4 * This file was part of the Independent JPEG Group's software:
5 * Copyright (C) 1994-1997, Thomas G. Lane.
6 * libjpeg-turbo Modifications:
7 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
8 * Copyright (C) 2010, 2015-2016, 2019-2020, 2022-2023, D. R. Commander.
9 * Copyright (C) 2015, 2020, Google, Inc.
10 * For conditions of distribution and use, see the accompanying README.ijg
13 * This file contains the coefficient buffer controller for decompression.
14 * This controller is the top level of the lossy JPEG decompressor proper.
15 * The coefficient buffer lies between entropy decoding and inverse-DCT steps.
17 * In buffered-image mode, this controller is the interface between
18 * input-oriented processing and output-oriented processing.
19 * Also, the input side (only) is used when reading a file for transcoding.
24 #include "jpegapicomp.h"
25 #include "jsamplecomp.h"
28 /* Forward declarations */
29 METHODDEF(int) decompress_onepass(j_decompress_ptr cinfo,
30 _JSAMPIMAGE output_buf);
31 #ifdef D_MULTISCAN_FILES_SUPPORTED
32 METHODDEF(int) decompress_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf);
34 #ifdef BLOCK_SMOOTHING_SUPPORTED
35 LOCAL(boolean) smoothing_ok(j_decompress_ptr cinfo);
36 METHODDEF(int) decompress_smooth_data(j_decompress_ptr cinfo,
37 _JSAMPIMAGE output_buf);
42 * Initialize for an input processing pass.
46 start_input_pass(j_decompress_ptr cinfo)
48 cinfo->input_iMCU_row = 0;
49 start_iMCU_row(cinfo);
54 * Initialize for an output processing pass.
58 start_output_pass(j_decompress_ptr cinfo)
60 #ifdef BLOCK_SMOOTHING_SUPPORTED
61 my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
63 /* If multipass, check to see whether to use block smoothing on this pass */
64 if (coef->pub.coef_arrays != NULL) {
65 if (cinfo->do_block_smoothing && smoothing_ok(cinfo))
66 coef->pub._decompress_data = decompress_smooth_data;
68 coef->pub._decompress_data = decompress_data;
71 cinfo->output_iMCU_row = 0;
76 * Decompress and return some data in the single-pass case.
77 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
78 * Input and output must run in lockstep since we have only a one-MCU buffer.
79 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
81 * NB: output_buf contains a plane for each component in image,
82 * which we index according to the component's SOF position.
86 decompress_onepass(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf)
88 my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
89 JDIMENSION MCU_col_num; /* index of current MCU within row */
90 JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1;
91 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
92 int blkn, ci, xindex, yindex, yoffset, useful_width;
93 _JSAMPARRAY output_ptr;
94 JDIMENSION start_col, output_col;
95 jpeg_component_info *compptr;
96 _inverse_DCT_method_ptr inverse_DCT;
98 /* region decoding. this limits decode to the set of blocks +- 1 outside
99 * bounding blocks around the desired region to decode */
100 int blk1 = 0, blk2 = 0, skip = 0;
102 if ((cinfo->region_w > 0) && (cinfo->region_h > 0)) {
103 int bsz_w = 0, bsz_h = 0;
105 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
106 compptr = cinfo->cur_comp_info[ci];
107 if (compptr->MCU_sample_width > bsz_w)
108 bsz_w = compptr->MCU_sample_width;
109 if ((compptr->MCU_height * 8) > bsz_h)
110 bsz_h = compptr->MCU_height * 8;
112 int _region_y = (int)cinfo->region_y;
113 _region_y = (_region_y>>1)<<1;
114 if (((int)cinfo->output_scanline < (_region_y - bsz_h - 1)) ||
115 ((int)cinfo->output_scanline > (_region_y + cinfo->region_h + bsz_h)))
118 blk1 = (cinfo->region_x / bsz_w) - 1;
122 blk2 = ((cinfo->region_x + cinfo->region_w + bsz_w - 1) / bsz_w) + 1;
128 /* Loop to process as much as one whole iMCU row */
129 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
131 for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col;
134 /* see if we need to skip this MCU or not */
135 if ((cinfo->region_w > 0) && (cinfo->region_h > 0)) {
