jdcoefct.c revision 3147fbe7688fc353e6ae03825a37cf101a4ee01d
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  /* Loop to process as much as one whole iMCU row */
160  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
161       yoffset++) {
162    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
163	 MCU_col_num++) {
164      /* Try to fetch an MCU.  Entropy decoder expects buffer to be zeroed. */
165      jzero_far((void FAR *) coef->MCU_buffer[0],
166		(size_t) (cinfo->blocks_in_MCU * SIZEOF(JBLOCK)));
167      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
168	/* Suspension forced; update state counters and exit */
169	coef->MCU_vert_offset = yoffset;
170	coef->MCU_ctr = MCU_col_num;
171	return JPEG_SUSPENDED;
172      }
173      /* Determine where data should go in output_buf and do the IDCT thing.
174       * We skip dummy blocks at the right and bottom edges (but blkn gets
175       * incremented past them!).  Note the inner loop relies on having
176       * allocated the MCU_buffer[] blocks sequentially.
177       */
178      blkn = 0;			/* index of current DCT block within MCU */
179      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
180	compptr = cinfo->cur_comp_info[ci];
181	/* Don't bother to IDCT an uninteresting component. */
182	if (! compptr->component_needed) {
183	  blkn += compptr->MCU_blocks;
184	  continue;
185	}
186	inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index];
187	useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
188						    : compptr->last_col_width;
189	output_ptr = output_buf[compptr->component_index] +
190	  yoffset * compptr->DCT_scaled_size;
191	start_col = MCU_col_num * compptr->MCU_sample_width;
192	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
193	  if (cinfo->input_iMCU_row < last_iMCU_row ||
194	      yoffset+yindex < compptr->last_row_height) {
195	    output_col = start_col;
196	    for (xindex = 0; xindex < useful_width; xindex++) {
197	      (*inverse_DCT) (cinfo, compptr,
198			      (JCOEFPTR) coef->MCU_buffer[blkn+xindex],
199			      output_ptr, output_col);
200	      output_col += compptr->DCT_scaled_size;
201	    }
202	  }
203	  blkn += compptr->MCU_width;
204	  output_ptr += compptr->DCT_scaled_size;
205	}
206      }
207    }
208    /* Completed an MCU row, but perhaps not an iMCU row */
209    coef->MCU_ctr = 0;
210  }
211  /* Completed the iMCU row, advance counters for next one */
212  cinfo->output_iMCU_row++;
213  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
214    start_iMCU_row(cinfo);
215    return JPEG_ROW_COMPLETED;
216  }
217  /* Completed the scan */
218  (*cinfo->inputctl->finish_input_pass) (cinfo);
219  return JPEG_SCAN_COMPLETED;
220}
221
222
223/*
224 * Dummy consume-input routine for single-pass operation.
225 */
226
227METHODDEF(int)
228dummy_consume_data (j_decompress_ptr cinfo)
229{
230  return JPEG_SUSPENDED;	/* Always indicate nothing was done */
231}
232
233
234#ifdef D_MULTISCAN_FILES_SUPPORTED
235
236/*
237 * Consume input data and store it in the full-image coefficient buffer.
238 * We read as much as one fully interleaved MCU row ("iMCU" row) per call,
239 * ie, v_samp_factor block rows for each component in the scan.
240 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
241 */
242
243METHODDEF(int)
244consume_data (j_decompress_ptr cinfo)
245{
246  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
247  JDIMENSION MCU_col_num;	/* index of current MCU within row */
248  int blkn, ci, xindex, yindex, yoffset;
249  JDIMENSION start_col;
250  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
251  JBLOCKROW buffer_ptr;
252  jpeg_component_info *compptr;
253
254  /* Align the virtual buffers for the components used in this scan. */
255  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
256    compptr = cinfo->cur_comp_info[ci];
257    buffer[ci] = (*cinfo->mem->access_virt_barray)
258      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
259       cinfo->input_iMCU_row * compptr->v_samp_factor,
260       (JDIMENSION) compptr->v_samp_factor, TRUE);
261    /* Note: entropy decoder expects buffer to be zeroed,
262     * but this is handled automatically by the memory manager
263     * because we requested a pre-zeroed array.
