jdcoefct.c revision f5b94eebe742df1a9bb3941fc0a0ec0137e936ef
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 =
276        (cinfo->coef->column_right_boundary - cinfo->coef->column_left_boundary)
277        * cinfo->entropy->index->MCU_sample_size * cinfo->max_h_samp_factor;
278    MCUs_per_row = jmin(MCUs_per_row, cinfo->MCUs_per_row);
279  }
280#endif
281
282  /* Loop to process one whole iMCU row */
283  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
284       yoffset++) {
285#ifdef ANDROID_TILE_BASED_DECODE
286    if (cinfo->tile_decode) {
287      huffman_scan_header scan_header =
288            cinfo->entropy->index->scan[cinfo->input_scan_number];
289      int col_offset = cinfo->coef->column_left_boundary;
290      (*cinfo->entropy->configure_huffman_decoder) (cinfo,
291              scan_header.offset[cinfo->input_iMCU_row]
292              [col_offset + yoffset * scan_header.MCUs_per_row]);
293    }
294#endif
295    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < MCUs_per_row;
296	 MCU_col_num++) {
297      /* Construct list of pointers to DCT blocks belonging to this MCU */
298      blkn = 0;			/* index of current DCT block within MCU */
299      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
300        compptr = cinfo->cur_comp_info[ci];
301        start_col = MCU_col_num * compptr->MCU_width;
302        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
303          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
304          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
305            coef->MCU_buffer[blkn++] = buffer_ptr++;
306#ifdef ANDROID_TILE_BASED_DECODE
307            if (cinfo->tile_decode && cinfo->input_scan_number == 0) {
308              // need to do pre-zero ourself.
309              jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
310                        (size_t) (SIZEOF(JBLOCK)));
311            }
312#endif
313          }
314        }
315      }
316      /* Try to fetch the MCU. */
317      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
318	/* Suspension forced; update state counters and exit */
319	coef->MCU_vert_offset = yoffset;
320	coef->MCU_ctr = MCU_col_num;
321	return JPEG_SUSPENDED;
322      }
323    }
324    /* Completed an MCU row, but perhaps not an iMCU row */
325    coef->MCU_ctr = 0;
326  }
327  /* Completed the iMCU row, advance counters for next one */
328  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
329    start_iMCU_row(cinfo);
330    return JPEG_ROW_COMPLETED;
331  }
332  /* Completed the scan */
333  (*cinfo->inputctl->finish_input_pass) (cinfo);
334  return JPEG_SCAN_COMPLETED;
335}
336
337/*
338 * Consume input data and store it in the coefficient buffer.
339 * Read one fully interleaved MCU row ("iMCU" row) per call.
340 */
341
342METHODDEF(int)
343consume_data_multi_scan (j_decompress_ptr cinfo)
344{
345  huffman_index *index = cinfo->entropy->index;
346  int i, retcode, ci;
347  int mcu = cinfo->input_iMCU_row;
348  jinit_phuff_decoder(cinfo);
349  for (i = 0; i < index->scan_count; i++) {
350    (*cinfo->inputctl->finish_input_pass) (cinfo);
351    jset_input_stream_position(cinfo, index->scan[i].bitstream_offset);
352    cinfo->output_iMCU_row = mcu;
353    cinfo->unread_marker = 0;
354    // Consume SOS and DHT headers
355    retcode = (*cinfo->inputctl->consume_markers) (cinfo, index, i);
356    cinfo->input_iMCU_row = mcu;
357    cinfo->input_scan_number = i;
358    cinfo->entropy->index = index;
359    // Consume scan block data
360    consume_data(cinfo);
361  }
362  cinfo->input_iMCU_row = mcu + 1;
363  cinfo->input_scan_number = 0;
364  cinfo->output_scan_number = 0;
365  return JPEG_ROW_COMPLETED;
366}
367
368/*
369 * Same as consume_data, expect for saving the Huffman decode information
370 * - bitstream offset and DC coefficient to index.
