blob: 84ce56fb7a0154cafb1324c0aec6fe1c19265b23 [file] [log] [blame]
Dees_Troy51a0e822012-09-05 15:24:24 -04001/*
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 < (unsigned int)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 int iMCU_width_To_MCU_width;
281 if (cinfo->comps_in_scan > 1) {
282 // Interleaved
283 iMCU_width_To_MCU_width = 1;
284 } else {
285 // Non-intervleaved
286 iMCU_width_To_MCU_width = cinfo->cur_comp_info[0]->h_samp_factor;
287 }
288 MCUs_per_row = jmin(MCUs_per_row,
289 (cinfo->coef->column_right_boundary - cinfo->coef->column_left_boundary)
290 * cinfo->entropy->index->MCU_sample_size * iMCU_width_To_MCU_width);
291 }
292#endif
293
294 /* Loop to process one whole iMCU row */
295 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
296 yoffset++) {
297 // configure huffman decoder
298#ifdef ANDROID_TILE_BASED_DECODE
299 if (cinfo->tile_decode) {
300 huffman_scan_header scan_header =
301 cinfo->entropy->index->scan[cinfo->input_scan_number];
302 int col_offset = cinfo->coef->column_left_boundary;
303 (*cinfo->entropy->configure_huffman_decoder) (cinfo,
304 scan_header.offset[cinfo->input_iMCU_row]
305 [col_offset + yoffset * scan_header.MCUs_per_row]);
306 }
307#endif
308
309 // zero all blocks
310 for (MCU_col_num = coef->MCU_ctr; MCU_col_num < MCUs_per_row;
311 MCU_col_num++) {
312 /* Construct list of pointers to DCT blocks belonging to this MCU */
313 blkn = 0; /* index of current DCT block within MCU */
314 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
315 compptr = cinfo->cur_comp_info[ci];
316 start_col = MCU_col_num * compptr->MCU_width;
317 for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
318 buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
319 for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
320 coef->MCU_buffer[blkn++] = buffer_ptr++;
321#ifdef ANDROID_TILE_BASED_DECODE
322 if (cinfo->tile_decode && cinfo->input_scan_number == 0) {
323 // need to do pre-zero ourselves.
324 jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
325 (size_t) (SIZEOF(JBLOCK)));
326 }
327#endif
328 }
329 }
330 }
331
332
333 /* Try to fetch the MCU. */
334 if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
335 /* Suspension forced; update state counters and exit */
336 coef->MCU_vert_offset = yoffset;
337 coef->MCU_ctr = MCU_col_num;
338 return JPEG_SUSPENDED;
339 }
340 }
341 /* Completed an MCU row, but perhaps not an iMCU row */
342 coef->MCU_ctr = 0;
343 }
344 /* Completed the iMCU row, advance counters for next one */
345 if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
346 start_iMCU_row(cinfo);
347 return JPEG_ROW_COMPLETED;
348 }
349 /* Completed the scan */
350 (*cinfo->inputctl->finish_input_pass) (cinfo);
351 return JPEG_SCAN_COMPLETED;
352}
353
354/*
355 * Consume input data and store it in the coefficient buffer.
356 * Read one fully interleaved MCU row ("iMCU" row) per call.
357 */
358
359METHODDEF(int)
360consume_data_multi_scan (j_decompress_ptr cinfo)
361{
362 huffman_index *index = cinfo->entropy->index;
363 int i, retcode, ci;
364 int mcu = cinfo->input_iMCU_row;
365 jinit_phuff_decoder(cinfo);
366 for (i = 0; i < index->scan_count; i++) {
367 (*cinfo->inputctl->finish_input_pass) (cinfo);
368 jset_input_stream_position(cinfo, index->scan[i].bitstream_offset);
369 cinfo->output_iMCU_row = mcu;
370 cinfo->unread_marker = 0;
371 // Consume SOS and DHT headers
372 retcode = (*cinfo->inputctl->consume_markers) (cinfo, index, i);
373 cinfo->input_iMCU_row = mcu;
374 cinfo->input_scan_number = i;
375 cinfo->entropy->index = index;
376 // Consume scan block data
377 consume_data(cinfo);
378 }
379 cinfo->input_iMCU_row = mcu + 1;
380 cinfo->input_scan_number = 0;
381 cinfo->output_scan_number = 0;
382 return JPEG_ROW_COMPLETED;
383}
384
385/*
386 * Same as consume_data, expect for saving the Huffman decode information
387 * - bitstream offset and DC coefficient to index.
