/src/libjpeg-turbo.2.0.x/jdcoefct.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * jdcoefct.c |
3 | | * |
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, D. R. Commander. |
9 | | * Copyright (C) 2015, 2020, Google, Inc. |
10 | | * For conditions of distribution and use, see the accompanying README.ijg |
11 | | * file. |
12 | | * |
13 | | * This file contains the coefficient buffer controller for decompression. |
14 | | * This controller is the top level of the JPEG decompressor proper. |
15 | | * The coefficient buffer lies between entropy decoding and inverse-DCT steps. |
16 | | * |
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. |
20 | | */ |
21 | | |
22 | | #include "jinclude.h" |
23 | | #include "jdcoefct.h" |
24 | | #include "jpegcomp.h" |
25 | | |
26 | | |
27 | | /* Forward declarations */ |
28 | | METHODDEF(int) decompress_onepass(j_decompress_ptr cinfo, |
29 | | JSAMPIMAGE output_buf); |
30 | | #ifdef D_MULTISCAN_FILES_SUPPORTED |
31 | | METHODDEF(int) decompress_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf); |
32 | | #endif |
33 | | #ifdef BLOCK_SMOOTHING_SUPPORTED |
34 | | LOCAL(boolean) smoothing_ok(j_decompress_ptr cinfo); |
35 | | METHODDEF(int) decompress_smooth_data(j_decompress_ptr cinfo, |
36 | | JSAMPIMAGE output_buf); |
37 | | #endif |
38 | | |
39 | | |
40 | | /* |
41 | | * Initialize for an input processing pass. |
42 | | */ |
43 | | |
44 | | METHODDEF(void) |
45 | | start_input_pass(j_decompress_ptr cinfo) |
46 | 239k | { |
47 | 239k | cinfo->input_iMCU_row = 0; |
48 | 239k | start_iMCU_row(cinfo); |
49 | 239k | } |
50 | | |
51 | | |
52 | | /* |
53 | | * Initialize for an output processing pass. |
54 | | */ |
55 | | |
56 | | METHODDEF(void) |
57 | | start_output_pass(j_decompress_ptr cinfo) |
58 | 21.4k | { |
59 | 21.4k | #ifdef BLOCK_SMOOTHING_SUPPORTED |
60 | 21.4k | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
61 | | |
62 | | /* If multipass, check to see whether to use block smoothing on this pass */ |
63 | 21.4k | if (coef->pub.coef_arrays != NULL) { |
64 | 15.9k | if (cinfo->do_block_smoothing && smoothing_ok(cinfo)) |
65 | 5.10k | coef->pub.decompress_data = decompress_smooth_data; |
66 | 10.8k | else |
67 | 10.8k | coef->pub.decompress_data = decompress_data; |
68 | 15.9k | } |
69 | 21.4k | #endif |
70 | 21.4k | cinfo->output_iMCU_row = 0; |
71 | 21.4k | } |
72 | | |
73 | | |
74 | | /* |
75 | | * Decompress and return some data in the single-pass case. |
76 | | * Always attempts to emit one fully interleaved MCU row ("iMCU" row). |
77 | | * Input and output must run in lockstep since we have only a one-MCU buffer. |
78 | | * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. |
79 | | * |
80 | | * NB: output_buf contains a plane for each component in image, |
81 | | * which we index according to the component's SOF position. |
82 | | */ |
83 | | |
84 | | METHODDEF(int) |
85 | | decompress_onepass(j_decompress_ptr cinfo, JSAMPIMAGE output_buf) |
86 | 591k | { |
87 | 591k | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
88 | 591k | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
89 | 591k | JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; |
90 | 591k | JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; |
91 | 591k | int blkn, ci, xindex, yindex, yoffset, useful_width; |
92 | 591k | JSAMPARRAY output_ptr; |
93 | 591k | JDIMENSION start_col, output_col; |
94 | 591k | jpeg_component_info *compptr; |
95 | 591k | inverse_DCT_method_ptr inverse_DCT; |
96 | | |
97 | | /* Loop to process as much as one whole iMCU row */ |
98 | 1.39M | for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; |
99 | 802k | yoffset++) { |
100 | 4.78M | for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col; |
101 | 3.97M | MCU_col_num++) { |
102 | | /* Try to fetch an MCU. Entropy decoder expects buffer to be zeroed. */ |
