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