388 */
389
390METHODDEF(int)
391consume_data_build_huffman_index_baseline (j_decompress_ptr cinfo,
392 huffman_index *index, int current_scan)
393{
394 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
395 JDIMENSION MCU_col_num; /* index of current MCU within row */
396 int ci, xindex, yindex, yoffset;
397 JDIMENSION start_col;
398 JBLOCKROW buffer_ptr;
399
400 huffman_scan_header *scan_header = index->scan + current_scan;
401 scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
402
403 size_t allocate_size = coef->MCU_rows_per_iMCU_row
404 * jdiv_round_up(cinfo->MCUs_per_row, index->MCU_sample_size)
405 * sizeof(huffman_offset_data);
406 scan_header->offset[cinfo->input_iMCU_row] =
407 (huffman_offset_data*)malloc(allocate_size);
408 index->mem_used += allocate_size;
409
410 huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
411
412 /* Loop to process one whole iMCU row */
413 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
414 yoffset++) {
415 for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
416 MCU_col_num++) {
417 // Record huffman bit offset
418 if (MCU_col_num % index->MCU_sample_size == 0) {
419 (*cinfo->entropy->get_huffman_decoder_configuration)
420 (cinfo, offset_data);
421 ++offset_data;
422 }
423
424 /* Try to fetch the MCU. */
425 if (! (*cinfo->entropy->decode_mcu_discard_coef) (cinfo)) {
426 /* Suspension forced; update state counters and exit */
427 coef->MCU_vert_offset = yoffset;
428 coef->MCU_ctr = MCU_col_num;
429 return JPEG_SUSPENDED;
430 }
431 }
432 /* Completed an MCU row, but perhaps not an iMCU row */
433 coef->MCU_ctr = 0;
434 }
435 /* Completed the iMCU row, advance counters for next one */
436 if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
437 start_iMCU_row(cinfo);
438 return JPEG_ROW_COMPLETED;
439 }
440 /* Completed the scan */
441 (*cinfo->inputctl->finish_input_pass) (cinfo);
442 return JPEG_SCAN_COMPLETED;
443}
444
445/*
446 * Same as consume_data, expect for saving the Huffman decode information
447 * - bitstream offset and DC coefficient to index.
448 */
449
450METHODDEF(int)
451consume_data_build_huffman_index_progressive (j_decompress_ptr cinfo,
452 huffman_index *index, int current_scan)
453{
454 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
455 JDIMENSION MCU_col_num; /* index of current MCU within row */
456 int blkn, ci, xindex, yindex, yoffset;
457 JDIMENSION start_col;
458 JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN];
459 JBLOCKROW buffer_ptr;
460 jpeg_component_info *compptr;
461
462 int factor = 4; // maximum factor is 4.