264     */
265  }
266  /* Loop to process one whole iMCU row */
267  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
268       yoffset++) {
269    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
270	 MCU_col_num++) {
271      /* Construct list of pointers to DCT blocks belonging to this MCU */
272      blkn = 0;			/* index of current DCT block within MCU */
273      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
274        compptr = cinfo->cur_comp_info[ci];
275        start_col = MCU_col_num * compptr->MCU_width;
276        for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
277          buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
278          for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
279            coef->MCU_buffer[blkn++] = buffer_ptr++;
280          }
281        }
282      }
283      /* Try to fetch the MCU. */
284      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
285	/* Suspension forced; update state counters and exit */
286	coef->MCU_vert_offset = yoffset;
287	coef->MCU_ctr = MCU_col_num;
288	return JPEG_SUSPENDED;
289      }
290    }
291    /* Completed an MCU row, but perhaps not an iMCU row */
292    coef->MCU_ctr = 0;
293  }
294  /* Completed the iMCU row, advance counters for next one */
295  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
296    start_iMCU_row(cinfo);
297    return JPEG_ROW_COMPLETED;
298  }
299  /* Completed the scan */
300  (*cinfo->inputctl->finish_input_pass) (cinfo);
301  return JPEG_SCAN_COMPLETED;
302}
303
304#define  rounded_division(A,B) ((A+B-1)/(B))
305/*
306 * Same as consume_data, expect for saving the Huffman decode information
307 * - bitstream offset and DC coefficient to index.
308 */
309
310METHODDEF(int)
311consume_data_with_huffman_index (j_decompress_ptr cinfo, huffman_index *index,
312        int current_scan)
313{
314  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
315  JDIMENSION MCU_col_num;	/* index of current MCU within row */
316  int ci, xindex, yindex, yoffset;
317  JDIMENSION start_col;
318  JBLOCKROW buffer_ptr;
319
320  huffman_scan_header current_header = index->scan[current_scan];
321  current_header.MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
322  current_header.MCUs_per_row = cinfo->MCUs_per_row;
323  current_header.comps_in_scan = cinfo->comps_in_scan;
324
325  size_t allocate_size = coef->MCU_rows_per_iMCU_row
326      * rounded_division(cinfo->MCUs_per_row, index->MCU_sample_size)
327      * sizeof(huffman_offset_data);
328  current_header.offset[cinfo->input_iMCU_row] = (huffman_offset_data*)malloc(allocate_size);
329  index->mem_used += allocate_size;
330
331  huffman_offset_data *offset_data = current_header.offset[cinfo->input_iMCU_row];
332
333  /* Loop to process one whole iMCU row */
334  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
335       yoffset++) {
336    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
337	 MCU_col_num++) {
338      // Record huffman bit offset
339      if (MCU_col_num % index->MCU_sample_size == 0) {
340        jpeg_get_huffman_decoder_configuration(cinfo,
341                &offset_data->bitstream_offset, offset_data->prev_dc);
342        ++offset_data;
343      }
344
345      /* Try to fetch the MCU. */
346      if (! (*cinfo->entropy->decode_mcu_discard_coef) (cinfo)) {
347        /* Suspension forced; update state counters and exit */
348        coef->MCU_vert_offset = yoffset;
349        coef->MCU_ctr = MCU_col_num;
350        return JPEG_SUSPENDED;
351      }
352    }
353    /* Completed an MCU row, but perhaps not an iMCU row */
354    coef->MCU_ctr = 0;
355  }
356  /* Completed the iMCU row, advance counters for next one */
357  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
358    start_iMCU_row(cinfo);
359    return JPEG_ROW_COMPLETED;
360  }
361  /* Completed the scan */
362  (*cinfo->inputctl->finish_input_pass) (cinfo);
363  return JPEG_SCAN_COMPLETED;
364}
365
366
367/*
368 * Decompress and return some data in the multi-pass case.
