1#if !defined(_FX_JPEG_TURBO_)
2/*
3 * jdcoefct.c
4 *
5 * Copyright (C) 1994-1997, Thomas G. Lane.
6 * This file is part of the Independent JPEG Group's software.
7 * For conditions of distribution and use, see the accompanying README file.
8 *
9 * This file contains the coefficient buffer controller for decompression.
10 * This controller is the top level of the JPEG decompressor proper.
11 * The coefficient buffer lies between entropy decoding and inverse-DCT steps.
12 *
13 * In buffered-image mode, this controller is the interface between
14 * input-oriented processing and output-oriented processing.
15 * Also, the input side (only) is used when reading a file for transcoding.
16 */
17
18#define JPEG_INTERNALS
19#include "jinclude.h"
20#include "jpeglib.h"
21
22/* Block smoothing is only applicable for progressive JPEG, so: */
23#ifndef D_PROGRESSIVE_SUPPORTED
24#undef BLOCK_SMOOTHING_SUPPORTED
25#endif
26
27/* Private buffer controller object */
28
29typedef struct {
30  struct jpeg_d_coef_controller pub; /* public fields */
31
32  /* These variables keep track of the current location of the input side. */
33  /* cinfo->input_iMCU_row is also used for this. */
34  JDIMENSION MCU_ctr;		/* counts MCUs processed in current row */
35  int MCU_vert_offset;		/* counts MCU rows within iMCU row */
36  int MCU_rows_per_iMCU_row;	/* number of such rows needed */
37
38  /* The output side's location is represented by cinfo->output_iMCU_row. */
39
40  /* In single-pass modes, it's sufficient to buffer just one MCU.
41   * We allocate a workspace of D_MAX_BLOCKS_IN_MCU coefficient blocks,
42   * and let the entropy decoder write into that workspace each time.
43   * (On 80x86, the workspace is FAR even though it's not really very big;
44   * this is to keep the module interfaces unchanged when a large coefficient
45   * buffer is necessary.)
46   * In multi-pass modes, this array points to the current MCU's blocks
47   * within the virtual arrays; it is used only by the input side.
48   */
49  JBLOCKROW MCU_buffer[D_MAX_BLOCKS_IN_MCU];
50
51#ifdef D_MULTISCAN_FILES_SUPPORTED
52  /* In multi-pass modes, we need a virtual block array for each component. */
53  jvirt_barray_ptr whole_image[MAX_COMPONENTS];
54#endif
55
56#ifdef BLOCK_SMOOTHING_SUPPORTED
57  /* When doing block smoothing, we latch coefficient Al values here */
58  int * coef_bits_latch;
59#define SAVED_COEFS  6		/* we save coef_bits[0..5] */
60#endif
61} my_coef_controller;
62
63typedef my_coef_controller * my_coef_ptr;
64
65/* Forward declarations */
66METHODDEF(int) decompress_onepass
67	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
68#ifdef D_MULTISCAN_FILES_SUPPORTED
69METHODDEF(int) decompress_data
70	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
71#endif
72#ifdef BLOCK_SMOOTHING_SUPPORTED
73LOCAL(boolean) smoothing_ok JPP((j_decompress_ptr cinfo));
74METHODDEF(int) decompress_smooth_data
75	JPP((j_decompress_ptr cinfo, JSAMPIMAGE output_buf));
76#endif
77
78
79LOCAL(void)
80start_iMCU_row (j_decompress_ptr cinfo)
81/* Reset within-iMCU-row counters for a new row (input side) */
82{
83  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
84
85  /* In an interleaved scan, an MCU row is the same as an iMCU row.
86   * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows.
87   * But at the bottom of the image, process only what's left.