463 for (ci = 0; ci < cinfo->comps_in_scan; ci++)
464 factor = jmin(factor, cinfo->cur_comp_info[ci]->h_samp_factor);
465
466 int sample_size = index->MCU_sample_size * factor;
467 huffman_scan_header *scan_header = index->scan + current_scan;
468 scan_header->MCU_rows_per_iMCU_row = coef->MCU_rows_per_iMCU_row;
469 scan_header->MCUs_per_row = jdiv_round_up(cinfo->MCUs_per_row, sample_size);
470 scan_header->comps_in_scan = cinfo->comps_in_scan;
471
472 size_t allocate_size = coef->MCU_rows_per_iMCU_row
473 * scan_header->MCUs_per_row * sizeof(huffman_offset_data);
474 scan_header->offset[cinfo->input_iMCU_row] =
475 (huffman_offset_data*)malloc(allocate_size);
476 index->mem_used += allocate_size;
477
478 huffman_offset_data *offset_data = scan_header->offset[cinfo->input_iMCU_row];
479
480 /* Align the virtual buffers for the components used in this scan. */
481 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
482 compptr = cinfo->cur_comp_info[ci];
483 buffer[ci] = (*cinfo->mem->access_virt_barray)
484 ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index],
485 0, // Only need one row buffer
486 (JDIMENSION) compptr->v_samp_factor, TRUE);
487 }
488 /* Loop to process one whole iMCU row */
489 for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row;
490 yoffset++) {
491 for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row;
492 MCU_col_num++) {
493 /* For each MCU, we loop through different color components.
494 * Then, for each color component we will get a list of pointers to DCT
495 * blocks in the virtual buffer.
496 */
497 blkn = 0; /* index of current DCT block within MCU */
498 for (ci = 0; ci < cinfo->comps_in_scan; ci++) {
499 compptr = cinfo->cur_comp_info[ci];
500 start_col = MCU_col_num * compptr->MCU_width;
501 /* Get the list of pointers to DCT blocks in
502 * the virtual buffer in a color component of the MCU.
503 */
504 for (yindex = 0; yindex < compptr->MCU_height; yindex++) {
505 buffer_ptr = buffer[ci][yindex+yoffset] + start_col;
506 for (xindex = 0; xindex < compptr->MCU_width; xindex++) {
507 coef->MCU_buffer[blkn++] = buffer_ptr++;
508 if (cinfo->input_scan_number == 0) {
509 // need to do pre-zero by ourself.
510 jzero_far((void FAR *) coef->MCU_buffer[blkn-1],
511 (size_t) (SIZEOF(JBLOCK)));
512 }
513 }
514 }
515 }
516 // Record huffman bit offset
517 if (MCU_col_num % sample_size == 0) {
518 (*cinfo->entropy->get_huffman_decoder_configuration)
519 (cinfo, offset_data);
520 ++offset_data;
521 }
522 /* Try to fetch the MCU. */
523 if (! (*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) {
524 /* Suspension forced; update state counters and exit */
525 coef->MCU_vert_offset = yoffset;
526 coef->MCU_ctr = MCU_col_num;
527 return JPEG_SUSPENDED;
528 }
529 }
530 /* Completed an MCU row, but perhaps not an iMCU row */
531 coef->MCU_ctr = 0;
532 }
533 (*cinfo->entropy->get_huffman_decoder_configuration)
534 (cinfo, &scan_header->prev_MCU_offset);
535 /* Completed the iMCU row, advance counters for next one */
536 if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) {
537 start_iMCU_row(cinfo);
538 return JPEG_ROW_COMPLETED;
539 }
540 /* Completed the scan */
541 (*cinfo->inputctl->finish_input_pass) (cinfo);
542 return JPEG_SCAN_COMPLETED;
543}
544
545/*
546 * Decompress and return some data in the multi-pass case.
547 * Always attempts to emit one fully interleaved MCU row ("iMCU" row).
548 * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED.
549 *
550 * NB: output_buf contains a plane for each component in image.