136 if (!((MCU_col_num < blk1) || (MCU_col_num > blk2) || skip))
139 /* if we are not skipping this MCU, zero it ready for huffman decode */
141 jzero_far((void FAR *) coef->MCU_buffer[0],
142 (size_t) (cinfo->blocks_in_MCU * sizeof(JBLOCK)));
144 /* Try to fetch an MCU. Entropy decoder expects buffer to be zeroed. */
146 jzero_far((void FAR *) coef->MCU_buffer[0],
147 (size_t) (cinfo->blocks_in_MCU * sizeof(JBLOCK)));
149 jzero_far((void *)coef->MCU_buffer[0],
150 (size_t)(cinfo->blocks_in_MCU * sizeof(JBLOCK)));
152 if (!cinfo->entropy->insufficient_data)
153 cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row;
154 if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
155 /* Suspension forced; update state counters and exit */
156 coef->MCU_vert_offset = yoffset;
157 coef->MCU_ctr = MCU_col_num;
158 return JPEG_SUSPENDED;
161 /* region decoding. this limits decode to the set of blocks +- 1 outside
162 * bounding blocks around the desired region to decode */
167 /* Only perform the IDCT on blocks that are contained within the desired
170 if (MCU_col_num >= cinfo->master->first_iMCU_col &&
171 MCU_col_num <= cinfo->master->last_iMCU_col) {
172 /* Determine where data should go in output_buf and do the IDCT thing.
173 * We skip dummy blocks at the right and bottom edges (but blkn gets
174 * incremented past them!). Note the inner loop relies on having
175 * allocated the MCU_buffer[] blocks sequentially.
177 blkn = 0; /* index of current DCT block within MCU */
178 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
179 compptr = cinfo->cur_comp_info[ci];
180 /* Don't bother to IDCT an uninteresting component. */
181 if (!compptr->component_needed) {
182 blkn += compptr->MCU_blocks;
185 inverse_DCT = cinfo->idct->_inverse_DCT[compptr->component_index];
186 useful_width = (MCU_col_num < last_MCU_col) ?
187 compptr->MCU_width : compptr->last_col_width;
188 output_ptr = output_buf[compptr->component_index] +
189 yoffset * compptr->_DCT_scaled_size;
190 start_col = (MCU_col_num - cinfo->master->first_iMCU_col) *
191 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;
203 blkn += compptr->MCU_width;
204 output_ptr += compptr->_DCT_scaled_size;
209 /* Completed an MCU row, but perhaps not an iMCU row */
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;
218 /* Completed the scan */
219 (*cinfo->inputctl->finish_input_pass) (cinfo);
220 return JPEG_SCAN_COMPLETED;
225 * Dummy consume-input routine for single-pass operation.
229 dummy_consume_data(j_decompress_ptr cinfo)
231 return JPEG_SUSPENDED; /* Always indicate nothing was done */
235 #ifdef D_MULTISCAN_FILES_SUPPORTED
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.
245 consume_data(j_decompress_ptr cinfo)
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;
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.
268 /* Loop to process one whole iMCU row */
269 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
271 for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
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++;
285 if (!cinfo->entropy->insufficient_data)
286 cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row;
287 /* Try to fetch the MCU. */
288 if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
289 /* Suspension forced; update state counters and exit */
290 coef->MCU_vert_offset = yoffset;
291 coef->MCU_ctr = MCU_col_num;
292 return JPEG_SUSPENDED;
295 /* Completed an MCU row, but perhaps not an iMCU row */
298 /* Completed the iMCU row, advance counters for next one */
299 if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
300 start_iMCU_row(cinfo);
301 return JPEG_ROW_COMPLETED;
303 /* Completed the scan */
304 (*cinfo->inputctl->finish_input_pass) (cinfo);
305 return JPEG_SCAN_COMPLETED;
310 * Decompress and return some data in the multi-pass case.
311 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
312 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
314 * NB: output_buf contains a plane for each component in image.
