1 | /* |
2 | * reserved comment block |
3 | * DO NOT REMOVE OR ALTER! |
4 | */ |
5 | /* |
6 | * jccoefct.c |
7 | * |
8 | * Copyright (C) 1994-1997, Thomas G. Lane. |
9 | * This file is part of the Independent JPEG Group's software. |
10 | * For conditions of distribution and use, see the accompanying README file. |
11 | * |
12 | * This file contains the coefficient buffer controller for compression. |
13 | * This controller is the top level of the JPEG compressor proper. |
14 | * The coefficient buffer lies between forward-DCT and entropy encoding steps. |
15 | */ |
16 | |
17 | #define JPEG_INTERNALS |
18 | #include "jinclude.h" |
19 | #include "jpeglib.h" |
20 | |
21 | |
22 | /* We use a full-image coefficient buffer when doing Huffman optimization, |
23 | * and also for writing multiple-scan JPEG files. In all cases, the DCT |
24 | * step is run during the first pass, and subsequent passes need only read |
25 | * the buffered coefficients. |
26 | */ |
27 | #ifdef ENTROPY_OPT_SUPPORTED |
28 | #define FULL_COEF_BUFFER_SUPPORTED |
29 | #else |
30 | #ifdef C_MULTISCAN_FILES_SUPPORTED |
31 | #define FULL_COEF_BUFFER_SUPPORTED |
32 | #endif |
33 | #endif |
34 | |
35 | |
36 | /* Private buffer controller object */ |
37 | |
38 | typedef struct { |
39 | struct jpeg_c_coef_controller pub; /* public fields */ |
40 | |
41 | JDIMENSION iMCU_row_num; /* iMCU row # within image */ |
42 | JDIMENSION mcu_ctr; /* counts MCUs processed in current row */ |
43 | int MCU_vert_offset; /* counts MCU rows within iMCU row */ |
44 | int MCU_rows_per_iMCU_row; /* number of such rows needed */ |
45 | |
46 | /* For single-pass compression, it's sufficient to buffer just one MCU |
47 | * (although this may prove a bit slow in practice). We allocate a |
48 | * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each |
49 | * MCU constructed and sent. (On 80x86, the workspace is FAR even though |
50 | * it's not really very big; this is to keep the module interfaces unchanged |
51 | * when a large coefficient buffer is necessary.) |
52 | * In multi-pass modes, this array points to the current MCU's blocks |
53 | * within the virtual arrays. |
54 | */ |
55 | JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU]; |
56 | |
57 | /* In multi-pass modes, we need a virtual block array for each component. */ |
58 | jvirt_barray_ptr whole_image[MAX_COMPONENTS]; |
59 | } my_coef_controller; |
60 | |
61 | typedef my_coef_controller * my_coef_ptr; |
62 | |
63 | |
64 | /* Forward declarations */ |
65 | METHODDEF(boolean) compress_data |
66 | JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf)); |
67 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
68 | METHODDEF(boolean) compress_first_pass |
69 | JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf)); |
70 | METHODDEF(boolean) compress_output |
71 | JPP((j_compress_ptr cinfo, JSAMPIMAGE input_buf)); |
72 | #endif |
73 | |
74 | |
75 | LOCAL(void) |
76 | start_iMCU_row (j_compress_ptr cinfo) |
77 | /* Reset within-iMCU-row counters for a new row */ |
78 | { |
79 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
80 | |
81 | /* In an interleaved scan, an MCU row is the same as an iMCU row. |
82 | * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. |
83 | * But at the bottom of the image, process only what's left. |
84 | */ |
85 | if (cinfo->comps_in_scan > 1) { |
86 | coef->MCU_rows_per_iMCU_row = 1; |
87 | } else { |
88 | if (coef->iMCU_row_num < (cinfo->total_iMCU_rows-1)) |
89 | coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; |
90 | else |
91 | coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; |
92 | } |
93 | |
94 | coef->mcu_ctr = 0; |
95 | coef->MCU_vert_offset = 0; |
96 | } |
97 | |
98 | |
99 | /* |
100 | * Initialize for a processing pass. |
101 | */ |
102 | |
103 | METHODDEF(void) |
104 | start_pass_coef (j_compress_ptr cinfo, J_BUF_MODE pass_mode) |
105 | { |
106 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
107 | |
108 | coef->iMCU_row_num = 0; |
109 | start_iMCU_row(cinfo); |
110 | |
111 | switch (pass_mode) { |
112 | case JBUF_PASS_THRU: |
113 | if (coef->whole_image[0] != NULL) |
114 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
115 | coef->pub.compress_data = compress_data; |
116 | break; |