103 | 3.97M | jzero_far((void *)coef->MCU_buffer[0], |
104 | 3.97M | (size_t)(cinfo->blocks_in_MCU * sizeof(JBLOCK))); |
105 | 3.97M | if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) { |
106 | | /* Suspension forced; update state counters and exit */ |
107 | 0 | coef->MCU_vert_offset = yoffset; |
108 | 0 | coef->MCU_ctr = MCU_col_num; |
109 | 0 | return JPEG_SUSPENDED; |
110 | 0 | } |
111 | | |
112 | | /* Only perform the IDCT on blocks that are contained within the desired |
113 | | * cropping region. |
114 | | */ |
115 | 3.97M | if (MCU_col_num >= cinfo->master->first_iMCU_col && |
116 | 3.97M | MCU_col_num <= cinfo->master->last_iMCU_col) { |
117 | | /* Determine where data should go in output_buf and do the IDCT thing. |
118 | | * We skip dummy blocks at the right and bottom edges (but blkn gets |
119 | | * incremented past them!). Note the inner loop relies on having |
120 | | * allocated the MCU_buffer[] blocks sequentially. |
121 | | */ |
122 | 3.97M | blkn = 0; /* index of current DCT block within MCU */ |
123 | 8.36M | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
124 | 4.38M | compptr = cinfo->cur_comp_info[ci]; |
125 | | /* Don't bother to IDCT an uninteresting component. */ |
126 | 4.38M | if (!compptr->component_needed) { |
127 | 20.0k | blkn += compptr->MCU_blocks; |
128 | 20.0k | continue; |
129 | 20.0k | } |
130 | 4.36M | inverse_DCT = cinfo->idct->inverse_DCT[compptr->component_index]; |
131 | 4.36M | useful_width = (MCU_col_num < last_MCU_col) ? |
132 | 3.52M | compptr->MCU_width : compptr->last_col_width; |
133 | 4.36M | output_ptr = output_buf[compptr->component_index] + |
134 | 4.36M | yoffset * compptr->_DCT_scaled_size; |
135 | 4.36M | start_col = (MCU_col_num - cinfo->master->first_iMCU_col) * |
136 | 4.36M | compptr->MCU_sample_width; |
137 | 8.88M | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
138 | 4.52M | if (cinfo->input_iMCU_row < last_iMCU_row || |
139 | 4.52M | yoffset + yindex < compptr->last_row_height) { |
140 | 4.51M | output_col = start_col; |
141 | 9.22M | for (xindex = 0; xindex < useful_width; xindex++) { |
142 | 4.70M | (*inverse_DCT) (cinfo, compptr, |
143 | 4.70M | (JCOEFPTR)coef->MCU_buffer[blkn + xindex], |
144 | 4.70M | output_ptr, output_col); |
145 | 4.70M | output_col += compptr->_DCT_scaled_size; |
146 | 4.70M | } |
147 | 4.51M | } |
148 | 4.52M | blkn += compptr->MCU_width; |
149 | 4.52M | output_ptr += compptr->_DCT_scaled_size; |
150 | 4.52M | } |
151 | 4.36M | } |
152 | 3.97M | } |
153 | 3.97M | } |
154 | | /* Completed an MCU row, but perhaps not an iMCU row */ |
155 | 802k | coef->MCU_ctr = 0; |
156 | 802k | } |
157 | | /* Completed the iMCU row, advance counters for next one */ |
158 | 591k | cinfo->output_iMCU_row++; |
159 | 591k | if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) { |
160 | 589k | start_iMCU_row(cinfo); |
161 | 589k | return JPEG_ROW_COMPLETED; |
162 | 589k | } |
163 | | /* Completed the scan */ |
164 | 1.86k | (*cinfo->inputctl->finish_input_pass) (cinfo); |
165 | 1.86k | return JPEG_SCAN_COMPLETED; |
166 | 591k | } |
167 | | |
168 | | |
169 | | /* |
170 | | * Dummy consume-input routine for single-pass operation. |
171 | | */ |
172 | | |
173 | | METHODDEF(int) |
174 | | dummy_consume_data(j_decompress_ptr cinfo) |
175 | 0 | { |
176 | 0 | return JPEG_SUSPENDED; /* Always indicate nothing was done */ |
177 | 0 | } |
178 | | |
179 | | |
180 | | #ifdef D_MULTISCAN_FILES_SUPPORTED |
181 | | |
182 | | /* |
183 | | * Consume input data and store it in the full-image coefficient buffer. |
184 | | * We read as much as one fully interleaved MCU row ("iMCU" row) per call, |
185 | | * ie, v_samp_factor block rows for each component in the scan. |
186 | | * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. |
187 | | */ |
188 | | |