318 decompress_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf)
320 my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
321 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
322 JDIMENSION block_num;
323 int ci, block_row, block_rows;
325 JBLOCKROW buffer_ptr;
326 _JSAMPARRAY output_ptr;
327 JDIMENSION output_col;
328 jpeg_component_info *compptr;
329 _inverse_DCT_method_ptr inverse_DCT;
331 /* Force some input to be done if we are getting ahead of the input. */
332 while (cinfo->input_scan_number < cinfo->output_scan_number ||
333 (cinfo->input_scan_number == cinfo->output_scan_number &&
334 cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
335 if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED)
336 return JPEG_SUSPENDED;
339 /* OK, output from the virtual arrays. */
340 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
342 /* Don't bother to IDCT an uninteresting component. */
343 if (!compptr->component_needed)
345 /* Align the virtual buffer for this component. */
346 buffer = (*cinfo->mem->access_virt_barray)
347 ((j_common_ptr)cinfo, coef->whole_image[ci],
348 cinfo->output_iMCU_row * compptr->v_samp_factor,
349 (JDIMENSION)compptr->v_samp_factor, FALSE);
350 /* Count non-dummy DCT block rows in this iMCU row. */
351 if (cinfo->output_iMCU_row < last_iMCU_row)
352 block_rows = compptr->v_samp_factor;
354 /* NB: can't use last_row_height here; it is input-side-dependent! */
355 block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor);
356 if (block_rows == 0) block_rows = compptr->v_samp_factor;
358 inverse_DCT = cinfo->idct->_inverse_DCT[ci];
359 output_ptr = output_buf[ci];
360 /* Loop over all DCT blocks to be processed. */
361 for (block_row = 0; block_row < block_rows; block_row++) {
362 buffer_ptr = buffer[block_row] + cinfo->master->first_MCU_col[ci];
364 for (block_num = cinfo->master->first_MCU_col[ci];
365 block_num <= cinfo->master->last_MCU_col[ci]; block_num++) {
366 (*inverse_DCT) (cinfo, compptr, (JCOEFPTR)buffer_ptr, output_ptr,
369 output_col += compptr->_DCT_scaled_size;
371 output_ptr += compptr->_DCT_scaled_size;
375 if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
376 return JPEG_ROW_COMPLETED;
377 return JPEG_SCAN_COMPLETED;
380 #endif /* D_MULTISCAN_FILES_SUPPORTED */
383 #ifdef BLOCK_SMOOTHING_SUPPORTED
386 * This code applies interblock smoothing; the first 9 AC coefficients are
387 * estimated from the DC values of a DCT block and its 24 neighboring blocks.
388 * We apply smoothing only for progressive JPEG decoding, and only if
389 * the coefficients it can estimate are not yet known to full precision.
392 /* Natural-order array positions of the first 9 zigzag-order coefficients */
404 * Determine whether block smoothing is applicable and safe.
405 * We also latch the current states of the coef_bits[] entries for the
406 * AC coefficients; otherwise, if the input side of the decompressor
407 * advances into a new scan, we might think the coefficients are known
408 * more accurately than they really are.
412 smoothing_ok(j_decompress_ptr cinfo)
414 my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
415 boolean smoothing_useful = FALSE;
417 jpeg_component_info *compptr;
419 int *coef_bits, *prev_coef_bits;
420 int *coef_bits_latch, *prev_coef_bits_latch;
422 if (!cinfo->progressive_mode || cinfo->coef_bits == NULL)
425 /* Allocate latch area if not already done */
426 if (coef->coef_bits_latch == NULL)
427 coef->coef_bits_latch = (int *)
428 (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
429 cinfo->num_components * 2 *
430 (SAVED_COEFS * sizeof(int)));
431 coef_bits_latch = coef->coef_bits_latch;
432 prev_coef_bits_latch =
433 &coef->coef_bits_latch[cinfo->num_components * SAVED_COEFS];
435 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
437 /* All components' quantization values must already be latched. */
438 if ((qtable = compptr->quant_table) == NULL)
440 /* Verify DC & first 9 AC quantizers are nonzero to avoid zero-divide. */
441 if (qtable->quantval[0] == 0 ||
442 qtable->quantval[Q01_POS] == 0 ||
443 qtable->quantval[Q10_POS] == 0 ||
444 qtable->quantval[Q20_POS] == 0 ||
445 qtable->quantval[Q11_POS] == 0 ||
446 qtable->quantval[Q02_POS] == 0 ||
447 qtable->quantval[Q03_POS] == 0 ||
448 qtable->quantval[Q12_POS] == 0 ||
449 qtable->quantval[Q21_POS] == 0 ||
450 qtable->quantval[Q30_POS] == 0)
452 /* DC values must be at least partly known for all components. */
453 coef_bits = cinfo->coef_bits[ci];
454 prev_coef_bits = cinfo->coef_bits[ci + cinfo->num_components];
455 if (coef_bits[0] < 0)
457 coef_bits_latch[0] = coef_bits[0];
458 /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
459 for (coefi = 1; coefi < SAVED_COEFS; coefi++) {
460 if (cinfo->input_scan_number > 1)
461 prev_coef_bits_latch[coefi] = prev_coef_bits[coefi];
463 prev_coef_bits_latch[coefi] = -1;
464 coef_bits_latch[coefi] = coef_bits[coefi];
465 if (coef_bits[coefi] != 0)
466 smoothing_useful = TRUE;
468 coef_bits_latch += SAVED_COEFS;
469 prev_coef_bits_latch += SAVED_COEFS;
472 return smoothing_useful;
477 * Variant of decompress_data for use when doing block smoothing.
