_space) {
      cinfo->out_color_components = cinfo->num_components;
      cconvert->pub.color_convert = null_convert;
    } else          /* unsupported non-null conversion */
      ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL);
    break;
  }

  if (cinfo->quantize_colors)
    cinfo->output_components = 1; /* single colormapped output component */
  else
    cinfo->output_components = cinfo->out_color_components;
}
/*
 * jcomapi.c
 *
 * Copyright (C) 1994-1997, 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 application interface routines that are used for both
 * compression and decompression.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/*
 * Abort processing of a JPEG compression or decompression operation,
 * but don't destroy the object itself.
 *
 * For this, we merely clean up all the nonpermanent memory pools.
 * Note that temp files (virtual arrays) are not allowed to belong to
 * the permanent pool, so we will be able to close all temp files here.
 * Closing a data source or destination, if necessary, is the application's
 * responsibility.
 */

GLOBAL(void)
jpeg_abort (j_common_ptr cinfo)
{
  int pool;

  /* Do nothing if called on a not-initialized or destroyed JPEG object. */
  if (cinfo->mem == NULL)
    return;

  /* Releasing pools in reverse order might help avoid fragmentation
   * with some (brain-damaged) malloc libraries.
   */
  for (pool = JPOOL_NUMPOOLS-1; pool > JPOOL_PERMANENT; pool--) {
    (*cinfo->mem->free_pool) (cinfo, pool);
  }

  /* Reset overall state for possible reuse of object */
  if (cinfo->is_decompressor) {
    cinfo->global_state = DSTATE_START;
    /* Try to keep application from accessing now-deleted marker list.
     * A bit kludgy to do it here, but this is the most central place.
     */
    ((j_decompress_ptr) cinfo)->marker_list = NULL;
  } else {
    cinfo->global_state = CSTATE_START;
  }
}


/*
 * Destruction of a JPEG object.
 *
 * Everything gets deallocated except the master jpeg_compress_struct itself
 * and the error manager struct.  Both of these are supplied by the application
 * and must be freed, if necessary, by the application.  (Often they are on
 * the stack and so don't need to be freed anyway.)
 * Closing a data source or destination, if necessary, is the application's
 * responsibility.
 */

GLOBAL(void)
jpeg_destroy (j_common_ptr cinfo)
{
  /* We need only tell the memory manager to release everything. */
  /* NB: mem pointer is NULL if memory mgr failed to initialize. */
  if (cinfo->mem != NULL)
    (*cinfo->mem->self_destruct) (cinfo);
  cinfo->mem = NULL;        /* be safe if jpeg_destroy is called twice */
  cinfo->global_state = 0;  /* mark it destroyed */
}


/*
 * Convenience routines for allocating quantization and Huffman tables.
 * (Would jutils.c be a more reasonable place to put these?)
 */

GLOBAL(JQUANT_TBL *)
jpeg_alloc_quant_table (j_common_ptr cinfo)
{
  JQUANT_TBL *tbl;

  tbl = (JQUANT_TBL *)
    (*cinfo->mem->alloc_small) (cinfo, JPOOL_PERMANENT, SIZEOF(JQUANT_TBL));
  tbl->sent_table = FALSE;  /* make sure this is false in any new table */
  return tbl;
}


GLOBAL(JHUFF_TBL *)
jpeg_alloc_huff_table (j_common_ptr cinfo)
{
  JHUFF_TBL *tbl;

  tbl = (JHUFF_TBL *)
    (*cinfo->mem->alloc_small) (cinfo, JPOOL_PERMANENT, SIZEOF(JHUFF_TBL));
  tbl->sent_table = FALSE;  /* make sure this is false in any new table */
  return tbl;
}
/*
 * jutils.c
 *
 * Copyright (C) 1991-1996, 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 tables and miscellaneous utility routines needed
 * for both compression and decompression.
 * Note we prefix all global names with "j" to minimize conflicts with
 * a surrounding application.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"


/*
 * jpeg_zigzag_order[i] is the zigzag-order position of the i'th element
 * of a DCT block read in natural order (left to right, top to bottom).
 */

#if 0               /* This table is not actually needed in v6a */

const int jpeg_zigzag_order[DCTSIZE2] = {
   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
};

#endif

/*
 * jpeg_natural_order[i] is the natural-order position of the i'th element
 * of zigzag order.
 *
 * When reading corrupted data, the Huffman decoders could attempt
 * to reference an entry beyond the end of this array (if the decoded
 * zero run length reaches past the end of the block).  To prevent
 * wild stores without adding an inner-loop test, we put some extra
 * "63"s after the real entries.  This will cause the extra coefficient
 * to be stored in location 63 of the block, not somewhere random.
 * The worst case would be a run-length of 15, which means we need 16
 * fake entries.
 */

const int jpeg_natural_order[DCTSIZE2+16] = {
  0,  1,  8, 16,  9,  2,  3, 10,
 17, 24, 32, 25, 18, 11,  4,  5,
 12, 19, 26, 33, 40, 48, 41, 34,
 27, 20, 13,  6,  7, 14, 21, 28,
 35, 42, 49, 56, 57, 50, 43, 36,
 29, 22, 15, 23, 30, 37, 44, 51,
 58, 59, 52, 45, 38, 31, 39, 46,
 53, 60, 61, 54, 47, 55, 62, 63,
 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */
 63, 63, 63, 63, 63, 63, 63, 63
};


/*
 * Arithmetic utilities
 */

GLOBAL(long)
jdiv_round_up (long a, long b)
/* Compute a/b rounded up to next integer, ie, ceil(a/b) */
/* Assumes a >= 0, b > 0 */
{
  return (a + b - 1L) / b;
}


GLOBAL(long)
jround_up (long a, long b)
/* Compute a rounded up to next multiple of b, ie, ceil(a/b)*b */
/* Assumes a >= 0, b > 0 */
{
  a += b - 1L;
  return a - (a % b);
}


/* On normal machines we can apply MEMCOPY() and MEMZERO() to sample arrays
 * and coefficient-block arrays.  This won't work on 80x86 because the arrays
 * are FAR and we're assuming a small-pointer memory model.  However, some
 * DOS compilers provide far-pointer versions of memcpy() and memset() even
 * in the small-model libraries.  These will be used if USE_FMEM is defined.
 * Otherwise, the routines below do it the hard way.  (The performance cost
 * is not all that great, because these routines aren't very heavily used.)
 */

#ifndef NEED_FAR_POINTERS   /* normal case, same as regular macros */
#define jui_jui_FMEMCOPY(dest,src,size) MEMCOPY(dest,src,size)
#define jui_jui_FMEMZERO(target,size)   MEMZERO(target,size)
#else               /* 80x86 case, define if we can */
#ifdef USE_FMEM
#define jui_jui_FMEMCOPY(dest,src,size) _fmemcpy((void FAR *)(dest), (const void FAR *)(src), (size_t)(size))
#define jui_jui_FMEMZERO(target,size)   _fmemset((void FAR *)(target), 0, (size_t)(size))
#endif
#endif


GLOBAL(void)
jcopy_sample_rows (JSAMPARRAY input_array, int source_row,
           JSAMPARRAY output_array, int dest_row,
           int num_rows, JDIMENSION num_cols)
/* Copy some rows of samples from one place to another.
 * num_rows rows are copied from input_array[source_row++]
 * to output_array[dest_row++]; these areas may overlap for duplication.
 * The source and destination arrays must be at least as wide as num_cols.
 */
{
  JSAMPROW inptr, outptr;
#ifdef FMEMCOPY
  size_t count = (size_t) (num_cols * SIZEOF(JSAMPLE));
#else
  JDIMENSION count;
#endif
  int row;

  input_array += source_row;
  output_array += dest_row;

  for (row = num_rows; row > 0; row--) {
    inptr = *input_array++;
    outptr = *output_array++;
#ifdef FMEMCOPY
    jui_jui_FMEMCOPY(outptr, inptr, count);
#else
    for (count = num_cols; count > 0; count--)
      *outptr++ = *inptr++; /* needn't bother with GETJSAMPLE() here */
#endif
  }
}


GLOBAL(void)
jcopy_block_row (JBLOCKROW input_row, JBLOCKROW output_row,
         JDIMENSION num_blocks)
/* Copy a row of coefficient blocks from one place to another. */
{
#ifdef FMEMCOPY
  jui_jui_FMEMCOPY(output_row, input_row, num_blocks * (DCTSIZE2 * SIZEOF(JCOEF)));
#else
  JCOEFPTR inptr, outptr;
  long count;

  inptr = (JCOEFPTR) input_row;
  outptr = (JCOEFPTR) output_row;
  for (count = (long) num_blocks * DCTSIZE2; count > 0; count--) {
    *outptr++ = *inptr++;
  }
#endif
}


GLOBAL(void)
jzero_far (void FAR * target, size_t bytestozero)
/* Zero out a chunk of FAR memory. */
/* This might be sample-array data, block-array data, or alloc_large data. */
{
#ifdef FMEMZERO
  jui_jui_FMEMZERO(target, bytestozero);
#else
  char FAR * ptr = (char FAR *) target;
  size_t count;

  for (count = bytestozero; count > 0; count--) {
    *ptr++ = 0;
  }
#endif
}
/*
 * jerror.c
 *
 * Copyright (C) 1991-1998, 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 simple error-reporting and trace-message routines.
 * These are suitable for Unix-like systems and others where writing to
 * stderr is the right thing to do.  Many applications will want to replace
 * some or all of these routines.
 *
 * If you define USE_WINDOWS_MESSAGEBOX in jconfig.h or in the makefile,
 * you get a Windows-specific hack to display error messages in a dialog box.
 * It ain't much, but it beats dropping error messages into the bit bucket,
 * which is what happens to output to stderr under most Windows C compilers.
 *
 * These routines are used by both the compression and decompression code.
 */

/*
 * Error exit handler: must not return to caller.
 *
 * Applications may override this if they want to get control back after
 * an error.  Typically one would longjmp somewhere instead of exiting.
 * The setjmp buffer can be made a private field within an expanded error
 * handler object.  Note that the info needed to generate an error message
 * is stored in the error object, so you can generate the message now or
 * later, at your convenience.
 * You should make sure that the JPEG object is cleaned up (with jpeg_abort
 * or jpeg_destroy) at some point.
 */

jmp_buf jpeg_state;

METHODDEF(void)
error_exit (j_common_ptr cinfo)
{
    jpeg_destroy(cinfo); /* don't just abort, actually destroy the object */

    longjmp(jpeg_state, 1);
}


/*
 * Actual output of an error or trace message.
 * Applications may override this method to send JPEG messages somewhere
 * other than stderr.
 *
 * On Windows, printing to stderr is generally completely useless,
 * so we provide optional code to produce an error-dialog popup.
 * Most Windows applications will still prefer to override this routine,
 * but if they don't, it'll do something at least marginally useful.
 *
 * NOTE: to use the library in an environment that doesn't support the
 * C stdio library, you may have to delete the call to fprintf() entirely,
 * not just not use this routine.
 */

METHODDEF(void)
output_message (j_common_ptr cinfo)
{
}


/*
 * Decide whether to emit a trace or warning message.
 * msg_level is one of:
 *   -1: recoverable corrupt-data warning, may want to abort.
 *    0: important advisory messages (always display to user).
 *    1: first level of tracing detail.
 *    2,3,...: successively more detailed tracing messages.
 * An application might override this method if it wanted to abort on warnings
 * or change the policy about which messages to display.
 */

METHODDEF(void)
emit_message (j_common_ptr cinfo, int msg_level)
{
}


/*
 * Format a message string for the most recent JPEG error or message.
 * The message is stored into buffer, which should be at least JMSG_LENGTH_MAX
 * characters.  Note that no '\n' character is added to the string.
 * Few applications should need to override this method.
 */

METHODDEF(void)
format_message (j_common_ptr cinfo, char * buffer)
{
}


/*
 * Reset error state variables at start of a new image.
 * This is called during compression startup to reset trace/error
 * processing to default state, without losing any application-specific
 * method pointers.  An application might possibly want to override
 * this method if it has additional error processing state.
 */

METHODDEF(void)
reset_error_mgr (j_common_ptr cinfo)
{
  cinfo->err->num_warnings = 0;
  /* trace_level is not reset since it is an application-supplied parameter */
  cinfo->err->msg_code = 0; /* may be useful as a flag for "no error" */
}


/*
 * Fill in the standard error-handling methods in a jpeg_error_mgr object.
 * Typical call is:
 *  struct jpeg_compress_struct cinfo;
 *  struct jpeg_error_mgr err;
 *
 *  cinfo.err = jpeg_std_error(&err);
 * after which the application may override some of the methods.
 */