88   */
89  if (cinfo->comps_in_scan > 1) {
90    coef->MCU_rows_per_iMCU_row = 1;
91  } else {
92    if (cinfo->input_iMCU_row < (cinfo->total_iMCU_rows-1))
93      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor;
94    else
95      coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height;
96  }
97
98  coef->MCU_ctr = 0;
99  coef->MCU_vert_offset = 0;
100}
101
102
103/*
104 * Initialize for an input processing pass.
105 */
106
107METHODDEF(void)
108start_input_pass (j_decompress_ptr cinfo)
109{
110  cinfo->input_iMCU_row = 0;
111  start_iMCU_row(cinfo);
112}
113
114
115/*
116 * Initialize for an output processing pass.
117 */
118
119METHODDEF(void)
120start_output_pass (j_decompress_ptr cinfo)
121{
122#ifdef BLOCK_SMOOTHING_SUPPORTED
123  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
124
125  /* If multipass, check to see whether to use block smoothing on this pass */
126  if (coef->pub.coef_arrays != NULL) {
127    if (cinfo->do_block_smoothing && smoothing_ok(cinfo))
128      coef->pub.decompress_data = decompress_smooth_data;
129    else
130      coef->pub.decompress_data = decompress_data;
131  }
132#endif
133  cinfo->output_iMCU_row = 0;
134}
135
136
137/*
138 * Decompress and return some data in the single-pass case.
139 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
140 * Input and output must run in lockstep since we have only a one-MCU buffer.
141 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
142 *
143 * NB: output_buf contains a plane for each component in image,
144 * which we index according to the component's SOF position.
145 */
146
147METHODDEF(int)
148decompress_onepass (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
149{
150  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
151  JDIMENSION MCU_col_num;	/* index of current MCU within row */
152  JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
153  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
154  int blkn, ci, xindex, yindex, yoffset, useful_width;
155  JSAMPARRAY output_ptr;
156  JDIMENSION start_col, output_col;
157  jpeg_component_info *compptr;
158  inverse_DCT_method_ptr inverse_DCT;
159
160  /* Loop to process as much as one whole iMCU row */
161  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
162       yoffset++) {
163    for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
164	 MCU_col_num++) {
165      /* Try to fetch an MCU.  Entropy decoder expects buffer to be zeroed. */
166      jzero_far((void FAR *) coef->MCU_buffer[0],
167		(size_t) (cinfo->blocks_in_MCU * SIZEOF(JBLOCK)));
168      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
169	/* Suspension forced; update state counters and exit */
170	coef->MCU_vert_offset = yoffset;
171	coef->MCU_ctr = MCU_col_num;
172	return JPEG_SUSPENDED;
173      }
174      /* Determine where data should go in output_buf and do the IDCT thing.
175       * We skip dummy blocks at the right and bottom edges (but blkn gets
176       * incremented past them!).  Note the inner loop relies on having
177       * allocated the MCU_buffer[] blocks sequentially.
178       */
179      blkn = 0;			/* index of current DCT block within MCU */
180      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
181	compptr = cinfo->cur_comp_info[ci];
182	/* Don't bother to IDCT an uninteresting component. */
183	if (! compptr->component_needed) {
184	  blkn += compptr->MCU_blocks;
185	  continue;
186	}
187	inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index];
188	useful_width = (MCU_col_num < last_MCU_col) ? compptr->MCU_width
189						    : compptr->last_col_width;
190	output_ptr = output_buf[compptr->component_index] +
191	  yoffset * compptr->DCT_scaled_size;
192	start_col = MCU_col_num * compptr->MCU_sample_width;
193	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
194	  if (cinfo->input_iMCU_row < last_iMCU_row ||
195	      yoffset+yindex < compptr->last_row_height) {
196	    output_col = start_col;
197	    for (xindex = 0; xindex < useful_width; xindex++) {
198	      (*inverse_DCT) (cinfo, compptr,
199			      (JCOEFPTR) coef->MCU_buffer[blkn+xindex],
200			      output_ptr, output_col);
201	      output_col += compptr->DCT_scaled_size;
202	    }
203	  }
204	  blkn += compptr->MCU_width;
205	  output_ptr += compptr->DCT_scaled_size;
206	}
207      }
208    }
209    /* Completed an MCU row, but perhaps not an iMCU row */
210    coef->MCU_ctr = 0;
211  }
212  /* Completed the iMCU row, advance counters for next one */
213  cinfo->output_iMCU_row++;
214  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
215    start_iMCU_row(cinfo);
216    return JPEG_ROW_COMPLETED;
217  }
218  /* Completed the scan */
219  (*cinfo->inputctl->finish_input_pass) (cinfo);
220  return JPEG_SCAN_COMPLETED;
221}
222
223
224/*
225 * Dummy consume-input routine for single-pass operation.
