jdcoefct.c revision e0eadaa39b72e33f032220246c771d7302ebeaf8
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      if (MCU_col_num < coef->pub.MCU_columns_to_skip) {
174        (*cinfo->entropy->decode_mcu_discard_coef) (cinfo);
175        continue;
176      } else {
177        jzero_far((void FAR *) coef->MCU_buffer[0],
178		(size_t) (cinfo->blocks_in_MCU * SIZEOF(JBLOCK)));
179        if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
180	  /* Suspension forced; update state counters and exit */
181	  coef->MCU_vert_offset = yoffset;
182	  coef->MCU_ctr = MCU_col_num;
183	  return JPEG_SUSPENDED;
184        }
185      }
186      /* Determine where data should go in output_buf and do the IDCT thing.
187       * We skip dummy blocks at the right and bottom edges (but blkn gets
188       * incremented past them!).  Note the inner loop relies on having
189       * allocated the MCU_buffer[] blocks sequentially.
190       */
191      blkn = 0;			/* index of current DCT block within MCU */
192      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
193	compptr = cinfo->cur_comp_info[ci];
194	/* Don't bother to IDCT an uninteresting component. */
195	if (! compptr->component_needed) {
196	  blkn += compptr->MCU_blocks;
197	  continue;
198	}
199	inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index];
200	useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
201						    : compptr->last_col_width;
202	output_ptr = output_buf[compptr->component_index] +
203	  yoffset * compptr->DCT_scaled_size;
204	start_col = MCU_col_num * compptr->MCU_sample_width;
205	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
206	  if (cinfo->input_iMCU_row < last_iMCU_row ||
207	      yoffset+yindex < compptr->last_row_height) {
208	    output_col = start_col;
209	    for (xindex = 0; xindex < useful_width; xindex++) {
210	      (*inverse_DCT) (cinfo, compptr,
211		        (JCOEFPTR) coef->MCU_buffer[blkn+xindex],
212		        output_ptr, output_col);
213	      output_col += compptr->DCT_scaled_size;
214	    }
215	  }
216	  blkn += compptr->MCU_width;
217	  output_ptr += compptr->DCT_scaled_size;
218	}
219      }
220    }
221    /* Completed an MCU row, but perhaps not an iMCU row */
222    coef->MCU_ctr = 0;
223  }
224  /* Completed the iMCU row, advance counters for next one */
225  cinfo->output_iMCU_row++;
226  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
227    start_iMCU_row(cinfo);
228    return JPEG_ROW_COMPLETED;
229  }
230  /* Completed the scan */
231  (*cinfo->inputctl->finish_input_pass) (cinfo);
232  return JPEG_SCAN_COMPLETED;
233}
234
235
236/*
237 * Dummy consume-input routine for single-pass operation.
238 */
239
240METHODDEF(int)
241dummy_consume_data (j_decompress_ptr cinfo)
242{
243  return JPEG_SUSPENDED;	/* Always indicate nothing was done */
244}
245
246#ifdef D_MULTISCAN_FILES_SUPPORTED
247/*
248 * Consume input data and store it in the full-image coefficient buffer.
249 * We read as much as one fully interleaved MCU row ("iMCU" row) per call,
250 * ie, v_samp_factor block rows for each component in the scan.
251 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
252 */
253
254METHODDEF(int)
255consume_data (j_decompress_ptr cinfo)
256{
257  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
258  JDIMENSION MCU_col_num;	/* index of current MCU within row */
259  int blkn, ci, xindex, yindex, yoffset;
260  JDIMENSION start_col;
261  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
262  JBLOCKROW buffer_ptr;
263  jpeg_component_info *compptr;
264
265  /* Align the virtual buffers for the components used in this scan. */
266  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
267    compptr = cinfo->cur_comp_info[ci];
268    buffer[ci] = (*cinfo->mem->access_virt_barray)
269      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
270       cinfo->tile_decode ? 0 : cinfo->input_iMCU_row * compptr->v_samp_factor,
271       (JDIMENSION) compptr->v_samp_factor, TRUE);
272    /* Note: entropy decoder expects buffer to be zeroed,
273     * but this is handled automatically by the memory manager
274     * because we requested a pre-zeroed array.
