1 | /* |
2 | * jccoefct.c |
3 | * |
4 | * This file was part of the Independent JPEG Group's software: |
5 | * Copyright (C) 1994-1997, Thomas G. Lane. |
6 | * It was modified by The libjpeg-turbo Project to include only code and |
7 | * information relevant to libjpeg-turbo. |
8 | * For conditions of distribution and use, see the accompanying README.ijg |
9 | * file. |
10 | * |
11 | * This file contains the coefficient buffer controller for compression. |
12 | * This controller is the top level of the JPEG compressor proper. |
13 | * The coefficient buffer lies between forward-DCT and entropy encoding steps. |
14 | */ |
15 | |
16 | #define JPEG_INTERNALS |
17 | #include "jinclude.h" |
18 | #include "jpeglib.h" |
19 | |
20 | |
21 | /* We use a full-image coefficient buffer when doing Huffman optimization, |
22 | * and also for writing multiple-scan JPEG files. In all cases, the DCT |
23 | * step is run during the first pass, and subsequent passes need only read |
24 | * the buffered coefficients. |
25 | */ |
26 | #ifdef ENTROPY_OPT_SUPPORTED |
27 | #define FULL_COEF_BUFFER_SUPPORTED |
28 | #else |
29 | #ifdef C_MULTISCAN_FILES_SUPPORTED |
30 | #define FULL_COEF_BUFFER_SUPPORTED |
31 | #endif |
32 | #endif |
33 | |
34 | |
35 | /* Private buffer controller object */ |
36 | |
37 | typedef struct { |
38 | struct jpeg_c_coef_controller pub; /* public fields */ |
39 | |
40 | JDIMENSION iMCU_row_num; /* iMCU row # within image */ |
41 | JDIMENSION mcu_ctr; /* counts MCUs processed in current row */ |
42 | int MCU_vert_offset; /* counts MCU rows within iMCU row */ |
43 | int MCU_rows_per_iMCU_row; /* number of such rows needed */ |
44 | |
45 | /* For single-pass compression, it's sufficient to buffer just one MCU |
46 | * (although this may prove a bit slow in practice). We allocate a |
47 | * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each |
48 | * MCU constructed and sent. In multi-pass modes, this array points to the |
49 | * current MCU's blocks within the virtual arrays. |
50 | */ |
51 | JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU]; |
52 | |
53 | /* In multi-pass modes, we need a virtual block array for each component. */ |
54 | jvirt_barray_ptr whole_image[MAX_COMPONENTS]; |
55 | } my_coef_controller; |
56 | |
57 | typedef my_coef_controller *my_coef_ptr; |
58 | |
59 | |
60 | /* Forward declarations */ |
61 | METHODDEF(boolean) compress_data(j_compress_ptr cinfo, JSAMPIMAGE input_buf); |
62 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
63 | METHODDEF(boolean) compress_first_pass(j_compress_ptr cinfo, |
64 | JSAMPIMAGE input_buf); |
65 | METHODDEF(boolean) compress_output(j_compress_ptr cinfo, JSAMPIMAGE input_buf); |
66 | #endif |
67 | |
68 | |
69 | LOCAL(void) |
70 | start_iMCU_row(j_compress_ptr cinfo) |
71 | /* Reset within-iMCU-row counters for a new row */ |
72 | { |
73 | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
74 | |
75 | /* In an interleaved scan, an MCU row is the same as an iMCU row. |
76 | * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. |
77 | * But at the bottom of the image, process only what's left. |
78 | */ |
79 | if (cinfo->comps_in_scan > 1) { |
80 | coef->MCU_rows_per_iMCU_row = 1; |
81 | } else { |
82 | if (coef->iMCU_row_num < (cinfo->total_iMCU_rows - 1)) |
83 | coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; |
84 | else |
85 | coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; |
86 | } |
87 | |
88 | coef->mcu_ctr = 0; |
89 | coef->MCU_vert_offset = 0; |
90 | } |
91 | |
92 | |
93 | /* |
94 | * Initialize for a processing pass. |
95 | */ |
96 | |
97 | METHODDEF(void) |
98 | start_pass_coef(j_compress_ptr cinfo, J_BUF_MODE pass_mode) |
99 | { |