226 */
227
228METHODDEF(int)
229dummy_consume_data (j_decompress_ptr cinfo)
230{
231  return JPEG_SUSPENDED;	/* Always indicate nothing was done */
232}
233
234
235#ifdef D_MULTISCAN_FILES_SUPPORTED
236
237/*
238 * Consume input data and store it in the full-image coefficient buffer.
239 * We read as much as one fully interleaved MCU row ("iMCU" row) per call,
240 * ie, v_samp_factor block rows for each component in the scan.
241 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
242 */
243
244METHODDEF(int)
245consume_data (j_decompress_ptr cinfo)
246{
247  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
248  JDIMENSION MCU_col_num;	/* index of current MCU within row */
249  int blkn, ci, xindex, yindex, yoffset;
250  JDIMENSION start_col;
251  JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
252  JBLOCKROW buffer_ptr;
253  jpeg_component_info *compptr;
254
255  /* Align the virtual buffers for the components used in this scan. */
256  for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
257    compptr = cinfo->cur_comp_info[ci];
258    buffer[ci] = (*cinfo->mem->access_virt_barray)
259      ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
260       cinfo->input_iMCU_row * compptr->v_samp_factor,
261       (JDIMENSION) compptr->v_samp_factor, TRUE);
262    /* Note: entropy decoder expects buffer to be zeroed,
263     * but this is handled automatically by the memory manager
264     * because we requested a pre-zeroed array.
265     */
266  }
267
268  /* Loop to process one whole iMCU row */
269  for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
270       yoffset++) {
271    for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
272	 MCU_col_num++) {
273      /* Construct list of pointers to DCT blocks belonging to this MCU */
274      blkn = 0;			/* index of current DCT block within MCU */
275      for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
276	compptr = cinfo->cur_comp_info[ci];
277	start_col = MCU_col_num * compptr->MCU_width;
278	for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
279	  buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
280	  for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
281	    coef->MCU_buffer[blkn++] = buffer_ptr++;
282	  }
283	}
284      }
285      /* Try to fetch the MCU. */
286      if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
287	/* Suspension forced; update state counters and exit */
288	coef->MCU_vert_offset = yoffset;
289	coef->MCU_ctr = MCU_col_num;
290	return JPEG_SUSPENDED;
291      }
292    }
293    /* Completed an MCU row, but perhaps not an iMCU row */
294    coef->MCU_ctr = 0;
295  }
296  /* Completed the iMCU row, advance counters for next one */
297  if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
298    start_iMCU_row(cinfo);
299    return JPEG_ROW_COMPLETED;
300  }
301  /* Completed the scan */
302  (*cinfo->inputctl->finish_input_pass) (cinfo);
303  return JPEG_SCAN_COMPLETED;
304}
305
306
307/*
308 * Decompress and return some data in the multi-pass case.
309 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
310 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
311 *
312 * NB: output_buf contains a plane for each component in image.