551 */
552
553METHODDEF(int)
554decompress_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
555{
556 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
557 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
558 JDIMENSION block_num;
559 int ci, block_row, block_rows;
560 JBLOCKARRAY buffer;
561 JBLOCKROW buffer_ptr;
562 JSAMPARRAY output_ptr;
563 JDIMENSION output_col;
564 jpeg_component_info *compptr;
565 inverse_DCT_method_ptr inverse_DCT;
566
567 /* Force some input to be done if we are getting ahead of the input. */
568 while (cinfo->input_scan_number < cinfo->output_scan_number ||
569 (cinfo->input_scan_number == cinfo->output_scan_number &&
570 cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) {
571 if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
572 return JPEG_SUSPENDED;
573 }
574
575 /* OK, output from the virtual arrays. */
576 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
577 ci++, compptr++) {
578 /* Don't bother to IDCT an uninteresting component. */
579 if (! compptr->component_needed)
580 continue;
581 /* Align the virtual buffer for this component. */
582 buffer = (*cinfo->mem->access_virt_barray)
583 ((j_common_ptr) cinfo, coef->whole_image[ci],
584 cinfo->tile_decode ? 0 : cinfo->output_iMCU_row * compptr->v_samp_factor,
585 (JDIMENSION) compptr->v_samp_factor, FALSE);
586 /* Count non-dummy DCT block rows in this iMCU row. */
587 if (cinfo->output_iMCU_row < last_iMCU_row)
588 block_rows = compptr->v_samp_factor;
589 else {
590 /* NB: can't use last_row_height here; it is input-side-dependent! */
591 block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
592 if (block_rows == 0) block_rows = compptr->v_samp_factor;
593 }
594 inverse_DCT = cinfo->idct->inverse_DCT[ci];
595 output_ptr = output_buf[ci];
596 int width_in_blocks = compptr->width_in_blocks;
597 int start_block = 0;
598#if ANDROID_TILE_BASED_DECODE
599 if (cinfo->tile_decode) {
600 // width_in_blocks for a component depends on its h_samp_factor.
601 width_in_blocks = jmin(width_in_blocks,
602 (cinfo->coef->MCU_column_right_boundary -
603 cinfo->coef->MCU_column_left_boundary) *
604 compptr->h_samp_factor);
605 start_block = coef->pub.MCU_columns_to_skip *
606 compptr->h_samp_factor;
607 }
608#endif
609 /* Loop over all DCT blocks to be processed. */
610 for (block_row = 0; block_row < block_rows; block_row++) {
611 buffer_ptr = buffer[block_row];
612 output_col = start_block * compptr->DCT_scaled_size;
613 buffer_ptr += start_block;
614 for (block_num = start_block; block_num < (unsigned int)width_in_blocks; block_num++) {
615 (*inverse_DCT) (cinfo, compptr, (JCOEFPTR) buffer_ptr,
616 output_ptr, output_col);
617 buffer_ptr++;
618 output_col += compptr->DCT_scaled_size;
619 }
620 output_ptr += compptr->DCT_scaled_size;
621 }
622 }
623
624 if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
625 return JPEG_ROW_COMPLETED;
626 return JPEG_SCAN_COMPLETED;
627}
628
629#endif /* D_MULTISCAN_FILES_SUPPORTED */
630
631
632#ifdef BLOCK_SMOOTHING_SUPPORTED
633
634/*
635 * This code applies interblock smoothing as described by section K.8
636 * of the JPEG standard: the first 5 AC coefficients are estimated from
637 * the DC values of a DCT block and its 8 neighboring blocks.
638 * We apply smoothing only for progressive JPEG decoding, and only if
639 * the coefficients it can estimate are not yet known to full precision.
640 */
641
642/* Natural-order array positions of the first 5 zigzag-order coefficients */
643#define Q01_POS 1
644#define Q10_POS 8
645#define Q20_POS 16
646#define Q11_POS 9
647#define Q02_POS 2
648
649/*
650 * Determine whether block smoothing is applicable and safe.
651 * We also latch the current states of the coef_bits[] entries for the
652 * AC coefficients; otherwise, if the input side of the decompressor
653 * advances into a new scan, we might think the coefficients are known
654 * more accurately than they really are.