GLOBAL(struct jpeg_error_mgr *)
jpeg_std_error (struct jpeg_error_mgr * err)
{
  err->error_exit = error_exit;
  err->emit_message = emit_message;
  err->output_message = output_message;
  err->format_message = format_message;
  err->reset_error_mgr = reset_error_mgr;

  err->trace_level = 0;     /* default = no tracing */
  err->num_warnings = 0;    /* no warnings emitted yet */
  err->msg_code = 0;        /* may be useful as a flag for "no error" */

  return err;
}
/*
 * jmemmgr.c
 *
 * Copyright (C) 1991-1997, 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 JPEG system-independent memory management
 * routines.  This code is usable across a wide variety of machines; most
 * of the system dependencies have been isolated in a separate file.
 * The major functions provided here are:
 *   * pool-based allocation and freeing of memory;
 *   * policy decisions about how to divide available memory among the
 *     virtual arrays;
 *   * control logic for swapping virtual arrays between main memory and
 *     backing storage.
 * The separate system-dependent file provides the actual backing-storage
 * access code, and it contains the policy decision about how much total
 * main memory to use.
 * This file is system-dependent in the sense that some of its functions
 * are unnecessary in some systems.  For example, if there is enough virtual
 * memory so that backing storage will never be used, much of the virtual
 * array control logic could be removed.  (Of course, if you have that much
 * memory then you shouldn't care about a little bit of unused code...)
 */

#define JPEG_INTERNALS
#define AM_MEMORY_MANAGER   /* we define jvirt_Xarray_control structs */
//#include "jinclude.h"
//#include "jpeglib.h"
//#include "jmemsys.h"      /* import the system-dependent declarations */

#ifndef NO_GETENV
#ifndef HAVE_STDLIB_H       /* <stdlib.h> should declare getenv() */
extern char * getenv JPP((const char * name));
#endif
#endif


/*
 * Some important notes:
 *   The allocation routines provided here must never return NULL.
 *   They should exit to error_exit if unsuccessful.
 *
 *   It's not a good idea to try to merge the sarray and barray routines,
 *   even though they are textually almost the same, because samples are
 *   usually stored as bytes while coefficients are shorts or ints.  Thus,
 *   in machines where byte pointers have a different representation from
 *   word pointers, the resulting machine code could not be the same.
 */


/*
 * Many machines require storage alignment: longs must start on 4-byte
 * boundaries, doubles on 8-byte boundaries, etc.  On such machines, malloc()
 * always returns pointers that are multiples of the worst-case alignment
 * requirement, and we had better do so too.
 * There isn't any really portable way to determine the worst-case alignment
 * requirement.  This module assumes that the alignment requirement is
 * multiples of sizeof(ALIGN_TYPE).
 * By default, we define ALIGN_TYPE as double.  This is necessary on some
 * workstations (where doubles really do need 8-byte alignment) and will work
 * fine on nearly everything.  If your machine has lesser alignment needs,
 * you can save a few bytes by making ALIGN_TYPE smaller.
 * The only place I know of where this will NOT work is certain Macintosh
 * 680x0 compilers that define double as a 10-byte IEEE extended float.
 * Doing 10-byte alignment is counterproductive because longwords won't be
 * aligned well.  Put "#define ALIGN_TYPE long" in jconfig.h if you have
 * such a compiler.
 */

#ifndef ALIGN_TYPE      /* so can override from jconfig.h */
#define ALIGN_TYPE  double
#endif


/*
 * We allocate objects from "pools", where each pool is gotten with a single
 * request to jpeg_get_small() or jpeg_get_large().  There is no per-object
 * overhead within a pool, except for alignment padding.  Each pool has a
 * header with a link to the next pool of the same class.
 * Small and large pool headers are identical except that the latter's
 * link pointer must be FAR on 80x86 machines.
 * Notice that the "real" header fields are union'ed with a dummy ALIGN_TYPE
 * field.  This forces the compiler to make SIZEOF(small_pool_hdr) a multiple
 * of the alignment requirement of ALIGN_TYPE.
 */

typedef union small_pool_struct * small_pool_ptr;

typedef union small_pool_struct {
  struct {
    small_pool_ptr next;    /* next in list of pools */
    size_t bytes_used;      /* how many bytes already used within pool */
    size_t bytes_left;      /* bytes still available in this pool */
  } hdr;
  ALIGN_TYPE dummy;     /* included in union to ensure alignment */
} small_pool_hdr;

typedef union large_pool_struct FAR * large_pool_ptr;

typedef union large_pool_struct {
  struct {
    large_pool_ptr next;    /* next in list of pools */
    size_t bytes_used;      /* how many bytes already used within pool */
    size_t bytes_left;      /* bytes still available in this pool */
  } hdr;
  ALIGN_TYPE dummy;     /* included in union to ensure alignment */
} large_pool_hdr;


/*
 * Here is the full definition of a memory manager object.
 */

typedef struct {
  struct jpeg_memory_mgr pub;   /* public fields */

  /* Each pool identifier (lifetime class) names a linked list of pools. */
  small_pool_ptr small_list[JPOOL_NUMPOOLS];
  large_pool_ptr large_list[JPOOL_NUMPOOLS];

  /* Since we only have one lifetime class of virtual arrays, only one
   * linked list is necessary (for each datatype).  Note that the virtual
   * array control blocks being linked together are actually stored somewhere
   * in the small-pool list.
   */
  jvirt_sarray_ptr virt_sarray_list;
  jvirt_barray_ptr virt_barray_list;

  /* This counts total space obtained from jpeg_get_small/large */
  long total_space_allocated;

  /* alloc_sarray and alloc_barray set this value for use by virtual
   * array routines.
   */
  JDIMENSION last_rowsperchunk; /* from most recent alloc_sarray/barray */
} my_memory_mgr;

typedef my_memory_mgr * my_mem_ptr;


/*
 * The control blocks for virtual arrays.
 * Note that these blocks are allocated in the "small" pool area.
 * System-dependent info for the associated backing store (if any) is hidden
 * inside the backing_store_info struct.
 */

struct jvirt_sarray_control {
  JSAMPARRAY mem_buffer;    /* => the in-memory buffer */
  JDIMENSION rows_in_array; /* total virtual array height */
  JDIMENSION samplesperrow; /* width of array (and of memory buffer) */
  JDIMENSION maxaccess;     /* max rows accessed by access_virt_sarray */
  JDIMENSION rows_in_mem;   /* height of memory buffer */
  JDIMENSION rowsperchunk;  /* allocation chunk size in mem_buffer */
  JDIMENSION cur_start_row; /* first logical row # in the buffer */
  JDIMENSION first_undef_row;   /* row # of first uninitialized row */
  boolean pre_zero;     /* pre-zero mode requested? */
  boolean dirty;        /* do current buffer contents need written? */
  boolean b_s_open;     /* is backing-store data valid? */
  jvirt_sarray_ptr next;    /* link to next virtual sarray control block */
  backing_store_info b_s_info;  /* System-dependent control info */
};

struct jvirt_barray_control {
  JBLOCKARRAY mem_buffer;   /* => the in-memory buffer */
  JDIMENSION rows_in_array; /* total virtual array height */
  JDIMENSION blocksperrow;  /* width of array (and of memory buffer) */
  JDIMENSION maxaccess;     /* max rows accessed by access_virt_barray */
  JDIMENSION rows_in_mem;   /* height of memory buffer */
  JDIMENSION rowsperchunk;  /* allocation chunk size in mem_buffer */
  JDIMENSION cur_start_row; /* first logical row # in the buffer */
  JDIMENSION first_undef_row;   /* row # of first uninitialized row */
  boolean pre_zero;     /* pre-zero mode requested? */
  boolean dirty;        /* do current buffer contents need written? */
  boolean b_s_open;     /* is backing-store data valid? */
  jvirt_barray_ptr next;    /* link to next virtual barray control block */
  backing_store_info b_s_info;  /* System-dependent control info */
};


#ifdef MEM_STATS        /* optional extra stuff for statistics */

LOCAL(void)
print_mem_stats (j_common_ptr cinfo, int pool_id)
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  small_pool_ptr shdr_ptr;
  large_pool_ptr lhdr_ptr;

  /* Since this is only a debugging stub, we can cheat a little by using
   * fprintf directly rather than going through the trace message code.
   * This is helpful because message parm array can't handle longs.
   */
  fprintf(stderr, "Freeing pool %d, total space = %ld\n",
      pool_id, mem->total_space_allocated);

  for (lhdr_ptr = mem->large_list[pool_id]; lhdr_ptr != NULL;
       lhdr_ptr = lhdr_ptr->hdr.next) {
    fprintf(stderr, "  Large chunk used %ld\n",
        (long) lhdr_ptr->hdr.bytes_used);
  }

  for (shdr_ptr = mem->small_list[pool_id]; shdr_ptr != NULL;
       shdr_ptr = shdr_ptr->hdr.next) {
    fprintf(stderr, "  Small chunk used %ld free %ld\n",
        (long) shdr_ptr->hdr.bytes_used,
        (long) shdr_ptr->hdr.bytes_left);
  }
}

#endif /* MEM_STATS */

/* coverity[+kill] */
LOCAL(void)
out_of_memory (j_common_ptr cinfo, int which)
/* Report an out-of-memory error and stop execution */
/* If we compiled MEM_STATS support, report alloc requests before dying */
{
#ifdef MEM_STATS
  cinfo->err->trace_level = 2;  /* force self_destruct to report stats */
#endif
  ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, which);
}


/*
 * Allocation of "small" objects.
 *
 * For these, we use pooled storage.  When a new pool must be created,
 * we try to get enough space for the current request plus a "slop" factor,
 * where the slop will be the amount of leftover space in the new pool.
 * The speed vs. space tradeoff is largely determined by the slop values.
 * A different slop value is provided for each pool class (lifetime),
 * and we also distinguish the first pool of a class from later ones.
 * NOTE: the values given work fairly well on both 16- and 32-bit-int
 * machines, but may be too small if longs are 64 bits or more.
 */

static const size_t first_pool_slop[JPOOL_NUMPOOLS] =
{
    1600,           /* first PERMANENT pool */
    16000           /* first IMAGE pool */
};

static const size_t extra_pool_slop[JPOOL_NUMPOOLS] =
{
    0,          /* additional PERMANENT pools */
    5000            /* additional IMAGE pools */
};

#define MIN_SLOP  50        /* greater than 0 to avoid futile looping */


METHODDEF(void *)
alloc_small (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
/* Allocate a "small" object */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  small_pool_ptr hdr_ptr, prev_hdr_ptr;
  char * data_ptr;
  size_t odd_bytes, min_request, slop;

  /* Check for unsatisfiable request (do now to ensure no overflow below) */
  if (sizeofobject > (size_t) (MAX_ALLOC_CHUNK-SIZEOF(small_pool_hdr)))
    out_of_memory(cinfo, 1);    /* request exceeds malloc's ability */

  /* Round up the requested size to a multiple of SIZEOF(ALIGN_TYPE) */
  odd_bytes = sizeofobject % SIZEOF(ALIGN_TYPE);
  if (odd_bytes > 0)
    sizeofobject += SIZEOF(ALIGN_TYPE) - odd_bytes;

  /* See if space is available in any existing pool */
  if (pool_id < 0 || pool_id >= JPOOL_NUMPOOLS)
    ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */
  prev_hdr_ptr = NULL;
  hdr_ptr = mem->small_list[pool_id];
  while (hdr_ptr != NULL) {
    if (hdr_ptr->hdr.bytes_left >= sizeofobject)
      break;            /* found pool with enough space */
    prev_hdr_ptr = hdr_ptr;
    hdr_ptr = hdr_ptr->hdr.next;
  }