313 */
314
315METHODDEF(int)
316decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
317{
318  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
319  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
320  JDIMENSION block_num;
321  int ci, block_row, block_rows;
322  JBLOCKARRAY buffer;
323  JBLOCKROW buffer_ptr;
324  JSAMPARRAY output_ptr;
325  JDIMENSION output_col;
326  jpeg_component_info *compptr;
327  inverse_DCT_method_ptr inverse_DCT;
328
329  /* Force some input to be done if we are getting ahead of the input. */
330  while (cinfo->input_scan_number < cinfo->output_scan_number ||
331	 (cinfo->input_scan_number == cinfo->output_scan_number &&
332	  cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
333    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
334      return JPEG_SUSPENDED;
335  }
336
337  /* OK, output from the virtual arrays. */
338  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
339       ci++, compptr++) {
340    /* Don't bother to IDCT an uninteresting component. */
341    if (! compptr->component_needed)
342      continue;
343    /* Align the virtual buffer for this component. */
344    buffer = (*cinfo->mem->access_virt_barray)
345      ((j_common_ptr) cinfo, coef->whole_image[ci],
346       cinfo->output_iMCU_row * compptr->v_samp_factor,
347       (JDIMENSION) compptr->v_samp_factor, FALSE);
348    /* Count non-dummy DCT block rows in this iMCU row. */
349    if (cinfo->output_iMCU_row < last_iMCU_row)
350      block_rows = compptr->v_samp_factor;
351    else {
352      /* NB: can't use last_row_height here; it is input-side-dependent! */
353      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
354      if (block_rows == 0) block_rows = compptr->v_samp_factor;
355    }
356    inverse_DCT = cinfo->idct->inverse_DCT[ci];
357    output_ptr = output_buf[ci];
358    /* Loop over all DCT blocks to be processed. */
359    for (block_row = 0; block_row < block_rows; block_row++) {
360      buffer_ptr = buffer[block_row];
361      output_col = 0;
362      for (block_num = 0; block_num < compptr->width_in_blocks; block_num++) {
363	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
364			output_ptr, output_col);
365	buffer_ptr++;
366	output_col += compptr->DCT_scaled_size;
367      }
368      output_ptr += compptr->DCT_scaled_size;
369    }
370  }
371
372  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
373    return JPEG_ROW_COMPLETED;
374  return JPEG_SCAN_COMPLETED;
375}
376
377#endif /* D_MULTISCAN_FILES_SUPPORTED */
378
379
380#ifdef BLOCK_SMOOTHING_SUPPORTED
381
382/*
383 * This code applies interblock smoothing as described by section K.8
384 * of the JPEG standard: the first 5 AC coefficients are estimated from
385 * the DC values of a DCT block and its 8 neighboring blocks.
386 * We apply smoothing only for progressive JPEG decoding, and only if
387 * the coefficients it can estimate are not yet known to full precision.
388 */
389
390/* Natural-order array positions of the first 5 zigzag-order coefficients */
391#define Q01_POS  1
392#define Q10_POS  8
393#define Q20_POS  16
394#define Q11_POS  9
395#define Q02_POS  2
396
397/*
398 * Determine whether block smoothing is applicable and safe.
399 * We also latch the current states of the coef_bits[] entries for the
400 * AC coefficients; otherwise, if the input side of the decompressor
401 * advances into a new scan, we might think the coefficients are known
402 * more accurately than they really are.