655 */
656
657LOCAL(boolean)
658smoothing_ok (j_decompress_ptr cinfo)
659{
660 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
661 boolean smoothing_useful = FALSE;
662 int ci, coefi;
663 jpeg_component_info *compptr;
664 JQUANT_TBL * qtable;
665 int * coef_bits;
666 int * coef_bits_latch;
667
668 if (! cinfo->progressive_mode || cinfo->coef_bits == NULL)
669 return FALSE;
670
671 /* Allocate latch area if not already done */
672 if (coef->coef_bits_latch == NULL)
673 coef->coef_bits_latch = (int *)
674 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
675 cinfo->num_components *
676 (SAVED_COEFS * SIZEOF(int)));
677 coef_bits_latch = coef->coef_bits_latch;
678
679 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
680 ci++, compptr++) {
681 /* All components' quantization values must already be latched. */
682 if ((qtable = compptr->quant_table) == NULL)
683 return FALSE;
684 /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */
685 if (qtable->quantval[0] == 0 ||
686 qtable->quantval[Q01_POS] == 0 ||
687 qtable->quantval[Q10_POS] == 0 ||
688 qtable->quantval[Q20_POS] == 0 ||
689 qtable->quantval[Q11_POS] == 0 ||
690 qtable->quantval[Q02_POS] == 0)
691 return FALSE;
692 /* DC values must be at least partly known for all components. */
693 coef_bits = cinfo->coef_bits[ci];
694 if (coef_bits[0] < 0)
695 return FALSE;
696 /* Block smoothing is helpful if some AC coefficients remain inaccurate. */
697 for (coefi = 1; coefi <= 5; coefi++) {
698 coef_bits_latch[coefi] = coef_bits[coefi];
699 if (coef_bits[coefi] != 0)
700 smoothing_useful = TRUE;
701 }
702 coef_bits_latch += SAVED_COEFS;
703 }
704
705 return smoothing_useful;
706}
707
708
709/*
710 * Variant of decompress_data for use when doing block smoothing.
711 */
712
713METHODDEF(int)
714decompress_smooth_data (j_decompress_ptr cinfo, JSAMPIMAGE output_buf)
715{
716 my_coef_ptr coef = (my_coef_ptr) cinfo->coef;
717 JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1;
718 JDIMENSION block_num, last_block_column;
719 int ci, block_row, block_rows, access_rows;
720 JBLOCKARRAY buffer;
721 JBLOCKROW buffer_ptr, prev_block_row, next_block_row;
722 JSAMPARRAY output_ptr;
723 JDIMENSION output_col;
724 jpeg_component_info *compptr;
725 inverse_DCT_method_ptr inverse_DCT;
726 boolean first_row, last_row;
727 JBLOCK workspace;
728 int *coef_bits;
729 JQUANT_TBL *quanttbl;
730 INT32 Q00,Q01,Q02,Q10,Q11,Q20, num;
731 int DC1,DC2,DC3,DC4,DC5,DC6,DC7,DC8,DC9;
732 int Al, pred;
733
734 /* Force some input to be done if we are getting ahead of the input. */
735 while (cinfo->input_scan_number <= cinfo->output_scan_number &&
736 ! cinfo->inputctl->eoi_reached) {
737 if (cinfo->input_scan_number == cinfo->output_scan_number) {
738 /* If input is working on current scan, we ordinarily want it to
739 * have completed the current row. But if input scan is DC,
740 * we want it to keep one row ahead so that next block row's DC
741 * values are up to date.