275     */
276  }
277  unsigned int MCUs_per_row = cinfo->MCUs_per_row;
278#ifdef ANDROID_TILE_BASED_DECODE
279  if (cinfo->tile_decode) {
280    MCUs_per_row = jmin(MCUs_per_row,
281        (cinfo->coef->column_right_boundary - cinfo->coef->column_left_boundary)
282        * cinfo->entropy->index->MCU_sample_size * cinfo->max_h_samp_factor);
283  }
284#endif
285
286  /* Loop to process one whole iMCU row */
287  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
288       yoffset++) {
289#ifdef ANDROID_TILE_BASED_DECODE
290    if (cinfo->tile_decode) {
291      huffman_scan_header scan_header =
292            cinfo->entropy->index->scan[cinfo->input_scan_number];
293      int col_offset = cinfo->coef->column_left_boundary;
294      (*cinfo->entropy->configure_huffman_decoder) (cinfo,
295              scan_header.offset[cinfo->input_iMCU_row]
296              [col_offset + yoffset * scan_header.MCUs_per_row]);
297    }
298#endif
299    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < MCUs_per_row;
300	 MCU_col_num++) {
301      /* Construct list of pointers to DCT blocks belonging to this MCU */
302      blkn = 0;			/* index of current DCT block within MCU */
303      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
304        compptr = cinfo->cur_comp_info[ci];
305        start_col = MCU_col_num * compptr->MCU_width;
306        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
307          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
308          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
309            coef->MCU_buffer[blkn++] = buffer_ptr++;
310#ifdef ANDROID_TILE_BASED_DECODE
311            if (cinfo->tile_decode && cinfo->input_scan_number == 0) {
312              // need to do pre-zero ourself.
313              jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
314                        (size_t) (SIZEOF(JBLOCK)));
315            }
316#endif
317          }
318        }
319      }
320      /* Try to fetch the MCU. */
321      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
322	/* Suspension forced; update state counters and exit */
323	coef->MCU_vert_offset = yoffset;
324	coef->MCU_ctr = MCU_col_num;
325	return JPEG_SUSPENDED;
326      }
327    }
328    /* Completed an MCU row, but perhaps not an iMCU row */
329    coef->MCU_ctr = 0;
330  }
331  /* Completed the iMCU row, advance counters for next one */
332  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
333    start_iMCU_row(cinfo);
334    return JPEG_ROW_COMPLETED;
335  }
336  /* Completed the scan */
337  (*cinfo->inputctl->finish_input_pass) (cinfo);
338  return JPEG_SCAN_COMPLETED;
339}
340
341/*
342 * Consume input data and store it in the coefficient buffer.
343 * Read one fully interleaved MCU row ("iMCU" row) per call.
344 */
345
346METHODDEF(int)
347consume_data_multi_scan (j_decompress_ptr cinfo)
348{
349  huffman_index *index = cinfo->entropy->index;
350  int i, retcode, ci;
351  int mcu = cinfo->input_iMCU_row;
352  jinit_phuff_decoder(cinfo);
353  for (i = 0; i < index->scan_count; i++) {
354    (*cinfo->inputctl->finish_input_pass) (cinfo);
355    jset_input_stream_position(cinfo, index->scan[i].bitstream_offset);
356    cinfo->output_iMCU_row = mcu;
357    cinfo->unread_marker = 0;
358    // Consume SOS and DHT headers
359    retcode = (*cinfo->inputctl->consume_markers) (cinfo, index, i);
360    cinfo->input_iMCU_row = mcu;
361    cinfo->input_scan_number = i;
362    cinfo->entropy->index = index;
363    // Consume scan block data
364    consume_data(cinfo);
365  }
366  cinfo->input_iMCU_row = mcu + 1;
367  cinfo->input_scan_number = 0;
368  cinfo->output_scan_number = 0;
369  return JPEG_ROW_COMPLETED;
370}
371
372/*
373 * Same as consume_data, expect for saving the Huffman decode information
374 * - bitstream offset and DC coefficient to index.