371 */
372
373METHODDEF(int)
374consume_data_build_huffman_index_baseline (j_decompress_ptr cinfo,
375        huffman_index *index, int current_scan)
376{
377  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
378  JDIMENSION MCU_col_num;	/* index of current MCU within row */
379  int ci, xindex, yindex, yoffset;
380  JDIMENSION start_col;
381  JBLOCKROW buffer_ptr;
382
383  huffman_scan_header *scan_header = index->scan + current_scan;
384  scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
385
386  size_t allocate_size = coef->MCU_rows_per_iMCU_row
387      * jdiv_round_up(cinfo->MCUs_per_row, index->MCU_sample_size)
388      * sizeof(huffman_offset_data);
389  scan_header->offset[cinfo->input_iMCU_row] =
390        (huffman_offset_data*)malloc(allocate_size);
391  index->mem_used += allocate_size;
392
393  huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
394
395  /* Loop to process one whole iMCU row */
396  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
397       yoffset++) {
398    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
399	 MCU_col_num++) {
400      // Record huffman bit offset
401      if (MCU_col_num % index->MCU_sample_size == 0) {
402        (*cinfo->entropy->get_huffman_decoder_configuration)
403                (cinfo, offset_data);
404        ++offset_data;
405      }
406
407      /* Try to fetch the MCU. */
408      if (! (*cinfo->entropy->decode_mcu_discard_coef) (cinfo)) {
409        /* Suspension forced; update state counters and exit */
410        coef->MCU_vert_offset = yoffset;
411        coef->MCU_ctr = MCU_col_num;
412        return JPEG_SUSPENDED;
413      }
414    }
415    /* Completed an MCU row, but perhaps not an iMCU row */
416    coef->MCU_ctr = 0;
417  }
418  /* Completed the iMCU row, advance counters for next one */
419  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
420    start_iMCU_row(cinfo);
421    return JPEG_ROW_COMPLETED;
422  }
423  /* Completed the scan */
424  (*cinfo->inputctl->finish_input_pass) (cinfo);
425  return JPEG_SCAN_COMPLETED;
426}
427
428/*
429 * Same as consume_data, expect for saving the Huffman decode information
430 * - bitstream offset and DC coefficient to index.
431 */
432
433METHODDEF(int)
434consume_data_build_huffman_index_progressive (j_decompress_ptr cinfo,
435        huffman_index *index, int current_scan)
436{
437  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
438  JDIMENSION MCU_col_num;	/* index of current MCU within row */
439  int blkn, ci, xindex, yindex, yoffset;
440  JDIMENSION start_col;
441  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
442  JBLOCKROW buffer_ptr;
443  jpeg_component_info *compptr;
444
445  int factor = 4; // maximum factor is 4.
446  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
447    factor = jmin(factor, cinfo->cur_comp_info[ci]->h_samp_factor);
448
449  int sample_size = index->MCU_sample_size * factor;
450  huffman_scan_header *scan_header = index->scan + current_scan;
451  scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
452  scan_header->MCUs_per_row = jdiv_round_up(cinfo->MCUs_per_row, sample_size);
453  scan_header->comps_in_scan = cinfo->comps_in_scan;
454
455  size_t allocate_size = coef->MCU_rows_per_iMCU_row
456      * scan_header->MCUs_per_row * sizeof(huffman_offset_data);
457  scan_header->offset[cinfo->input_iMCU_row] =
458        (huffman_offset_data*)malloc(allocate_size);
459  index->mem_used += allocate_size;
460
461  huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
462
463  /* Align the virtual buffers for the components used in this scan. */
464  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
465    compptr = cinfo->cur_comp_info[ci];
466    buffer[ci] = (*cinfo->mem->access_virt_barray)
467      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
468       0, // Only need one row buffer
469       (JDIMENSION) compptr->v_samp_factor, TRUE);
470  }
471  /* Loop to process one whole iMCU row */
472  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
473       yoffset++) {
474    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
475	 MCU_col_num++) {
476      /* For each MCU, we loop through different color components.