742 */
743 JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0;
744 if (cinfo->input_iMCU_row > cinfo->output_iMCU_row+delta)
745 break;
746 }
747 if ((*cinfo->inputctl->consume_input)(cinfo) == JPEG_SUSPENDED)
748 return JPEG_SUSPENDED;
749 }
750
751 /* OK, output from the virtual arrays. */
752 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
753 ci++, compptr++) {
754 /* Don't bother to IDCT an uninteresting component. */
755 if (! compptr->component_needed)
756 continue;
757 /* Count non-dummy DCT block rows in this iMCU row. */
758 if (cinfo->output_iMCU_row < last_iMCU_row) {
759 block_rows = compptr->v_samp_factor;
760 access_rows = block_rows * 2; /* this and next iMCU row */
761 last_row = FALSE;
762 } else {
763 /* NB: can't use last_row_height here; it is input-side-dependent! */
764 block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor);
765 if (block_rows == 0) block_rows = compptr->v_samp_factor;
766 access_rows = block_rows; /* this iMCU row only */
767 last_row = TRUE;
768 }
769 /* Align the virtual buffer for this component. */
770 if (cinfo->output_iMCU_row > 0) {
771 access_rows += compptr->v_samp_factor; /* prior iMCU row too */
772 buffer = (*cinfo->mem->access_virt_barray)
773 ((j_common_ptr) cinfo, coef->whole_image[ci],
774 (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor,
775 (JDIMENSION) access_rows, FALSE);
776 buffer += compptr->v_samp_factor; /* point to current iMCU row */
777 first_row = FALSE;
778 } else {
779 buffer = (*cinfo->mem->access_virt_barray)
780 ((j_common_ptr) cinfo, coef->whole_image[ci],
781 (JDIMENSION) 0, (JDIMENSION) access_rows, FALSE);
782 first_row = TRUE;
783 }
784 /* Fetch component-dependent info */
785 coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS);
786 quanttbl = compptr->quant_table;
787 Q00 = quanttbl->quantval[0];
788 Q01 = quanttbl->quantval[Q01_POS];
789 Q10 = quanttbl->quantval[Q10_POS];
790 Q20 = quanttbl->quantval[Q20_POS];
791 Q11 = quanttbl->quantval[Q11_POS];
792 Q02 = quanttbl->quantval[Q02_POS];
793 inverse_DCT = cinfo->idct->inverse_DCT[ci];
794 output_ptr = output_buf[ci];
795 /* Loop over all DCT blocks to be processed. */
796 for (block_row = 0; block_row < block_rows; block_row++) {
797 buffer_ptr = buffer[block_row];
798 if (first_row && block_row == 0)
799 prev_block_row = buffer_ptr;
800 else
801 prev_block_row = buffer[block_row-1];
802 if (last_row && block_row == block_rows-1)
803 next_block_row = buffer_ptr;
804 else
805 next_block_row = buffer[block_row+1];
806 /* We fetch the surrounding DC values using a sliding-register approach.
807 * Initialize all nine here so as to do the right thing on narrow pics.
808 */
809 DC1 = DC2 = DC3 = (int) prev_block_row[0][0];
810 DC4 = DC5 = DC6 = (int) buffer_ptr[0][0];
811 DC7 = DC8 = DC9 = (int) next_block_row[0][0];
812 output_col = 0;
813 last_block_column = compptr->width_in_blocks - 1;
814 for (block_num = 0; block_num <= last_block_column; block_num++) {
815 /* Fetch current DCT block into workspace so we can modify it. */
816 jcopy_block_row(buffer_ptr, (JBLOCKROW) workspace, (JDIMENSION) 1);
817 /* Update DC values */
818 if (block_num < last_block_column) {
819 DC3 = (int) prev_block_row[1][0];
820 DC6 = (int) buffer_ptr[1][0];
821 DC9 = (int) next_block_row[1][0];
822 }
823 /* Compute coefficient estimates per K.8.
824 * An estimate is applied only if coefficient is still zero,
825 * and is not known to be fully accurate.
