jdcoefct.c revision cc66ecf44d1407039b05ffd7b3342389f95c17b7
1/*
2 * jdcoefct.c
3 *
4 * Copyright (C) 1994-1997, Thomas G. Lane.
5 * This file is part of the Independent JPEG Group's software.
6 * For conditions of distribution and use, see the accompanying README file.
7 *
8 * This file contains the coefficient buffer controller for decompression.
9 * This controller is the top level of the JPEG decompressor proper.
10 * The coefficient buffer lies between entropy decoding and inverse-DCT steps.
11 *
12 * In buffered-image mode, this controller is the interface between
13 * input-oriented processing and output-oriented processing.
14 * Also, the input side (only) is used when reading a file for transcoding.
15 */
16
17#define JPEG_INTERNALS
18#include "jinclude.h"
19#include "jpeglib.h"
20
21/* Block smoothing is only applicable for progressive JPEG, so: */
22#ifndef D_PROGRESSIVE_SUPPORTED
23#undef BLOCK_SMOOTHING_SUPPORTED
24#endif
25
26/* Private buffer controller object */
27
28typedef struct {
29  struct jpeg_d_coef_controller pub; /* public fields */
30
31  /* These variables keep track of the current location of the input side. */
32  /* cinfo->input_iMCU_row is also used for this. */
33  JDIMENSION MCU_ctr;		/* counts MCUs processed in current row */
34  int MCU_vert_offset;		/* counts MCU rows within iMCU row */
35  int MCU_rows_per_iMCU_row;	/* number of such rows needed */
36
37  /* The output side's location is represented by cinfo->output_iMCU_row. */
38
39  /* In single-pass modes, it's sufficient to buffer just one MCU.
40   * We allocate a workspace of D_MAX_BLOCKS_IN_MCU coefficient blocks,
41   * and let the entropy decoder write into that workspace each time.
42   * (On 80x86, the workspace is FAR even though it's not really very big;
43   * this is to keep the module interfaces unchanged when a large coefficient
44   * buffer is necessary.)
45   * In multi-pass modes, this array points to the current MCU's blocks
46   * within the virtual arrays; it is used only by the input side.
47   */
48  JBLOCKROW MCU_buffer[D_MAX_BLOCKS_IN_MCU];
49
50#ifdef D_MULTISCAN_FILES_SUPPORTED
51  /* In multi-pass modes, we need a virtual block array for each component. */
52  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
53#endif
54
55#ifdef BLOCK_SMOOTHING_SUPPORTED
56  /* When doing block smoothing, we latch coefficient Al values here */
57  int * coef_bits_latch;
58#define SAVED_COEFS  6		/* we save coef_bits[0..5] */
59#endif
60} my_coef_controller;
61
62typedef my_coef_controller * my_coef_ptr;
63
64/* Forward declarations */
65METHODDEF(int) decompress_onepass
66	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
67#ifdef D_MULTISCAN_FILES_SUPPORTED
68METHODDEF(int) decompress_data
69	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
70#endif
71#ifdef BLOCK_SMOOTHING_SUPPORTED
72LOCAL(boolean) smoothing_ok JPP((j_decompress_ptr cinfo));
73METHODDEF(int) decompress_smooth_data
74	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
75#endif
76
77
78LOCAL(void)
79start_iMCU_row (j_decompress_ptr cinfo)
80/* Reset within-iMCU-row counters for a new row (input side) */
81{
82  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
83
84  /* In an interleaved scan, an MCU row is the same as an iMCU row.
85   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
86   * But at the bottom of the image, process only what's left.
87   */
88  if (cinfo->comps_in_scan > 1) {
89    coef->MCU_rows_per_iMCU_row = 1;
90  } else {
91    if (cinfo->input_iMCU_row < (cinfo->total_iMCU_rows-1))
92      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
93    else
94      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
95  }
96
97  coef->MCU_ctr = 0;
98  coef->MCU_vert_offset = 0;
99}
100
101
102/*
103 * Initialize for an input processing pass.
104 */
105
106METHODDEF(void)
107start_input_pass (j_decompress_ptr cinfo)
108{
109  cinfo->input_iMCU_row = 0;
110  start_iMCU_row(cinfo);
111}
112
113
114/*
115 * Initialize for an output processing pass.
116 */
117
118METHODDEF(void)
119start_output_pass (j_decompress_ptr cinfo)
120{
121#ifdef BLOCK_SMOOTHING_SUPPORTED
122  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
123
124  /* If multipass, check to see whether to use block smoothing on this pass */
125  if (coef->pub.coef_arrays != NULL) {
126    if (cinfo->do_block_smoothing && smoothing_ok(cinfo))
127      coef->pub.decompress_data = decompress_smooth_data;
128    else
129      coef->pub.decompress_data = decompress_data;
130  }
131#endif
132  cinfo->output_iMCU_row = 0;
133}
134
135
136/*
137 * Decompress and return some data in the single-pass case.