  /* Time to make a new pool? */
  if (hdr_ptr == NULL) {
    /* min_request is what we need now, slop is what will be leftover */
    min_request = sizeofobject + SIZEOF(small_pool_hdr);
    if (prev_hdr_ptr == NULL)   /* first pool in class? */
      slop = first_pool_slop[pool_id];
    else
      slop = extra_pool_slop[pool_id];
    /* Don't ask for more than MAX_ALLOC_CHUNK */
    if (slop > (size_t) (MAX_ALLOC_CHUNK-min_request))
      slop = (size_t) (MAX_ALLOC_CHUNK-min_request);
    /* Try to get space, if fail reduce slop and try again */
    for (;;) {
      hdr_ptr = (small_pool_ptr) jpeg_get_small(cinfo, min_request + slop);
      if (hdr_ptr != NULL)
    break;
      slop /= 2;
      if (slop < MIN_SLOP)  /* give up when it gets real small */
    out_of_memory(cinfo, 2); /* jpeg_get_small failed */
    }
    mem->total_space_allocated += min_request + slop;
    /* Success, initialize the new pool header and add to end of list */
    hdr_ptr->hdr.next = NULL;
    hdr_ptr->hdr.bytes_used = 0;
    hdr_ptr->hdr.bytes_left = sizeofobject + slop;
    if (prev_hdr_ptr == NULL)   /* first pool in class? */
      mem->small_list[pool_id] = hdr_ptr;
    else
      prev_hdr_ptr->hdr.next = hdr_ptr;
  }

  /* OK, allocate the object from the current pool */
  data_ptr = (char *) (hdr_ptr + 1); /* point to first data byte in pool */
  data_ptr += hdr_ptr->hdr.bytes_used; /* point to place for object */
  hdr_ptr->hdr.bytes_used += sizeofobject;
  hdr_ptr->hdr.bytes_left -= sizeofobject;

  return (void *) data_ptr;
}


/*
 * Allocation of "large" objects.
 *
 * The external semantics of these are the same as "small" objects,
 * except that FAR pointers are used on 80x86.  However the pool
 * management heuristics are quite different.  We assume that each
 * request is large enough that it may as well be passed directly to
 * jpeg_get_large; the pool management just links everything together
 * so that we can free it all on demand.
 * Note: the major use of "large" objects is in JSAMPARRAY and JBLOCKARRAY
 * structures.  The routines that create these structures (see below)
 * deliberately bunch rows together to ensure a large request size.
 */

METHODDEF(void FAR *)
alloc_large (j_common_ptr cinfo, int pool_id, size_t sizeofobject)
/* Allocate a "large" object */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  large_pool_ptr hdr_ptr;
  size_t odd_bytes;

  /* Check for unsatisfiable request (do now to ensure no overflow below) */
  if (sizeofobject > (size_t) (MAX_ALLOC_CHUNK-SIZEOF(large_pool_hdr)))
    out_of_memory(cinfo, 3);    /* request exceeds malloc's ability */

  /* Round up the requested size to a multiple of SIZEOF(ALIGN_TYPE) */
  odd_bytes = sizeofobject % SIZEOF(ALIGN_TYPE);
  if (odd_bytes > 0)
    sizeofobject += SIZEOF(ALIGN_TYPE) - odd_bytes;

  /* Always make a new pool */
  if (pool_id < 0 || pool_id >= JPOOL_NUMPOOLS)
    ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */

  hdr_ptr = (large_pool_ptr) jpeg_get_large(cinfo, sizeofobject +
                        SIZEOF(large_pool_hdr));
  if (hdr_ptr == NULL)
    out_of_memory(cinfo, 4);    /* jpeg_get_large failed */
  mem->total_space_allocated += sizeofobject + SIZEOF(large_pool_hdr);

  /* Success, initialize the new pool header and add to list */
  hdr_ptr->hdr.next = mem->large_list[pool_id];
  /* We maintain space counts in each pool header for statistical purposes,
   * even though they are not needed for allocation.
   */
  hdr_ptr->hdr.bytes_used = sizeofobject;
  hdr_ptr->hdr.bytes_left = 0;
  mem->large_list[pool_id] = hdr_ptr;

  return (void FAR *) (hdr_ptr + 1); /* point to first data byte in pool */
}


/*
 * Creation of 2-D sample arrays.
 * The pointers are in near heap, the samples themselves in FAR heap.
 *
 * To minimize allocation overhead and to allow I/O of large contiguous
 * blocks, we allocate the sample rows in groups of as many rows as possible
 * without exceeding MAX_ALLOC_CHUNK total bytes per allocation request.
 * NB: the virtual array control routines, later in this file, know about
 * this chunking of rows.  The rowsperchunk value is left in the mem manager
 * object so that it can be saved away if this sarray is the workspace for
 * a virtual array.
 */

METHODDEF(JSAMPARRAY)
alloc_sarray (j_common_ptr cinfo, int pool_id,
          JDIMENSION samplesperrow, JDIMENSION numrows)
/* Allocate a 2-D sample array */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  JSAMPARRAY result;
  JSAMPROW workspace;
  JDIMENSION rowsperchunk, currow, i;
  long ltemp;

  /* Calculate max # of rows allowed in one allocation chunk */
  ltemp = (MAX_ALLOC_CHUNK-SIZEOF(large_pool_hdr)) /
      ((long) samplesperrow * SIZEOF(JSAMPLE));
  if (ltemp <= 0)
    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);
  if (ltemp < (long) numrows)
    rowsperchunk = (JDIMENSION) ltemp;
  else
    rowsperchunk = numrows;
  mem->last_rowsperchunk = rowsperchunk;

  /* Get space for row pointers (small object) */
  result = (JSAMPARRAY) alloc_small(cinfo, pool_id,
                    (size_t) (numrows * SIZEOF(JSAMPROW)));

  /* Get the rows themselves (large objects) */
  currow = 0;
  while (currow < numrows) {
    rowsperchunk = MIN(rowsperchunk, numrows - currow);
    workspace = (JSAMPROW) alloc_large(cinfo, pool_id,
    (size_t) ((size_t) rowsperchunk * (size_t) samplesperrow
          * SIZEOF(JSAMPLE)));
    for (i = rowsperchunk; i > 0; i--) {
      result[currow++] = workspace;
      workspace += samplesperrow;
    }
  }

  return result;
}


/*
 * Creation of 2-D coefficient-block arrays.
 * This is essentially the same as the code for sample arrays, above.
 */

METHODDEF(JBLOCKARRAY)
alloc_barray (j_common_ptr cinfo, int pool_id,
          JDIMENSION blocksperrow, JDIMENSION numrows)
/* Allocate a 2-D coefficient-block array */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  JBLOCKARRAY result;
  JBLOCKROW workspace;
  JDIMENSION rowsperchunk, currow, i;
  long ltemp;

  /* Calculate max # of rows allowed in one allocation chunk */
  ltemp = (MAX_ALLOC_CHUNK-SIZEOF(large_pool_hdr)) /
      ((long) blocksperrow * SIZEOF(JBLOCK));
  if (ltemp <= 0)
    ERREXIT(cinfo, JERR_WIDTH_OVERFLOW);
  if (ltemp < (long) numrows)
    rowsperchunk = (JDIMENSION) ltemp;
  else
    rowsperchunk = numrows;
  mem->last_rowsperchunk = rowsperchunk;

  /* Get space for row pointers (small object) */
  result = (JBLOCKARRAY) alloc_small(cinfo, pool_id,
                     (size_t) (numrows * SIZEOF(JBLOCKROW)));

  /* Get the rows themselves (large objects) */
  currow = 0;
  while (currow < numrows) {
    rowsperchunk = MIN(rowsperchunk, numrows - currow);
    workspace = (JBLOCKROW) alloc_large(cinfo, pool_id,
    (size_t) ((size_t) rowsperchunk * (size_t) blocksperrow
          * SIZEOF(JBLOCK)));
    for (i = rowsperchunk; i > 0; i--) {
      result[currow++] = workspace;
      workspace += blocksperrow;
    }
  }

  return result;
}


/*
 * About virtual array management:
 *
 * The above "normal" array routines are only used to allocate strip buffers
 * (as wide as the image, but just a few rows high).  Full-image-sized buffers
 * are handled as "virtual" arrays.  The array is still accessed a strip at a
 * time, but the memory manager must save the whole array for repeated
 * accesses.  The intended implementation is that there is a strip buffer in
 * memory (as high as is possible given the desired memory limit), plus a
 * backing file that holds the rest of the array.
 *
 * The request_virt_array routines are told the total size of the image and
 * the maximum number of rows that will be accessed at once.  The in-memory
 * buffer must be at least as large as the maxaccess value.
 *
 * The request routines create control blocks but not the in-memory buffers.
 * That is postponed until realize_virt_arrays is called.  At that time the
 * total amount of space needed is known (approximately, anyway), so free
 * memory can be divided up fairly.
 *
 * The access_virt_array routines are responsible for making a specific strip
 * area accessible (after reading or writing the backing file, if necessary).
 * Note that the access routines are told whether the caller intends to modify
 * the accessed strip; during a read-only pass this saves having to rewrite
 * data to disk.  The access routines are also responsible for pre-zeroing
 * any newly accessed rows, if pre-zeroing was requested.
 *
 * In current usage, the access requests are usually for nonoverlapping
 * strips; that is, successive access start_row numbers differ by exactly
 * num_rows = maxaccess.  This means we can get good performance with simple
 * buffer dump/reload logic, by making the in-memory buffer be a multiple
 * of the access height; then there will never be accesses across bufferload
 * boundaries.  The code will still work with overlapping access requests,
 * but it doesn't handle bufferload overlaps very efficiently.
 */


METHODDEF(jvirt_sarray_ptr)
request_virt_sarray (j_common_ptr cinfo, int pool_id, boolean pre_zero,
             JDIMENSION samplesperrow, JDIMENSION numrows,
             JDIMENSION maxaccess)
/* Request a virtual 2-D sample array */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  jvirt_sarray_ptr result;

  /* Only IMAGE-lifetime virtual arrays are currently supported */
  if (pool_id != JPOOL_IMAGE)
    ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */

  /* get control block */
  result = (jvirt_sarray_ptr) alloc_small(cinfo, pool_id,
                      SIZEOF(struct jvirt_sarray_control));

  result->mem_buffer = NULL;    /* marks array not yet realized */
  result->rows_in_array = numrows;
  result->samplesperrow = samplesperrow;
  result->maxaccess = maxaccess;
  result->pre_zero = pre_zero;
  result->b_s_open = FALSE; /* no associated backing-store object */
  result->next = mem->virt_sarray_list; /* add to list of virtual arrays */
  mem->virt_sarray_list = result;

  return result;
}


METHODDEF(jvirt_barray_ptr)
request_virt_barray (j_common_ptr cinfo, int pool_id, boolean pre_zero,
             JDIMENSION blocksperrow, JDIMENSION numrows,
             JDIMENSION maxaccess)
/* Request a virtual 2-D coefficient-block array */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  jvirt_barray_ptr result;

  /* Only IMAGE-lifetime virtual arrays are currently supported */
  if (pool_id != JPOOL_IMAGE)
    ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */

  /* get control block */
  result = (jvirt_barray_ptr) alloc_small(cinfo, pool_id,
                      SIZEOF(struct jvirt_barray_control));

  result->mem_buffer = NULL;    /* marks array not yet realized */
  result->rows_in_array = numrows;
  result->blocksperrow = blocksperrow;
  result->maxaccess = maxaccess;
  result->pre_zero = pre_zero;
  result->b_s_open = FALSE; /* no associated backing-store object */
  result->next = mem->virt_barray_list; /* add to list of virtual arrays */
  mem->virt_barray_list = result;

  return result;
}


METHODDEF(void)
realize_virt_arrays (j_common_ptr cinfo)
/* Allocate the in-memory buffers for any unrealized virtual arrays */
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  long space_per_minheight, maximum_space, avail_mem;
  long minheights, max_minheights;
  jvirt_sarray_ptr sptr;
  jvirt_barray_ptr bptr;

  /* Compute the minimum space needed (maxaccess rows in each buffer)
   * and the maximum space needed (full image height in each buffer).
   * These may be of use to the system-dependent jpeg_mem_available routine.
   */
  space_per_minheight = 0;
  maximum_space = 0;
  for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) {
    if (sptr->mem_buffer == NULL) { /* if not realized yet */
      space_per_minheight += (long) sptr->maxaccess *
                 (long) sptr->samplesperrow * SIZEOF(JSAMPLE);
      maximum_space += (long) sptr->rows_in_array *
               (long) sptr->samplesperrow * SIZEOF(JSAMPLE);
    }
  }
  for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) {
    if (bptr->mem_buffer == NULL) { /* if not realized yet */
      space_per_minheight += (long) bptr->maxaccess *
                 (long) bptr->blocksperrow * SIZEOF(JBLOCK);
      maximum_space += (long) bptr->rows_in_array *
               (long) bptr->blocksperrow * SIZEOF(JBLOCK);
    }
  }

  if (space_per_minheight <= 0)
    return;         /* no unrealized arrays, no work */

  /* Determine amount of memory to actually use; this is system-dependent. */
  avail_mem = jpeg_mem_available(cinfo, space_per_minheight, maximum_space,
                 mem->total_space_allocated);