369 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
370 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
371 *
372 * NB: output_buf contains a plane for each component in image.
373 */
374
375METHODDEF(int)
376decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
377{
378  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
379  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
380  JDIMENSION block_num;
381  int ci, block_row, block_rows;
382  JBLOCKARRAY buffer;
383  JBLOCKROW buffer_ptr;
384  JSAMPARRAY output_ptr;
385  JDIMENSION output_col;
386  jpeg_component_info *compptr;
387  inverse_DCT_method_ptr inverse_DCT;
388
389  /* Force some input to be done if we are getting ahead of the input. */
390  while (cinfo->input_scan_number < cinfo->output_scan_number ||
391	 (cinfo->input_scan_number == cinfo->output_scan_number &&
392	  cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
393    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
394      return JPEG_SUSPENDED;
395  }
396
397  /* OK, output from the virtual arrays. */
398  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
399       ci++, compptr++) {
400    /* Don't bother to IDCT an uninteresting component. */
401    if (! compptr->component_needed)
402      continue;
403    /* Align the virtual buffer for this component. */
404    buffer = (*cinfo->mem->access_virt_barray)
405      ((j_common_ptr) cinfo, coef->whole_image[ci],
406       cinfo->output_iMCU_row * compptr->v_samp_factor,
407       (JDIMENSION) compptr->v_samp_factor, FALSE);
408    /* Count non-dummy DCT block rows in this iMCU row. */
409    if (cinfo->output_iMCU_row < last_iMCU_row)
410      block_rows = compptr->v_samp_factor;
411    else {
412      /* NB: can't use last_row_height here; it is input-side-dependent! */
413      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
414      if (block_rows == 0) block_rows = compptr->v_samp_factor;
415    }
416    inverse_DCT = cinfo->idct->inverse_DCT[ci];
417    output_ptr = output_buf[ci];
418    /* Loop over all DCT blocks to be processed. */
419    for (block_row = 0; block_row < block_rows; block_row++) {
420      buffer_ptr = buffer[block_row];
421      output_col = 0;
422      for (block_num = 0; block_num < compptr->width_in_blocks; block_num++) {
423	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
424			output_ptr, output_col);
425	buffer_ptr++;
426	output_col += compptr->DCT_scaled_size;
427      }
428      output_ptr += compptr->DCT_scaled_size;
429    }
430  }
431
432  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
433    return JPEG_ROW_COMPLETED;
434  return JPEG_SCAN_COMPLETED;
435}
436
437#endif /* D_MULTISCAN_FILES_SUPPORTED */
438
439
440#ifdef BLOCK_SMOOTHING_SUPPORTED
441
442/*
443 * This code applies interblock smoothing as described by section K.8
444 * of the JPEG standard: the first 5 AC coefficients are estimated from
445 * the DC values of a DCT block and its 8 neighboring blocks.
446 * We apply smoothing only for progressive JPEG decoding, and only if
447 * the coefficients it can estimate are not yet known to full precision.
448 */
449
450/* Natural-order array positions of the first 5 zigzag-order coefficients */
451#define Q01_POS  1
452#define Q10_POS  8
453#define Q20_POS  16
454#define Q11_POS  9
455#define Q02_POS  2
456
457/*
458 * Determine whether block smoothing is applicable and safe.
459 * We also latch the current states of the coef_bits[] entries for the
460 * AC coefficients; otherwise, if the input side of the decompressor
461 * advances into a new scan, we might think the coefficients are known
462 * more accurately than they really are.