138 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
139 * Input and output must run in lockstep since we have only a one-MCU buffer.
140 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
141 *
142 * NB: output_buf contains a plane for each component in image,
143 * which we index according to the component's SOF position.
144 */
145
146METHODDEF(int)
147decompress_onepass (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
148{
149  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
150  JDIMENSION MCU_col_num;	/* index of current MCU within row */
151  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
152  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
153  int blkn, ci, xindex, yindex, yoffset, useful_width;
154  JSAMPARRAY output_ptr;
155  JDIMENSION start_col, output_col;
156  jpeg_component_info *compptr;
157  inverse_DCT_method_ptr inverse_DCT;
158
159#ifdef ANDROID_TILE_BASED_DECODE
160  if (cinfo->tile_decode) {
161    last_MCU_col =
162        (cinfo->coef->MCU_column_right_boundary -
163         cinfo->coef->MCU_column_left_boundary) - 1;
164  }
165#endif
166
167  /* Loop to process as much as one whole iMCU row */
168  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
169       yoffset++) {
170    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
171	 MCU_col_num++) {
172      /* Try to fetch an MCU.  Entropy decoder expects buffer to be zeroed. */
173      jzero_far((void FAR *) coef->MCU_buffer[0],
174		(size_t) (cinfo->blocks_in_MCU * SIZEOF(JBLOCK)));
175      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
176	/* Suspension forced; update state counters and exit */
177	coef->MCU_vert_offset = yoffset;
178	coef->MCU_ctr = MCU_col_num;
179	return JPEG_SUSPENDED;
180      }
181      /* Determine where data should go in output_buf and do the IDCT thing.
182       * We skip dummy blocks at the right and bottom edges (but blkn gets
183       * incremented past them!).  Note the inner loop relies on having
184       * allocated the MCU_buffer[] blocks sequentially.
185       */
186      blkn = 0;			/* index of current DCT block within MCU */
187      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
188	compptr = cinfo->cur_comp_info[ci];
189	/* Don't bother to IDCT an uninteresting component. */
190	if (! compptr->component_needed) {
191	  blkn += compptr->MCU_blocks;
192	  continue;
193	}
194	inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index];
195	useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
196						    : compptr->last_col_width;
197	output_ptr = output_buf[compptr->component_index] +
198	  yoffset * compptr->DCT_scaled_size;
199	start_col = MCU_col_num * compptr->MCU_sample_width;
200	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
201	  if (cinfo->input_iMCU_row < last_iMCU_row ||
202	      yoffset+yindex < compptr->last_row_height) {
203	    output_col = start_col;
204	    for (xindex = 0; xindex < useful_width; xindex++) {
205	      (*inverse_DCT) (cinfo, compptr,
206			      (JCOEFPTR) coef->MCU_buffer[blkn+xindex],
207			      output_ptr, output_col);
208	      output_col += compptr->DCT_scaled_size;
209	    }
210	  }
211	  blkn += compptr->MCU_width;
212	  output_ptr += compptr->DCT_scaled_size;
213	}
214      }
215    }
216    /* Completed an MCU row, but perhaps not an iMCU row */
217    coef->MCU_ctr = 0;
218  }
219  /* Completed the iMCU row, advance counters for next one */
220  cinfo->output_iMCU_row++;
221  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
222    start_iMCU_row(cinfo);
223    return JPEG_ROW_COMPLETED;
224  }
225  /* Completed the scan */
226  (*cinfo->inputctl->finish_input_pass) (cinfo);
227  return JPEG_SCAN_COMPLETED;
228}
229
230
231/*
232 * Dummy consume-input routine for single-pass operation.
233 */
234
235METHODDEF(int)
236dummy_consume_data (j_decompress_ptr cinfo)
237{
238  return JPEG_SUSPENDED;	/* Always indicate nothing was done */
239}
240
241#ifdef D_MULTISCAN_FILES_SUPPORTED
242/*
243 * Consume input data and store it in the full-image coefficient buffer.
244 * We read as much as one fully interleaved MCU row ("iMCU" row) per call,
245 * ie, v_samp_factor block rows for each component in the scan.