375 */
376
377METHODDEF(int)
378consume_data_build_huffman_index_baseline (j_decompress_ptr cinfo,
379        huffman_index *index, int current_scan)
380{
381  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
382  JDIMENSION MCU_col_num;	/* index of current MCU within row */
383  int ci, xindex, yindex, yoffset;
384  JDIMENSION start_col;
385  JBLOCKROW buffer_ptr;
386
387  huffman_scan_header *scan_header = index->scan + current_scan;
388  scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
389
390  size_t allocate_size = coef->MCU_rows_per_iMCU_row
391      * jdiv_round_up(cinfo->MCUs_per_row, index->MCU_sample_size)
392      * sizeof(huffman_offset_data);
393  scan_header->offset[cinfo->input_iMCU_row] =
394        (huffman_offset_data*)malloc(allocate_size);
395  index->mem_used += allocate_size;
396
397  huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
398
399  /* Loop to process one whole iMCU row */
400  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
401       yoffset++) {
402    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
403	 MCU_col_num++) {
404      // Record huffman bit offset
405      if (MCU_col_num % index->MCU_sample_size == 0) {
406        (*cinfo->entropy->get_huffman_decoder_configuration)
407                (cinfo, offset_data);
408        ++offset_data;
409      }
410
411      /* Try to fetch the MCU. */
412      if (! (*cinfo->entropy->decode_mcu_discard_coef) (cinfo)) {
413        /* Suspension forced; update state counters and exit */
414        coef->MCU_vert_offset = yoffset;
415        coef->MCU_ctr = MCU_col_num;
416        return JPEG_SUSPENDED;
417      }
418    }
419    /* Completed an MCU row, but perhaps not an iMCU row */
420    coef->MCU_ctr = 0;
421  }
422  /* Completed the iMCU row, advance counters for next one */
423  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
424    start_iMCU_row(cinfo);
425    return JPEG_ROW_COMPLETED;
426  }
427  /* Completed the scan */
428  (*cinfo->inputctl->finish_input_pass) (cinfo);
429  return JPEG_SCAN_COMPLETED;
430}
431
432/*
433 * Same as consume_data, expect for saving the Huffman decode information
434 * - bitstream offset and DC coefficient to index.
435 */
436
437METHODDEF(int)
438consume_data_build_huffman_index_progressive (j_decompress_ptr cinfo,
439        huffman_index *index, int current_scan)
440{
441  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
442  JDIMENSION MCU_col_num;	/* index of current MCU within row */
443  int blkn, ci, xindex, yindex, yoffset;
444  JDIMENSION start_col;
445  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
446  JBLOCKROW buffer_ptr;
447  jpeg_component_info *compptr;
448
449  int factor = 4; // maximum factor is 4.
450  for (ci = 0; ci < cinfo->comps_in_scan; ci++)
451    factor = jmin(factor, cinfo->cur_comp_info[ci]->h_samp_factor);
452
453  int sample_size = index->MCU_sample_size * factor;
454  huffman_scan_header *scan_header = index->scan + current_scan;
455  scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
456  scan_header->MCUs_per_row = jdiv_round_up(cinfo->MCUs_per_row, sample_size);
457  scan_header->comps_in_scan = cinfo->comps_in_scan;
458
459  size_t allocate_size = coef->MCU_rows_per_iMCU_row
460      * scan_header->MCUs_per_row * sizeof(huffman_offset_data);
461  scan_header->offset[cinfo->input_iMCU_row] =
462        (huffman_offset_data*)malloc(allocate_size);
463  index->mem_used += allocate_size;
464
465  huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
466
467  /* Align the virtual buffers for the components used in this scan. */
468  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
469    compptr = cinfo->cur_comp_info[ci];
470    buffer[ci] = (*cinfo->mem->access_virt_barray)
471      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
472       0, // Only need one row buffer
473       (JDIMENSION) compptr->v_samp_factor, TRUE);
474  }
475  /* Loop to process one whole iMCU row */
476  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
477       yoffset++) {
478    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
479	 MCU_col_num++) {
480      /* For each MCU, we loop through different color components.