189 | | METHODDEF(int) |
190 | | consume_data(j_decompress_ptr cinfo) |
191 | 24.8M | { |
192 | 24.8M | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
193 | 24.8M | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
194 | 24.8M | int blkn, ci, xindex, yindex, yoffset; |
195 | 24.8M | JDIMENSION start_col; |
196 | 24.8M | JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; |
197 | 24.8M | JBLOCKROW buffer_ptr; |
198 | 24.8M | jpeg_component_info *compptr; |
199 | | |
200 | | /* Align the virtual buffers for the components used in this scan. */ |
201 | 53.0M | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
202 | 28.2M | compptr = cinfo->cur_comp_info[ci]; |
203 | 28.2M | buffer[ci] = (*cinfo->mem->access_virt_barray) |
204 | 28.2M | ((j_common_ptr)cinfo, coef->whole_image[compptr->component_index], |
205 | 28.2M | cinfo->input_iMCU_row * compptr->v_samp_factor, |
206 | 28.2M | (JDIMENSION)compptr->v_samp_factor, TRUE); |
207 | | /* Note: entropy decoder expects buffer to be zeroed, |
208 | | * but this is handled automatically by the memory manager |
209 | | * because we requested a pre-zeroed array. |
210 | | */ |
211 | 28.2M | } |
212 | | |
213 | | /* Loop to process one whole iMCU row */ |
214 | 53.5M | for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; |
215 | 28.7M | yoffset++) { |
216 | 229M | for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row; |
217 | 200M | MCU_col_num++) { |
218 | | /* Construct list of pointers to DCT blocks belonging to this MCU */ |
219 | 200M | blkn = 0; /* index of current DCT block within MCU */ |
220 | 418M | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
221 | 217M | compptr = cinfo->cur_comp_info[ci]; |
222 | 217M | start_col = MCU_col_num * compptr->MCU_width; |
223 | 441M | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
224 | 224M | buffer_ptr = buffer[ci][yindex + yoffset] + start_col; |
225 | 463M | for (xindex = 0; xindex < compptr->MCU_width; xindex++) { |
226 | 239M | coef->MCU_buffer[blkn++] = buffer_ptr++; |
227 | 239M | } |
228 | 224M | } |
229 | 217M | } |
230 | | /* Try to fetch the MCU. */ |
231 | 200M | if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) { |
232 | | /* Suspension forced; update state counters and exit */ |
233 | 0 | coef->MCU_vert_offset = yoffset; |
234 | 0 | coef->MCU_ctr = MCU_col_num; |
235 | 0 | return JPEG_SUSPENDED; |
236 | 0 | } |
237 | 200M | } |
238 | | /* Completed an MCU row, but perhaps not an iMCU row */ |
239 | 28.7M | coef->MCU_ctr = 0; |
240 | 28.7M | } |
241 | | /* Completed the iMCU row, advance counters for next one */ |
242 | 24.8M | if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) { |
243 | 24.7M | start_iMCU_row(cinfo); |
244 | 24.7M | return JPEG_ROW_COMPLETED; |
245 | 24.7M | } |
246 | | /* Completed the scan */ |
247 | 71.2k | (*cinfo->inputctl->finish_input_pass) (cinfo); |
248 | 71.2k | return JPEG_SCAN_COMPLETED; |
249 | 24.8M | } |
250 | | |
251 | | |
252 | | /* |
253 | | * Decompress and return some data in the multi-pass case. |
254 | | * Always attempts to emit one fully interleaved MCU row ("iMCU" row). |
255 | | * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. |
256 | | * |
257 | | * NB: output_buf contains a plane for each component in image. |
258 | | */ |
259 | | |
260 | | METHODDEF(int) |
261 | | decompress_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf) |
262 | 4.51M | { |
263 | 4.51M | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
264 | 4.51M | JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; |
265 | 4.51M | JDIMENSION block_num; |
266 | 4.51M | int ci, block_row, block_rows; |
267 | 4.51M | JBLOCKARRAY buffer; |
268 | 4.51M | JBLOCKROW buffer_ptr; |
269 | 4.51M | JSAMPARRAY output_ptr; |
270 | 4.51M | JDIMENSION output_col; |
271 | 4.51M | jpeg_component_info *compptr; |