246 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
247 */
248
249METHODDEF(int)
250consume_data (j_decompress_ptr cinfo)
251{
252  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
253  JDIMENSION MCU_col_num;	/* index of current MCU within row */
254  int blkn, ci, xindex, yindex, yoffset;
255  JDIMENSION start_col;
256  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
257  JBLOCKROW buffer_ptr;
258  jpeg_component_info *compptr;
259
260  /* Align the virtual buffers for the components used in this scan. */
261  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
262    compptr = cinfo->cur_comp_info[ci];
263    buffer[ci] = (*cinfo->mem->access_virt_barray)
264      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
265       cinfo->tile_decode ? 0 : cinfo->input_iMCU_row * compptr->v_samp_factor,
266       (JDIMENSION) compptr->v_samp_factor, TRUE);
267    /* Note: entropy decoder expects buffer to be zeroed,
268     * but this is handled automatically by the memory manager
269     * because we requested a pre-zeroed array.
270     */
271  }
272  unsigned int MCUs_per_row = cinfo->MCUs_per_row;
273#ifdef ANDROID_TILE_BASED_DECODE
274  if (cinfo->tile_decode) {
275    MCUs_per_row = jmin(MCUs_per_row,
276        (cinfo->coef->column_right_boundary - cinfo->coef->column_left_boundary)
277        * cinfo->entropy->index->MCU_sample_size * cinfo->max_h_samp_factor);
278  }
279#endif
280
281  /* Loop to process one whole iMCU row */
282  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
283       yoffset++) {
284#ifdef ANDROID_TILE_BASED_DECODE
285    if (cinfo->tile_decode) {
286      huffman_scan_header scan_header =
287            cinfo->entropy->index->scan[cinfo->input_scan_number];
288      int col_offset = cinfo->coef->column_left_boundary;
289      (*cinfo->entropy->configure_huffman_decoder) (cinfo,
290              scan_header.offset[cinfo->input_iMCU_row]
291              [col_offset + yoffset * scan_header.MCUs_per_row]);
292    }
293#endif
294    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < MCUs_per_row;
295	 MCU_col_num++) {
296      /* Construct list of pointers to DCT blocks belonging to this MCU */
297      blkn = 0;			/* index of current DCT block within MCU */
298      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
299        compptr = cinfo->cur_comp_info[ci];
300        start_col = MCU_col_num * compptr->MCU_width;
301        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
302          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
303          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
304            coef->MCU_buffer[blkn++] = buffer_ptr++;
305#ifdef ANDROID_TILE_BASED_DECODE
306            if (cinfo->tile_decode && cinfo->input_scan_number == 0) {
307              // need to do pre-zero ourself.
308              jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
309                        (size_t) (SIZEOF(JBLOCK)));
310            }
311#endif
312          }
313        }
314      }
315      /* Try to fetch the MCU. */
316      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
317	/* Suspension forced; update state counters and exit */
318	coef->MCU_vert_offset = yoffset;
319	coef->MCU_ctr = MCU_col_num;
320	return JPEG_SUSPENDED;
321      }
322    }
323    /* Completed an MCU row, but perhaps not an iMCU row */
324    coef->MCU_ctr = 0;
325  }
326  /* Completed the iMCU row, advance counters for next one */
327  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
328    start_iMCU_row(cinfo);
329    return JPEG_ROW_COMPLETED;
330  }
331  /* Completed the scan */
332  (*cinfo->inputctl->finish_input_pass) (cinfo);
333  return JPEG_SCAN_COMPLETED;
334}
335
336/*
337 * Consume input data and store it in the coefficient buffer.
338 * Read one fully interleaved MCU row ("iMCU" row) per call.
339 */
340
341METHODDEF(int)
342consume_data_multi_scan (j_decompress_ptr cinfo)
343{
344  huffman_index *index = cinfo->entropy->index;
345  int i, retcode, ci;
346  int mcu = cinfo->input_iMCU_row;
347  jinit_phuff_decoder(cinfo);
348  for (i = 0; i < index->scan_count; i++) {
349    (*cinfo->inputctl->finish_input_pass) (cinfo);
350    jset_input_stream_position(cinfo, index->scan[i].bitstream_offset);
351    cinfo->output_iMCU_row = mcu;
352    cinfo->unread_marker = 0;
353    // Consume SOS and DHT headers
354    retcode = (*cinfo->inputctl->consume_markers) (cinfo, index, i);
355    cinfo->input_iMCU_row = mcu;
356    cinfo->input_scan_number = i;
357    cinfo->entropy->index = index;
358    // Consume scan block data
359    consume_data(cinfo);
360  }
361  cinfo->input_iMCU_row = mcu + 1;
362  cinfo->input_scan_number = 0;
363  cinfo->output_scan_number = 0;
364  return JPEG_ROW_COMPLETED;
365}
366
367/*
368 * Same as consume_data, expect for saving the Huffman decode information
369 * - bitstream offset and DC coefficient to index.