481       * Then, for each color component we will get a list of pointers to DCT
482       * blocks in the virtual buffer.
483       */
484      blkn = 0; /* index of current DCT block within MCU */
485      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
486        compptr = cinfo->cur_comp_info[ci];
487        start_col = MCU_col_num * compptr->MCU_width;
488        /* Get the list of pointers to DCT blocks in
489         * the virtual buffer in a color component of the MCU.
490         */
491        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
492          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
493          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
494            coef->MCU_buffer[blkn++] = buffer_ptr++;
495            if (cinfo->input_scan_number == 0) {
496              // need to do pre-zero by ourself.
497              jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
498                        (size_t) (SIZEOF(JBLOCK)));
499            }
500          }
501        }
502      }
503      // Record huffman bit offset
504      if (MCU_col_num % sample_size == 0) {
505        (*cinfo->entropy->get_huffman_decoder_configuration)
506                (cinfo, offset_data);
507        ++offset_data;
508      }
509      /* Try to fetch the MCU. */
510      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
511	/* Suspension forced; update state counters and exit */
512	coef->MCU_vert_offset = yoffset;
513	coef->MCU_ctr = MCU_col_num;
514	return JPEG_SUSPENDED;
515      }
516    }
517    /* Completed an MCU row, but perhaps not an iMCU row */
518    coef->MCU_ctr = 0;
519  }
520  (*cinfo->entropy->get_huffman_decoder_configuration)
521        (cinfo, &scan_header->prev_MCU_offset);
522  /* Completed the iMCU row, advance counters for next one */
523  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
524    start_iMCU_row(cinfo);
525    return JPEG_ROW_COMPLETED;
526  }
527  /* Completed the scan */
528  (*cinfo->inputctl->finish_input_pass) (cinfo);
529  return JPEG_SCAN_COMPLETED;
530}
531
532/*
533 * Decompress and return some data in the multi-pass case.
534 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
535 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
536 *
537 * NB: output_buf contains a plane for each component in image.
538 */
539
540METHODDEF(int)
541decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
542{
543  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
544  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
545  JDIMENSION block_num;
546  int ci, block_row, block_rows;
547  JBLOCKARRAY buffer;
548  JBLOCKROW buffer_ptr;
549  JSAMPARRAY output_ptr;
550  JDIMENSION output_col;
551  jpeg_component_info *compptr;
552  inverse_DCT_method_ptr inverse_DCT;
553
554  /* Force some input to be done if we are getting ahead of the input. */
555  while (cinfo->input_scan_number < cinfo->output_scan_number ||
556	 (cinfo->input_scan_number == cinfo->output_scan_number &&
557	  cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
558    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
559      return JPEG_SUSPENDED;
560  }
561
562  /* OK, output from the virtual arrays. */
563  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
564       ci++, compptr++) {
565    /* Don't bother to IDCT an uninteresting component. */
566    if (! compptr->component_needed)
567      continue;
568    /* Align the virtual buffer for this component. */
569    buffer = (*cinfo->mem->access_virt_barray)
570      ((j_common_ptr) cinfo, coef->whole_image[ci],
571       cinfo->tile_decode ? 0 : cinfo->output_iMCU_row * compptr->v_samp_factor,
572       (JDIMENSION) compptr->v_samp_factor, FALSE);
573    /* Count non-dummy DCT block rows in this iMCU row. */
574    if (cinfo->output_iMCU_row < last_iMCU_row)
575      block_rows = compptr->v_samp_factor;
576    else {
577      /* NB: can't use last_row_height here; it is input-side-dependent! */
578      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
579      if (block_rows == 0) block_rows = compptr->v_samp_factor;
580    }