  /* If the maximum space needed is available, make all the buffers full
   * height; otherwise parcel it out with the same number of minheights
   * in each buffer.
   */
  if (avail_mem >= maximum_space)
    max_minheights = 1000000000L;
  else {
    max_minheights = avail_mem / space_per_minheight;
    /* If there doesn't seem to be enough space, try to get the minimum
     * anyway.  This allows a "stub" implementation of jpeg_mem_available().
     */
    if (max_minheights <= 0)
      max_minheights = 1;
  }

  /* Allocate the in-memory buffers and initialize backing store as needed. */

  for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) {
    if (sptr->mem_buffer == NULL) { /* if not realized yet */
      minheights = ((long) sptr->rows_in_array - 1L) / sptr->maxaccess + 1L;
      if (minheights <= max_minheights) {
    /* This buffer fits in memory */
    sptr->rows_in_mem = sptr->rows_in_array;
      } else {
    /* It doesn't fit in memory, create backing store. */
    sptr->rows_in_mem = (JDIMENSION) (max_minheights * sptr->maxaccess);
    jpeg_open_backing_store(cinfo, & sptr->b_s_info,
                (long) sptr->rows_in_array *
                (long) sptr->samplesperrow *
                (long) SIZEOF(JSAMPLE));
    sptr->b_s_open = TRUE;
      }
      sptr->mem_buffer = alloc_sarray(cinfo, JPOOL_IMAGE,
                      sptr->samplesperrow, sptr->rows_in_mem);
      sptr->rowsperchunk = mem->last_rowsperchunk;
      sptr->cur_start_row = 0;
      sptr->first_undef_row = 0;
      sptr->dirty = FALSE;
    }
  }

  for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) {
    if (bptr->mem_buffer == NULL) { /* if not realized yet */
      minheights = ((long) bptr->rows_in_array - 1L) / bptr->maxaccess + 1L;
      if (minheights <= max_minheights) {
    /* This buffer fits in memory */
    bptr->rows_in_mem = bptr->rows_in_array;
      } else {
    /* It doesn't fit in memory, create backing store. */
    bptr->rows_in_mem = (JDIMENSION) (max_minheights * bptr->maxaccess);
    jpeg_open_backing_store(cinfo, & bptr->b_s_info,
                (long) bptr->rows_in_array *
                (long) bptr->blocksperrow *
                (long) SIZEOF(JBLOCK));
    bptr->b_s_open = TRUE;
      }
      bptr->mem_buffer = alloc_barray(cinfo, JPOOL_IMAGE,
                      bptr->blocksperrow, bptr->rows_in_mem);
      bptr->rowsperchunk = mem->last_rowsperchunk;
      bptr->cur_start_row = 0;
      bptr->first_undef_row = 0;
      bptr->dirty = FALSE;
    }
  }
}


LOCAL(void)
do_sarray_io (j_common_ptr cinfo, jvirt_sarray_ptr ptr, boolean writing)
/* Do backing store read or write of a virtual sample array */
{
  long bytesperrow, file_offset, byte_count, rows, thisrow, i;

  bytesperrow = (long) ptr->samplesperrow * SIZEOF(JSAMPLE);
  file_offset = ptr->cur_start_row * bytesperrow;
  /* Loop to read or write each allocation chunk in mem_buffer */
  for (i = 0; i < (long) ptr->rows_in_mem; i += ptr->rowsperchunk) {
    /* One chunk, but check for short chunk at end of buffer */
    rows = MIN((long) ptr->rowsperchunk, (long) ptr->rows_in_mem - i);
    /* Transfer no more than is currently defined */
    thisrow = (long) ptr->cur_start_row + i;
    rows = MIN(rows, (long) ptr->first_undef_row - thisrow);
    /* Transfer no more than fits in file */
    rows = MIN(rows, (long) ptr->rows_in_array - thisrow);
    if (rows <= 0)      /* this chunk might be past end of file! */
      break;
    byte_count = rows * bytesperrow;
    if (writing)
      (*ptr->b_s_info.write_backing_store) (cinfo, & ptr->b_s_info,
                        (void FAR *) ptr->mem_buffer[i],
                        file_offset, byte_count);
    else
      (*ptr->b_s_info.read_backing_store) (cinfo, & ptr->b_s_info,
                       (void FAR *) ptr->mem_buffer[i],
                       file_offset, byte_count);
    file_offset += byte_count;
  }
}


LOCAL(void)
do_barray_io (j_common_ptr cinfo, jvirt_barray_ptr ptr, boolean writing)
/* Do backing store read or write of a virtual coefficient-block array */
{
  long bytesperrow, file_offset, byte_count, rows, thisrow, i;

  bytesperrow = (long) ptr->blocksperrow * SIZEOF(JBLOCK);
  file_offset = ptr->cur_start_row * bytesperrow;
  /* Loop to read or write each allocation chunk in mem_buffer */
  for (i = 0; i < (long) ptr->rows_in_mem; i += ptr->rowsperchunk) {
    /* One chunk, but check for short chunk at end of buffer */
    rows = MIN((long) ptr->rowsperchunk, (long) ptr->rows_in_mem - i);
    /* Transfer no more than is currently defined */
    thisrow = (long) ptr->cur_start_row + i;
    rows = MIN(rows, (long) ptr->first_undef_row - thisrow);
    /* Transfer no more than fits in file */
    rows = MIN(rows, (long) ptr->rows_in_array - thisrow);
    if (rows <= 0)      /* this chunk might be past end of file! */
      break;
    byte_count = rows * bytesperrow;
    if (writing)
      (*ptr->b_s_info.write_backing_store) (cinfo, & ptr->b_s_info,
                        (void FAR *) ptr->mem_buffer[i],
                        file_offset, byte_count);
    else
      (*ptr->b_s_info.read_backing_store) (cinfo, & ptr->b_s_info,
                       (void FAR *) ptr->mem_buffer[i],
                       file_offset, byte_count);
    file_offset += byte_count;
  }
}


METHODDEF(JSAMPARRAY)
access_virt_sarray (j_common_ptr cinfo, jvirt_sarray_ptr ptr,
            JDIMENSION start_row, JDIMENSION num_rows,
            boolean writable)
/* Access the part of a virtual sample array starting at start_row */
/* and extending for num_rows rows.  writable is true if  */
/* caller intends to modify the accessed area. */
{
  JDIMENSION end_row = start_row + num_rows;
  JDIMENSION undef_row;

  /* debugging check */
  if (end_row > ptr->rows_in_array || num_rows > ptr->maxaccess ||
      ptr->mem_buffer == NULL)
    ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS);

  /* Make the desired part of the virtual array accessible */
  if (start_row < ptr->cur_start_row ||
      end_row > ptr->cur_start_row+ptr->rows_in_mem) {
    if (! ptr->b_s_open)
      ERREXIT(cinfo, JERR_VIRTUAL_BUG);
    /* Flush old buffer contents if necessary */
    if (ptr->dirty) {
      do_sarray_io(cinfo, ptr, TRUE);
      ptr->dirty = FALSE;
    }
    /* Decide what part of virtual array to access.
     * Algorithm: if target address > current window, assume forward scan,
     * load starting at target address.  If target address < current window,
     * assume backward scan, load so that target area is top of window.
     * Note that when switching from forward write to forward read, will have
     * start_row = 0, so the limiting case applies and we load from 0 anyway.
     */
    if (start_row > ptr->cur_start_row) {
      ptr->cur_start_row = start_row;
    } else {
      /* use long arithmetic here to avoid overflow & unsigned problems */
      long ltemp;

      ltemp = (long) end_row - (long) ptr->rows_in_mem;
      if (ltemp < 0)
    ltemp = 0;      /* don't fall off front end of file */
      ptr->cur_start_row = (JDIMENSION) ltemp;
    }
    /* Read in the selected part of the array.
     * During the initial write pass, we will do no actual read
     * because the selected part is all undefined.
     */
    do_sarray_io(cinfo, ptr, FALSE);
  }
  /* Ensure the accessed part of the array is defined; prezero if needed.
   * To improve locality of access, we only prezero the part of the array
   * that the caller is about to access, not the entire in-memory array.
   */
  if (ptr->first_undef_row < end_row) {
    if (ptr->first_undef_row < start_row) {
      if (writable)     /* writer skipped over a section of array */
    ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS);
      undef_row = start_row;    /* but reader is allowed to read ahead */
    } else {
      undef_row = ptr->first_undef_row;
    }
    if (writable)
      ptr->first_undef_row = end_row;
    if (ptr->pre_zero) {
      size_t bytesperrow = (size_t) ptr->samplesperrow * SIZEOF(JSAMPLE);
      undef_row -= ptr->cur_start_row; /* make indexes relative to buffer */
      end_row -= ptr->cur_start_row;
      while (undef_row < end_row) {
    jzero_far((void FAR *) ptr->mem_buffer[undef_row], bytesperrow);
    undef_row++;
      }
    } else {
      if (! writable)       /* reader looking at undefined data */
    ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS);
    }
  }
  /* Flag the buffer dirty if caller will write in it */
  if (writable)
    ptr->dirty = TRUE;
  /* Return address of proper part of the buffer */
  return ptr->mem_buffer + (start_row - ptr->cur_start_row);
}


METHODDEF(JBLOCKARRAY)
access_virt_barray (j_common_ptr cinfo, jvirt_barray_ptr ptr,
            JDIMENSION start_row, JDIMENSION num_rows,
            boolean writable)
/* Access the part of a virtual block array starting at start_row */
/* and extending for num_rows rows.  writable is true if  */
/* caller intends to modify the accessed area. */
{
  JDIMENSION end_row = start_row + num_rows;
  JDIMENSION undef_row;

  /* debugging check */
  if (end_row > ptr->rows_in_array || num_rows > ptr->maxaccess ||
      ptr->mem_buffer == NULL)
    ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS);

  /* Make the desired part of the virtual array accessible */
  if (start_row < ptr->cur_start_row ||
      end_row > ptr->cur_start_row+ptr->rows_in_mem) {
    if (! ptr->b_s_open)
      ERREXIT(cinfo, JERR_VIRTUAL_BUG);
    /* Flush old buffer contents if necessary */
    if (ptr->dirty) {
      do_barray_io(cinfo, ptr, TRUE);
      ptr->dirty = FALSE;
    }
    /* Decide what part of virtual array to access.
     * Algorithm: if target address > current window, assume forward scan,
     * load starting at target address.  If target address < current window,
     * assume backward scan, load so that target area is top of window.
     * Note that when switching from forward write to forward read, will have
     * start_row = 0, so the limiting case applies and we load from 0 anyway.
     */
    if (start_row > ptr->cur_start_row) {
      ptr->cur_start_row = start_row;
    } else {
      /* use long arithmetic here to avoid overflow & unsigned problems */
      long ltemp;

      ltemp = (long) end_row - (long) ptr->rows_in_mem;
      if (ltemp < 0)
    ltemp = 0;      /* don't fall off front end of file */
      ptr->cur_start_row = (JDIMENSION) ltemp;
    }
    /* Read in the selected part of the array.
     * During the initial write pass, we will do no actual read
     * because the selected part is all undefined.
     */
    do_barray_io(cinfo, ptr, FALSE);
  }
  /* Ensure the accessed part of the array is defined; prezero if needed.
   * To improve locality of access, we only prezero the part of the array
   * that the caller is about to access, not the entire in-memory array.
   */
  if (ptr->first_undef_row < end_row) {
    if (ptr->first_undef_row < start_row) {
      if (writable)     /* writer skipped over a section of array */
    ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS);
      undef_row = start_row;    /* but reader is allowed to read ahead */
    } else {
      undef_row = ptr->first_undef_row;
    }
    if (writable)
      ptr->first_undef_row = end_row;
    if (ptr->pre_zero) {
      size_t bytesperrow = (size_t) ptr->blocksperrow * SIZEOF(JBLOCK);
      undef_row -= ptr->cur_start_row; /* make indexes relative to buffer */
      end_row -= ptr->cur_start_row;
      while (undef_row < end_row) {
    jzero_far((void FAR *) ptr->mem_buffer[undef_row], bytesperrow);
    undef_row++;
      }
    } else {
      if (! writable)       /* reader looking at undefined data */
    ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS);
    }
  }
  /* Flag the buffer dirty if caller will write in it */
  if (writable)
    ptr->dirty = TRUE;
  /* Return address of proper part of the buffer */
  return ptr->mem_buffer + (start_row - ptr->cur_start_row);
}