100 | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
101 | |
102 | coef->iMCU_row_num = 0; |
103 | start_iMCU_row(cinfo); |
104 | |
105 | switch (pass_mode) { |
106 | case JBUF_PASS_THRU: |
107 | if (coef->whole_image[0] != NULL) |
108 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
109 | coef->pub.compress_data = compress_data; |
110 | break; |
111 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
112 | case JBUF_SAVE_AND_PASS: |
113 | if (coef->whole_image[0] == NULL) |
114 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
115 | coef->pub.compress_data = compress_first_pass; |
116 | break; |
117 | case JBUF_CRANK_DEST: |
118 | if (coef->whole_image[0] == NULL) |
119 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
120 | coef->pub.compress_data = compress_output; |
121 | break; |
122 | #endif |
123 | default: |
124 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
125 | break; |
126 | } |
127 | } |
128 | |
129 | |
130 | /* |
131 | * Process some data in the single-pass case. |
132 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
133 | * per call, ie, v_samp_factor block rows for each component in the image. |
134 | * Returns TRUE if the iMCU row is completed, FALSE if suspended. |
135 | * |
136 | * NB: input_buf contains a plane for each component in image, |
137 | * which we index according to the component's SOF position. |
138 | */ |
139 | |
140 | METHODDEF(boolean) |
141 | compress_data(j_compress_ptr cinfo, JSAMPIMAGE input_buf) |
142 | { |
143 | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
144 | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
145 | JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; |
146 | JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; |
147 | int blkn, bi, ci, yindex, yoffset, blockcnt; |
148 | JDIMENSION ypos, xpos; |
149 | jpeg_component_info *compptr; |
150 | |
151 | /* Loop to write as much as one whole iMCU row */ |
152 | for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; |
153 | yoffset++) { |
154 | for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col; |
155 | MCU_col_num++) { |
156 | /* Determine where data comes from in input_buf and do the DCT thing. |
157 | * Each call on forward_DCT processes a horizontal row of DCT blocks |
158 | * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks |
159 | * sequentially. Dummy blocks at the right or bottom edge are filled in |
160 | * specially. The data in them does not matter for image reconstruction, |
161 | * so we fill them with values that will encode to the smallest amount of |
162 | * data, viz: all zeroes in the AC entries, DC entries equal to previous |
163 | * block's DC value. (Thanks to Thomas Kinsman for this idea.) |
164 | */ |
165 | blkn = 0; |
166 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
167 | compptr = cinfo->cur_comp_info[ci]; |
168 | blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width : |
169 | compptr->last_col_width; |
170 | xpos = MCU_col_num * compptr->MCU_sample_width; |
171 | ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */ |
172 | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
173 | if (coef->iMCU_row_num < last_iMCU_row || |
174 | yoffset + yindex < compptr->last_row_height) { |
175 | (*cinfo->fdct->forward_DCT) (cinfo, compptr, |
176 | input_buf[compptr->component_index], |
177 | coef->MCU_buffer[blkn], |
178 | ypos, xpos, (JDIMENSION)blockcnt); |
179 | if (blockcnt < compptr->MCU_width) { |
180 | /* Create some dummy blocks at the right edge of the image. */ |
181 | jzero_far((void *)coef->MCU_buffer[blkn + blockcnt], |
182 | (compptr->MCU_width - blockcnt) * sizeof(JBLOCK)); |
183 | for (bi = blockcnt; bi < compptr->MCU_width; bi++) { |