370 */
371
372METHODDEF(int)
373consume_data_build_huffman_index_baseline (j_decompress_ptr cinfo,
374        huffman_index *index, int current_scan)
375{
376  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
377  JDIMENSION MCU_col_num;	/* index of current MCU within row */
378  int ci, xindex, yindex, yoffset;
379  JDIMENSION start_col;
380  JBLOCKROW buffer_ptr;
381
382  huffman_scan_header *scan_header = index->scan + current_scan;
383  scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
384
385  size_t allocate_size = coef->MCU_rows_per_iMCU_row
386      * jdiv_round_up(cinfo->MCUs_per_row, index->MCU_sample_size)
387      * sizeof(huffman_offset_data);
388  scan_header->offset[cinfo->input_iMCU_row] =
389        (huffman_offset_data*)malloc(allocate_size);
390  index->mem_used += allocate_size;
391
392  huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
393
394  /* Loop to process one whole iMCU row */
395  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
396       yoffset++) {
397    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
398	 MCU_col_num++) {
399      // Record huffman bit offset
400      if (MCU_col_num % index->MCU_sample_size == 0) {
401        (*cinfo->entropy->get_huffman_decoder_configuration)
402                (cinfo, offset_data);
403        ++offset_data;
404      }
405
406      /* Try to fetch the MCU. */
407      if (! (*cinfo->entropy->decode_mcu_discard_coef) (cinfo)) {
408        /* Suspension forced; update state counters and exit */
409        coef->MCU_vert_offset = yoffset;
410        coef->MCU_ctr = MCU_col_num;
411        return JPEG_SUSPENDED;
412      }
413    }
414    /* Completed an MCU row, but perhaps not an iMCU row */
415    coef->MCU_ctr = 0;
416  }
417  /* Completed the iMCU row, advance counters for next one */
418  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
419    start_iMCU_row(cinfo);
420    return JPEG_ROW_COMPLETED;
421  }
422  /* Completed the scan */
423  (*cinfo->inputctl->finish_input_pass) (cinfo);
424  return JPEG_SCAN_COMPLETED;
425}
426
427/*
428 * Same as consume_data, expect for saving the Huffman decode information
429 * - bitstream offset and DC coefficient to index.
430 */
431
432METHODDEF(int)
433consume_data_build_huffman_index_progressive (j_decompress_ptr cinfo,
434        huffman_index *index, int current_scan)
435{
436  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
437  JDIMENSION MCU_col_num;	/* index of current MCU within row */
438  int blkn, ci, xindex, yindex, yoffset;
439  JDIMENSION start_col;
440  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
441  JBLOCKROW buffer_ptr;
442  jpeg_component_info *compptr;
443
444  int factor = 4; // maximum factor is 4.
445  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
446    factor = jmin(factor, cinfo->cur_comp_info[ci]->h_samp_factor);
447
448  int sample_size = index->MCU_sample_size * factor;
449  huffman_scan_header *scan_header = index->scan + current_scan;
450  scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
451  scan_header->MCUs_per_row = jdiv_round_up(cinfo->MCUs_per_row, sample_size);
452  scan_header->comps_in_scan = cinfo->comps_in_scan;
453
454  size_t allocate_size = coef->MCU_rows_per_iMCU_row
455      * scan_header->MCUs_per_row * sizeof(huffman_offset_data);
456  scan_header->offset[cinfo->input_iMCU_row] =
457        (huffman_offset_data*)malloc(allocate_size);
458  index->mem_used += allocate_size;
459
460  huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
461
462  /* Align the virtual buffers for the components used in this scan. */
463  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
464    compptr = cinfo->cur_comp_info[ci];
465    buffer[ci] = (*cinfo->mem->access_virt_barray)
466      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
467       0, // Only need one row buffer
468       (JDIMENSION) compptr->v_samp_factor, TRUE);
469  }
470  /* Loop to process one whole iMCU row */
471  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
472       yoffset++) {
473    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
474	 MCU_col_num++) {
475      /* For each MCU, we loop through different color components.
476       * Then, for each color component we will get a list of pointers to DCT
477       * blocks in the virtual buffer.
478       */
479      blkn = 0; /* index of current DCT block within MCU */
480      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
481        compptr = cinfo->cur_comp_info[ci];
482        start_col = MCU_col_num * compptr->MCU_width;
483        /* Get the list of pointers to DCT blocks in
484         * the virtual buffer in a color component of the MCU.
485         */
486        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
487          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
488          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
489            coef->MCU_buffer[blkn++] = buffer_ptr++;
490            if (cinfo->input_scan_number == 0) {
491              // need to do pre-zero by ourself.