403 */
404
405LOCAL(boolean)
406smoothing_ok (j_decompress_ptr cinfo)
407{
408  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
409  boolean smoothing_useful = FALSE;
410  int ci, coefi;
411  jpeg_component_info *compptr;
412  JQUANT_TBL * qtable;
413  int * coef_bits;
414  int * coef_bits_latch;
415
416  if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
417    return FALSE;
418
419  /* Allocate latch area if not already done */
420  if (coef->coef_bits_latch == NULL)
421    coef->coef_bits_latch = (int *)
422      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
423				  cinfo->num_components *
424				  (SAVED_COEFS * SIZEOF(int)));
425  coef_bits_latch = coef->coef_bits_latch;
426
427  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
428       ci++, compptr++) {
429    /* All components' quantization values must already be latched. */
430    if ((qtable = compptr->quant_table) == NULL)
431      return FALSE;
432    /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
433    if (qtable->quantval[0] == 0 ||
434	qtable->quantval[Q01_POS] == 0 ||
435	qtable->quantval[Q10_POS] == 0 ||
436	qtable->quantval[Q20_POS] == 0 ||
437	qtable->quantval[Q11_POS] == 0 ||
438	qtable->quantval[Q02_POS] == 0)
439      return FALSE;
440    /* DC values must be at least partly known for all components. */
441    coef_bits = cinfo->coef_bits[ci];
442    if (coef_bits[0] < 0)
443      return FALSE;
444    /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
445    for (coefi = 1; coefi <= 5; coefi++) {
446      coef_bits_latch[coefi] = coef_bits[coefi];
447      if (coef_bits[coefi] != 0)
448	smoothing_useful = TRUE;
449    }
450    coef_bits_latch += SAVED_COEFS;
451  }
452
453  return smoothing_useful;
454}
455
456
457/*
458 * Variant of decompress_data for use when doing block smoothing.
459 */
460
461METHODDEF(int)
462decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
463{
464  my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
465  JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
466  JDIMENSION block_num, last_block_column;
467  int ci, block_row, block_rows, access_rows;
468  JBLOCKARRAY buffer;
469  JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
470  JSAMPARRAY output_ptr;
471  JDIMENSION output_col;
472  jpeg_component_info *compptr;
473  inverse_DCT_method_ptr inverse_DCT;
474  boolean first_row, last_row;
475  JBLOCK workspace;
476  int *coef_bits;
477  JQUANT_TBL *quanttbl;
478  INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
479  int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
480  int Al, pred;
481
482  /* Force some input to be done if we are getting ahead of the input. */
483  while (cinfo->input_scan_number <= cinfo->output_scan_number &&
484	 ! cinfo->inputctl->eoi_reached) {
485    if (cinfo->input_scan_number == cinfo->output_scan_number) {
486      /* If input is working on current scan, we ordinarily want it to
487       * have completed the current row.  But if input scan is DC,
488       * we want it to keep one row ahead so that next block row's DC
489       * values are up to date.
490       */
491      JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
492      if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
493	break;
494    }
495    if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
496      return JPEG_SUSPENDED;
497  }
498
499  /* OK, output from the virtual arrays. */
500  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
501       ci++, compptr++) {
502    /* Don't bother to IDCT an uninteresting component. */
503    if (! compptr->component_needed)
504      continue;
505    /* Count non-dummy DCT block rows in this iMCU row. */
506    if (cinfo->output_iMCU_row < last_iMCU_row) {
507      block_rows = compptr->v_samp_factor;
508      access_rows = block_rows * 2; /* this and next iMCU row */
509      last_row = FALSE;
510    } else {
511      /* NB: can't use last_row_height here; it is input-side-dependent! */
512      block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
513      if (block_rows == 0) block_rows = compptr->v_samp_factor;
514      access_rows = block_rows; /* this iMCU row only */
515      last_row = TRUE;
516    }
517    /* Align the virtual buffer for this component. */
518    if (cinfo->output_iMCU_row > 0) {
519      access_rows += compptr->v_samp_factor; /* prior iMCU row too */
520      buffer = (*cinfo->mem->access_virt_barray)
521	((j_common_ptr) cinfo, coef->whole_image[ci],
522	 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
523	 (JDIMENSION) access_rows, FALSE);
524      buffer += compptr->v_samp_factor;	/* point to current iMCU row */
525      first_row = FALSE;
526    } else {
527      buffer = (*cinfo->mem->access_virt_barray)
528	((j_common_ptr) cinfo, coef->whole_image[ci],
529	 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
530      first_row = TRUE;
531    }
532    /* Fetch component-dependent info */
533    coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
534    quanttbl = compptr->quant_table;
535    Q00 = quanttbl->quantval[0];
536    Q01 = quanttbl->quantval[Q01_POS];
537    Q10 = quanttbl->quantval[Q10_POS];
538    Q20 = quanttbl->quantval[Q20_POS];
539    Q11 = quanttbl->quantval[Q11_POS];
540    Q02 = quanttbl->quantval[Q02_POS];
541    inverse_DCT = cinfo->idct->inverse_DCT[ci];
542    output_ptr = output_buf[ci];
543    /* Loop over all DCT blocks to be processed. */
544    for (block_row = 0; block_row < block_rows; block_row++) {
545      buffer_ptr = buffer[block_row];
546      if (first_row && block_row == 0)
547	prev_block_row = buffer_ptr;
548      else
549	prev_block_row = buffer[block_row-1];
550      if (last_row && block_row == block_rows-1)
551	next_block_row = buffer_ptr;
552      else
553	next_block_row = buffer[block_row+1];
554      /* We fetch the surrounding DC values using a sliding-register approach.