481 decompress_smooth_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf)
483 my_coef_ptr coef = (my_coef_ptr)cinfo->coef;
484 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
485 JDIMENSION block_num, last_block_column;
486 int ci, block_row, block_rows, access_rows, image_block_row,
489 JBLOCKROW buffer_ptr, prev_prev_block_row, prev_block_row;
490 JBLOCKROW next_block_row, next_next_block_row;
491 _JSAMPARRAY output_ptr;
492 JDIMENSION output_col;
493 jpeg_component_info *compptr;
494 _inverse_DCT_method_ptr inverse_DCT;
498 JQUANT_TBL *quanttbl;
499 JLONG Q00, Q01, Q02, Q03 = 0, Q10, Q11, Q12 = 0, Q20, Q21 = 0, Q30 = 0, num;
500 int DC01, DC02, DC03, DC04, DC05, DC06, DC07, DC08, DC09, DC10, DC11, DC12,
501 DC13, DC14, DC15, DC16, DC17, DC18, DC19, DC20, DC21, DC22, DC23, DC24,
505 /* Keep a local variable to avoid looking it up more than once */
506 workspace = coef->workspace;
508 /* Force some input to be done if we are getting ahead of the input. */
509 while (cinfo->input_scan_number <= cinfo->output_scan_number &&
510 !cinfo->inputctl->eoi_reached) {
511 if (cinfo->input_scan_number == cinfo->output_scan_number) {
512 /* If input is working on current scan, we ordinarily want it to
513 * have completed the current row. But if input scan is DC,
514 * we want it to keep two rows ahead so that next two block rows' DC
515 * values are up to date.
517 JDIMENSION delta = (cinfo->Ss == 0) ? 2 : 0;
518 if (cinfo->input_iMCU_row > cinfo->output_iMCU_row + delta)
521 if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED)
522 return JPEG_SUSPENDED;
525 /* OK, output from the virtual arrays. */
526 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
528 /* Don't bother to IDCT an uninteresting component. */
529 if (!compptr->component_needed)
531 /* Count non-dummy DCT block rows in this iMCU row. */
532 if (cinfo->output_iMCU_row + 1 < last_iMCU_row) {
533 block_rows = compptr->v_samp_factor;
534 access_rows = block_rows * 3; /* this and next two iMCU rows */
535 } else if (cinfo->output_iMCU_row < last_iMCU_row) {
536 block_rows = compptr->v_samp_factor;
537 access_rows = block_rows * 2; /* this and next iMCU row */
539 /* NB: can't use last_row_height here; it is input-side-dependent! */
540 block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor);
541 if (block_rows == 0) block_rows = compptr->v_samp_factor;
542 access_rows = block_rows; /* this iMCU row only */
544 /* Align the virtual buffer for this component. */
545 if (cinfo->output_iMCU_row > 1) {
546 access_rows += 2 * compptr->v_samp_factor; /* prior two iMCU rows too */
547 buffer = (*cinfo->mem->access_virt_barray)
548 ((j_common_ptr)cinfo, coef->whole_image[ci],
549 (cinfo->output_iMCU_row - 2) * compptr->v_samp_factor,
550 (JDIMENSION)access_rows, FALSE);
551 buffer += 2 * compptr->v_samp_factor; /* point to current iMCU row */
552 } else if (cinfo->output_iMCU_row > 0) {
553 access_rows += compptr->v_samp_factor; /* prior iMCU row too */
554 buffer = (*cinfo->mem->access_virt_barray)
555 ((j_common_ptr)cinfo, coef->whole_image[ci],
556 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
557 (JDIMENSION)access_rows, FALSE);
558 buffer += compptr->v_samp_factor; /* point to current iMCU row */
560 buffer = (*cinfo->mem->access_virt_barray)
561 ((j_common_ptr)cinfo, coef->whole_image[ci],
562 (JDIMENSION)0, (JDIMENSION)access_rows, FALSE);
564 /* Fetch component-dependent info.