184 | coef->MCU_buffer[blkn + bi][0][0] = |
185 | coef->MCU_buffer[blkn + bi - 1][0][0]; |
186 | } |
187 | } |
188 | } else { |
189 | /* Create a row of dummy blocks at the bottom of the image. */ |
190 | jzero_far((void *)coef->MCU_buffer[blkn], |
191 | compptr->MCU_width * sizeof(JBLOCK)); |
192 | for (bi = 0; bi < compptr->MCU_width; bi++) { |
193 | coef->MCU_buffer[blkn + bi][0][0] = |
194 | coef->MCU_buffer[blkn - 1][0][0]; |
195 | } |
196 | } |
197 | blkn += compptr->MCU_width; |
198 | ypos += DCTSIZE; |
199 | } |
200 | } |
201 | /* Try to write the MCU. In event of a suspension failure, we will |
202 | * re-DCT the MCU on restart (a bit inefficient, could be fixed...) |
203 | */ |
204 | if (!(*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { |
205 | /* Suspension forced; update state counters and exit */ |
206 | coef->MCU_vert_offset = yoffset; |
207 | coef->mcu_ctr = MCU_col_num; |
208 | return FALSE; |
209 | } |
210 | } |
211 | /* Completed an MCU row, but perhaps not an iMCU row */ |
212 | coef->mcu_ctr = 0; |
213 | } |
214 | /* Completed the iMCU row, advance counters for next one */ |
215 | coef->iMCU_row_num++; |
216 | start_iMCU_row(cinfo); |
217 | return TRUE; |
218 | } |
219 | |
220 | |
221 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
222 | |
223 | /* |
224 | * Process some data in the first pass of a multi-pass case. |
225 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
226 | * per call, ie, v_samp_factor block rows for each component in the image. |
227 | * This amount of data is read from the source buffer, DCT'd and quantized, |
228 | * and saved into the virtual arrays. We also generate suitable dummy blocks |
229 | * as needed at the right and lower edges. (The dummy blocks are constructed |
230 | * in the virtual arrays, which have been padded appropriately.) This makes |
231 | * it possible for subsequent passes not to worry about real vs. dummy blocks. |
232 | * |
233 | * We must also emit the data to the entropy encoder. This is conveniently |
234 | * done by calling compress_output() after we've loaded the current strip |
235 | * of the virtual arrays. |
236 | * |
237 | * NB: input_buf contains a plane for each component in image. All |
238 | * components are DCT'd and loaded into the virtual arrays in this pass. |
239 | * However, it may be that only a subset of the components are emitted to |
240 | * the entropy encoder during this first pass; be careful about looking |
241 | * at the scan-dependent variables (MCU dimensions, etc). |
242 | */ |
243 | |
244 | METHODDEF(boolean) |
245 | compress_first_pass(j_compress_ptr cinfo, JSAMPIMAGE input_buf) |
246 | { |
247 | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
248 | JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; |
249 | JDIMENSION blocks_across, MCUs_across, MCUindex; |
250 | int bi, ci, h_samp_factor, block_row, block_rows, ndummy; |
251 | JCOEF lastDC; |
252 | jpeg_component_info *compptr; |
253 | JBLOCKARRAY buffer; |
254 | JBLOCKROW thisblockrow, lastblockrow; |
255 | |
256 | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
257 | ci++, compptr++) { |
258 | /* Align the virtual buffer for this component. */ |
259 | buffer = (*cinfo->mem->access_virt_barray) |
260 | ((j_common_ptr)cinfo, coef->whole_image[ci], |
261 | coef->iMCU_row_num * compptr->v_samp_factor, |
262 | (JDIMENSION)compptr->v_samp_factor, TRUE); |
263 | /* Count non-dummy DCT block rows in this iMCU row. */ |
264 | if (coef->iMCU_row_num < last_iMCU_row) |
265 | block_rows = compptr->v_samp_factor; |
266 | else { |
267 | /* NB: can't use last_row_height here, since may not be set! */ |
268 | block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); |