492              jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
493                        (size_t) (SIZEOF(JBLOCK)));
494            }
495          }
496        }
497      }
498      // Record huffman bit offset
499      if (MCU_col_num % sample_size == 0) {
500        (*cinfo->entropy->get_huffman_decoder_configuration)
501                (cinfo, offset_data);
502        ++offset_data;
503      }
504      /* Try to fetch the MCU. */
505      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
506	/* Suspension forced; update state counters and exit */
507	coef->MCU_vert_offset = yoffset;
508	coef->MCU_ctr = MCU_col_num;
509	return JPEG_SUSPENDED;
510      }
511    }
512    /* Completed an MCU row, but perhaps not an iMCU row */
513    coef->MCU_ctr = 0;
514  }
515  (*cinfo->entropy->get_huffman_decoder_configuration)
516        (cinfo, &scan_header->prev_MCU_offset);
517  /* Completed the iMCU row, advance counters for next one */
518  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
519    start_iMCU_row(cinfo);
520    return JPEG_ROW_COMPLETED;
521  }
522  /* Completed the scan */
523  (*cinfo->inputctl->finish_input_pass) (cinfo);
524  return JPEG_SCAN_COMPLETED;
525}
526
527/*
528 * Decompress and return some data in the multi-pass case.
529 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
530 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
531 *
532 * NB: output_buf contains a plane for each component in image.
533 */
534
535METHODDEF(int)
536decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
537{
538  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
539  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
540  JDIMENSION block_num;
541  int ci, block_row, block_rows;
542  JBLOCKARRAY buffer;
543  JBLOCKROW buffer_ptr;
544  JSAMPARRAY output_ptr;
545  JDIMENSION output_col;
546  jpeg_component_info *compptr;
547  inverse_DCT_method_ptr inverse_DCT;
548
549  /* Force some input to be done if we are getting ahead of the input. */
550  while (cinfo->input_scan_number < cinfo->output_scan_number ||
551	 (cinfo->input_scan_number == cinfo->output_scan_number &&
552	  cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
553    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
554      return JPEG_SUSPENDED;
555  }
556
557  /* OK, output from the virtual arrays. */
558  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
559       ci++, compptr++) {
560    /* Don't bother to IDCT an uninteresting component. */
561    if (! compptr->component_needed)
562      continue;
563    /* Align the virtual buffer for this component. */
564    buffer = (*cinfo->mem->access_virt_barray)
565      ((j_common_ptr) cinfo, coef->whole_image[ci],
566       cinfo->tile_decode ? 0 : cinfo->output_iMCU_row * compptr->v_samp_factor,
567       (JDIMENSION) compptr->v_samp_factor, FALSE);
568    /* Count non-dummy DCT block rows in this iMCU row. */
569    if (cinfo->output_iMCU_row < last_iMCU_row)
570      block_rows = compptr->v_samp_factor;
571    else {
572      /* NB: can't use last_row_height here; it is input-side-dependent! */
573      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
574      if (block_rows == 0) block_rows = compptr->v_samp_factor;
575    }
576    inverse_DCT = cinfo->idct->inverse_DCT[ci];
577    output_ptr = output_buf[ci];
578    /* Loop over all DCT blocks to be processed. */
579    for (block_row = 0; block_row < block_rows; block_row++) {
580      buffer_ptr = buffer[block_row];
581      output_col = 0;
582      for (block_num = 0; block_num < compptr->width_in_blocks; block_num++) {
583	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
584			output_ptr, output_col);
585	buffer_ptr++;
586	output_col += compptr->DCT_scaled_size;
587      }
588      output_ptr += compptr->DCT_scaled_size;
589    }
590  }
591
592  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
593    return JPEG_ROW_COMPLETED;
594  return JPEG_SCAN_COMPLETED;
595}
596
597#endif /* D_MULTISCAN_FILES_SUPPORTED */
598
599
600#ifdef BLOCK_SMOOTHING_SUPPORTED
601
602/*
603 * This code applies interblock smoothing as described by section K.8
604 * of the JPEG standard: the first 5 AC coefficients are estimated from
605 * the DC values of a DCT block and its 8 neighboring blocks.
606 * We apply smoothing only for progressive JPEG decoding, and only if
607 * the coefficients it can estimate are not yet known to full precision.
608 */
609
610/* Natural-order array positions of the first 5 zigzag-order coefficients */
611#define Q01_POS  1
612#define Q10_POS  8
613#define Q20_POS  16
614#define Q11_POS  9
615#define Q02_POS  2
616
617/*
618 * Determine whether block smoothing is applicable and safe.