565 * If the current scan is incomplete, then we use the component-dependent
566 * info from the previous scan.
568 if (cinfo->output_iMCU_row > cinfo->master->last_good_iMCU_row)
570 coef->coef_bits_latch + ((ci + cinfo->num_components) * SAVED_COEFS);
572 coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
574 /* We only do DC interpolation if no AC coefficient data is available. */
576 coef_bits[1] == -1 && coef_bits[2] == -1 && coef_bits[3] == -1 &&
577 coef_bits[4] == -1 && coef_bits[5] == -1 && coef_bits[6] == -1 &&
578 coef_bits[7] == -1 && coef_bits[8] == -1 && coef_bits[9] == -1;
580 quanttbl = compptr->quant_table;
581 Q00 = quanttbl->quantval[0];
582 Q01 = quanttbl->quantval[Q01_POS];
583 Q10 = quanttbl->quantval[Q10_POS];
584 Q20 = quanttbl->quantval[Q20_POS];
585 Q11 = quanttbl->quantval[Q11_POS];
586 Q02 = quanttbl->quantval[Q02_POS];
588 Q03 = quanttbl->quantval[Q03_POS];
589 Q12 = quanttbl->quantval[Q12_POS];
590 Q21 = quanttbl->quantval[Q21_POS];
591 Q30 = quanttbl->quantval[Q30_POS];
593 inverse_DCT = cinfo->idct->_inverse_DCT[ci];
594 output_ptr = output_buf[ci];
595 /* Loop over all DCT blocks to be processed. */
596 image_block_rows = block_rows * cinfo->total_iMCU_rows;
597 for (block_row = 0; block_row < block_rows; block_row++) {
598 image_block_row = cinfo->output_iMCU_row * block_rows + block_row;
599 buffer_ptr = buffer[block_row] + cinfo->master->first_MCU_col[ci];
601 if (image_block_row > 0)
603 buffer[block_row - 1] + cinfo->master->first_MCU_col[ci];
605 prev_block_row = buffer_ptr;
607 if (image_block_row > 1)
608 prev_prev_block_row =
609 buffer[block_row - 2] + cinfo->master->first_MCU_col[ci];
611 prev_prev_block_row = prev_block_row;
613 if (image_block_row < image_block_rows - 1)
615 buffer[block_row + 1] + cinfo->master->first_MCU_col[ci];
617 next_block_row = buffer_ptr;
619 if (image_block_row < image_block_rows - 2)
620 next_next_block_row =
621 buffer[block_row + 2] + cinfo->master->first_MCU_col[ci];
623 next_next_block_row = next_block_row;
625 /* We fetch the surrounding DC values using a sliding-register approach.
626 * Initialize all 25 here so as to do the right thing on narrow pics.