581    inverse_DCT = cinfo->idct->inverse_DCT[ci];
582    output_ptr = output_buf[ci];
583    int width_in_blocks = compptr->width_in_blocks;
584    int start_block = 0;
585#if ANDROID_TILE_BASED_DECODE
586    if (cinfo->tile_decode) {
587      width_in_blocks = jmin(width_in_blocks,
588        (cinfo->coef->MCU_column_right_boundary -
589         cinfo->coef->MCU_column_left_boundary) *
590         cinfo->max_h_samp_factor /
591         compptr->h_samp_factor);
592      start_block = coef->pub.MCU_columns_to_skip *
593        cinfo->max_h_samp_factor / compptr->h_samp_factor;
594    }
595#endif
596    /* Loop over all DCT blocks to be processed. */
597    for (block_row = 0; block_row < block_rows; block_row++) {
598      buffer_ptr = buffer[block_row];
599      output_col = start_block * compptr->DCT_scaled_size;
600      buffer_ptr += start_block;
601      for (block_num = start_block; block_num < width_in_blocks; block_num++) {
602	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
603			output_ptr, output_col);
604	buffer_ptr++;
605	output_col += compptr->DCT_scaled_size;
606      }
607      output_ptr += compptr->DCT_scaled_size;
608    }
609  }
610
611  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
612    return JPEG_ROW_COMPLETED;
613  return JPEG_SCAN_COMPLETED;
614}
615
616#endif /* D_MULTISCAN_FILES_SUPPORTED */
617
618
619#ifdef BLOCK_SMOOTHING_SUPPORTED
620
621/*
622 * This code applies interblock smoothing as described by section K.8
623 * of the JPEG standard: the first 5 AC coefficients are estimated from
624 * the DC values of a DCT block and its 8 neighboring blocks.
625 * We apply smoothing only for progressive JPEG decoding, and only if
626 * the coefficients it can estimate are not yet known to full precision.
627 */
628
629/* Natural-order array positions of the first 5 zigzag-order coefficients */
630#define Q01_POS  1
631#define Q10_POS  8
632#define Q20_POS  16
633#define Q11_POS  9
634#define Q02_POS  2
635
636/*
637 * Determine whether block smoothing is applicable and safe.
638 * We also latch the current states of the coef_bits[] entries for the
639 * AC coefficients; otherwise, if the input side of the decompressor
640 * advances into a new scan, we might think the coefficients are known
641 * more accurately than they really are.
642 */
643
644LOCAL(boolean)
645smoothing_ok (j_decompress_ptr cinfo)
646{
647  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
648  boolean smoothing_useful = FALSE;
649  int ci, coefi;
650  jpeg_component_info *compptr;
651  JQUANT_TBL * qtable;
652  int * coef_bits;
653  int * coef_bits_latch;
654
655  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
656    return FALSE;
657
658  /* Allocate latch area if not already done */
659  if (coef->coef_bits_latch == NULL)
660    coef->coef_bits_latch = (int *)
661      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
662				  cinfo->num_components *
663				  (SAVED_COEFS * SIZEOF(int)));
664  coef_bits_latch = coef->coef_bits_latch;
665
666  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
667       ci++, compptr++) {
668    /* All components' quantization values must already be latched. */
669    if ((qtable = compptr->quant_table) == NULL)
670      return FALSE;
671    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
672    if (qtable->quantval[0] == 0 ||
673	qtable->quantval[Q01_POS] == 0 ||
674	qtable->quantval[Q10_POS] == 0 ||
675	qtable->quantval[Q20_POS] == 0 ||
676	qtable->quantval[Q11_POS] == 0 ||
677	qtable->quantval[Q02_POS] == 0)
678      return FALSE;
679    /* DC values must be at least partly known for all components. */
680    coef_bits = cinfo->coef_bits[ci];
681    if (coef_bits[0] < 0)
682      return FALSE;
683    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
684    for (coefi = 1; coefi <= 5; coefi++) {
685      coef_bits_latch[coefi] = coef_bits[coefi];
686      if (coef_bits[coefi] != 0)
687	smoothing_useful = TRUE;
688    }
689    coef_bits_latch += SAVED_COEFS;
690  }
691
692  return smoothing_useful;
693}
694
695
696/*
697 * Variant of decompress_data for use when doing block smoothing.