463 */
464
465LOCAL(boolean)
466smoothing_ok (j_decompress_ptr cinfo)
467{
468  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
469  boolean smoothing_useful = FALSE;
470  int ci, coefi;
471  jpeg_component_info *compptr;
472  JQUANT_TBL * qtable;
473  int * coef_bits;
474  int * coef_bits_latch;
475
476  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
477    return FALSE;
478
479  /* Allocate latch area if not already done */
480  if (coef->coef_bits_latch == NULL)
481    coef->coef_bits_latch = (int *)
482      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
483				  cinfo->num_components *
484				  (SAVED_COEFS * SIZEOF(int)));
485  coef_bits_latch = coef->coef_bits_latch;
486
487  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
488       ci++, compptr++) {
489    /* All components' quantization values must already be latched. */
490    if ((qtable = compptr->quant_table) == NULL)
491      return FALSE;
492    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
493    if (qtable->quantval[0] == 0 ||
494	qtable->quantval[Q01_POS] == 0 ||
495	qtable->quantval[Q10_POS] == 0 ||
496	qtable->quantval[Q20_POS] == 0 ||
497	qtable->quantval[Q11_POS] == 0 ||
498	qtable->quantval[Q02_POS] == 0)
499      return FALSE;
500    /* DC values must be at least partly known for all components. */
501    coef_bits = cinfo->coef_bits[ci];
502    if (coef_bits[0] < 0)
503      return FALSE;
504    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
505    for (coefi = 1; coefi <= 5; coefi++) {
506      coef_bits_latch[coefi] = coef_bits[coefi];
507      if (coef_bits[coefi] != 0)
508	smoothing_useful = TRUE;
509    }
510    coef_bits_latch += SAVED_COEFS;
511  }
512
513  return smoothing_useful;
514}
515
516
517/*
518 * Variant of decompress_data for use when doing block smoothing.
519 */
520
521METHODDEF(int)
522decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
523{
524  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
525  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
526  JDIMENSION block_num, last_block_column;
527  int ci, block_row, block_rows, access_rows;
528  JBLOCKARRAY buffer;
529  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
530  JSAMPARRAY output_ptr;
531  JDIMENSION output_col;
532  jpeg_component_info *compptr;
533  inverse_DCT_method_ptr inverse_DCT;
534  boolean first_row, last_row;
535  JBLOCK workspace;
536  int *coef_bits;
537  JQUANT_TBL *quanttbl;
538  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
539  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
540  int Al, pred;
541
542  /* Force some input to be done if we are getting ahead of the input. */
543  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
544	 ! cinfo->inputctl->eoi_reached) {
545    if (cinfo->input_scan_number == cinfo->output_scan_number) {
546      /* If input is working on current scan, we ordinarily want it to
547       * have completed the current row.  But if input scan is DC,
548       * we want it to keep one row ahead so that next block row's DC
549       * values are up to date.
550       */
551      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
552      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
553	break;
554    }
555    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
556      return JPEG_SUSPENDED;
557  }
558
559  /* OK, output from the virtual arrays. */
560  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
561       ci++, compptr++) {
562    /* Don't bother to IDCT an uninteresting component. */
563    if (! compptr->component_needed)
564      continue;
565    /* Count non-dummy DCT block rows in this iMCU row. */
566    if (cinfo->output_iMCU_row < last_iMCU_row) {
567      block_rows = compptr->v_samp_factor;
568      access_rows = block_rows * 2; /* this and next iMCU row */
569      last_row = FALSE;
570    } else {
571      /* NB: can't use last_row_height here; it is input-side-dependent! */
572      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
573      if (block_rows == 0) block_rows = compptr->v_samp_factor;