628 DC01 = DC02 = DC03 = DC04 = DC05 = (int)prev_prev_block_row[0][0];
629 DC06 = DC07 = DC08 = DC09 = DC10 = (int)prev_block_row[0][0];
630 DC11 = DC12 = DC13 = DC14 = DC15 = (int)buffer_ptr[0][0];
631 DC16 = DC17 = DC18 = DC19 = DC20 = (int)next_block_row[0][0];
632 DC21 = DC22 = DC23 = DC24 = DC25 = (int)next_next_block_row[0][0];
634 last_block_column = compptr->width_in_blocks - 1;
635 for (block_num = cinfo->master->first_MCU_col[ci];
636 block_num <= cinfo->master->last_MCU_col[ci]; block_num++) {
637 /* Fetch current DCT block into workspace so we can modify it. */
638 jcopy_block_row(buffer_ptr, (JBLOCKROW)workspace, (JDIMENSION)1);
639 /* Update DC values */
640 if (block_num == cinfo->master->first_MCU_col[ci] &&
641 block_num < last_block_column) {
642 DC04 = DC05 = (int)prev_prev_block_row[1][0];
643 DC09 = DC10 = (int)prev_block_row[1][0];
644 DC14 = DC15 = (int)buffer_ptr[1][0];
645 DC19 = DC20 = (int)next_block_row[1][0];
646 DC24 = DC25 = (int)next_next_block_row[1][0];
648 if (block_num + 1 < last_block_column) {
649 DC05 = (int)prev_prev_block_row[2][0];
650 DC10 = (int)prev_block_row[2][0];
651 DC15 = (int)buffer_ptr[2][0];
652 DC20 = (int)next_block_row[2][0];
653 DC25 = (int)next_next_block_row[2][0];
655 /* If DC interpolation is enabled, compute coefficient estimates using
656 * a Gaussian-like kernel, keeping the averages of the DC values.
658 * If DC interpolation is disabled, compute coefficient estimates using
659 * an algorithm similar to the one described in Section K.8 of the JPEG
660 * standard, except applied to a 5x5 window rather than a 3x3 window.
662 * An estimate is applied only if the coefficient is still zero and is
663 * not known to be fully accurate.
666 if ((Al = coef_bits[1]) != 0 && workspace[1] == 0) {
667 num = Q00 * (change_dc ?
668 (-DC01 - DC02 + DC04 + DC05 - 3 * DC06 + 13 * DC07 -
669 13 * DC09 + 3 * DC10 - 3 * DC11 + 38 * DC12 - 38 * DC14 +
670 3 * DC15 - 3 * DC16 + 13 * DC17 - 13 * DC19 + 3 * DC20 -
671 DC21 - DC22 + DC24 + DC25) :
672 (-7 * DC11 + 50 * DC12 - 50 * DC14 + 7 * DC15));
674 pred = (int)(((Q01 << 7) + num) / (Q01 << 8));
675 if (Al > 0 && pred >= (1 << Al))
676 pred = (1 << Al) - 1;
678 pred = (int)(((Q01 << 7) - num) / (Q01 << 8));
679 if (Al > 0 && pred >= (1 << Al))
680 pred = (1 << Al) - 1;
683 workspace[1] = (JCOEF)pred;
686 if ((Al = coef_bits[2]) != 0 && workspace[8] == 0) {
687 num = Q00 * (change_dc ?
688 (-DC01 - 3 * DC02 - 3 * DC03 - 3 * DC04 - DC05 - DC06 +
689 13 * DC07 + 38 * DC08 + 13 * DC09 - DC10 + DC16 -
690 13 * DC17 - 38 * DC18 - 13 * DC19 + DC20 + DC21 +
691 3 * DC22 + 3 * DC23 + 3 * DC24 + DC25) :
692 (-7 * DC03 + 50 * DC08 - 50 * DC18 + 7 * DC23));
694 pred = (int)(((Q10 << 7) + num) / (Q10 << 8));
695 if (Al > 0 && pred >= (1 << Al))
696 pred = (1 << Al) - 1;
698 pred = (int)(((Q10 << 7) - num) / (Q10 << 8));
699 if (Al > 0 && pred >= (1 << Al))
700 pred = (1 << Al) - 1;
703 workspace[8] = (JCOEF)pred;
706 if ((Al = coef_bits[3]) != 0 && workspace[16] == 0) {
707 num = Q00 * (change_dc ?
708 (DC03 + 2 * DC07 + 7 * DC08 + 2 * DC09 - 5 * DC12 - 14 * DC13 -
709 5 * DC14 + 2 * DC17 + 7 * DC18 + 2 * DC19 + DC23) :
710 (-DC03 + 13 * DC08 - 24 * DC13 + 13 * DC18 - DC23));
712 pred = (int)(((Q20 << 7) + num) / (Q20 << 8));
713 if (Al > 0 && pred >= (1 << Al))
714 pred = (1 << Al) - 1;
716 pred = (int)(((Q20 << 7) - num) / (Q20 << 8));
717 if (Al > 0 && pred >= (1 << Al))
718 pred = (1 << Al) - 1;
721 workspace[16] = (JCOEF)pred;
724 if ((Al = coef_bits[4]) != 0 && workspace[9] == 0) {
725 num = Q00 * (change_dc ?