272 | 4.51M | inverse_DCT_method_ptr inverse_DCT; |
273 | | |
274 | | /* Force some input to be done if we are getting ahead of the input. */ |
275 | 4.51M | while (cinfo->input_scan_number < cinfo->output_scan_number || |
276 | 4.51M | (cinfo->input_scan_number == cinfo->output_scan_number && |
277 | 4.51M | cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) { |
278 | 0 | if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) |
279 | 0 | return JPEG_SUSPENDED; |
280 | 0 | } |
281 | | |
282 | | /* OK, output from the virtual arrays. */ |
283 | 14.0M | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
284 | 9.49M | ci++, compptr++) { |
285 | | /* Don't bother to IDCT an uninteresting component. */ |
286 | 9.49M | if (!compptr->component_needed) |
287 | 184k | continue; |
288 | | /* Align the virtual buffer for this component. */ |
289 | 9.31M | buffer = (*cinfo->mem->access_virt_barray) |
290 | 9.31M | ((j_common_ptr)cinfo, coef->whole_image[ci], |
291 | 9.31M | cinfo->output_iMCU_row * compptr->v_samp_factor, |
292 | 9.31M | (JDIMENSION)compptr->v_samp_factor, FALSE); |
293 | | /* Count non-dummy DCT block rows in this iMCU row. */ |
294 | 9.31M | if (cinfo->output_iMCU_row < last_iMCU_row) |
295 | 9.28M | block_rows = compptr->v_samp_factor; |
296 | 22.0k | else { |
297 | | /* NB: can't use last_row_height here; it is input-side-dependent! */ |
298 | 22.0k | block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); |
299 | 22.0k | if (block_rows == 0) block_rows = compptr->v_samp_factor; |
300 | 22.0k | } |
301 | 9.31M | inverse_DCT = cinfo->idct->inverse_DCT[ci]; |
302 | 9.31M | output_ptr = output_buf[ci]; |
303 | | /* Loop over all DCT blocks to be processed. */ |
304 | 21.9M | for (block_row = 0; block_row < block_rows; block_row++) { |
305 | 12.6M | buffer_ptr = buffer[block_row] + cinfo->master->first_MCU_col[ci]; |
306 | 12.6M | output_col = 0; |
307 | 12.6M | for (block_num = cinfo->master->first_MCU_col[ci]; |
308 | 82.6M | block_num <= cinfo->master->last_MCU_col[ci]; block_num++) { |
309 | 70.0M | (*inverse_DCT) (cinfo, compptr, (JCOEFPTR)buffer_ptr, output_ptr, |
310 | 70.0M | output_col); |
311 | 70.0M | buffer_ptr++; |
312 | 70.0M | output_col += compptr->_DCT_scaled_size; |
313 | 70.0M | } |
314 | 12.6M | output_ptr += compptr->_DCT_scaled_size; |
315 | 12.6M | } |
316 | 9.31M | } |
317 | | |
318 | 4.51M | if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows) |
319 | 4.50M | return JPEG_ROW_COMPLETED; |
320 | 10.6k | return JPEG_SCAN_COMPLETED; |
321 | 4.51M | } |
322 | | |
323 | | #endif /* D_MULTISCAN_FILES_SUPPORTED */ |
324 | | |
325 | | |
326 | | #ifdef BLOCK_SMOOTHING_SUPPORTED |
327 | | |
328 | | /* |
329 | | * This code applies interblock smoothing as described by section K.8 |
330 | | * of the JPEG standard: the first 5 AC coefficients are estimated from |
331 | | * the DC values of a DCT block and its 8 neighboring blocks. |
332 | | * We apply smoothing only for progressive JPEG decoding, and only if |
333 | | * the coefficients it can estimate are not yet known to full precision. |
334 | | */ |
335 | | |
336 | | /* Natural-order array positions of the first 5 zigzag-order coefficients */ |
337 | 661k | #define Q01_POS 1 |
338 | 661k | #define Q10_POS 8 |
339 | 661k | #define Q20_POS 16 |
340 | 660k | #define Q11_POS 9 |
341 | 660k | #define Q02_POS 2 |
342 | | |
343 | | /* |
344 | | * Determine whether block smoothing is applicable and safe. |
345 | | * We also latch the current states of the coef_bits[] entries for the |
346 | | * AC coefficients; otherwise, if the input side of the decompressor |
347 | | * advances into a new scan, we might think the coefficients are known |
348 | | * more accurately than they really are. |
349 | | */ |
350 | | |
351 | | LOCAL(boolean) |
352 | | smoothing_ok(j_decompress_ptr cinfo) |