/*
 * Release all objects belonging to a specified pool.
 */

METHODDEF(void)
free_pool (j_common_ptr cinfo, int pool_id)
{
  my_mem_ptr mem = (my_mem_ptr) cinfo->mem;
  small_pool_ptr shdr_ptr;
  large_pool_ptr lhdr_ptr;
  size_t space_freed;

  if (pool_id < 0 || pool_id >= JPOOL_NUMPOOLS)
    ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */

#ifdef MEM_STATS
  if (cinfo->err->trace_level > 1)
    print_mem_stats(cinfo, pool_id); /* print pool's memory usage statistics */
#endif

  /* If freeing IMAGE pool, close any virtual arrays first */
  if (pool_id == JPOOL_IMAGE) {
    jvirt_sarray_ptr sptr;
    jvirt_barray_ptr bptr;

    for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) {
      if (sptr->b_s_open) { /* there may be no backing store */
    sptr->b_s_open = FALSE; /* prevent recursive close if error */
    (*sptr->b_s_info.close_backing_store) (cinfo, & sptr->b_s_info);
      }
    }
    mem->virt_sarray_list = NULL;
    for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) {
      if (bptr->b_s_open) { /* there may be no backing store */
    bptr->b_s_open = FALSE; /* prevent recursive close if error */
    (*bptr->b_s_info.close_backing_store) (cinfo, & bptr->b_s_info);
      }
    }
    mem->virt_barray_list = NULL;
  }

  /* Release large objects */
  lhdr_ptr = mem->large_list[pool_id];
  mem->large_list[pool_id] = NULL;

  while (lhdr_ptr != NULL) {
    large_pool_ptr next_lhdr_ptr = lhdr_ptr->hdr.next;
    space_freed = lhdr_ptr->hdr.bytes_used +
          lhdr_ptr->hdr.bytes_left +
          SIZEOF(large_pool_hdr);
    jpeg_free_large(cinfo, (void FAR *) lhdr_ptr, space_freed);
    mem->total_space_allocated -= space_freed;
    lhdr_ptr = next_lhdr_ptr;
  }

  /* Release small objects */
  shdr_ptr = mem->small_list[pool_id];
  mem->small_list[pool_id] = NULL;

  while (shdr_ptr != NULL) {
    small_pool_ptr next_shdr_ptr = shdr_ptr->hdr.next;
    space_freed = shdr_ptr->hdr.bytes_used +
          shdr_ptr->hdr.bytes_left +
          SIZEOF(small_pool_hdr);
    jpeg_free_small(cinfo, (void *) shdr_ptr, space_freed);
    mem->total_space_allocated -= space_freed;
    shdr_ptr = next_shdr_ptr;
  }
}


/*
 * Close up shop entirely.
 * Note that this cannot be called unless cinfo->mem is non-NULL.
 */

METHODDEF(void)
self_destruct (j_common_ptr cinfo)
{
  int pool;

  /* Close all backing store, release all memory.
   * Releasing pools in reverse order might help avoid fragmentation
   * with some (brain-damaged) malloc libraries.
   */
  for (pool = JPOOL_NUMPOOLS-1; pool >= JPOOL_PERMANENT; pool--) {
    free_pool(cinfo, pool);
  }

  /* Release the memory manager control block too. */
  jpeg_free_small(cinfo, (void *) cinfo->mem, SIZEOF(my_memory_mgr));
  cinfo->mem = NULL;        /* ensures I will be called only once */

  jpeg_mem_term(cinfo);     /* system-dependent cleanup */
}


/*
 * Memory manager initialization.
 * When this is called, only the error manager pointer is valid in cinfo!
 */

GLOBAL(void)
jinit_memory_mgr (j_common_ptr cinfo)
{
  my_mem_ptr mem;
  long max_to_use;
  int pool;
  size_t test_mac;

  cinfo->mem = NULL;        /* for safety if init fails */

  /* Check for configuration errors.
   * SIZEOF(ALIGN_TYPE) should be a power of 2; otherwise, it probably
   * doesn't reflect any real hardware alignment requirement.
   * The test is a little tricky: for X>0, X and X-1 have no one-bits
   * in common if and only if X is a power of 2, ie has only one one-bit.
   * Some compilers may give an "unreachable code" warning here; ignore it.
   */
  if ((SIZEOF(ALIGN_TYPE) & (SIZEOF(ALIGN_TYPE)-1)) != 0)
    ERREXIT(cinfo, JERR_BAD_ALIGN_TYPE);
  /* MAX_ALLOC_CHUNK must be representable as type size_t, and must be
   * a multiple of SIZEOF(ALIGN_TYPE).
   * Again, an "unreachable code" warning may be ignored here.
   * But a "constant too large" warning means you need to fix MAX_ALLOC_CHUNK.
   */
  test_mac = (size_t) MAX_ALLOC_CHUNK;
  if ((long) test_mac != MAX_ALLOC_CHUNK ||
      (MAX_ALLOC_CHUNK % SIZEOF(ALIGN_TYPE)) != 0)
    ERREXIT(cinfo, JERR_BAD_ALLOC_CHUNK);

  max_to_use = jpeg_mem_init(cinfo); /* system-dependent initialization */

  /* Attempt to allocate memory manager's control block */
  mem = (my_mem_ptr) jpeg_get_small(cinfo, SIZEOF(my_memory_mgr));

  if (mem == NULL) {
    jpeg_mem_term(cinfo);   /* system-dependent cleanup */
    ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 0);
  }

  /* OK, fill in the method pointers */
  mem->pub.alloc_small = alloc_small;
  mem->pub.alloc_large = alloc_large;
  mem->pub.alloc_sarray = alloc_sarray;
  mem->pub.alloc_barray = alloc_barray;
  mem->pub.request_virt_sarray = request_virt_sarray;
  mem->pub.request_virt_barray = request_virt_barray;
  mem->pub.realize_virt_arrays = realize_virt_arrays;
  mem->pub.access_virt_sarray = access_virt_sarray;
  mem->pub.access_virt_barray = access_virt_barray;
  mem->pub.free_pool = free_pool;
  mem->pub.self_destruct = self_destruct;

  /* Make MAX_ALLOC_CHUNK accessible to other modules */
  mem->pub.max_alloc_chunk = MAX_ALLOC_CHUNK;

  /* Initialize working state */
  mem->pub.max_memory_to_use = max_to_use;

  for (pool = JPOOL_NUMPOOLS-1; pool >= JPOOL_PERMANENT; pool--) {
    mem->small_list[pool] = NULL;
    mem->large_list[pool] = NULL;
  }
  mem->virt_sarray_list = NULL;
  mem->virt_barray_list = NULL;

  mem->total_space_allocated = SIZEOF(my_memory_mgr);

  /* Declare ourselves open for business */
  cinfo->mem = & mem->pub;

  /* Check for an environment variable JPEGMEM; if found, override the
   * default max_memory setting from jpeg_mem_init.  Note that the
   * surrounding application may again override this value.
   * If your system doesn't support getenv(), define NO_GETENV to disable
   * this feature.
   */
#ifndef NO_GETENV
  { char * memenv;

    if ((memenv = getenv("JPEGMEM")) != NULL) {
      char ch = 'x';

      if (sscanf(memenv, "%ld%c", &max_to_use, &ch) > 0) {
    if (ch == 'm' || ch == 'M')
      max_to_use *= 1000L;
    mem->pub.max_memory_to_use = max_to_use * 1000L;
      }
    }
  }
#endif

}
/*
 * jmemnobs.c
 *
 * Copyright (C) 1992-1996, 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 provides a really simple implementation of the system-
 * dependent portion of the JPEG memory manager.  This implementation
 * assumes that no backing-store files are needed: all required space
 * can be obtained from malloc().
 * This is very portable in the sense that it'll compile on almost anything,
 * but you'd better have lots of main memory (or virtual memory) if you want
 * to process big images.
 * Note that the max_memory_to_use option is ignored by this implementation.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"
//#include "jmemsys.h"      /* import the system-dependent declarations */

#ifndef HAVE_STDLIB_H       /* <stdlib.h> should declare malloc(),free() */
extern void * malloc JPP((size_t size));
extern void free JPP((void *ptr));
#endif


/*
 * Memory allocation and freeing are controlled by the regular library
 * routines malloc() and free().
 */

GLOBAL(void *)
jpeg_get_small (j_common_ptr cinfo, size_t sizeofobject)
{
  return (void *) malloc(sizeofobject);
}

GLOBAL(void)
jpeg_free_small (j_common_ptr cinfo, void * object, size_t sizeofobject)
{
  free(object);
}


/*
 * "Large" objects are treated the same as "small" ones.
 * NB: although we include FAR keywords in the routine declarations,
 * this file won't actually work in 80x86 small/medium model; at least,
 * you probably won't be able to process useful-size images in only 64KB.
 */

GLOBAL(void FAR *)
jpeg_get_large (j_common_ptr cinfo, size_t sizeofobject)
{
  return (void FAR *) malloc(sizeofobject);
}

GLOBAL(void)
jpeg_free_large (j_common_ptr cinfo, void FAR * object, size_t sizeofobject)
{
  free(object);
}


/*
 * This routine computes the total memory space available for allocation.
 * Here we always say, "we got all you want bud!"
 */

GLOBAL(long)
jpeg_mem_available (j_common_ptr cinfo, long min_bytes_needed,
            long max_bytes_needed, long already_allocated)
{
  return max_bytes_needed;
}


/*
 * Backing store (temporary file) management.
 * Since jpeg_mem_available always promised the moon,
 * this should never be called and we can just error out.
 */

GLOBAL(void)
jpeg_open_backing_store (j_common_ptr cinfo, backing_store_ptr info,
             long total_bytes_needed)
{
  ERREXIT(cinfo, JERR_NO_BACKING_STORE);
}


/*
 * These routines take care of any system-dependent initialization and
 * cleanup required.  Here, there isn't any.
 */

GLOBAL(long)
jpeg_mem_init (j_common_ptr cinfo)
{
  return 0;         /* just set max_memory_to_use to 0 */
}

GLOBAL(void)
jpeg_mem_term (j_common_ptr cinfo)
{
  /* no work */
}

/*
 * jquant1.c
 *
 * Copyright (C) 1991-1996, 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 1-pass color quantization (color mapping) routines.
 * These routines provide mapping to a fixed color map using equally spaced
 * color values.  Optional Floyd-Steinberg or ordered dithering is available.
 */

#define JPEG_INTERNALS
//#include "jinclude.h"
//#include "jpeglib.h"

#ifdef QUANT_1PASS_SUPPORTED


/*
 * The main purpose of 1-pass quantization is to provide a fast, if not very
 * high quality, colormapped output capability.  A 2-pass quantizer usually
 * gives better visual quality; however, for quantized grayscale output this
 * quantizer is perfectly adequate.  Dithering is highly recommended with this
 * quantizer, though you can turn it off if you really want to.
 *
 * In 1-pass quantization the colormap must be chosen in advance of seeing the
 * image.  We use a map consisting of all combinations of Ncolors[i] color
 * values for the i'th component.  The Ncolors[] values are chosen so that
 * their product, the total number of colors, is no more than that requested.
 * (In most cases, the product will be somewhat less.)
 *
 * Since the colormap is orthogonal, the representative value for each color
 * component can be determined without considering the other components;
 * then these indexes can be combined into a colormap index by a standard
 * N-dimensional-array-subscript calculation.  Most of the arithmetic involved
 * can be precalculated and stored in the lookup table colorindex[].
 * colorindex[i][j] maps pixel value j in component i to the nearest
 * representative value (grid plane) for that component; this index is
 * multiplied by the array stride for component i, so that the
 * index of the colormap entry closest to a given pixel value is just
 *    sum( colorindex[component-number][pixel-component-value] )
 * Aside from being fast, this scheme allows for variable spacing between
 * representative values with no additional lookup cost.
 *
 * If gamma correction has been applied in color conversion, it might be wise
 * to adjust the color grid spacing so that the representative colors are
 * equidistant in linear space.  At this writing, gamma correction is not
 * implemented by jdcolor, so nothing is done here.
 */


/* Declarations for ordered dithering.
 *
 * We use a standard 16x16 ordered dither array.  The basic concept of ordered
 * dithering is described in many references, for instance Dale Schumacher's
 * chapter II.2 of Graphics Gems II (James Arvo, ed. Academic Press, 1991).
 * In place of Schumacher's comparisons against a "threshold" value, we add a
 * "dither" value to the input pixel and then round the result to the nearest
 * output value.  The dither value is equivalent to (0.5 - threshold) times
 * the distance between output values.  For ordered dithering, we assume that
 * the output colors are equally spaced; if not, results will probably be
 * worse, since the dither may be too much or too little at a given point.
 *
 * The normal calculation would be to form pixel value + dither, range-limit
 * this to 0..MAXJSAMPLE, and then index into the colorindex table as usual.
 * We can skip the separate range-limiting step by extending the colorindex
 * table in both directions.
 */