555       * Initialize all nine here so as to do the right thing on narrow pics.
556       */
557      DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
558      DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
559      DC7 = DC8 = DC9 = (int) next_block_row[0][0];
560      output_col = 0;
561      last_block_column = compptr->width_in_blocks - 1;
562      for (block_num = 0; block_num <= last_block_column; block_num++) {
563	/* Fetch current DCT block into workspace so we can modify it. */
564	jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
565	/* Update DC values */
566	if (block_num < last_block_column) {
567	  DC3 = (int) prev_block_row[1][0];
568	  DC6 = (int) buffer_ptr[1][0];
569	  DC9 = (int) next_block_row[1][0];
570	}
571	/* Compute coefficient estimates per K.8.
572	 * An estimate is applied only if coefficient is still zero,
573	 * and is not known to be fully accurate.
574	 */
575	/* AC01 */
576	if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
577	  num = 36 * Q00 * (DC4 - DC6);
578	  if (num >= 0) {
579	    pred = (int) (((Q01<<7) + num) / (Q01<<8));
580	    if (Al > 0 && pred >= (1<<Al))
581	      pred = (1<<Al)-1;
582	  } else {
583	    pred = (int) (((Q01<<7) - num) / (Q01<<8));
584	    if (Al > 0 && pred >= (1<<Al))
585	      pred = (1<<Al)-1;
586	    pred = -pred;
587	  }
588	  workspace[1] = (JCOEF) pred;
589	}
590	/* AC10 */
591	if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
592	  num = 36 * Q00 * (DC2 - DC8);
593	  if (num >= 0) {
594	    pred = (int) (((Q10<<7) + num) / (Q10<<8));
595	    if (Al > 0 && pred >= (1<<Al))
596	      pred = (1<<Al)-1;
597	  } else {
598	    pred = (int) (((Q10<<7) - num) / (Q10<<8));
599	    if (Al > 0 && pred >= (1<<Al))
600	      pred = (1<<Al)-1;
601	    pred = -pred;
602	  }
603	  workspace[8] = (JCOEF) pred;
604	}
605	/* AC20 */
606	if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
607	  num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
608	  if (num >= 0) {
609	    pred = (int) (((Q20<<7) + num) / (Q20<<8));
610	    if (Al > 0 && pred >= (1<<Al))
611	      pred = (1<<Al)-1;
612	  } else {
613	    pred = (int) (((Q20<<7) - num) / (Q20<<8));
614	    if (Al > 0 && pred >= (1<<Al))
615	      pred = (1<<Al)-1;
616	    pred = -pred;
617	  }
618	  workspace[16] = (JCOEF) pred;
619	}
620	/* AC11 */
621	if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
622	  num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
623	  if (num >= 0) {
624	    pred = (int) (((Q11<<7) + num) / (Q11<<8));
625	    if (Al > 0 && pred >= (1<<Al))
626	      pred = (1<<Al)-1;
627	  } else {
628	    pred = (int) (((Q11<<7) - num) / (Q11<<8));
629	    if (Al > 0 && pred >= (1<<Al))
630	      pred = (1<<Al)-1;
631	    pred = -pred;
632	  }
633	  workspace[9] = (JCOEF) pred;
634	}
635	/* AC02 */
636	if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
637	  num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
638	  if (num >= 0) {
639	    pred = (int) (((Q02<<7) + num) / (Q02<<8));
640	    if (Al > 0 && pred >= (1<<Al))
641	      pred = (1<<Al)-1;
642	  } else {
643	    pred = (int) (((Q02<<7) - num) / (Q02<<8));