353 | 4.68k | { |
354 | 4.68k | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
355 | 4.68k | boolean smoothing_useful = FALSE; |
356 | 4.68k | int ci, coefi; |
357 | 4.68k | jpeg_component_info *compptr; |
358 | 4.68k | JQUANT_TBL *qtable; |
359 | 4.68k | int *coef_bits; |
360 | 4.68k | int *coef_bits_latch; |
361 | | |
362 | 4.68k | if (!cinfo->progressive_mode || cinfo->coef_bits == NULL) |
363 | 961 | return FALSE; |
364 | | |
365 | | /* Allocate latch area if not already done */ |
366 | 3.72k | if (coef->coef_bits_latch == NULL) |
367 | 3.72k | coef->coef_bits_latch = (int *) |
368 | 3.72k | (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, |
369 | 3.72k | cinfo->num_components * |
370 | 3.72k | (SAVED_COEFS * sizeof(int))); |
371 | 3.72k | coef_bits_latch = coef->coef_bits_latch; |
372 | | |
373 | 5.66k | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
374 | 4.28k | ci++, compptr++) { |
375 | | /* All components' quantization values must already be latched. */ |
376 | 4.28k | if ((qtable = compptr->quant_table) == NULL) |
377 | 917 | return FALSE; |
378 | | /* Verify DC & first 5 AC quantizers are nonzero to avoid zero-divide. */ |
379 | 3.36k | if (qtable->quantval[0] == 0 || |
380 | 3.36k | qtable->quantval[Q01_POS] == 0 || |
381 | 3.36k | qtable->quantval[Q10_POS] == 0 || |
382 | 3.36k | qtable->quantval[Q20_POS] == 0 || |
383 | 3.36k | qtable->quantval[Q11_POS] == 0 || |
384 | 3.36k | qtable->quantval[Q02_POS] == 0) |
385 | 1.19k | return FALSE; |
386 | | /* DC values must be at least partly known for all components. */ |
387 | 2.17k | coef_bits = cinfo->coef_bits[ci]; |
388 | 2.17k | if (coef_bits[0] < 0) |
389 | 232 | return FALSE; |
390 | | /* Block smoothing is helpful if some AC coefficients remain inaccurate. */ |
391 | 11.6k | for (coefi = 1; coefi <= 5; coefi++) { |
392 | 9.69k | coef_bits_latch[coefi] = coef_bits[coefi]; |
393 | 9.69k | if (coef_bits[coefi] != 0) |
394 | 9.28k | smoothing_useful = TRUE; |
395 | 9.69k | } |
396 | 1.93k | coef_bits_latch += SAVED_COEFS; |
397 | 1.93k | } |
398 | | |
399 | 1.37k | return smoothing_useful; |
400 | 3.72k | } |
401 | | |
402 | | |
403 | | /* |
404 | | * Variant of decompress_data for use when doing block smoothing. |
405 | | */ |
406 | | |
407 | | METHODDEF(int) |
408 | | decompress_smooth_data(j_decompress_ptr cinfo, JSAMPIMAGE output_buf) |
409 | 551k | { |
410 | 551k | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
411 | 551k | JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; |
412 | 551k | JDIMENSION block_num, last_block_column; |
413 | 551k | int ci, block_row, block_rows, access_rows; |
414 | 551k | JBLOCKARRAY buffer; |
415 | 551k | JBLOCKROW buffer_ptr, prev_block_row, next_block_row; |
416 | 551k | JSAMPARRAY output_ptr; |
417 | 551k | JDIMENSION output_col; |
418 | 551k | jpeg_component_info *compptr; |
419 | 551k | inverse_DCT_method_ptr inverse_DCT; |
420 | 551k | boolean first_row, last_row; |
421 | 551k | JCOEF *workspace; |
422 | 551k | int *coef_bits; |
423 | 551k | JQUANT_TBL *quanttbl; |
424 | 551k | JLONG Q00, Q01, Q02, Q10, Q11, Q20, num; |
425 | 551k | int DC1, DC2, DC3, DC4, DC5, DC6, DC7, DC8, DC9; |
426 | 551k | int Al, pred; |
427 | | |
428 | | /* Keep a local variable to avoid looking it up more than once */ |
429 | 551k | workspace = coef->workspace; |
430 | | |
431 | | /* Force some input to be done if we are getting ahead of the input. */ |
432 | 551k | while (cinfo->input_scan_number <= cinfo->output_scan_number && |
433 | 551k | !cinfo->inputctl->eoi_reached) { |
434 | 0 | if (cinfo->input_scan_number == cinfo->output_scan_number) { |
435 | | /* If input is working on current scan, we ordinarily want it to |
436 | | * have completed the current row. But if input scan is DC, |
437 | | * we want it to keep one row ahead so that next block row's DC |