117 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
118 | case JBUF_SAVE_AND_PASS: |
119 | if (coef->whole_image[0] == NULL) |
120 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
121 | coef->pub.compress_data = compress_first_pass; |
122 | break; |
123 | case JBUF_CRANK_DEST: |
124 | if (coef->whole_image[0] == NULL) |
125 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
126 | coef->pub.compress_data = compress_output; |
127 | break; |
128 | #endif |
129 | default: |
130 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
131 | break; |
132 | } |
133 | } |
134 | |
135 | |
136 | /* |
137 | * Process some data in the single-pass case. |
138 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
139 | * per call, ie, v_samp_factor block rows for each component in the image. |
140 | * Returns TRUE if the iMCU row is completed, FALSE if suspended. |
141 | * |
142 | * NB: input_buf contains a plane for each component in image, |
143 | * which we index according to the component's SOF position. |
144 | */ |
145 | |
146 | METHODDEF(boolean) |
147 | compress_data (j_compress_ptr cinfo, JSAMPIMAGE input_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, bi, ci, yindex, yoffset, blockcnt; |
154 | JDIMENSION ypos, xpos; |
155 | jpeg_component_info *compptr; |
156 | |
157 | /* Loop to write as much as one whole iMCU row */ |
158 | for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; |
159 | yoffset++) { |
160 | for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col; |
161 | MCU_col_num++) { |
162 | /* Determine where data comes from in input_buf and do the DCT thing. |
163 | * Each call on forward_DCT processes a horizontal row of DCT blocks |
164 | * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks |
165 | * sequentially. Dummy blocks at the right or bottom edge are filled in |
166 | * specially. The data in them does not matter for image reconstruction, |
167 | * so we fill them with values that will encode to the smallest amount of |
168 | * data, viz: all zeroes in the AC entries, DC entries equal to previous |
169 | * block's DC value. (Thanks to Thomas Kinsman for this idea.) |
170 | */ |
171 | blkn = 0; |
172 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
173 | compptr = cinfo->cur_comp_info[ci]; |
174 | blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width |
175 | : compptr->last_col_width; |
176 | xpos = MCU_col_num * compptr->MCU_sample_width; |
177 | ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */ |
178 | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
179 | if (coef->iMCU_row_num < last_iMCU_row || |
180 | yoffset+yindex < compptr->last_row_height) { |
181 | (*cinfo->fdct->forward_DCT) (cinfo, compptr, |
182 | input_buf[compptr->component_index], |
183 | coef->MCU_buffer[blkn], |
184 | ypos, xpos, (JDIMENSION) blockcnt); |
185 | if (blockcnt < compptr->MCU_width) { |
186 | /* Create some dummy blocks at the right edge of the image. */ |
187 | jzero_far((void FAR *) coef->MCU_buffer[blkn + blockcnt], |
188 | (compptr->MCU_width - blockcnt) * SIZEOF(JBLOCK)); |
189 | for (bi = blockcnt; bi < compptr->MCU_width; bi++) { |
190 | coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn+bi-1][0][0]; |
191 | } |
192 | } |
193 | } else { |
194 | /* Create a row of dummy blocks at the bottom of the image. */ |
195 | jzero_far((void FAR *) coef->MCU_buffer[blkn], |
196 | compptr->MCU_width * SIZEOF(JBLOCK)); |
197 | for (bi = 0; bi < compptr->MCU_width; bi++) { |
198 | coef->MCU_buffer[blkn+bi][0][0] = coef->MCU_buffer[blkn-1][0][0]; |
199 | } |
200 | } |
201 | blkn += compptr->MCU_width; |
202 | ypos += DCTSIZE; |
203 | } |
204 | } |
205 | /* Try to write the MCU. In event of a suspension failure, we will |
206 | * re-DCT the MCU on restart (a bit inefficient, could be fixed...) |
207 | */ |
208 | if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { |
209 | /* Suspension forced; update state counters and exit */ |
210 | coef->MCU_vert_offset = yoffset; |
211 | coef->mcu_ctr = MCU_col_num; |
212 | return FALSE; |
213 | } |
214 | } |
215 | /* Completed an MCU row, but perhaps not an iMCU row */ |
216 | coef->mcu_ctr = 0; |
217 | } |
218 | /* Completed the iMCU row, advance counters for next one */ |