#define ODITHER_SIZE  16    /* dimension of dither matrix */
/* NB: if ODITHER_SIZE is not a power of 2, ODITHER_MASK uses will break */
#define ODITHER_CELLS (ODITHER_SIZE*ODITHER_SIZE)   /* # cells in matrix */
#define ODITHER_MASK  (ODITHER_SIZE-1) /* mask for wrapping around counters */

typedef int ODITHER_MATRIX[ODITHER_SIZE][ODITHER_SIZE];
typedef int (*ODITHER_MATRIX_PTR)[ODITHER_SIZE];

static const UINT8 base_dither_matrix[ODITHER_SIZE][ODITHER_SIZE] = {
  /* Bayer's order-4 dither array.  Generated by the code given in
   * Stephen Hawley's article "Ordered Dithering" in Graphics Gems I.
   * The values in this array must range from 0 to ODITHER_CELLS-1.
   */
  {   0,192, 48,240, 12,204, 60,252,  3,195, 51,243, 15,207, 63,255 },
  { 128, 64,176,112,140, 76,188,124,131, 67,179,115,143, 79,191,127 },
  {  32,224, 16,208, 44,236, 28,220, 35,227, 19,211, 47,239, 31,223 },
  { 160, 96,144, 80,172,108,156, 92,163, 99,147, 83,175,111,159, 95 },
  {   8,200, 56,248,  4,196, 52,244, 11,203, 59,251,  7,199, 55,247 },
  { 136, 72,184,120,132, 68,180,116,139, 75,187,123,135, 71,183,119 },
  {  40,232, 24,216, 36,228, 20,212, 43,235, 27,219, 39,231, 23,215 },
  { 168,104,152, 88,164,100,148, 84,171,107,155, 91,167,103,151, 87 },
  {   2,194, 50,242, 14,206, 62,254,  1,193, 49,241, 13,205, 61,253 },
  { 130, 66,178,114,142, 78,190,126,129, 65,177,113,141, 77,189,125 },
  {  34,226, 18,210, 46,238, 30,222, 33,225, 17,209, 45,237, 29,221 },
  { 162, 98,146, 82,174,110,158, 94,161, 97,145, 81,173,109,157, 93 },
  {  10,202, 58,250,  6,198, 54,246,  9,201, 57,249,  5,197, 53,245 },
  { 138, 74,186,122,134, 70,182,118,137, 73,185,121,133, 69,181,117 },
  {  42,234, 26,218, 38,230, 22,214, 41,233, 25,217, 37,229, 21,213 },
  { 170,106,154, 90,166,102,150, 86,169,105,153, 89,165,101,149, 85 }
};


/* Declarations for Floyd-Steinberg dithering.
 *
 * Errors are accumulated into the array fserrors[], at a resolution of
 * 1/16th of a pixel count.  The error at a given pixel is propagated
 * to its not-yet-processed neighbors using the standard F-S fractions,
 *      ... (here)  7/16
 *      3/16    5/16    1/16
 * We work left-to-right on even rows, right-to-left on odd rows.
 *
 * We can get away with a single array (holding one row's worth of errors)
 * by using it to store the current row's errors at pixel columns not yet
 * processed, but the next row's errors at columns already processed.  We
 * need only a few extra variables to hold the errors immediately around the
 * current column.  (If we are lucky, those variables are in registers, but
 * even if not, they're probably cheaper to access than array elements are.)
 *
 * The fserrors[] array is indexed [component#][position].
 * We provide (#columns + 2) entries per component; the extra entry at each
 * end saves us from special-casing the first and last pixels.
 *
 * Note: on a wide image, we might not have enough room in a PC's near data
 * segment to hold the error array; so it is allocated with alloc_large.
 */

#if BITS_IN_JSAMPLE == 8
typedef INT16 FSERROR;      /* 16 bits should be enough */
typedef int LOCFSERROR;     /* use 'int' for calculation temps */
#else
typedef INT32 FSERROR;      /* may need more than 16 bits */
typedef INT32 LOCFSERROR;   /* be sure calculation temps are big enough */
#endif

typedef FSERROR FAR *FSERRPTR;  /* pointer to error array (in FAR storage!) */


/* Private subobject */

#define MAX_Q_COMPS 4       /* max components I can handle */

typedef struct {
  struct jpeg_color_quantizer pub; /* public fields */

  /* Initially allocated colormap is saved here */
  JSAMPARRAY sv_colormap;   /* The color map as a 2-D pixel array */
  int sv_actual;        /* number of entries in use */

  JSAMPARRAY colorindex;    /* Precomputed mapping for speed */
  /* colorindex[i][j] = index of color closest to pixel value j in component i,
   * premultiplied as described above.  Since colormap indexes must fit into
   * JSAMPLEs, the entries of this array will too.
   */
  boolean is_padded;        /* is the colorindex padded for odither? */

  int Ncolors[MAX_Q_COMPS]; /* # of values alloced to each component */

  /* Variables for ordered dithering */
  int row_index;        /* cur row's vertical index in dither matrix */
  ODITHER_MATRIX_PTR odither[MAX_Q_COMPS]; /* one dither array per component */

  /* Variables for Floyd-Steinberg dithering */
  FSERRPTR fserrors[MAX_Q_COMPS]; /* accumulated errors */
  boolean on_odd_row;       /* flag to remember which row we are on */
} my_cquantizer;

typedef my_cquantizer * my_cquantize_ptr;


/*
 * Policy-making subroutines for create_colormap and create_colorindex.
 * These routines determine the colormap to be used.  The rest of the module
 * only assumes that the colormap is orthogonal.
 *
 *  * select_ncolors decides how to divvy up the available colors
 *    among the components.
 *  * output_value defines the set of representative values for a component.
 *  * largest_input_value defines the mapping from input values to
 *    representative values for a component.
 * Note that the latter two routines may impose different policies for
 * different components, though this is not currently done.
 */


LOCAL(int)
select_ncolors (j_decompress_ptr cinfo, int Ncolors[])
/* Determine allocation of desired colors to components, */
/* and fill in Ncolors[] array to indicate choice. */
/* Return value is total number of colors (product of Ncolors[] values). */
{
  int nc = cinfo->out_color_components; /* number of color components */
  int max_colors = cinfo->desired_number_of_colors;
  int total_colors, iroot, i, j;
  boolean changed;
  long temp;
  static const int RGB_order[3] = { RGB_GREEN, RGB_RED, RGB_BLUE };

  /* We can allocate at least the nc'th root of max_colors per component. */
  /* Compute floor(nc'th root of max_colors). */
  iroot = 1;
  do {
    iroot++;
    temp = iroot;       /* set temp = iroot ** nc */
    for (i = 1; i < nc; i++)
      temp *= iroot;
  } while (temp <= (long) max_colors); /* repeat till iroot exceeds root */
  iroot--;          /* now iroot = floor(root) */

  /* Must have at least 2 color values per component */
  if (iroot < 2)
    ERREXIT1(cinfo, JERR_QUANT_FEW_COLORS, (int) temp);

  /* Initialize to iroot color values for each component */
  total_colors = 1;
  for (i = 0; i < nc; i++) {
    Ncolors[i] = iroot;
    total_colors *= iroot;
  }
  /* We may be able to increment the count for one or more components without
   * exceeding max_colors, though we know not all can be incremented.
   * Sometimes, the first component can be incremented more than once!
   * (Example: for 16 colors, we start at 2*2*2, go to 3*2*2, then 4*2*2.)
   * In RGB colorspace, try to increment G first, then R, then B.
   */
  do {
    changed = FALSE;
    for (i = 0; i < nc; i++) {
      j = (cinfo->out_color_space == JCS_RGB ? RGB_order[i] : i);
      /* calculate new total_colors if Ncolors[j] is incremented */
      temp = total_colors / Ncolors[j];
      temp *= Ncolors[j]+1; /* done in long arith to avoid oflo */
      if (temp > (long) max_colors)
    break;          /* won't fit, done with this pass */
      Ncolors[j]++;     /* OK, apply the increment */
      total_colors = (int) temp;
      changed = TRUE;
    }
  } while (changed);

  return total_colors;
}


LOCAL(int)
output_value (j_decompress_ptr cinfo, int ci, int j, int maxj)
/* Return j'th output value, where j will range from 0 to maxj */
/* The output values must fall in 0..MAXJSAMPLE in increasing order */
{
  /* We always provide values 0 and MAXJSAMPLE for each component;
   * any additional values are equally spaced between these limits.
   * (Forcing the upper and lower values to the limits ensures that
   * dithering can't produce a color outside the selected gamut.)
   */
  return (int) (((INT32) j * MAXJSAMPLE + maxj/2) / maxj);
}


LOCAL(int)
largest_input_value (j_decompress_ptr cinfo, int ci, int j, int maxj)
/* Return largest input value that should map to j'th output value */
/* Must have largest(j=0) >= 0, and largest(j=maxj) >= MAXJSAMPLE */
{
  /* Breakpoints are halfway between values returned by output_value */
  return (int) (((INT32) (2*j + 1) * MAXJSAMPLE + maxj) / (2*maxj));
}


/*
 * Create the colormap.
 */

LOCAL(void)
create_colormap (j_decompress_ptr cinfo)
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  JSAMPARRAY colormap;      /* Created colormap */
  int total_colors;     /* Number of distinct output colors */
  int i,j,k, nci, blksize, blkdist, ptr, val;

  /* Select number of colors for each component */
  total_colors = select_ncolors(cinfo, cquantize->Ncolors);

  /* Report selected color counts */
  if (cinfo->out_color_components == 3)
    TRACEMS4(cinfo, 1, JTRC_QUANT_3_NCOLORS,
         total_colors, cquantize->Ncolors[0],
         cquantize->Ncolors[1], cquantize->Ncolors[2]);
  else
    TRACEMS1(cinfo, 1, JTRC_QUANT_NCOLORS, total_colors);

  /* Allocate and fill in the colormap. */
  /* The colors are ordered in the map in standard row-major order, */
  /* i.e. rightmost (highest-indexed) color changes most rapidly. */

  colormap = (*cinfo->mem->alloc_sarray)
    ((j_common_ptr) cinfo, JPOOL_IMAGE,
     (JDIMENSION) total_colors, (JDIMENSION) cinfo->out_color_components);

  /* blksize is number of adjacent repeated entries for a component */
  /* blkdist is distance between groups of identical entries for a component */
  blkdist = total_colors;

  for (i = 0; i < cinfo->out_color_components; i++) {
    /* fill in colormap entries for i'th color component */
    nci = cquantize->Ncolors[i]; /* # of distinct values for this color */
    blksize = blkdist / nci;
    for (j = 0; j < nci; j++) {
      /* Compute j'th output value (out of nci) for component */
      val = output_value(cinfo, i, j, nci-1);
      /* Fill in all colormap entries that have this value of this component */
      for (ptr = j * blksize; ptr < total_colors; ptr += blkdist) {
    /* fill in blksize entries beginning at ptr */
    for (k = 0; k < blksize; k++)
      colormap[i][ptr+k] = (JSAMPLE) val;
      }
    }
    blkdist = blksize;      /* blksize of this color is blkdist of next */
  }

  /* Save the colormap in private storage,
   * where it will survive color quantization mode changes.
   */
  cquantize->sv_colormap = colormap;
  cquantize->sv_actual = total_colors;
}


/*
 * Create the color index table.
 */

LOCAL(void)
create_colorindex (j_decompress_ptr cinfo)
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  JSAMPROW indexptr;
  int i,j,k, nci, blksize, val, pad;

  /* For ordered dither, we pad the color index tables by MAXJSAMPLE in
   * each direction (input index values can be -MAXJSAMPLE .. 2*MAXJSAMPLE).
   * This is not necessary in the other dithering modes.  However, we
   * flag whether it was done in case user changes dithering mode.
   */
  if (cinfo->dither_mode == JDITHER_ORDERED) {
    pad = MAXJSAMPLE*2;
    cquantize->is_padded = TRUE;
  } else {
    pad = 0;
    cquantize->is_padded = FALSE;
  }

  cquantize->colorindex = (*cinfo->mem->alloc_sarray)
    ((j_common_ptr) cinfo, JPOOL_IMAGE,
     (JDIMENSION) (MAXJSAMPLE+1 + pad),
     (JDIMENSION) cinfo->out_color_components);