619 * We also latch the current states of the coef_bits[] entries for the
620 * AC coefficients; otherwise, if the input side of the decompressor
621 * advances into a new scan, we might think the coefficients are known
622 * more accurately than they really are.
623 */
624
625LOCAL(boolean)
626smoothing_ok (j_decompress_ptr cinfo)
627{
628  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
629  boolean smoothing_useful = FALSE;
630  int ci, coefi;
631  jpeg_component_info *compptr;
632  JQUANT_TBL * qtable;
633  int * coef_bits;
634  int * coef_bits_latch;
635
636  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
637    return FALSE;
638
639  /* Allocate latch area if not already done */
640  if (coef->coef_bits_latch == NULL)
641    coef->coef_bits_latch = (int *)
642      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
643				  cinfo->num_components *
644				  (SAVED_COEFS * SIZEOF(int)));
645  coef_bits_latch = coef->coef_bits_latch;
646
647  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
648       ci++, compptr++) {
649    /* All components' quantization values must already be latched. */
650    if ((qtable = compptr->quant_table) == NULL)
651      return FALSE;
652    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
653    if (qtable->quantval[0] == 0 ||
654	qtable->quantval[Q01_POS] == 0 ||
655	qtable->quantval[Q10_POS] == 0 ||
656	qtable->quantval[Q20_POS] == 0 ||
657	qtable->quantval[Q11_POS] == 0 ||
658	qtable->quantval[Q02_POS] == 0)
659      return FALSE;
660    /* DC values must be at least partly known for all components. */
661    coef_bits = cinfo->coef_bits[ci];
662    if (coef_bits[0] < 0)
663      return FALSE;
664    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
665    for (coefi = 1; coefi <= 5; coefi++) {
666      coef_bits_latch[coefi] = coef_bits[coefi];
667      if (coef_bits[coefi] != 0)
668	smoothing_useful = TRUE;
669    }
670    coef_bits_latch += SAVED_COEFS;
671  }
672
673  return smoothing_useful;
674}
675
676
677/*
678 * Variant of decompress_data for use when doing block smoothing.
679 */
680
681METHODDEF(int)
682decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
683{
684  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
685  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
686  JDIMENSION block_num, last_block_column;
687  int ci, block_row, block_rows, access_rows;
688  JBLOCKARRAY buffer;
689  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
690  JSAMPARRAY output_ptr;
691  JDIMENSION output_col;
692  jpeg_component_info *compptr;
693  inverse_DCT_method_ptr inverse_DCT;
694  boolean first_row, last_row;
695  JBLOCK workspace;
696  int *coef_bits;
697  JQUANT_TBL *quanttbl;
698  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
699  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
700  int Al, pred;
701
702  /* Force some input to be done if we are getting ahead of the input. */
703  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
704	 ! cinfo->inputctl->eoi_reached) {
705    if (cinfo->input_scan_number == cinfo->output_scan_number) {
706      /* If input is working on current scan, we ordinarily want it to
707       * have completed the current row.  But if input scan is DC,
708       * we want it to keep one row ahead so that next block row's DC
709       * values are up to date.
710       */
711      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
712      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
713	break;
714    }
715    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
716      return JPEG_SUSPENDED;
717  }
718
719  /* OK, output from the virtual arrays. */
720  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
721       ci++, compptr++) {
722    /* Don't bother to IDCT an uninteresting component. */
723    if (! compptr->component_needed)
724      continue;
725    /* Count non-dummy DCT block rows in this iMCU row. */
726    if (cinfo->output_iMCU_row < last_iMCU_row) {
727      block_rows = compptr->v_samp_factor;
728      access_rows = block_rows * 2; /* this and next iMCU row */
729      last_row = FALSE;
730    } else {
731      /* NB: can't use last_row_height here; it is input-side-dependent! */
732      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
733      if (block_rows == 0) block_rows = compptr->v_samp_factor;
734      access_rows = block_rows; /* this iMCU row only */
735      last_row = TRUE;
736    }
737    /* Align the virtual buffer for this component. */
738    if (cinfo->output_iMCU_row > 0) {
739      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
740      buffer = (*cinfo->mem->access_virt_barray)
741	((j_common_ptr) cinfo, coef->whole_image[ci],
742	 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
743	 (JDIMENSION) access_rows, FALSE);
744      buffer += compptr->v_samp_factor;	/* point to current iMCU row */
745      first_row = FALSE;
746    } else {
747      buffer = (*cinfo->mem->access_virt_barray)
748	((j_common_ptr) cinfo, coef->whole_image[ci],
749	 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
750      first_row = TRUE;
751    }
752    /* Fetch component-dependent info */
753    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
754    quanttbl = compptr->quant_table;
755    Q00 = quanttbl->quantval[0];
756    Q01 = quanttbl->quantval[Q01_POS];
757    Q10 = quanttbl->quantval[Q10_POS];
758    Q20 = quanttbl->quantval[Q20_POS];
759    Q11 = quanttbl->quantval[Q11_POS];
760    Q02 = quanttbl->quantval[Q02_POS];
761    inverse_DCT = cinfo->idct->inverse_DCT[ci];
762    output_ptr = output_buf[ci];
763    /* Loop over all DCT blocks to be processed. */
764    for (block_row = 0; block_row < block_rows; block_row++) {
765      buffer_ptr = buffer[block_row];
766      if (first_row && block_row == 0)
767	prev_block_row = buffer_ptr;
768      else
769	prev_block_row = buffer[block_row-1];
770      if (last_row && block_row == block_rows-1)
771	next_block_row = buffer_ptr;
772      else
773	next_block_row = buffer[block_row+1];
774      /* We fetch the surrounding DC values using a sliding-register approach.