698 */
699
700METHODDEF(int)
701decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
702{
703  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
704  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
705  JDIMENSION block_num, last_block_column;
706  int ci, block_row, block_rows, access_rows;
707  JBLOCKARRAY buffer;
708  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
709  JSAMPARRAY output_ptr;
710  JDIMENSION output_col;
711  jpeg_component_info *compptr;
712  inverse_DCT_method_ptr inverse_DCT;
713  boolean first_row, last_row;
714  JBLOCK workspace;
715  int *coef_bits;
716  JQUANT_TBL *quanttbl;
717  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
718  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
719  int Al, pred;
720
721  /* Force some input to be done if we are getting ahead of the input. */
722  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
723	 ! cinfo->inputctl->eoi_reached) {
724    if (cinfo->input_scan_number == cinfo->output_scan_number) {
725      /* If input is working on current scan, we ordinarily want it to
726       * have completed the current row.  But if input scan is DC,
727       * we want it to keep one row ahead so that next block row's DC
728       * values are up to date.
729       */
730      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
731      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
732	break;
733    }
734    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
735      return JPEG_SUSPENDED;
736  }
737
738  /* OK, output from the virtual arrays. */
739  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
740       ci++, compptr++) {
741    /* Don't bother to IDCT an uninteresting component. */
742    if (! compptr->component_needed)
743      continue;
744    /* Count non-dummy DCT block rows in this iMCU row. */
745    if (cinfo->output_iMCU_row < last_iMCU_row) {
746      block_rows = compptr->v_samp_factor;
747      access_rows = block_rows * 2; /* this and next iMCU row */
748      last_row = FALSE;
749    } else {
750      /* NB: can't use last_row_height here; it is input-side-dependent! */
751      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
752      if (block_rows == 0) block_rows = compptr->v_samp_factor;
753      access_rows = block_rows; /* this iMCU row only */
754      last_row = TRUE;
755    }
756    /* Align the virtual buffer for this component. */
757    if (cinfo->output_iMCU_row > 0) {
758      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
759      buffer = (*cinfo->mem->access_virt_barray)
760	((j_common_ptr) cinfo, coef->whole_image[ci],
761	 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
762	 (JDIMENSION) access_rows, FALSE);
763      buffer += compptr->v_samp_factor;	/* point to current iMCU row */
764      first_row = FALSE;
765    } else {
766      buffer = (*cinfo->mem->access_virt_barray)
767	((j_common_ptr) cinfo, coef->whole_image[ci],
768	 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
769      first_row = TRUE;
770    }
771    /* Fetch component-dependent info */
772    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
773    quanttbl = compptr->quant_table;
774    Q00 = quanttbl->quantval[0];
775    Q01 = quanttbl->quantval[Q01_POS];
776    Q10 = quanttbl->quantval[Q10_POS];
777    Q20 = quanttbl->quantval[Q20_POS];
778    Q11 = quanttbl->quantval[Q11_POS];
779    Q02 = quanttbl->quantval[Q02_POS];
780    inverse_DCT = cinfo->idct->inverse_DCT[ci];
781    output_ptr = output_buf[ci];
782    /* Loop over all DCT blocks to be processed. */
783    for (block_row = 0; block_row < block_rows; block_row++) {
784      buffer_ptr = buffer[block_row];
785      if (first_row && block_row == 0)
786	prev_block_row = buffer_ptr;
787      else
788	prev_block_row = buffer[block_row-1];
789      if (last_row && block_row == block_rows-1)
790	next_block_row = buffer_ptr;
791      else
792	next_block_row = buffer[block_row+1];
793      /* We fetch the surrounding DC values using a sliding-register approach.
794       * Initialize all nine here so as to do the right thing on narrow pics.