  /* blksize is number of adjacent repeated entries for a component */
  blksize = cquantize->sv_actual;

  for (i = 0; i < cinfo->out_color_components; i++) {
    /* fill in colorindex entries for i'th color component */
    nci = cquantize->Ncolors[i]; /* # of distinct values for this color */
    blksize = blksize / nci;

    /* adjust colorindex pointers to provide padding at negative indexes. */
    if (pad)
      cquantize->colorindex[i] += MAXJSAMPLE;

    /* in loop, val = index of current output value, */
    /* and k = largest j that maps to current val */
    indexptr = cquantize->colorindex[i];
    val = 0;
    k = largest_input_value(cinfo, i, 0, nci-1);
    for (j = 0; j <= MAXJSAMPLE; j++) {
      while (j > k)     /* advance val if past boundary */
    k = largest_input_value(cinfo, i, ++val, nci-1);
      /* premultiply so that no multiplication needed in main processing */
      indexptr[j] = (JSAMPLE) (val * blksize);
    }
    /* Pad at both ends if necessary */
    if (pad)
      for (j = 1; j <= MAXJSAMPLE; j++) {
    indexptr[-j] = indexptr[0];
    indexptr[MAXJSAMPLE+j] = indexptr[MAXJSAMPLE];
      }
  }
}


/*
 * Create an ordered-dither array for a component having ncolors
 * distinct output values.
 */

LOCAL(ODITHER_MATRIX_PTR)
make_odither_array (j_decompress_ptr cinfo, int ncolors)
{
  ODITHER_MATRIX_PTR odither;
  int j,k;
  INT32 num,den;

  odither = (ODITHER_MATRIX_PTR)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(ODITHER_MATRIX));
  /* The inter-value distance for this color is MAXJSAMPLE/(ncolors-1).
   * Hence the dither value for the matrix cell with fill order f
   * (f=0..N-1) should be (N-1-2*f)/(2*N) * MAXJSAMPLE/(ncolors-1).
   * On 16-bit-int machine, be careful to avoid overflow.
   */
  den = 2 * ODITHER_CELLS * ((INT32) (ncolors - 1));
  for (j = 0; j < ODITHER_SIZE; j++) {
    for (k = 0; k < ODITHER_SIZE; k++) {
      num = ((INT32) (ODITHER_CELLS-1 - 2*((int)base_dither_matrix[j][k])))
        * MAXJSAMPLE;
      /* Ensure round towards zero despite C's lack of consistency
       * about rounding negative values in integer division...
       */
      odither[j][k] = (int) (num<0 ? -((-num)/den) : num/den);
    }
  }
  return odither;
}


/*
 * Create the ordered-dither tables.
 * Components having the same number of representative colors may
 * share a dither table.
 */

LOCAL(void)
create_odither_tables (j_decompress_ptr cinfo)
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  ODITHER_MATRIX_PTR odither;
  int i, j, nci;

  for (i = 0; i < cinfo->out_color_components; i++) {
    nci = cquantize->Ncolors[i]; /* # of distinct values for this color */
    odither = NULL;     /* search for matching prior component */
    for (j = 0; j < i; j++) {
      if (nci == cquantize->Ncolors[j]) {
    odither = cquantize->odither[j];
    break;
      }
    }
    if (odither == NULL)    /* need a new table? */
      odither = make_odither_array(cinfo, nci);
    cquantize->odither[i] = odither;
  }
}


/*
 * Map some rows of pixels to the output colormapped representation.
 */

METHODDEF(void)
color_quantize (j_decompress_ptr cinfo, JSAMPARRAY input_buf,
        JSAMPARRAY output_buf, int num_rows)
/* General case, no dithering */
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  JSAMPARRAY colorindex = cquantize->colorindex;
  int pixcode, ci;
  JSAMPROW ptrin, ptrout;
  int row;
  JDIMENSION col;
  JDIMENSION width = cinfo->output_width;
  int nc = cinfo->out_color_components;

  for (row = 0; row < num_rows; row++) {
    ptrin = input_buf[row];
    ptrout = output_buf[row];
    for (col = width; col > 0; col--) {
      pixcode = 0;
      for (ci = 0; ci < nc; ci++) {
    pixcode += GETJSAMPLE(colorindex[ci][GETJSAMPLE(*ptrin++)]);
      }
      *ptrout++ = (JSAMPLE) pixcode;
    }
  }
}


METHODDEF(void)
color_quantize3 (j_decompress_ptr cinfo, JSAMPARRAY input_buf,
         JSAMPARRAY output_buf, int num_rows)
/* Fast path for out_color_components==3, no dithering */
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  int pixcode;
  JSAMPROW ptrin, ptrout;
  JSAMPROW colorindex0 = cquantize->colorindex[0];
  JSAMPROW colorindex1 = cquantize->colorindex[1];
  JSAMPROW colorindex2 = cquantize->colorindex[2];
  int row;
  JDIMENSION col;
  JDIMENSION width = cinfo->output_width;

  for (row = 0; row < num_rows; row++) {
    ptrin = input_buf[row];
    ptrout = output_buf[row];
    for (col = width; col > 0; col--) {
      pixcode  = GETJSAMPLE(colorindex0[GETJSAMPLE(*ptrin++)]);
      pixcode += GETJSAMPLE(colorindex1[GETJSAMPLE(*ptrin++)]);
      pixcode += GETJSAMPLE(colorindex2[GETJSAMPLE(*ptrin++)]);
      *ptrout++ = (JSAMPLE) pixcode;
    }
  }
}


METHODDEF(void)
quantize_ord_dither (j_decompress_ptr cinfo, JSAMPARRAY input_buf,
             JSAMPARRAY output_buf, int num_rows)
/* General case, with ordered dithering */
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  JSAMPROW input_ptr;
  JSAMPROW output_ptr;
  JSAMPROW colorindex_ci;
  int * dither;         /* points to active row of dither matrix */
  int row_index, col_index; /* current indexes into dither matrix */
  int nc = cinfo->out_color_components;
  int ci;
  int row;
  JDIMENSION col;
  JDIMENSION width = cinfo->output_width;

  for (row = 0; row < num_rows; row++) {
    /* Initialize output values to 0 so can process components separately */
    jzero_far((void FAR *) output_buf[row],
          (size_t) (width * SIZEOF(JSAMPLE)));
    row_index = cquantize->row_index;
    for (ci = 0; ci < nc; ci++) {
      input_ptr = input_buf[row] + ci;
      output_ptr = output_buf[row];
      colorindex_ci = cquantize->colorindex[ci];
      dither = cquantize->odither[ci][row_index];
      col_index = 0;

      for (col = width; col > 0; col--) {
    /* Form pixel value + dither, range-limit to 0..MAXJSAMPLE,
     * select output value, accumulate into output code for this pixel.
     * Range-limiting need not be done explicitly, as we have extended
     * the colorindex table to produce the right answers for out-of-range
     * inputs.  The maximum dither is +- MAXJSAMPLE; this sets the
     * required amount of padding.
     */
    *output_ptr += colorindex_ci[GETJSAMPLE(*input_ptr)+dither[col_index]];
    input_ptr += nc;
    output_ptr++;
    col_index = (col_index + 1) & ODITHER_MASK;
      }
    }
    /* Advance row index for next row */
    row_index = (row_index + 1) & ODITHER_MASK;
    cquantize->row_index = row_index;
  }
}


METHODDEF(void)
quantize3_ord_dither (j_decompress_ptr cinfo, JSAMPARRAY input_buf,
              JSAMPARRAY output_buf, int num_rows)
/* Fast path for out_color_components==3, with ordered dithering */
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  int pixcode;
  JSAMPROW input_ptr;
  JSAMPROW output_ptr;
  JSAMPROW colorindex0 = cquantize->colorindex[0];
  JSAMPROW colorindex1 = cquantize->colorindex[1];
  JSAMPROW colorindex2 = cquantize->colorindex[2];
  int * dither0;        /* points to active row of dither matrix */
  int * dither1;
  int * dither2;
  int row_index, col_index; /* current indexes into dither matrix */
  int row;
  JDIMENSION col;
  JDIMENSION width = cinfo->output_width;

  for (row = 0; row < num_rows; row++) {
    row_index = cquantize->row_index;
    input_ptr = input_buf[row];
    output_ptr = output_buf[row];
    dither0 = cquantize->odither[0][row_index];
    dither1 = cquantize->odither[1][row_index];
    dither2 = cquantize->odither[2][row_index];
    col_index = 0;

    for (col = width; col > 0; col--) {
      pixcode  = GETJSAMPLE(colorindex0[GETJSAMPLE(*input_ptr++) +
                    dither0[col_index]]);
      pixcode += GETJSAMPLE(colorindex1[GETJSAMPLE(*input_ptr++) +
                    dither1[col_index]]);
      pixcode += GETJSAMPLE(colorindex2[GETJSAMPLE(*input_ptr++) +
                    dither2[col_index]]);
      *output_ptr++ = (JSAMPLE) pixcode;
      col_index = (col_index + 1) & ODITHER_MASK;
    }
    row_index = (row_index + 1) & ODITHER_MASK;
    cquantize->row_index = row_index;
  }
}


METHODDEF(void)
quantize_fs_dither (j_decompress_ptr cinfo, JSAMPARRAY input_buf,
            JSAMPARRAY output_buf, int num_rows)
/* General case, with Floyd-Steinberg dithering */
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  LOCFSERROR cur;   /* current error or pixel value */
  LOCFSERROR belowerr;      /* error for pixel below cur */
  LOCFSERROR bpreverr;      /* error for below/prev col */
  LOCFSERROR bnexterr;      /* error for below/next col */
  LOCFSERROR delta;
  FSERRPTR errorptr;    /* => fserrors[] at column before current */
  JSAMPROW input_ptr;
  JSAMPROW output_ptr;
  JSAMPROW colorindex_ci;
  JSAMPROW colormap_ci;
  int pixcode;
  int nc = cinfo->out_color_components;
  int dir;          /* 1 for left-to-right, -1 for right-to-left */
  int dirnc;            /* dir * nc */
  int ci;
  int row;
  JDIMENSION col;
  JDIMENSION width = cinfo->output_width;
  JSAMPLE *range_limit = cinfo->sample_range_limit;
  SHIFT_TEMPS

  for (row = 0; row < num_rows; row++) {
    /* Initialize output values to 0 so can process components separately */
    jzero_far((void FAR *) output_buf[row],
          (size_t) (width * SIZEOF(JSAMPLE)));
    for (ci = 0; ci < nc; ci++) {
      input_ptr = input_buf[row] + ci;
      output_ptr = output_buf[row];
      if (cquantize->on_odd_row) {
    /* work right to left in this row */
    input_ptr += (width-1) * nc; /* so point to rightmost pixel */
    output_ptr += width-1;
    dir = -1;
    dirnc = -nc;
    errorptr = cquantize->fserrors[ci] + (width+1); /* => entry after last column */
      } else {
    /* work left to right in this row */
    dir = 1;
    dirnc = nc;
    errorptr = cquantize->fserrors[ci]; /* => entry before first column */
      }
      colorindex_ci = cquantize->colorindex[ci];
      colormap_ci = cquantize->sv_colormap[ci];
      /* Preset error values: no error propagated to first pixel from left */
      cur = 0;
      /* and no error propagated to row below yet */
      belowerr = bpreverr = 0;

      for (col = width; col > 0; col--) {
    /* cur holds the error propagated from the previous pixel on the
     * current line.  Add the error propagated from the previous line
     * to form the complete error correction term for this pixel, and
     * round the error term (which is expressed * 16) to an integer.
     * RIGHT_SHIFT rounds towards minus infinity, so adding 8 is correct
     * for either sign of the error value.
     * Note: errorptr points to *previous* column's array entry.
     */
    cur = RIGHT_SHIFT(cur + errorptr[dir] + 8, 4);
    /* Form pixel value + error, and range-limit to 0..MAXJSAMPLE.
     * The maximum error is +- MAXJSAMPLE; this sets the required size
     * of the range_limit array.
     */
    cur += GETJSAMPLE(*input_ptr);
    cur = GETJSAMPLE(range_limit[cur]);
    /* Select output value, accumulate into output code for this pixel */
    pixcode = GETJSAMPLE(colorindex_ci[cur]);
    *output_ptr += (JSAMPLE) pixcode;
    /* Compute actual representation error at this pixel */
    /* Note: we can do this even though we don't have the final */
    /* pixel code, because the colormap is orthogonal. */
    cur -= GETJSAMPLE(colormap_ci[pixcode]);
    /* Compute error fractions to be propagated to adjacent pixels.
     * Add these into the running sums, and simultaneously shift the
     * next-line error sums left by 1 column.
     */
    bnexterr = cur;
    delta = cur * 2;
    cur += delta;       /* form error * 3 */
    errorptr[0] = (FSERROR) (bpreverr + cur);
    cur += delta;       /* form error * 5 */
    bpreverr = belowerr + cur;
    belowerr = bnexterr;
    cur += delta;       /* form error * 7 */
    /* At this point cur contains the 7/16 error value to be propagated
     * to the next pixel on the current line, and all the errors for the
     * next line have been shifted over. We are therefore ready to move on.
     */
    input_ptr += dirnc; /* advance input ptr to next column */
    output_ptr += dir;  /* advance output ptr to next column */
    errorptr += dir;    /* advance errorptr to current column */
      }
      /* Post-loop cleanup: we must unload the final error value into the
       * final fserrors[] entry.  Note we need not unload belowerr because
       * it is for the dummy column before or after the actual array.
       */
      errorptr[0] = (FSERROR) bpreverr; /* unload prev err into array */
    }
    cquantize->on_odd_row = (cquantize->on_odd_row ? FALSE : TRUE);
  }
}