644	    if (Al > 0 && pred >= (1<<Al))
645	      pred = (1<<Al)-1;
646	    pred = -pred;
647	  }
648	  workspace[2] = (JCOEF) pred;
649	}
650	/* OK, do the IDCT */
651	(*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
652			output_ptr, output_col);
653	/* Advance for next column */
654	DC1 = DC2; DC2 = DC3;
655	DC4 = DC5; DC5 = DC6;
656	DC7 = DC8; DC8 = DC9;
657	buffer_ptr++, prev_block_row++, next_block_row++;
658	output_col += compptr->DCT_scaled_size;
659      }
660      output_ptr += compptr->DCT_scaled_size;
661    }
662  }
663
664  if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
665    return JPEG_ROW_COMPLETED;
666  return JPEG_SCAN_COMPLETED;
667}
668
669#endif /* BLOCK_SMOOTHING_SUPPORTED */
670
671
672/*
673 * Initialize coefficient buffer controller.
674 */
675
676GLOBAL(void)
677jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
678{
679  my_coef_ptr coef;
680
681  coef = (my_coef_ptr)
682    (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
683				SIZEOF(my_coef_controller));
684  cinfo->coef = (struct jpeg_d_coef_controller *) coef;
685  coef->pub.start_input_pass = start_input_pass;
686  coef->pub.start_output_pass = start_output_pass;
687#ifdef BLOCK_SMOOTHING_SUPPORTED
688  coef->coef_bits_latch = NULL;
689#endif
690
691  /* Create the coefficient buffer. */
692  if (need_full_buffer) {
693#ifdef D_MULTISCAN_FILES_SUPPORTED
694    /* Allocate a full-image virtual array for each component, */
695    /* padded to a multiple of samp_factor DCT blocks in each direction. */
696    /* Note we ask for a pre-zeroed array. */
697    int ci, access_rows;
698    jpeg_component_info *compptr;
699
700    for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
701	 ci++, compptr++) {
702      access_rows = compptr->v_samp_factor;
703#ifdef BLOCK_SMOOTHING_SUPPORTED
704      /* If block smoothing could be used, need a bigger window */
705      if (cinfo->progressive_mode)
706	access_rows *= 3;
707#endif
708      coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
709	((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
710	 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
711				(long) compptr->h_samp_factor),
712	 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
713				(long) compptr->v_samp_factor),
714	 (JDIMENSION) access_rows);
715    }
716    coef->pub.consume_data = consume_data;
717    coef->pub.decompress_data = decompress_data;
718    coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
719#else
720    ERREXIT(cinfo, JERR_NOT_COMPILED);
721#endif
722  } else {
723    /* We only need a single-MCU buffer. */
724    JBLOCKROW buffer;
725    int i;
726
727    buffer = (JBLOCKROW)
728      (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
729				  D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
730    for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
731      coef->MCU_buffer[i] = buffer + i;
732    }
733    coef->pub.consume_data = dummy_consume_data;
734    coef->pub.decompress_data = decompress_onepass;
735    coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
736  }
737}
738
739#endif //_FX_JPEG_TURBO_
740