574      access_rows = block_rows; /* this iMCU row only */
575      last_row = TRUE;
576    }
577    /* Align the virtual buffer for this component. */
578    if (cinfo->output_iMCU_row > 0) {
579      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
580      buffer = (*cinfo->mem->access_virt_barray)
581	((j_common_ptr) cinfo, coef->whole_image[ci],
582	 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
583	 (JDIMENSION) access_rows, FALSE);
584      buffer += compptr->v_samp_factor;	/* point to current iMCU row */
585      first_row = FALSE;
586    } else {
587      buffer = (*cinfo->mem->access_virt_barray)
588	((j_common_ptr) cinfo, coef->whole_image[ci],
589	 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
590      first_row = TRUE;
591    }
592    /* Fetch component-dependent info */
593    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
594    quanttbl = compptr->quant_table;
595    Q00 = quanttbl->quantval[0];
596    Q01 = quanttbl->quantval[Q01_POS];
597    Q10 = quanttbl->quantval[Q10_POS];
598    Q20 = quanttbl->quantval[Q20_POS];
599    Q11 = quanttbl->quantval[Q11_POS];
600    Q02 = quanttbl->quantval[Q02_POS];
601    inverse_DCT = cinfo->idct->inverse_DCT[ci];
602    output_ptr = output_buf[ci];
603    /* Loop over all DCT blocks to be processed. */
604    for (block_row = 0; block_row < block_rows; block_row++) {
605      buffer_ptr = buffer[block_row];
606      if (first_row && block_row == 0)
607	prev_block_row = buffer_ptr;
608      else
609	prev_block_row = buffer[block_row-1];
610      if (last_row && block_row == block_rows-1)
611	next_block_row = buffer_ptr;
612      else
613	next_block_row = buffer[block_row+1];
614      /* We fetch the surrounding DC values using a sliding-register approach.
615       * Initialize all nine here so as to do the right thing on narrow pics.
616       */
617      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
618      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
619      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
620      output_col = 0;
621      last_block_column = compptr->width_in_blocks - 1;
622      for (block_num = 0; block_num <= last_block_column; block_num++) {
623	/* Fetch current DCT block into workspace so we can modify it. */
624	jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
625	/* Update DC values */
626	if (block_num < last_block_column) {
627	  DC3 = (int) prev_block_row[1][0];
628	  DC6 = (int) buffer_ptr[1][0];
629	  DC9 = (int) next_block_row[1][0];
630	}
631	/* Compute coefficient estimates per K.8.
632	 * An estimate is applied only if coefficient is still zero,
633	 * and is not known to be fully accurate.
634	 */
635	/* AC01 */
636	if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
637	  num = 36 * Q00 * (DC4 - DC6);
638	  if (num >= 0) {
639	    pred = (int) (((Q01<<7) + num) / (Q01<<8));
640	    if (Al > 0 && pred >= (1<<Al))
641	      pred = (1<<Al)-1;
642	  } else {
643	    pred = (int) (((Q01<<7) - num) / (Q01<<8));
644	    if (Al > 0 && pred >= (1<<Al))
645	      pred = (1<<Al)-1;
646	    pred = -pred;
647	  }
648	  workspace[1] = (JCOEF) pred;
649	}
650	/* AC10 */
651	if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
652	  num = 36 * Q00 * (DC2 - DC8);
653	  if (num >= 0) {
654	    pred = (int) (((Q10<<7) + num) / (Q10<<8));
655	    if (Al > 0 && pred >= (1<<Al))
656	      pred = (1<<Al)-1;
657	  } else {
658	    pred = (int) (((Q10<<7) - num) / (Q10<<8));
659	    if (Al > 0 && pred >= (1<<Al))
660	      pred = (1<<Al)-1;
661	    pred = -pred;
662	  }
663	  workspace[8] = (JCOEF) pred;
664	}
665	/* AC20 */
666	if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
667	  num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
668	  if (num >= 0) {
669	    pred = (int) (((Q20<<7) + num) / (Q20<<8));
670	    if (Al > 0 && pred >= (1<<Al))
671	      pred = (1<<Al)-1;
672	  } else {
673	    pred = (int) (((Q20<<7) - num) / (Q20<<8));
674	    if (Al > 0 && pred >= (1<<Al))
675	      pred = (1<<Al)-1;
676	    pred = -pred;
677	  }
678	  workspace[16] = (JCOEF) pred;
679	}