775       * Initialize all nine here so as to do the right thing on narrow pics.
776       */
777      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
778      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
779      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
780      output_col = 0;
781      last_block_column = compptr->width_in_blocks - 1;
782      for (block_num = 0; block_num <= last_block_column; block_num++) {
783	/* Fetch current DCT block into workspace so we can modify it. */
784	jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
785	/* Update DC values */
786	if (block_num < last_block_column) {
787	  DC3 = (int) prev_block_row[1][0];
788	  DC6 = (int) buffer_ptr[1][0];
789	  DC9 = (int) next_block_row[1][0];
790	}
791	/* Compute coefficient estimates per K.8.
792	 * An estimate is applied only if coefficient is still zero,
793	 * and is not known to be fully accurate.
794	 */
795	/* AC01 */
796	if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
797	  num = 36 * Q00 * (DC4 - DC6);
798	  if (num >= 0) {
799	    pred = (int) (((Q01<<7) + num) / (Q01<<8));
800	    if (Al > 0 && pred >= (1<<Al))
801	      pred = (1<<Al)-1;
802	  } else {
803	    pred = (int) (((Q01<<7) - num) / (Q01<<8));
804	    if (Al > 0 && pred >= (1<<Al))
805	      pred = (1<<Al)-1;
806	    pred = -pred;
807	  }
808	  workspace[1] = (JCOEF) pred;
809	}
810	/* AC10 */
811	if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
812	  num = 36 * Q00 * (DC2 - DC8);
813	  if (num >= 0) {
814	    pred = (int) (((Q10<<7) + num) / (Q10<<8));
815	    if (Al > 0 && pred >= (1<<Al))
816	      pred = (1<<Al)-1;
817	  } else {
818	    pred = (int) (((Q10<<7) - num) / (Q10<<8));
819	    if (Al > 0 && pred >= (1<<Al))
820	      pred = (1<<Al)-1;
821	    pred = -pred;
822	  }
823	  workspace[8] = (JCOEF) pred;
824	}
825	/* AC20 */
826	if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
827	  num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
828	  if (num >= 0) {
829	    pred = (int) (((Q20<<7) + num) / (Q20<<8));
830	    if (Al > 0 && pred >= (1<<Al))
831	      pred = (1<<Al)-1;
832	  } else {
833	    pred = (int) (((Q20<<7) - num) / (Q20<<8));
834	    if (Al > 0 && pred >= (1<<Al))
835	      pred = (1<<Al)-1;
836	    pred = -pred;
837	  }
838	  workspace[16] = (JCOEF) pred;
839	}
840	/* AC11 */
841	if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
842	  num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
843	  if (num >= 0) {
844	    pred = (int) (((Q11<<7) + num) / (Q11<<8));
845	    if (Al > 0 && pred >= (1<<Al))
846	      pred = (1<<Al)-1;
847	  } else {
848	    pred = (int) (((Q11<<7) - num) / (Q11<<8));
849	    if (Al > 0 && pred >= (1<<Al))
850	      pred = (1<<Al)-1;
851	    pred = -pred;
852	  }
853	  workspace[9] = (JCOEF) pred;
854	}
855	/* AC02 */
856	if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
857	  num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
858	  if (num >= 0) {
859	    pred = (int) (((Q02<<7) + num) / (Q02<<8));
860	    if (Al > 0 && pred >= (1<<Al))
861	      pred = (1<<Al)-1;
862	  } else {
863	    pred = (int) (((Q02<<7) - num) / (Q02<<8));
864	    if (Al > 0 && pred >= (1<<Al))
865	      pred = (1<<Al)-1;
866	    pred = -pred;
867	  }
868	  workspace[2] = (JCOEF) pred;
869	}
870	/* OK, do the IDCT */
871	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
872			output_ptr, output_col);
873	/* Advance for next column */
874	DC1 = DC2; DC2 = DC3;
875	DC4 = DC5; DC5 = DC6;
876	DC7 = DC8; DC8 = DC9;
877	buffer_ptr++, prev_block_row++, next_block_row++;
878	output_col += compptr->DCT_scaled_size;
879      }
880      output_ptr += compptr->DCT_scaled_size;
881    }
882  }
883
884  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
885    return JPEG_ROW_COMPLETED;
886  return JPEG_SCAN_COMPLETED;
887}
888
889#endif /* BLOCK_SMOOTHING_SUPPORTED */
890
891
892/*
893 * Initialize coefficient buffer controller.