795       */
796      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
797      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
798      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
799      output_col = 0;
800      last_block_column = compptr->width_in_blocks - 1;
801      for (block_num = 0; block_num <= last_block_column; block_num++) {
802	/* Fetch current DCT block into workspace so we can modify it. */
803	jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
804	/* Update DC values */
805	if (block_num < last_block_column) {
806	  DC3 = (int) prev_block_row[1][0];
807	  DC6 = (int) buffer_ptr[1][0];
808	  DC9 = (int) next_block_row[1][0];
809	}
810	/* Compute coefficient estimates per K.8.
811	 * An estimate is applied only if coefficient is still zero,
812	 * and is not known to be fully accurate.
813	 */
814	/* AC01 */
815	if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
816	  num = 36 * Q00 * (DC4 - DC6);
817	  if (num >= 0) {
818	    pred = (int) (((Q01<<7) + num) / (Q01<<8));
819	    if (Al > 0 && pred >= (1<<Al))
820	      pred = (1<<Al)-1;
821	  } else {
822	    pred = (int) (((Q01<<7) - num) / (Q01<<8));
823	    if (Al > 0 && pred >= (1<<Al))
824	      pred = (1<<Al)-1;
825	    pred = -pred;
826	  }
827	  workspace[1] = (JCOEF) pred;
828	}
829	/* AC10 */
830	if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
831	  num = 36 * Q00 * (DC2 - DC8);
832	  if (num >= 0) {
833	    pred = (int) (((Q10<<7) + num) / (Q10<<8));
834	    if (Al > 0 && pred >= (1<<Al))
835	      pred = (1<<Al)-1;
836	  } else {
837	    pred = (int) (((Q10<<7) - num) / (Q10<<8));
838	    if (Al > 0 && pred >= (1<<Al))
839	      pred = (1<<Al)-1;
840	    pred = -pred;
841	  }
842	  workspace[8] = (JCOEF) pred;
843	}
844	/* AC20 */
845	if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
846	  num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
847	  if (num >= 0) {
848	    pred = (int) (((Q20<<7) + num) / (Q20<<8));
849	    if (Al > 0 && pred >= (1<<Al))
850	      pred = (1<<Al)-1;
851	  } else {
852	    pred = (int) (((Q20<<7) - num) / (Q20<<8));
853	    if (Al > 0 && pred >= (1<<Al))
854	      pred = (1<<Al)-1;
855	    pred = -pred;
856	  }
857	  workspace[16] = (JCOEF) pred;
858	}
859	/* AC11 */
860	if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
861	  num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
862	  if (num >= 0) {
863	    pred = (int) (((Q11<<7) + num) / (Q11<<8));
864	    if (Al > 0 && pred >= (1<<Al))
865	      pred = (1<<Al)-1;
866	  } else {
867	    pred = (int) (((Q11<<7) - num) / (Q11<<8));
868	    if (Al > 0 && pred >= (1<<Al))
869	      pred = (1<<Al)-1;
870	    pred = -pred;
871	  }
872	  workspace[9] = (JCOEF) pred;
873	}
874	/* AC02 */
875	if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
876	  num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
877	  if (num >= 0) {
878	    pred = (int) (((Q02<<7) + num) / (Q02<<8));
879	    if (Al > 0 && pred >= (1<<Al))
880	      pred = (1<<Al)-1;
881	  } else {
882	    pred = (int) (((Q02<<7) - num) / (Q02<<8));
883	    if (Al > 0 && pred >= (1<<Al))
884	      pred = (1<<Al)-1;
885	    pred = -pred;
886	  }
887	  workspace[2] = (JCOEF) pred;
888	}
889	/* OK, do the IDCT */
890	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
891			output_ptr, output_col);
892	/* Advance for next column */
893	DC1 = DC2; DC2 = DC3;
894	DC4 = DC5; DC5 = DC6;
895	DC7 = DC8; DC8 = DC9;
896	buffer_ptr++, prev_block_row++, next_block_row++;
897	output_col += compptr->DCT_scaled_size;
898      }
899      output_ptr += compptr->DCT_scaled_size;
900    }
901  }
902
903  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
904    return JPEG_ROW_COMPLETED;
905  return JPEG_SCAN_COMPLETED;
906}
907
908#endif /* BLOCK_SMOOTHING_SUPPORTED */
909
910
911/*
912 * Initialize coefficient buffer controller.