/*
 * Allocate workspace for Floyd-Steinberg errors.
 */

LOCAL(void)
alloc_fs_workspace (j_decompress_ptr cinfo)
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  size_t arraysize;
  int i;

  arraysize = (size_t) ((cinfo->output_width + 2) * SIZEOF(FSERROR));
  for (i = 0; i < cinfo->out_color_components; i++) {
    cquantize->fserrors[i] = (FSERRPTR)
      (*cinfo->mem->alloc_large)((j_common_ptr) cinfo, JPOOL_IMAGE, arraysize);
  }
}


/*
 * Initialize for one-pass color quantization.
 */

METHODDEF(void)
start_pass_1_quant (j_decompress_ptr cinfo, boolean is_pre_scan)
{
  my_cquantize_ptr cquantize = (my_cquantize_ptr) cinfo->cquantize;
  size_t arraysize;
  int i;

  /* Install my colormap. */
  cinfo->colormap = cquantize->sv_colormap;
  cinfo->actual_number_of_colors = cquantize->sv_actual;

  /* Initialize for desired dithering mode. */
  switch (cinfo->dither_mode) {
  case JDITHER_BLANK:
    if (cinfo->out_color_components == 3)
      cquantize->pub.color_quantize = color_quantize3;
    else
      cquantize->pub.color_quantize = color_quantize;
    break;
  case JDITHER_ORDERED:
    if (cinfo->out_color_components == 3)
      cquantize->pub.color_quantize = quantize3_ord_dither;
    else
      cquantize->pub.color_quantize = quantize_ord_dither;
    cquantize->row_index = 0;   /* initialize state for ordered dither */
    /* If user changed to ordered dither from another mode,
     * we must recreate the color index table with padding.
     * This will cost extra space, but probably isn't very likely.
     */
    if (! cquantize->is_padded)
      create_colorindex(cinfo);
    /* Create ordered-dither tables if we didn't already. */
    if (cquantize->odither[0] == NULL)
      create_odither_tables(cinfo);
    break;
  case JDITHER_FS:
    cquantize->pub.color_quantize = quantize_fs_dither;
    cquantize->on_odd_row = FALSE; /* initialize state for F-S dither */
    /* Allocate Floyd-Steinberg workspace if didn't already. */
    if (cquantize->fserrors[0] == NULL)
      alloc_fs_workspace(cinfo);
    /* Initialize the propagated errors to zero. */
    arraysize = (size_t) ((cinfo->output_width + 2) * SIZEOF(FSERROR));
    for (i = 0; i < cinfo->out_color_components; i++)
      jzero_far((void FAR *) cquantize->fserrors[i], arraysize);
    break;
  default:
    ERREXIT(cinfo, JERR_NOT_COMPILED);
    break;
  }
}


/*
 * Finish up at the end of the pass.
 */

METHODDEF(void)
finish_pass_1_quant (j_decompress_ptr cinfo)
{
  /* no work in 1-pass case */
}


/*
 * Switch to a new external colormap between output passes.
 * Shouldn't get to this module!
 */

METHODDEF(void)
new_color_map_1_quant (j_decompress_ptr cinfo)
{
  ERREXIT(cinfo, JERR_MODE_CHANGE);
}


/*
 * Module initialization routine for 1-pass color quantization.
 */

GLOBAL(void)
jinit_1pass_quantizer (j_decompress_ptr cinfo)
{
  my_cquantize_ptr cquantize;

  cquantize = (my_cquantize_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(my_cquantizer));
  cinfo->cquantize = (struct jpeg_color_quantizer *) cquantize;
  cquantize->pub.start_pass = start_pass_1_quant;
  cquantize->pub.finish_pass = finish_pass_1_quant;
  cquantize->pub.new_color_map = new_color_map_1_quant;
  cquantize->fserrors[0] = NULL; /* Flag FS workspace not allocated */
  cquantize->odither[0] = NULL; /* Also flag odither arrays not allocated */

  /* Make sure my internal arrays won't overflow */
  if (cinfo->out_color_components > MAX_Q_COMPS)
    ERREXIT1(cinfo, JERR_QUANT_COMPONENTS, MAX_Q_COMPS);
  /* Make sure colormap indexes can be represented by JSAMPLEs */
  if (cinfo->desired_number_of_colors > (MAXJSAMPLE+1))
    ERREXIT1(cinfo, JERR_QUANT_MANY_COLORS, MAXJSAMPLE+1);

  /* Create the colormap and color index table. */
  create_colormap(cinfo);
  create_colorindex(cinfo);

  /* Allocate Floyd-Steinberg workspace now if requested.
   * We do this now since it is FAR storage and may affect the memory
   * manager's space calculations.  If the user changes to FS dither
   * mode in a later pass, we will allocate the space then, and will
   * possibly overrun the max_memory_to_use setting.
   */
  if (cinfo->dither_mode == JDITHER_FS)
    alloc_fs_workspace(cinfo);
}

#endif /* QUANT_1PASS_SUPPORTED */

/*
 * jdphuff.c
 *
 * Copyright (C) 1995-1997, 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 Huffman entropy decoding routines for progressive JPEG.
 *
 * 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"
//#include "jdhuff.h"       /* Declarations shared with jdhuff.c */


#ifdef D_PROGRESSIVE_SUPPORTED

/*
 * Expanded entropy decoder object for progressive 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 */
} p_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 */
  p_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 */

  /* 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 */
} phuff_entropy_decoder;

typedef phuff_entropy_decoder * phuff_entropy_ptr;

/* Forward declarations */
METHODDEF(boolean) decode_mcu_DC_first JPP((j_decompress_ptr cinfo,
                        JBLOCKROW *MCU_data));
METHODDEF(boolean) decode_mcu_AC_first JPP((j_decompress_ptr cinfo,
                        JBLOCKROW *MCU_data));
METHODDEF(boolean) decode_mcu_DC_refine JPP((j_decompress_ptr cinfo,
                         JBLOCKROW *MCU_data));
METHODDEF(boolean) decode_mcu_AC_refine JPP((j_decompress_ptr cinfo,
                         JBLOCKROW *MCU_data));


/*
 * Initialize for a Huffman-compressed scan.
 */

METHODDEF(void)
start_pass_phuff_decoder (j_decompress_ptr cinfo)
{
  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
  boolean is_DC_band, bad;
  int ci, coefi, tbl;
  int *coef_bit_ptr;
  jpeg_component_info * compptr;

  is_DC_band = (cinfo->Ss == 0);

  /* Validate scan parameters */
  bad = FALSE;
  if (is_DC_band) {
    if (cinfo->Se != 0)
      bad = TRUE;
  } else {
    /* need not check Ss/Se < 0 since they came from unsigned bytes */
    if (cinfo->Ss > cinfo->Se || cinfo->Se >= DCTSIZE2)
      bad = TRUE;
    /* AC scans may have only one component */
    if (cinfo->comps_in_scan != 1)
      bad = TRUE;
  }
  if (cinfo->Ah != 0) {
    /* Successive approximation refinement scan: must have Al = Ah-1. */
    if (cinfo->Al != cinfo->Ah-1)
      bad = TRUE;
  }
  if (cinfo->Al > 13)       /* need not check for < 0 */
    bad = TRUE;
  /* 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.
   */
  if (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 cindex = cinfo->cur_comp_info[ci]->component_index;
    coef_bit_ptr = & cinfo->coef_bits[cindex][0];
    if (!is_DC_band && 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 (is_DC_band)
      entropy->pub.decode_mcu = decode_mcu_DC_first;
    else
      entropy->pub.decode_mcu = decode_mcu_AC_first;
  } else {
    if (is_DC_band)
      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 (is_DC_band) {
      if (cinfo->Ah == 0) { /* DC refinement needs no table */
    tbl = compptr->dc_tbl_no;
    jpeg_make_d_derived_tbl(cinfo, TRUE, tbl,
                & entropy->derived_tbls[tbl]);
      }
    } else {
      tbl = compptr->ac_tbl_no;
      jpeg_make_d_derived_tbl(cinfo, FALSE, tbl,
                  & entropy->derived_tbls[tbl]);
      /* remember the single active table */
      entropy->ac_derived_tbl = entropy->derived_tbls[tbl];
    }
    /* Initialize DC predictions to 0 */
    entropy->saved.last_dc_val[ci] = 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 private state variables */
  entropy->saved.EOBRUN = 0;

  /* Initialize restart counter */
  entropy->restarts_to_go = cinfo->restart_interval;
}


/*
 * Figure F.12: extend sign bit.
 * On some machines, a shift and add will be faster than a table lookup.
 */

#ifdef AVOID_TABLES

#define HUFF_EXTEND(x,s)  ((x) < (1<<((s)-1)) ? (x) + (((-1)<<(s)) + 1) : (x))

#else

#define HUFF_EXTEND(x,s)  ((x) < p_extend_test[s] ? (x) + p_extend_offset[s] : (x))

static const int p_extend_test[16] =   /* entry n is 2**(n-1) */
  { 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080,
    0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 };

static const int p_extend_offset[16] = /* entry n is (-1 << n) + 1 */
  { 0, -(1<<1) + 1, -(1<<2) + 1, -(1<<3) + 1, -(1<<4) + 1,
    -(1<<5) + 1, -(1<<6) + 1, -(1<<7) + 1, -(1<<8) + 1,
    -(1<<9) + 1, -(1<<10) + 1, -(1<<11) + 1, -(1<<12) + 1,
    -(1<<13) + 1, -(1<<14) + 1, -(1<<15) + 1 };

#endif /* AVOID_TABLES */


/*
 * Check for a restart marker & resynchronize decoder.
 * Returns FALSE if must suspend.
 */

LOCAL(boolean)
p_process_restart (j_decompress_ptr cinfo)
{
  phuff_entropy_ptr entropy = (phuff_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.
 *
 * 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)
{
  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
  int Al = cinfo->Al;
  int s, r;
  int blkn, ci;
  JBLOCKROW block;
  BITREAD_STATE_VARS;
  p_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 (! p_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)
{
  phuff_entropy_ptr entropy = (phuff_entropy_ptr) cinfo->entropy;
  int Se = cinfo->Se;
  int Al = cinfo->Al;
  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 (! p_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)
{
  phuff_entropy_ptr entropy = (phuff_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 (! p_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)
{
  phuff_entropy_ptr entropy = (phuff_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 */
  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 (! p_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;
}


/*
 * Module initialization routine for progressive Huffman entropy decoding.
 */

GLOBAL(void)
jinit_phuff_decoder (j_decompress_ptr cinfo)
{
  phuff_entropy_ptr entropy;
  int *coef_bit_ptr;
  int ci, i;

  entropy = (phuff_entropy_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
                SIZEOF(phuff_entropy_decoder));
  cinfo->entropy = (struct jpeg_entropy_decoder *) entropy;
  entropy->pub.start_pass = start_pass_phuff_decoder;

  /* Mark derived tables unallocated */
  for (i = 0; i < NUM_HUFF_TBLS; i++) {
    entropy->derived_tbls[i] = NULL;
  }

  /* Create progression status table */
  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;
}

#endif /* D_PROGRESSIVE_SUPPORTED */