826 */
827 /* AC01 */
828 if ((Al=coef_bits[1]) != 0 && workspace[1] == 0) {
829 num = 36 * Q00 * (DC4 - DC6);
830 if (num >= 0) {
831 pred = (int) (((Q01<<7) + num) / (Q01<<8));
832 if (Al > 0 && pred >= (1<<Al))
833 pred = (1<<Al)-1;
834 } else {
835 pred = (int) (((Q01<<7) - num) / (Q01<<8));
836 if (Al > 0 && pred >= (1<<Al))
837 pred = (1<<Al)-1;
838 pred = -pred;
839 }
840 workspace[1] = (JCOEF) pred;
841 }
842 /* AC10 */
843 if ((Al=coef_bits[2]) != 0 && workspace[8] == 0) {
844 num = 36 * Q00 * (DC2 - DC8);
845 if (num >= 0) {
846 pred = (int) (((Q10<<7) + num) / (Q10<<8));
847 if (Al > 0 && pred >= (1<<Al))
848 pred = (1<<Al)-1;
849 } else {
850 pred = (int) (((Q10<<7) - num) / (Q10<<8));
851 if (Al > 0 && pred >= (1<<Al))
852 pred = (1<<Al)-1;
853 pred = -pred;
854 }
855 workspace[8] = (JCOEF) pred;
856 }
857 /* AC20 */
858 if ((Al=coef_bits[3]) != 0 && workspace[16] == 0) {
859 num = 9 * Q00 * (DC2 + DC8 - 2*DC5);
860 if (num >= 0) {
861 pred = (int) (((Q20<<7) + num) / (Q20<<8));
862 if (Al > 0 && pred >= (1<<Al))
863 pred = (1<<Al)-1;
864 } else {
865 pred = (int) (((Q20<<7) - num) / (Q20<<8));
866 if (Al > 0 && pred >= (1<<Al))
867 pred = (1<<Al)-1;
868 pred = -pred;
869 }
870 workspace[16] = (JCOEF) pred;
871 }
872 /* AC11 */
873 if ((Al=coef_bits[4]) != 0 && workspace[9] == 0) {
874 num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9);
875 if (num >= 0) {
876 pred = (int) (((Q11<<7) + num) / (Q11<<8));
877 if (Al > 0 && pred >= (1<<Al))
878 pred = (1<<Al)-1;
879 } else {
880 pred = (int) (((Q11<<7) - num) / (Q11<<8));
881 if (Al > 0 && pred >= (1<<Al))
882 pred = (1<<Al)-1;
883 pred = -pred;
884 }
885 workspace[9] = (JCOEF) pred;
886 }
887 /* AC02 */
888 if ((Al=coef_bits[5]) != 0 && workspace[2] == 0) {
889 num = 9 * Q00 * (DC4 + DC6 - 2*DC5);
890 if (num >= 0) {
891 pred = (int) (((Q02<<7) + num) / (Q02<<8));
892 if (Al > 0 && pred >= (1<<Al))
893 pred = (1<<Al)-1;
894 } else {
895 pred = (int) (((Q02<<7) - num) / (Q02<<8));
896 if (Al > 0 && pred >= (1<<Al))
897 pred = (1<<Al)-1;
898 pred = -pred;
899 }
900 workspace[2] = (JCOEF) pred;
901 }
902 /* OK, do the IDCT */
903 (*inverse_DCT) (cinfo, compptr, (JCOEFPTR) workspace,
904 output_ptr, output_col);
905 /* Advance for next column */
906 DC1 = DC2; DC2 = DC3;
907 DC4 = DC5; DC5 = DC6;
908 DC7 = DC8; DC8 = DC9;
909 buffer_ptr++, prev_block_row++, next_block_row++;
910 output_col += compptr->DCT_scaled_size;
911 }
912 output_ptr += compptr->DCT_scaled_size;
913 }
914 }
915
916 if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows)
917 return JPEG_ROW_COMPLETED;
918 return JPEG_SCAN_COMPLETED;
919}
920
921#endif /* BLOCK_SMOOTHING_SUPPORTED */
922
923
924/*
925 * Initialize coefficient buffer controller.