438 | | * values are up to date. |
439 | | */ |
440 | 0 | JDIMENSION delta = (cinfo->Ss == 0) ? 1 : 0; |
441 | 0 | if (cinfo->input_iMCU_row > cinfo->output_iMCU_row + delta) |
442 | 0 | break; |
443 | 0 | } |
444 | 0 | if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) |
445 | 0 | return JPEG_SUSPENDED; |
446 | 0 | } |
447 | | |
448 | | /* OK, output from the virtual arrays. */ |
449 | 1.21M | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
450 | 668k | ci++, compptr++) { |
451 | | /* Don't bother to IDCT an uninteresting component. */ |
452 | 668k | if (!compptr->component_needed) |
453 | 9.97k | continue; |
454 | | /* Count non-dummy DCT block rows in this iMCU row. */ |
455 | 658k | if (cinfo->output_iMCU_row < last_iMCU_row) { |
456 | 656k | block_rows = compptr->v_samp_factor; |
457 | 656k | access_rows = block_rows * 2; /* this and next iMCU row */ |
458 | 656k | last_row = FALSE; |
459 | 656k | } else { |
460 | | /* NB: can't use last_row_height here; it is input-side-dependent! */ |
461 | 1.73k | block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); |
462 | 1.73k | if (block_rows == 0) block_rows = compptr->v_samp_factor; |
463 | 1.73k | access_rows = block_rows; /* this iMCU row only */ |
464 | 1.73k | last_row = TRUE; |
465 | 1.73k | } |
466 | | /* Align the virtual buffer for this component. */ |
467 | 658k | if (cinfo->output_iMCU_row > 0) { |
468 | 656k | access_rows += compptr->v_samp_factor; /* prior iMCU row too */ |
469 | 656k | buffer = (*cinfo->mem->access_virt_barray) |
470 | 656k | ((j_common_ptr)cinfo, coef->whole_image[ci], |
471 | 656k | (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor, |
472 | 656k | (JDIMENSION)access_rows, FALSE); |
473 | 656k | buffer += compptr->v_samp_factor; /* point to current iMCU row */ |
474 | 656k | first_row = FALSE; |
475 | 656k | } else { |
476 | 1.73k | buffer = (*cinfo->mem->access_virt_barray) |
477 | 1.73k | ((j_common_ptr)cinfo, coef->whole_image[ci], |
478 | 1.73k | (JDIMENSION)0, (JDIMENSION)access_rows, FALSE); |
479 | 1.73k | first_row = TRUE; |
480 | 1.73k | } |
481 | | /* Fetch component-dependent info */ |
482 | 658k | coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS); |
483 | 658k | quanttbl = compptr->quant_table; |
484 | 658k | Q00 = quanttbl->quantval[0]; |
485 | 658k | Q01 = quanttbl->quantval[Q01_POS]; |
486 | 658k | Q10 = quanttbl->quantval[Q10_POS]; |
487 | 658k | Q20 = quanttbl->quantval[Q20_POS]; |
488 | 658k | Q11 = quanttbl->quantval[Q11_POS]; |
489 | 658k | Q02 = quanttbl->quantval[Q02_POS]; |
490 | 658k | inverse_DCT = cinfo->idct->inverse_DCT[ci]; |
491 | 658k | output_ptr = output_buf[ci]; |
492 | | /* Loop over all DCT blocks to be processed. */ |
493 | 1.60M | for (block_row = 0; block_row < block_rows; block_row++) { |
494 | 948k | buffer_ptr = buffer[block_row] + cinfo->master->first_MCU_col[ci]; |
495 | 948k | if (first_row && block_row == 0) |
496 | 1.73k | prev_block_row = buffer_ptr; |
497 | 947k | else |
498 | 947k | prev_block_row = buffer[block_row - 1] + |
499 | 947k | cinfo->master->first_MCU_col[ci]; |
500 | 948k | if (last_row && block_row == block_rows - 1) |
501 | 1.73k | next_block_row = buffer_ptr; |
502 | 947k | else |
503 | 947k | next_block_row = buffer[block_row + 1] + |
504 | 947k | cinfo->master->first_MCU_col[ci]; |
505 | | /* We fetch the surrounding DC values using a sliding-register approach. |
506 | | * Initialize all nine here so as to do the right thing on narrow pics. |
507 | | */ |
508 | 948k | DC1 = DC2 = DC3 = (int)prev_block_row[0][0]; |
509 | 948k | DC4 = DC5 = DC6 = (int)buffer_ptr[0][0]; |
510 | 948k | DC7 = DC8 = DC9 = (int)next_block_row[0][0]; |
511 | 948k | output_col = 0; |
512 | 948k | last_block_column = compptr->width_in_blocks - 1; |
513 | 948k | for (block_num = cinfo->master->first_MCU_col[ci]; |