477       * Then, for each color component we will get a list of pointers to DCT
478       * blocks in the virtual buffer.
479       */
480      blkn = 0; /* index of current DCT block within MCU */
481      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
482        compptr = cinfo->cur_comp_info[ci];
483        start_col = MCU_col_num * compptr->MCU_width;
484        /* Get the list of pointers to DCT blocks in
485         * the virtual buffer in a color component of the MCU.
486         */
487        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
488          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
489          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
490            coef->MCU_buffer[blkn++] = buffer_ptr++;
491            if (cinfo->input_scan_number == 0) {
492              // need to do pre-zero by ourself.
493              jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
494                        (size_t) (SIZEOF(JBLOCK)));
495            }
496          }
497        }
498      }
499      // Record huffman bit offset
500      if (MCU_col_num % sample_size == 0) {
501        (*cinfo->entropy->get_huffman_decoder_configuration)
502                (cinfo, offset_data);
503        ++offset_data;
504      }
505      /* Try to fetch the MCU. */
506      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
507	/* Suspension forced; update state counters and exit */
508	coef->MCU_vert_offset = yoffset;
509	coef->MCU_ctr = MCU_col_num;
510	return JPEG_SUSPENDED;
511      }
512    }
513    /* Completed an MCU row, but perhaps not an iMCU row */
514    coef->MCU_ctr = 0;
515  }
516  (*cinfo->entropy->get_huffman_decoder_configuration)
517        (cinfo, &scan_header->prev_MCU_offset);
518  /* Completed the iMCU row, advance counters for next one */
519  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
520    start_iMCU_row(cinfo);
521    return JPEG_ROW_COMPLETED;
522  }
523  /* Completed the scan */
524  (*cinfo->inputctl->finish_input_pass) (cinfo);
525  return JPEG_SCAN_COMPLETED;
526}
527
528/*
529 * Decompress and return some data in the multi-pass case.
530 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
531 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
532 *
533 * NB: output_buf contains a plane for each component in image.
534 */
535
536METHODDEF(int)
537decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
538{
539  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
540  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
541  JDIMENSION block_num;
542  int ci, block_row, block_rows;
543  JBLOCKARRAY buffer;
544  JBLOCKROW buffer_ptr;
545  JSAMPARRAY output_ptr;
546  JDIMENSION output_col;
547  jpeg_component_info *compptr;
548  inverse_DCT_method_ptr inverse_DCT;
549
550  /* Force some input to be done if we are getting ahead of the input. */
551  while (cinfo->input_scan_number < cinfo->output_scan_number ||
552	 (cinfo->input_scan_number == cinfo->output_scan_number &&
553	  cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
554    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
555      return JPEG_SUSPENDED;
556  }
557
558  /* OK, output from the virtual arrays. */
559  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
560       ci++, compptr++) {
561    /* Don't bother to IDCT an uninteresting component. */
562    if (! compptr->component_needed)
563      continue;
564    /* Align the virtual buffer for this component. */
565    buffer = (*cinfo->mem->access_virt_barray)
566      ((j_common_ptr) cinfo, coef->whole_image[ci],
567       cinfo->tile_decode ? 0 : cinfo->output_iMCU_row * compptr->v_samp_factor,
568       (JDIMENSION) compptr->v_samp_factor, FALSE);
569    /* Count non-dummy DCT block rows in this iMCU row. */
570    if (cinfo->output_iMCU_row < last_iMCU_row)
571      block_rows = compptr->v_samp_factor;
572    else {
573      /* NB: can't use last_row_height here; it is input-side-dependent! */
574      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
575      if (block_rows == 0) block_rows = compptr->v_samp_factor;
576    }
577    inverse_DCT = cinfo->idct->inverse_DCT[ci];
578    output_ptr = output_buf[ci];
579    /* Loop over all DCT blocks to be processed. */
580    for (block_row = 0; block_row < block_rows; block_row++) {
581      buffer_ptr = buffer[block_row];
582      output_col = 0;
583      for (block_num = 0; block_num < compptr->width_in_blocks; block_num++) {
584	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
585			output_ptr, output_col);
586	buffer_ptr++;
587	output_col += compptr->DCT_scaled_size;
588      }
589      output_ptr += compptr->DCT_scaled_size;
590    }
591  }
592
593  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
594    return JPEG_ROW_COMPLETED;
595  return JPEG_SCAN_COMPLETED;
596}
597
598#endif /* D_MULTISCAN_FILES_SUPPORTED */
599
600
601#ifdef BLOCK_SMOOTHING_SUPPORTED
602
603/*
604 * This code applies interblock smoothing as described by section K.8
605 * of the JPEG standard: the first 5 AC coefficients are estimated from
606 * the DC values of a DCT block and its 8 neighboring blocks.