680	/* AC11 */
681	if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
682	  num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
683	  if (num >= 0) {
684	    pred = (int) (((Q11<<7) + num) / (Q11<<8));
685	    if (Al > 0 && pred >= (1<<Al))
686	      pred = (1<<Al)-1;
687	  } else {
688	    pred = (int) (((Q11<<7) - num) / (Q11<<8));
689	    if (Al > 0 && pred >= (1<<Al))
690	      pred = (1<<Al)-1;
691	    pred = -pred;
692	  }
693	  workspace[9] = (JCOEF) pred;
694	}
695	/* AC02 */
696	if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
697	  num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
698	  if (num >= 0) {
699	    pred = (int) (((Q02<<7) + num) / (Q02<<8));
700	    if (Al > 0 && pred >= (1<<Al))
701	      pred = (1<<Al)-1;
702	  } else {
703	    pred = (int) (((Q02<<7) - num) / (Q02<<8));
704	    if (Al > 0 && pred >= (1<<Al))
705	      pred = (1<<Al)-1;
706	    pred = -pred;
707	  }
708	  workspace[2] = (JCOEF) pred;
709	}
710	/* OK, do the IDCT */
711	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
712			output_ptr, output_col);
713	/* Advance for next column */
714	DC1 = DC2; DC2 = DC3;
715	DC4 = DC5; DC5 = DC6;
716	DC7 = DC8; DC8 = DC9;
717	buffer_ptr++, prev_block_row++, next_block_row++;
718	output_col += compptr->DCT_scaled_size;
719      }
720      output_ptr += compptr->DCT_scaled_size;
721    }
722  }
723
724  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
725    return JPEG_ROW_COMPLETED;
726  return JPEG_SCAN_COMPLETED;
727}
728
729#endif /* BLOCK_SMOOTHING_SUPPORTED */
730
731
732/*
733 * Initialize coefficient buffer controller.
734 */
735
736GLOBAL(void)
737jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
738{
739  my_coef_ptr coef;
740
741  coef = (my_coef_ptr)
742    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
743				SIZEOF(my_coef_controller));
744  cinfo->coef = (struct jpeg_d_coef_controller *) coef;
745  coef->pub.start_input_pass = start_input_pass;
746  coef->pub.start_output_pass = start_output_pass;
747#ifdef BLOCK_SMOOTHING_SUPPORTED
748  coef->coef_bits_latch = NULL;
749#endif
750
751  /* Create the coefficient buffer. */
752  if (need_full_buffer) {
753#ifdef D_MULTISCAN_FILES_SUPPORTED
754    /* Allocate a full-image virtual array for each component, */
755    /* padded to a multiple of samp_factor DCT blocks in each direction. */
756    /* Note we ask for a pre-zeroed array. */
757    int ci, access_rows;
758    jpeg_component_info *compptr;
759
760    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
761	 ci++, compptr++) {
762      access_rows = compptr->v_samp_factor;
763#ifdef BLOCK_SMOOTHING_SUPPORTED
764      /* If block smoothing could be used, need a bigger window */
765      if (cinfo->progressive_mode)
766	access_rows *= 3;
767#endif
768      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
769	((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
770	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
771				(long) compptr->h_samp_factor),
772	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
773				(long) compptr->v_samp_factor),
774	 (JDIMENSION) access_rows);
775    }
776    coef->pub.consume_data_with_huffman_index = consume_data_with_huffman_index;
777    coef->pub.consume_data = consume_data;
778    coef->pub.decompress_data = decompress_data;
779    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
780#else
781    ERREXIT(cinfo, JERR_NOT_COMPILED);
782#endif
783  } else {
784    /* We only need a single-MCU buffer. */
785    JBLOCKROW buffer;
786    int i;
787
788    buffer = (JBLOCKROW)
789      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
790				  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
791    for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
792      coef->MCU_buffer[i] = buffer + i;
793    }
794    coef->pub.consume_data_with_huffman_index = consume_data_with_huffman_index;
795    coef->pub.consume_data = dummy_consume_data;
796    coef->pub.decompress_data = decompress_onepass;
797    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
798  }
799}
800