726 (-DC01 + DC05 + 9 * DC07 - 9 * DC09 - 9 * DC17 +
727 9 * DC19 + DC21 - DC25) :
728 (DC10 + DC16 - 10 * DC17 + 10 * DC19 - DC02 - DC20 + DC22 -
729 DC24 + DC04 - DC06 + 10 * DC07 - 10 * DC09));
731 pred = (int)(((Q11 << 7) + num) / (Q11 << 8));
732 if (Al > 0 && pred >= (1 << Al))
733 pred = (1 << Al) - 1;
735 pred = (int)(((Q11 << 7) - num) / (Q11 << 8));
736 if (Al > 0 && pred >= (1 << Al))
737 pred = (1 << Al) - 1;
740 workspace[9] = (JCOEF)pred;
743 if ((Al = coef_bits[5]) != 0 && workspace[2] == 0) {
744 num = Q00 * (change_dc ?
745 (2 * DC07 - 5 * DC08 + 2 * DC09 + DC11 + 7 * DC12 - 14 * DC13 +
746 7 * DC14 + DC15 + 2 * DC17 - 5 * DC18 + 2 * DC19) :
747 (-DC11 + 13 * DC12 - 24 * DC13 + 13 * DC14 - DC15));
749 pred = (int)(((Q02 << 7) + num) / (Q02 << 8));
750 if (Al > 0 && pred >= (1 << Al))
751 pred = (1 << Al) - 1;
753 pred = (int)(((Q02 << 7) - num) / (Q02 << 8));
754 if (Al > 0 && pred >= (1 << Al))
755 pred = (1 << Al) - 1;
758 workspace[2] = (JCOEF)pred;
762 if ((Al = coef_bits[6]) != 0 && workspace[3] == 0) {
763 num = Q00 * (DC07 - DC09 + 2 * DC12 - 2 * DC14 + DC17 - DC19);
765 pred = (int)(((Q03 << 7) + num) / (Q03 << 8));
766 if (Al > 0 && pred >= (1 << Al))
767 pred = (1 << Al) - 1;
769 pred = (int)(((Q03 << 7) - num) / (Q03 << 8));
770 if (Al > 0 && pred >= (1 << Al))
771 pred = (1 << Al) - 1;
774 workspace[3] = (JCOEF)pred;
777 if ((Al = coef_bits[7]) != 0 && workspace[10] == 0) {
778 num = Q00 * (DC07 - 3 * DC08 + DC09 - DC17 + 3 * DC18 - DC19);
780 pred = (int)(((Q12 << 7) + num) / (Q12 << 8));
781 if (Al > 0 && pred >= (1 << Al))
782 pred = (1 << Al) - 1;
784 pred = (int)(((Q12 << 7) - num) / (Q12 << 8));
785 if (Al > 0 && pred >= (1 << Al))
786 pred = (1 << Al) - 1;
789 workspace[10] = (JCOEF)pred;
792 if ((Al = coef_bits[8]) != 0 && workspace[17] == 0) {
793 num = Q00 * (DC07 - DC09 - 3 * DC12 + 3 * DC14 + DC17 - DC19);
795 pred = (int)(((Q21 << 7) + num) / (Q21 << 8));
796 if (Al > 0 && pred >= (1 << Al))
797 pred = (1 << Al) - 1;
799 pred = (int)(((Q21 << 7) - num) / (Q21 << 8));
800 if (Al > 0 && pred >= (1 << Al))
801 pred = (1 << Al) - 1;
804 workspace[17] = (JCOEF)pred;
807 if ((Al = coef_bits[9]) != 0 && workspace[24] == 0) {
808 num = Q00 * (DC07 + 2 * DC08 + DC09 - DC17 - 2 * DC18 - DC19);
810 pred = (int)(((Q30 << 7) + num) / (Q30 << 8));
811 if (Al > 0 && pred >= (1 << Al))
812 pred = (1 << Al) - 1;
814 pred = (int)(((Q30 << 7) - num) / (Q30 << 8));
815 if (Al > 0 && pred >= (1 << Al))
816 pred = (1 << Al) - 1;
819 workspace[24] = (JCOEF)pred;
821 /* coef_bits[0] is non-negative. Otherwise this function would not
825 (-2 * DC01 - 6 * DC02 - 8 * DC03 - 6 * DC04 - 2 * DC05 -
826 6 * DC06 + 6 * DC07 + 42 * DC08 + 6 * DC09 - 6 * DC10 -
827 8 * DC11 + 42 * DC12 + 152 * DC13 + 42 * DC14 - 8 * DC15 -
828 6 * DC16 + 6 * DC17 + 42 * DC18 + 6 * DC19 - 6 * DC20 -
829 2 * DC21 - 6 * DC22 - 8 * DC23 - 6 * DC24 - 2 * DC25);
831 pred = (int)(((Q00 << 7) + num) / (Q00 << 8));
833 pred = (int)(((Q00 << 7) - num) / (Q00 << 8));
836 workspace[0] = (JCOEF)pred;