926 */
927
928GLOBAL(void)
929jinit_d_coef_controller (j_decompress_ptr cinfo, boolean need_full_buffer)
930{
931 my_coef_ptr coef;
932
933 coef = (my_coef_ptr)
934 (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE,
935 SIZEOF(my_coef_controller));
936 cinfo->coef = (struct jpeg_d_coef_controller *) coef;
937 coef->pub.start_input_pass = start_input_pass;
938 coef->pub.start_output_pass = start_output_pass;
939 coef->pub.column_left_boundary = 0;
940 coef->pub.column_right_boundary = 0;
941 coef->pub.MCU_columns_to_skip = 0;
942#ifdef BLOCK_SMOOTHING_SUPPORTED
943 coef->coef_bits_latch = NULL;
944#endif
945
946#ifdef ANDROID_TILE_BASED_DECODE
947 if (cinfo->tile_decode) {
948 if (cinfo->progressive_mode) {
949 /* Allocate one iMCU row virtual array, coef->whole_image[ci],
950 * for each color component, padded to a multiple of h_samp_factor
951 * DCT blocks in the horizontal direction.
952 */
953 int ci, access_rows;
954 jpeg_component_info *compptr;
955
956 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
957 ci++, compptr++) {
958 access_rows = compptr->v_samp_factor;
959 coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
960 ((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
961 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
962 (long) compptr->h_samp_factor),
963 (JDIMENSION) compptr->v_samp_factor, // one iMCU row
964 (JDIMENSION) access_rows);
965 }
966 coef->pub.consume_data_build_huffman_index =
967 consume_data_build_huffman_index_progressive;
968 coef->pub.consume_data = consume_data_multi_scan;
969 coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
970 coef->pub.decompress_data = decompress_onepass;
971 } else {
972 /* We only need a single-MCU buffer. */
973 JBLOCKROW buffer;
974 int i;
975
976 buffer = (JBLOCKROW)
977 (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
978 D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
979 for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
980 coef->MCU_buffer[i] = buffer + i;
981 }
982 coef->pub.consume_data_build_huffman_index =
983 consume_data_build_huffman_index_baseline;
984 coef->pub.consume_data = dummy_consume_data;
985 coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
986 coef->pub.decompress_data = decompress_onepass;
987 }
988 return;
989 }
990#endif
991
992 /* Create the coefficient buffer. */
993 if (need_full_buffer) {
994#ifdef D_MULTISCAN_FILES_SUPPORTED
995 /* Allocate a full-image virtual array for each component, */
996 /* padded to a multiple of samp_factor DCT blocks in each direction. */
997 /* Note we ask for a pre-zeroed array. */
998 int ci, access_rows;
999 jpeg_component_info *compptr;
1000
1001 for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
1002 ci++, compptr++) {
1003 access_rows = compptr->v_samp_factor;
1004#ifdef BLOCK_SMOOTHING_SUPPORTED
1005 /* If block smoothing could be used, need a bigger window */
1006 if (cinfo->progressive_mode)
1007 access_rows *= 3;
1008#endif
1009 coef->whole_image[ci] = (*cinfo->mem->request_virt_barray)
1010 ((j_common_ptr) cinfo, JPOOL_IMAGE, TRUE,
1011 (JDIMENSION) jround_up((long) compptr->width_in_blocks,
1012 (long) compptr->h_samp_factor),
1013 (JDIMENSION) jround_up((long) compptr->height_in_blocks,
1014 (long) compptr->v_samp_factor),
1015 (JDIMENSION) access_rows);
1016 }
1017 coef->pub.consume_data = consume_data;
1018 coef->pub.decompress_data = decompress_data;
1019 coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */
1020#else
1021 ERREXIT(cinfo, JERR_NOT_COMPILED);
1022#endif
1023 } else {
1024 /* We only need a single-MCU buffer. */
1025 JBLOCKROW buffer;
1026 int i;
1027
1028 buffer = (JBLOCKROW)
1029 (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE,
1030 D_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK));
1031 for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) {
1032 coef->MCU_buffer[i] = buffer + i;
1033 }
1034 coef->pub.consume_data = dummy_consume_data;
1035 coef->pub.decompress_data = decompress_onepass;
1036 coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */
1037 }
1038}