607 * We apply smoothing only for progressive JPEG decoding, and only if
608 * the coefficients it can estimate are not yet known to full precision.
609 */
610
611/* Natural-order array positions of the first 5 zigzag-order coefficients */
612#define Q01_POS  1
613#define Q10_POS  8
614#define Q20_POS  16
615#define Q11_POS  9
616#define Q02_POS  2
617
618/*
619 * Determine whether block smoothing is applicable and safe.
620 * We also latch the current states of the coef_bits[] entries for the
621 * AC coefficients; otherwise, if the input side of the decompressor
622 * advances into a new scan, we might think the coefficients are known
623 * more accurately than they really are.
624 */
625
626LOCAL(boolean)
627smoothing_ok (j_decompress_ptr cinfo)
628{
629  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
630  boolean smoothing_useful = FALSE;
631  int ci, coefi;
632  jpeg_component_info *compptr;
633  JQUANT_TBL * qtable;
634  int * coef_bits;
635  int * coef_bits_latch;
636
637  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
638    return FALSE;
639
640  /* Allocate latch area if not already done */
641  if (coef->coef_bits_latch == NULL)
642    coef->coef_bits_latch = (int *)
643      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
644				  cinfo->num_components *
645				  (SAVED_COEFS * SIZEOF(int)));
646  coef_bits_latch = coef->coef_bits_latch;
647
648  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
649       ci++, compptr++) {
650    /* All components' quantization values must already be latched. */
651    if ((qtable = compptr->quant_table) == NULL)
652      return FALSE;
653    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
654    if (qtable->quantval[0] == 0 ||
655	qtable->quantval[Q01_POS] == 0 ||
656	qtable->quantval[Q10_POS] == 0 ||
657	qtable->quantval[Q20_POS] == 0 ||
658	qtable->quantval[Q11_POS] == 0 ||
659	qtable->quantval[Q02_POS] == 0)
660      return FALSE;
661    /* DC values must be at least partly known for all components. */
662    coef_bits = cinfo->coef_bits[ci];
663    if (coef_bits[0] < 0)
664      return FALSE;
665    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
666    for (coefi = 1; coefi <= 5; coefi++) {
667      coef_bits_latch[coefi] = coef_bits[coefi];
668      if (coef_bits[coefi] != 0)
669	smoothing_useful = TRUE;
670    }
671    coef_bits_latch += SAVED_COEFS;
672  }
673
674  return smoothing_useful;
675}
676
677
678/*
679 * Variant of decompress_data for use when doing block smoothing.