514 | 9.50M | block_num <= cinfo->master->last_MCU_col[ci]; block_num++) { |
515 | | /* Fetch current DCT block into workspace so we can modify it. */ |
516 | 8.55M | jcopy_block_row(buffer_ptr, (JBLOCKROW)workspace, (JDIMENSION)1); |
517 | | /* Update DC values */ |
518 | 8.55M | if (block_num < last_block_column) { |
519 | 7.60M | DC3 = (int)prev_block_row[1][0]; |
520 | 7.60M | DC6 = (int)buffer_ptr[1][0]; |
521 | 7.60M | DC9 = (int)next_block_row[1][0]; |
522 | 7.60M | } |
523 | | /* Compute coefficient estimates per K.8. |
524 | | * An estimate is applied only if coefficient is still zero, |
525 | | * and is not known to be fully accurate. |
526 | | */ |
527 | | /* AC01 */ |
528 | 8.55M | if ((Al = coef_bits[1]) != 0 && workspace[1] == 0) { |
529 | 8.23M | num = 36 * Q00 * (DC4 - DC6); |
530 | 8.23M | if (num >= 0) { |
531 | 6.64M | pred = (int)(((Q01 << 7) + num) / (Q01 << 8)); |
532 | 6.64M | if (Al > 0 && pred >= (1 << Al)) |
533 | 489k | pred = (1 << Al) - 1; |
534 | 6.64M | } else { |
535 | 1.58M | pred = (int)(((Q01 << 7) - num) / (Q01 << 8)); |
536 | 1.58M | if (Al > 0 && pred >= (1 << Al)) |
537 | 393k | pred = (1 << Al) - 1; |
538 | 1.58M | pred = -pred; |
539 | 1.58M | } |
540 | 8.23M | workspace[1] = (JCOEF)pred; |
541 | 8.23M | } |
542 | | /* AC10 */ |
543 | 8.55M | if ((Al = coef_bits[2]) != 0 && workspace[8] == 0) { |
544 | 8.32M | num = 36 * Q00 * (DC2 - DC8); |
545 | 8.32M | if (num >= 0) { |
546 | 5.93M | pred = (int)(((Q10 << 7) + num) / (Q10 << 8)); |
547 | 5.93M | if (Al > 0 && pred >= (1 << Al)) |
548 | 796k | pred = (1 << Al) - 1; |
549 | 5.93M | } else { |
550 | 2.38M | pred = (int)(((Q10 << 7) - num) / (Q10 << 8)); |
551 | 2.38M | if (Al > 0 && pred >= (1 << Al)) |
552 | 885k | pred = (1 << Al) - 1; |
553 | 2.38M | pred = -pred; |
554 | 2.38M | } |
555 | 8.32M | workspace[8] = (JCOEF)pred; |
556 | 8.32M | } |
557 | | /* AC20 */ |
558 | 8.55M | if ((Al = coef_bits[3]) != 0 && workspace[16] == 0) { |
559 | 8.32M | num = 9 * Q00 * (DC2 + DC8 - 2 * DC5); |
560 | 8.32M | if (num >= 0) { |
561 | 5.69M | pred = (int)(((Q20 << 7) + num) / (Q20 << 8)); |
562 | 5.69M | if (Al > 0 && pred >= (1 << Al)) |
563 | 653k | pred = (1 << Al) - 1; |
564 | 5.69M | } else { |
565 | 2.62M | pred = (int)(((Q20 << 7) - num) / (Q20 << 8)); |
566 | 2.62M | if (Al > 0 && pred >= (1 << Al)) |
567 | 651k | pred = (1 << Al) - 1; |
568 | 2.62M | pred = -pred; |
569 | 2.62M | } |
570 | 8.32M | workspace[16] = (JCOEF)pred; |
571 | 8.32M | } |
572 | | /* AC11 */ |
573 | 8.55M | if ((Al = coef_bits[4]) != 0 && workspace[9] == 0) { |
574 | 8.31M | num = 5 * Q00 * (DC1 - DC3 - DC7 + DC9); |
575 | 8.31M | if (num >= 0) { |
576 | 6.99M | pred = (int)(((Q11 << 7) + num) / (Q11 << 8)); |
577 | 6.99M | if (Al > 0 && pred >= (1 << Al)) |
578 | 267k | pred = (1 << Al) - 1; |
579 | 6.99M | } else { |
580 | 1.31M | pred = (int)(((Q11 << 7) - num) / (Q11 << 8)); |
581 | 1.31M | if (Al > 0 && pred >= (1 << Al)) |
582 | 278k | pred = (1 << Al) - 1; |
583 | 1.31M | pred = -pred; |
584 | 1.31M | } |
585 | 8.31M | workspace[9] = (JCOEF)pred; |
586 | 8.31M | } |
587 | | /* AC02 */ |
588 | 8.55M | if ((Al = coef_bits[5]) != 0 && workspace[2] == 0) { |
589 | 8.34M | num = 9 * Q00 * (DC4 + DC6 - 2 * DC5); |
590 | 8.34M | if (num >= 0) { |
591 | 5.81M | pred = (int)(((Q02 << 7) + num) / (Q02 << 8)); |
592 | 5.81M | if (Al > 0 && pred >= (1 << Al)) |
593 | 341k | pred = (1 << Al) - 1; |
594 | 5.81M | } else { |
595 | 2.52M | pred = (int)(((Q02 << 7) - num) / (Q02 << 8)); |
596 | 2.52M | if (Al > 0 && pred >= (1 << Al)) |
597 | 337k | pred = (1 << Al) - 1; |
598 | 2.52M | pred = -pred; |
599 | 2.52M | } |
600 | 8.34M | workspace[2] = (JCOEF)pred; |
601 | 8.34M | } |
602 | | /* OK, do the IDCT */ |
603 | 8.55M | (*inverse_DCT) (cinfo, compptr, (JCOEFPTR)workspace, output_ptr, |