269 | if (block_rows == 0) block_rows = compptr->v_samp_factor; |
270 | } |
271 | blocks_across = compptr->width_in_blocks; |
272 | h_samp_factor = compptr->h_samp_factor; |
273 | /* Count number of dummy blocks to be added at the right margin. */ |
274 | ndummy = (int)(blocks_across % h_samp_factor); |
275 | if (ndummy > 0) |
276 | ndummy = h_samp_factor - ndummy; |
277 | /* Perform DCT for all non-dummy blocks in this iMCU row. Each call |
278 | * on forward_DCT processes a complete horizontal row of DCT blocks. |
279 | */ |
280 | for (block_row = 0; block_row < block_rows; block_row++) { |
281 | thisblockrow = buffer[block_row]; |
282 | (*cinfo->fdct->forward_DCT) (cinfo, compptr, |
283 | input_buf[ci], thisblockrow, |
284 | (JDIMENSION)(block_row * DCTSIZE), |
285 | (JDIMENSION)0, blocks_across); |
286 | if (ndummy > 0) { |
287 | /* Create dummy blocks at the right edge of the image. */ |
288 | thisblockrow += blocks_across; /* => first dummy block */ |
289 | jzero_far((void *)thisblockrow, ndummy * sizeof(JBLOCK)); |
290 | lastDC = thisblockrow[-1][0]; |
291 | for (bi = 0; bi < ndummy; bi++) { |
292 | thisblockrow[bi][0] = lastDC; |
293 | } |
294 | } |
295 | } |
296 | /* If at end of image, create dummy block rows as needed. |
297 | * The tricky part here is that within each MCU, we want the DC values |
298 | * of the dummy blocks to match the last real block's DC value. |
299 | * This squeezes a few more bytes out of the resulting file... |
300 | */ |
301 | if (coef->iMCU_row_num == last_iMCU_row) { |
302 | blocks_across += ndummy; /* include lower right corner */ |
303 | MCUs_across = blocks_across / h_samp_factor; |
304 | for (block_row = block_rows; block_row < compptr->v_samp_factor; |
305 | block_row++) { |
306 | thisblockrow = buffer[block_row]; |
307 | lastblockrow = buffer[block_row - 1]; |
308 | jzero_far((void *)thisblockrow, |
309 | (size_t)(blocks_across * sizeof(JBLOCK))); |
310 | for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) { |
311 | lastDC = lastblockrow[h_samp_factor - 1][0]; |
312 | for (bi = 0; bi < h_samp_factor; bi++) { |
313 | thisblockrow[bi][0] = lastDC; |
314 | } |
315 | thisblockrow += h_samp_factor; /* advance to next MCU in row */ |
316 | lastblockrow += h_samp_factor; |
317 | } |
318 | } |
319 | } |
320 | } |
321 | /* NB: compress_output will increment iMCU_row_num if successful. |
322 | * A suspension return will result in redoing all the work above next time. |
323 | */ |
324 | |
325 | /* Emit data to the entropy encoder, sharing code with subsequent passes */ |
326 | return compress_output(cinfo, input_buf); |
327 | } |
328 | |
329 | |
330 | /* |
331 | * Process some data in subsequent passes of a multi-pass case. |
332 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
333 | * per call, ie, v_samp_factor block rows for each component in the scan. |
334 | * The data is obtained from the virtual arrays and fed to the entropy coder. |
335 | * Returns TRUE if the iMCU row is completed, FALSE if suspended. |
336 | * |
337 | * NB: input_buf is ignored; it is likely to be a NULL pointer. |
338 | */ |
339 | |
340 | METHODDEF(boolean) |
341 | compress_output(j_compress_ptr cinfo, JSAMPIMAGE input_buf) |
342 | { |
343 | my_coef_ptr coef = (my_coef_ptr)cinfo->coef; |
344 | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
345 | int blkn, ci, xindex, yindex, yoffset; |
346 | JDIMENSION start_col; |
347 | JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; |
348 | JBLOCKROW buffer_ptr; |
349 | jpeg_component_info *compptr; |
350 | |
351 | /* Align the virtual buffers for the components used in this scan. |
352 | * NB: during first pass, this is safe only because the buffers will |