680 */
681
682METHODDEF(int)
683decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
684{
685  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
686  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
687  JDIMENSION block_num, last_block_column;
688  int ci, block_row, block_rows, access_rows;
689  JBLOCKARRAY buffer;
690  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
691  JSAMPARRAY output_ptr;
692  JDIMENSION output_col;
693  jpeg_component_info *compptr;
694  inverse_DCT_method_ptr inverse_DCT;
695  boolean first_row, last_row;
696  JBLOCK workspace;
697  int *coef_bits;
698  JQUANT_TBL *quanttbl;
699  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
700  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
701  int Al, pred;
702
703  /* Force some input to be done if we are getting ahead of the input. */
704  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
705	 ! cinfo->inputctl->eoi_reached) {
706    if (cinfo->input_scan_number == cinfo->output_scan_number) {
707      /* If input is working on current scan, we ordinarily want it to
708       * have completed the current row.  But if input scan is DC,
709       * we want it to keep one row ahead so that next block row's DC
710       * values are up to date.
711       */
712      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
713      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
714	break;
715    }
716    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
717      return JPEG_SUSPENDED;
718  }
719
720  /* OK, output from the virtual arrays. */
721  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
722       ci++, compptr++) {
723    /* Don't bother to IDCT an uninteresting component. */
724    if (! compptr->component_needed)
725      continue;
726    /* Count non-dummy DCT block rows in this iMCU row. */
727    if (cinfo->output_iMCU_row < last_iMCU_row) {
728      block_rows = compptr->v_samp_factor;
729      access_rows = block_rows * 2; /* this and next iMCU row */
730      last_row = FALSE;
731    } else {
732      /* NB: can't use last_row_height here; it is input-side-dependent! */
733      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
734      if (block_rows == 0) block_rows = compptr->v_samp_factor;
735      access_rows = block_rows; /* this iMCU row only */
736      last_row = TRUE;
737    }
738    /* Align the virtual buffer for this component. */
739    if (cinfo->output_iMCU_row > 0) {
740      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
741      buffer = (*cinfo->mem->access_virt_barray)
742	((j_common_ptr) cinfo, coef->whole_image[ci],
743	 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
744	 (JDIMENSION) access_rows, FALSE);
745      buffer += compptr->v_samp_factor;	/* point to current iMCU row */
746      first_row = FALSE;
747    } else {
748      buffer = (*cinfo->mem->access_virt_barray)
749	((j_common_ptr) cinfo, coef->whole_image[ci],
750	 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
751      first_row = TRUE;
752    }
753    /* Fetch component-dependent info */
754    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
755    quanttbl = compptr->quant_table;
756    Q00 = quanttbl->quantval[0];
757    Q01 = quanttbl->quantval[Q01_POS];
758    Q10 = quanttbl->quantval[Q10_POS];
759    Q20 = quanttbl->quantval[Q20_POS];
760    Q11 = quanttbl->quantval[Q11_POS];
761    Q02 = quanttbl->quantval[Q02_POS];
762    inverse_DCT = cinfo->idct->inverse_DCT[ci];
763    output_ptr = output_buf[ci];
764    /* Loop over all DCT blocks to be processed. */
765    for (block_row = 0; block_row < block_rows; block_row++) {
766      buffer_ptr = buffer[block_row];
767      if (first_row && block_row == 0)
768	prev_block_row = buffer_ptr;
769      else
770	prev_block_row = buffer[block_row-1];
771      if (last_row && block_row == block_rows-1)
772	next_block_row = buffer_ptr;
773      else
774	next_block_row = buffer[block_row+1];
775      /* We fetch the surrounding DC values using a sliding-register approach.
776       * Initialize all nine here so as to do the right thing on narrow pics.
777       */
778      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
779      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
780      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
781      output_col = 0;
782      last_block_column = compptr->width_in_blocks - 1;
783      for (block_num = 0; block_num <= last_block_column; block_num++) {
784	/* Fetch current DCT block into workspace so we can modify it. */
785	jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
786	/* Update DC values */
787	if (block_num < last_block_column) {
788	  DC3 = (int) prev_block_row[1][0];
789	  DC6 = (int) buffer_ptr[1][0];
790	  DC9 = (int) next_block_row[1][0];
791	}
792	/* Compute coefficient estimates per K.8.