604 | 8.55M | output_col); |
605 | | /* Advance for next column */ |
606 | 8.55M | DC1 = DC2; DC2 = DC3; |
607 | 8.55M | DC4 = DC5; DC5 = DC6; |
608 | 8.55M | DC7 = DC8; DC8 = DC9; |
609 | 8.55M | buffer_ptr++, prev_block_row++, next_block_row++; |
610 | 8.55M | output_col += compptr->_DCT_scaled_size; |
611 | 8.55M | } |
612 | 948k | output_ptr += compptr->_DCT_scaled_size; |
613 | 948k | } |
614 | 658k | } |
615 | | |
616 | 551k | if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows) |
617 | 550k | return JPEG_ROW_COMPLETED; |
618 | 1.33k | return JPEG_SCAN_COMPLETED; |
619 | 551k | } |
620 | | |
621 | | #endif /* BLOCK_SMOOTHING_SUPPORTED */ |
622 | | |
623 | | |
624 | | /* |
625 | | * Initialize coefficient buffer controller. |
626 | | */ |
627 | | |
628 | | GLOBAL(void) |
629 | | jinit_d_coef_controller(j_decompress_ptr cinfo, boolean need_full_buffer) |
630 | 30.3k | { |
631 | 30.3k | my_coef_ptr coef; |
632 | | |
633 | 30.3k | coef = (my_coef_ptr) |
634 | 30.3k | (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, |
635 | 30.3k | sizeof(my_coef_controller)); |
636 | 30.3k | cinfo->coef = (struct jpeg_d_coef_controller *)coef; |
637 | 30.3k | coef->pub.start_input_pass = start_input_pass; |
638 | 30.3k | coef->pub.start_output_pass = start_output_pass; |
639 | 30.3k | #ifdef BLOCK_SMOOTHING_SUPPORTED |
640 | 30.3k | coef->coef_bits_latch = NULL; |
641 | 30.3k | #endif |
642 | | |
643 | | /* Create the coefficient buffer. */ |
644 | 30.3k | if (need_full_buffer) { |
645 | 24.4k | #ifdef D_MULTISCAN_FILES_SUPPORTED |
646 | | /* Allocate a full-image virtual array for each component, */ |
647 | | /* padded to a multiple of samp_factor DCT blocks in each direction. */ |
648 | | /* Note we ask for a pre-zeroed array. */ |
649 | 24.4k | int ci, access_rows; |
650 | 24.4k | jpeg_component_info *compptr; |
651 | | |
652 | 73.7k | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
653 | 49.3k | ci++, compptr++) { |
654 | 49.3k | access_rows = compptr->v_samp_factor; |
655 | 49.3k | #ifdef BLOCK_SMOOTHING_SUPPORTED |
656 | | /* If block smoothing could be used, need a bigger window */ |
657 | 49.3k | if (cinfo->progressive_mode) |
658 | 27.7k | access_rows *= 3; |
659 | 49.3k | #endif |
660 | 49.3k | coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) |
661 | 49.3k | ((j_common_ptr)cinfo, JPOOL_IMAGE, TRUE, |
662 | 49.3k | (JDIMENSION)jround_up((long)compptr->width_in_blocks, |
663 | 49.3k | (long)compptr->h_samp_factor), |
664 | 49.3k | (JDIMENSION)jround_up((long)compptr->height_in_blocks, |
665 | 49.3k | (long)compptr->v_samp_factor), |
666 | 49.3k | (JDIMENSION)access_rows); |
667 | 49.3k | } |
668 | 24.4k | coef->pub.consume_data = consume_data; |
669 | 24.4k | coef->pub.decompress_data = decompress_data; |
670 | 24.4k | coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */ |
671 | | #else |
672 | | ERREXIT(cinfo, JERR_NOT_COMPILED); |
673 | | #endif |
674 | 24.4k | } else { |
675 | | /* We only need a single-MCU buffer. */ |
676 | 5.93k | JBLOCKROW buffer; |
677 | 5.93k | int i; |
678 | | |
679 | 5.93k | buffer = (JBLOCKROW) |
680 | 5.93k | (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, |
681 | 5.93k | D_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK)); |
682 | 65.2k | for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) { |
683 | 59.3k | coef->MCU_buffer[i] = buffer + i; |
684 | 59.3k | } |
685 | 5.93k | coef->pub.consume_data = dummy_consume_data; |
686 | 5.93k | coef->pub.decompress_data = decompress_onepass; |
687 | 5.93k | coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */ |
688 | 5.93k | } |
689 | | |
690 | | /* Allocate the workspace buffer */ |
691 | 30.3k | coef->workspace = (JCOEF *) |
692 | 30.3k | (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, |
693 | 30.3k | sizeof(JCOEF) * DCTSIZE2); |
694 | 30.3k | } |