894 */
895
896GLOBAL(void)
897jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
898{
899  my_coef_ptr coef;
900
901  coef = (my_coef_ptr)
902    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
903				SIZEOF(my_coef_controller));
904  cinfo->coef = (struct jpeg_d_coef_controller *) coef;
905  coef->pub.start_input_pass = start_input_pass;
906  coef->pub.start_output_pass = start_output_pass;
907  coef->pub.column_left_boundary = 0;
908  coef->pub.column_right_boundary = 0;
909#ifdef BLOCK_SMOOTHING_SUPPORTED
910  coef->coef_bits_latch = NULL;
911#endif
912
913#ifdef ANDROID_TILE_BASED_DECODE
914  if (cinfo->tile_decode) {
915    if (cinfo->progressive_mode) {
916      /* Allocate one iMCU row virtual array, coef->whole_image[ci],
917       * for each color component, padded to a multiple of h_samp_factor
918       * DCT blocks in the horizontal direction.
919       */
920      int ci, access_rows;
921      jpeg_component_info *compptr;
922
923      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
924	   ci++, compptr++) {
925        access_rows = compptr->v_samp_factor;
926        coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
927	  ((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
928	   (JDIMENSION) jround_up((long) compptr->width_in_blocks,
929				(long) compptr->h_samp_factor),
930	   (JDIMENSION) compptr->v_samp_factor, // one iMCU row
931	   (JDIMENSION) access_rows);
932      }
933      coef->pub.consume_data_build_huffman_index =
934            consume_data_build_huffman_index_progressive;
935      coef->pub.consume_data = consume_data_multi_scan;
936      coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
937      coef->pub.decompress_data = decompress_onepass;
938    } else {
939      /* We only need a single-MCU buffer. */
940      JBLOCKROW buffer;
941      int i;
942
943      buffer = (JBLOCKROW)
944      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
945				  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
946      for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
947        coef->MCU_buffer[i] = buffer + i;
948      }
949      coef->pub.consume_data_build_huffman_index =
950            consume_data_build_huffman_index_baseline;
951      coef->pub.consume_data = dummy_consume_data;
952      coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
953      coef->pub.decompress_data = decompress_onepass;
954    }
955    return;
956  }
957#endif
958
959  /* Create the coefficient buffer. */
960  if (need_full_buffer) {
961#ifdef D_MULTISCAN_FILES_SUPPORTED
962    /* Allocate a full-image virtual array for each component, */
963    /* padded to a multiple of samp_factor DCT blocks in each direction. */
964    /* Note we ask for a pre-zeroed array. */
965    int ci, access_rows;
966    jpeg_component_info *compptr;
967
968    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
969	 ci++, compptr++) {
970      access_rows = compptr->v_samp_factor;
971#ifdef BLOCK_SMOOTHING_SUPPORTED
972      /* If block smoothing could be used, need a bigger window */
973      if (cinfo->progressive_mode)
974	access_rows *= 3;
975#endif
976      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
977	((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
978	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
979				(long) compptr->h_samp_factor),
980	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
981				(long) compptr->v_samp_factor),
982	 (JDIMENSION) access_rows);
983    }
984    coef->pub.consume_data = consume_data;
985    coef->pub.decompress_data = decompress_data;
986    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
987#else
988    ERREXIT(cinfo, JERR_NOT_COMPILED);
989#endif
990  } else {
991    /* We only need a single-MCU buffer. */
992    JBLOCKROW buffer;
993    int i;
994
995    buffer = (JBLOCKROW)
996      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
997		  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
998    for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
999      coef->MCU_buffer[i] = buffer + i;
1000    }
1001    coef->pub.consume_data = dummy_consume_data;
1002    coef->pub.decompress_data = decompress_onepass;
1003    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
1004  }
1005}
1006