793	 * An estimate is applied only if coefficient is still zero,
794	 * and is not known to be fully accurate.
795	 */
796	/* AC01 */
797	if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
798	  num = 36 * Q00 * (DC4 - DC6);
799	  if (num >= 0) {
800	    pred = (int) (((Q01<<7) + num) / (Q01<<8));
801	    if (Al > 0 && pred >= (1<<Al))
802	      pred = (1<<Al)-1;
803	  } else {
804	    pred = (int) (((Q01<<7) - num) / (Q01<<8));
805	    if (Al > 0 && pred >= (1<<Al))
806	      pred = (1<<Al)-1;
807	    pred = -pred;
808	  }
809	  workspace[1] = (JCOEF) pred;
810	}
811	/* AC10 */
812	if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
813	  num = 36 * Q00 * (DC2 - DC8);
814	  if (num >= 0) {
815	    pred = (int) (((Q10<<7) + num) / (Q10<<8));
816	    if (Al > 0 && pred >= (1<<Al))
817	      pred = (1<<Al)-1;
818	  } else {
819	    pred = (int) (((Q10<<7) - num) / (Q10<<8));
820	    if (Al > 0 && pred >= (1<<Al))
821	      pred = (1<<Al)-1;
822	    pred = -pred;
823	  }
824	  workspace[8] = (JCOEF) pred;
825	}
826	/* AC20 */
827	if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
828	  num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
829	  if (num >= 0) {
830	    pred = (int) (((Q20<<7) + num) / (Q20<<8));
831	    if (Al > 0 && pred >= (1<<Al))
832	      pred = (1<<Al)-1;
833	  } else {
834	    pred = (int) (((Q20<<7) - num) / (Q20<<8));
835	    if (Al > 0 && pred >= (1<<Al))
836	      pred = (1<<Al)-1;
837	    pred = -pred;
838	  }
839	  workspace[16] = (JCOEF) pred;
840	}
841	/* AC11 */
842	if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
843	  num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
844	  if (num >= 0) {
845	    pred = (int) (((Q11<<7) + num) / (Q11<<8));
846	    if (Al > 0 && pred >= (1<<Al))
847	      pred = (1<<Al)-1;
848	  } else {
849	    pred = (int) (((Q11<<7) - num) / (Q11<<8));
850	    if (Al > 0 && pred >= (1<<Al))
851	      pred = (1<<Al)-1;
852	    pred = -pred;
853	  }
854	  workspace[9] = (JCOEF) pred;
855	}
856	/* AC02 */
857	if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
858	  num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
859	  if (num >= 0) {
860	    pred = (int) (((Q02<<7) + num) / (Q02<<8));
861	    if (Al > 0 && pred >= (1<<Al))
862	      pred = (1<<Al)-1;
863	  } else {
864	    pred = (int) (((Q02<<7) - num) / (Q02<<8));
865	    if (Al > 0 && pred >= (1<<Al))
866	      pred = (1<<Al)-1;
867	    pred = -pred;
868	  }
869	  workspace[2] = (JCOEF) pred;
870	}
871	/* OK, do the IDCT */
872	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
873			output_ptr, output_col);
874	/* Advance for next column */
875	DC1 = DC2; DC2 = DC3;
876	DC4 = DC5; DC5 = DC6;
877	DC7 = DC8; DC8 = DC9;
878	buffer_ptr++, prev_block_row++, next_block_row++;
879	output_col += compptr->DCT_scaled_size;
880      }
881      output_ptr += compptr->DCT_scaled_size;
882    }
883  }
884
885  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
886    return JPEG_ROW_COMPLETED;
887  return JPEG_SCAN_COMPLETED;
888}
889
890#endif /* BLOCK_SMOOTHING_SUPPORTED */
891
892
893/*
894 * Initialize coefficient buffer controller.