353 | * already be aligned properly, so jmemmgr.c won't need to do any I/O. |
354 | */ |
355 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
356 | compptr = cinfo->cur_comp_info[ci]; |
357 | buffer[ci] = (*cinfo->mem->access_virt_barray) |
358 | ((j_common_ptr)cinfo, coef->whole_image[compptr->component_index], |
359 | coef->iMCU_row_num * compptr->v_samp_factor, |
360 | (JDIMENSION)compptr->v_samp_factor, FALSE); |
361 | } |
362 | |
363 | /* Loop to process one whole iMCU row */ |
364 | for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; |
365 | yoffset++) { |
366 | for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row; |
367 | MCU_col_num++) { |
368 | /* Construct list of pointers to DCT blocks belonging to this MCU */ |
369 | blkn = 0; /* index of current DCT block within MCU */ |
370 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
371 | compptr = cinfo->cur_comp_info[ci]; |
372 | start_col = MCU_col_num * compptr->MCU_width; |
373 | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
374 | buffer_ptr = buffer[ci][yindex + yoffset] + start_col; |
375 | for (xindex = 0; xindex < compptr->MCU_width; xindex++) { |
376 | coef->MCU_buffer[blkn++] = buffer_ptr++; |
377 | } |
378 | } |
379 | } |
380 | /* Try to write the MCU. */ |
381 | if (!(*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { |
382 | /* Suspension forced; update state counters and exit */ |
383 | coef->MCU_vert_offset = yoffset; |
384 | coef->mcu_ctr = MCU_col_num; |
385 | return FALSE; |
386 | } |
387 | } |
388 | /* Completed an MCU row, but perhaps not an iMCU row */ |
389 | coef->mcu_ctr = 0; |
390 | } |
391 | /* Completed the iMCU row, advance counters for next one */ |
392 | coef->iMCU_row_num++; |
393 | start_iMCU_row(cinfo); |
394 | return TRUE; |
395 | } |
396 | |
397 | #endif /* FULL_COEF_BUFFER_SUPPORTED */ |
398 | |
399 | |
400 | /* |
401 | * Initialize coefficient buffer controller. |
402 | */ |
403 | |
404 | GLOBAL(void) |
405 | jinit_c_coef_controller(j_compress_ptr cinfo, boolean need_full_buffer) |
406 | { |
407 | my_coef_ptr coef; |
408 | |
409 | coef = (my_coef_ptr) |
410 | (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, |
411 | sizeof(my_coef_controller)); |
412 | cinfo->coef = (struct jpeg_c_coef_controller *)coef; |
413 | coef->pub.start_pass = start_pass_coef; |
414 | |
415 | /* Create the coefficient buffer. */ |
416 | if (need_full_buffer) { |
417 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
418 | /* Allocate a full-image virtual array for each component, */ |
419 | /* padded to a multiple of samp_factor DCT blocks in each direction. */ |
420 | int ci; |
421 | jpeg_component_info *compptr; |
422 | |
423 | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
424 | ci++, compptr++) { |
425 | coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) |
426 | ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE, |
427 | (JDIMENSION)jround_up((long)compptr->width_in_blocks, |
428 | (long)compptr->h_samp_factor), |
429 | (JDIMENSION)jround_up((long)compptr->height_in_blocks, |
430 | (long)compptr->v_samp_factor), |
431 | (JDIMENSION)compptr->v_samp_factor); |
432 | } |
433 | #else |
434 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
435 | #endif |
436 | } else { |
437 | /* We only need a single-MCU buffer. */ |
438 | JBLOCKROW buffer; |
439 | int i; |
440 | |
441 | buffer = (JBLOCKROW) |
442 | (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, |
443 | C_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK)); |
444 | for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) { |
445 | coef->MCU_buffer[i] = buffer + i; |
446 | } |
447 | coef->whole_image[0] = NULL; /* flag for no virtual arrays */ |
448 | } |
449 | } |
450 | |