913 */
914
915GLOBAL(void)
916jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
917{
918  my_coef_ptr coef;
919
920  coef = (my_coef_ptr)
921    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
922				SIZEOF(my_coef_controller));
923  cinfo->coef = (struct jpeg_d_coef_controller *) coef;
924  coef->pub.start_input_pass = start_input_pass;
925  coef->pub.start_output_pass = start_output_pass;
926  coef->pub.column_left_boundary = 0;
927  coef->pub.column_right_boundary = 0;
928  coef->pub.MCU_columns_to_skip = 0;
929#ifdef BLOCK_SMOOTHING_SUPPORTED
930  coef->coef_bits_latch = NULL;
931#endif
932
933#ifdef ANDROID_TILE_BASED_DECODE
934  if (cinfo->tile_decode) {
935    if (cinfo->progressive_mode) {
936      /* Allocate one iMCU row virtual array, coef->whole_image[ci],
937       * for each color component, padded to a multiple of h_samp_factor
938       * DCT blocks in the horizontal direction.
939       */
940      int ci, access_rows;
941      jpeg_component_info *compptr;
942
943      for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
944	   ci++, compptr++) {
945        access_rows = compptr->v_samp_factor;
946        coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
947	  ((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
948	   (JDIMENSION) jround_up((long) compptr->width_in_blocks,
949				(long) compptr->h_samp_factor),
950	   (JDIMENSION) compptr->v_samp_factor, // one iMCU row
951	   (JDIMENSION) access_rows);
952      }
953      coef->pub.consume_data_build_huffman_index =
954            consume_data_build_huffman_index_progressive;
955      coef->pub.consume_data = consume_data_multi_scan;
956      coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
957      coef->pub.decompress_data = decompress_onepass;
958    } else {
959      /* We only need a single-MCU buffer. */
960      JBLOCKROW buffer;
961      int i;
962
963      buffer = (JBLOCKROW)
964      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
965				  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
966      for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
967        coef->MCU_buffer[i] = buffer + i;
968      }
969      coef->pub.consume_data_build_huffman_index =
970            consume_data_build_huffman_index_baseline;
971      coef->pub.consume_data = dummy_consume_data;
972      coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
973      coef->pub.decompress_data = decompress_onepass;
974    }
975    return;
976  }
977#endif
978
979  /* Create the coefficient buffer. */
980  if (need_full_buffer) {
981#ifdef D_MULTISCAN_FILES_SUPPORTED
982    /* Allocate a full-image virtual array for each component, */
983    /* padded to a multiple of samp_factor DCT blocks in each direction. */
984    /* Note we ask for a pre-zeroed array. */
985    int ci, access_rows;
986    jpeg_component_info *compptr;
987
988    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
989	 ci++, compptr++) {
990      access_rows = compptr->v_samp_factor;
991#ifdef BLOCK_SMOOTHING_SUPPORTED
992      /* If block smoothing could be used, need a bigger window */
993      if (cinfo->progressive_mode)
994	access_rows *= 3;
995#endif
996      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
997	((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
998	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
999				(long) compptr->h_samp_factor),
1000	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
1001				(long) compptr->v_samp_factor),
1002	 (JDIMENSION) access_rows);
1003    }
1004    coef->pub.consume_data = consume_data;
1005    coef->pub.decompress_data = decompress_data;
1006    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
1007#else
1008    ERREXIT(cinfo, JERR_NOT_COMPILED);
1009#endif
1010  } else {
1011    /* We only need a single-MCU buffer. */
1012    JBLOCKROW buffer;
1013    int i;
1014
1015    buffer = (JBLOCKROW)
1016      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
1017		  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
1018    for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
1019      coef->MCU_buffer[i] = buffer + i;
1020    }
1021    coef->pub.consume_data = dummy_consume_data;
1022    coef->pub.decompress_data = decompress_onepass;
1023    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
1024  }
1025}
1026