895 */
896
897GLOBAL(void)
898jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
899{
900  my_coef_ptr coef;
901
902  coef = (my_coef_ptr)
903    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
904				SIZEOF(my_coef_controller));
905  cinfo->coef = (struct jpeg_d_coef_controller *) coef;
906  coef->pub.start_input_pass = start_input_pass;
907  coef->pub.start_output_pass = start_output_pass;
908  coef->pub.column_left_boundary = 0;
909  coef->pub.column_right_boundary = 0;
910#ifdef BLOCK_SMOOTHING_SUPPORTED
911  coef->coef_bits_latch = NULL;
912#endif
913
914#ifdef ANDROID_TILE_BASED_DECODE
915  if (cinfo->tile_decode) {
916    if (cinfo->progressive_mode) {
917      /* Allocate one iMCU row virtual array, coef->whole_image[ci],
918       * for each color component, padded to a multiple of h_samp_factor
919       * DCT blocks in the horizontal direction.
920       */
921      int ci, access_rows;
922      jpeg_component_info *compptr;
923
924      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
925	   ci++, compptr++) {
926        access_rows = compptr->v_samp_factor;
927        coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
928	  ((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
929	   (JDIMENSION) jround_up((long) compptr->width_in_blocks,
930				(long) compptr->h_samp_factor),
931	   (JDIMENSION) compptr->v_samp_factor, // one iMCU row
932	   (JDIMENSION) access_rows);
933      }
934      coef->pub.consume_data_build_huffman_index =
935            consume_data_build_huffman_index_progressive;
936      coef->pub.consume_data = consume_data_multi_scan;
937      coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
938      coef->pub.decompress_data = decompress_onepass;
939    } else {
940      /* We only need a single-MCU buffer. */
941      JBLOCKROW buffer;
942      int i;
943
944      buffer = (JBLOCKROW)
945      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
946				  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
947      for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
948        coef->MCU_buffer[i] = buffer + i;
949      }
950      coef->pub.consume_data_build_huffman_index =
951            consume_data_build_huffman_index_baseline;
952      coef->pub.consume_data = dummy_consume_data;
953      coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
954      coef->pub.decompress_data = decompress_onepass;
955    }
956    return;
957  }
958#endif
959
960  /* Create the coefficient buffer. */
961  if (need_full_buffer) {
962#ifdef D_MULTISCAN_FILES_SUPPORTED
963    /* Allocate a full-image virtual array for each component, */
964    /* padded to a multiple of samp_factor DCT blocks in each direction. */
965    /* Note we ask for a pre-zeroed array. */
966    int ci, access_rows;
967    jpeg_component_info *compptr;
968
969    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
970	 ci++, compptr++) {
971      access_rows = compptr->v_samp_factor;
972#ifdef BLOCK_SMOOTHING_SUPPORTED
973      /* If block smoothing could be used, need a bigger window */
974      if (cinfo->progressive_mode)
975	access_rows *= 3;
976#endif
977      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
978	((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
979	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
980				(long) compptr->h_samp_factor),
981	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
982				(long) compptr->v_samp_factor),
983	 (JDIMENSION) access_rows);
984    }
985    coef->pub.consume_data = consume_data;
986    coef->pub.decompress_data = decompress_data;
987    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
988#else
989    ERREXIT(cinfo, JERR_NOT_COMPILED);
990#endif
991  } else {
992    /* We only need a single-MCU buffer. */
993    JBLOCKROW buffer;
994    int i;
995
996    buffer = (JBLOCKROW)
997      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
998		  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
999    for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
1000      coef->MCU_buffer[i] = buffer + i;
1001    }
1002    coef->pub.consume_data = dummy_consume_data;
1003    coef->pub.decompress_data = decompress_onepass;
1004    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
1005  }
1006}
1007