839 /* OK, do the IDCT */
840 (*inverse_DCT) (cinfo, compptr, (JCOEFPTR)workspace, output_ptr,
842 /* Advance for next column */
843 DC01 = DC02; DC02 = DC03; DC03 = DC04; DC04 = DC05;
844 DC06 = DC07; DC07 = DC08; DC08 = DC09; DC09 = DC10;
845 DC11 = DC12; DC12 = DC13; DC13 = DC14; DC14 = DC15;
846 DC16 = DC17; DC17 = DC18; DC18 = DC19; DC19 = DC20;
847 DC21 = DC22; DC22 = DC23; DC23 = DC24; DC24 = DC25;
848 buffer_ptr++, prev_block_row++, next_block_row++,
849 prev_prev_block_row++, next_next_block_row++;
850 output_col += compptr->_DCT_scaled_size;
852 output_ptr += compptr->_DCT_scaled_size;
856 if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
857 return JPEG_ROW_COMPLETED;
858 return JPEG_SCAN_COMPLETED;
861 #endif /* BLOCK_SMOOTHING_SUPPORTED */
865 * Initialize coefficient buffer controller.
869 _jinit_d_coef_controller(j_decompress_ptr cinfo, boolean need_full_buffer)
873 if (cinfo->data_precision != BITS_IN_JSAMPLE)
874 ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision);
877 (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
878 sizeof(my_coef_controller));
879 cinfo->coef = (struct jpeg_d_coef_controller *)coef;
880 coef->pub.start_input_pass = start_input_pass;
881 coef->pub.start_output_pass = start_output_pass;
882 #ifdef BLOCK_SMOOTHING_SUPPORTED
883 coef->coef_bits_latch = NULL;
886 /* Create the coefficient buffer. */
887 if (need_full_buffer) {
888 #ifdef D_MULTISCAN_FILES_SUPPORTED
889 /* Allocate a full-image virtual array for each component, */
890 /* padded to a multiple of samp_factor DCT blocks in each direction. */
891 /* Note we ask for a pre-zeroed array. */
893 jpeg_component_info *compptr;
895 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
897 access_rows = compptr->v_samp_factor;
898 #ifdef BLOCK_SMOOTHING_SUPPORTED
899 /* If block smoothing could be used, need a bigger window */
900 if (cinfo->progressive_mode)
903 coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
904 ((j_common_ptr)cinfo, JPOOL_IMAGE, TRUE,
905 (JDIMENSION)jround_up((long)compptr->width_in_blocks,
906 (long)compptr->h_samp_factor),
907 (JDIMENSION)jround_up((long)compptr->height_in_blocks,
908 (long)compptr->v_samp_factor),
909 (JDIMENSION)access_rows);
911 coef->pub.consume_data = consume_data;
912 coef->pub._decompress_data = decompress_data;
913 coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
915 ERREXIT(cinfo, JERR_NOT_COMPILED);
918 /* We only need a single-MCU buffer. */
923 (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE,
924 D_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK));
925 for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
926 coef->MCU_buffer[i] = buffer + i;
928 coef->pub.consume_data = dummy_consume_data;
929 coef->pub._decompress_data = decompress_onepass;
930 coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
933 /* Allocate the workspace buffer */
934 coef->workspace = (JCOEF *)
935 (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
936 sizeof(JCOEF) * DCTSIZE2);