219 | coef->iMCU_row_num++; |
220 | start_iMCU_row(cinfo); |
221 | return TRUE; |
222 | } |
223 | |
224 | |
225 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
226 | |
227 | /* |
228 | * Process some data in the first pass of a multi-pass case. |
229 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
230 | * per call, ie, v_samp_factor block rows for each component in the image. |
231 | * This amount of data is read from the source buffer, DCT'd and quantized, |
232 | * and saved into the virtual arrays. We also generate suitable dummy blocks |
233 | * as needed at the right and lower edges. (The dummy blocks are constructed |
234 | * in the virtual arrays, which have been padded appropriately.) This makes |
235 | * it possible for subsequent passes not to worry about real vs. dummy blocks. |
236 | * |
237 | * We must also emit the data to the entropy encoder. This is conveniently |
238 | * done by calling compress_output() after we've loaded the current strip |
239 | * of the virtual arrays. |
240 | * |
241 | * NB: input_buf contains a plane for each component in image. All |
242 | * components are DCT'd and loaded into the virtual arrays in this pass. |
243 | * However, it may be that only a subset of the components are emitted to |
244 | * the entropy encoder during this first pass; be careful about looking |
245 | * at the scan-dependent variables (MCU dimensions, etc). |
246 | */ |
247 | |
248 | METHODDEF(boolean) |
249 | compress_first_pass (j_compress_ptr cinfo, JSAMPIMAGE input_buf) |
250 | { |
251 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
252 | JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; |
253 | JDIMENSION blocks_across, MCUs_across, MCUindex; |
254 | int bi, ci, h_samp_factor, block_row, block_rows, ndummy; |
255 | JCOEF lastDC; |
256 | jpeg_component_info *compptr; |
257 | JBLOCKARRAY buffer; |
258 | JBLOCKROW thisblockrow, lastblockrow; |
259 | |
260 | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
261 | ci++, compptr++) { |
262 | /* Align the virtual buffer for this component. */ |
263 | buffer = (*cinfo->mem->access_virt_barray) |
264 | ((j_common_ptr) cinfo, coef->whole_image[ci], |
265 | coef->iMCU_row_num * compptr->v_samp_factor, |
266 | (JDIMENSION) compptr->v_samp_factor, TRUE); |
267 | /* Count non-dummy DCT block rows in this iMCU row. */ |
268 | if (coef->iMCU_row_num < last_iMCU_row) |
269 | block_rows = compptr->v_samp_factor; |
270 | else { |
271 | /* NB: can't use last_row_height here, since may not be set! */ |
272 | block_rows = (int) (compptr->height_in_blocks % compptr->v_samp_factor); |
273 | if (block_rows == 0) block_rows = compptr->v_samp_factor; |
274 | } |
275 | blocks_across = compptr->width_in_blocks; |
276 | h_samp_factor = compptr->h_samp_factor; |
277 | /* Count number of dummy blocks to be added at the right margin. */ |
278 | ndummy = (int) (blocks_across % h_samp_factor); |
279 | if (ndummy > 0) |
280 | ndummy = h_samp_factor - ndummy; |
281 | /* Perform DCT for all non-dummy blocks in this iMCU row. Each call |
282 | * on forward_DCT processes a complete horizontal row of DCT blocks. |
283 | */ |
284 | for (block_row = 0; block_row < block_rows; block_row++) { |
285 | thisblockrow = buffer[block_row]; |
286 | (*cinfo->fdct->forward_DCT) (cinfo, compptr, |
287 | input_buf[ci], thisblockrow, |
288 | (JDIMENSION) (block_row * DCTSIZE), |
289 | (JDIMENSION) 0, blocks_across); |
290 | if (ndummy > 0) { |
291 | /* Create dummy blocks at the right edge of the image. */ |
292 | thisblockrow += blocks_across; /* => first dummy block */ |
293 | jzero_far((void FAR *) thisblockrow, ndummy * SIZEOF(JBLOCK)); |
294 | lastDC = thisblockrow[-1][0]; |
295 | for (bi = 0; bi < ndummy; bi++) { |
296 | thisblockrow[bi][0] = lastDC; |
297 | } |
298 | } |
299 | } |
300 | /* If at end of image, create dummy block rows as needed. |
301 | * The tricky part here is that within each MCU, we want the DC values |
302 | * of the dummy blocks to match the last real block's DC value. |
303 | * This squeezes a few more bytes out of the resulting file... |
304 | */ |
305 | if (coef->iMCU_row_num == last_iMCU_row) { |
306 | blocks_across += ndummy; /* include lower right corner */ |
307 | MCUs_across = blocks_across / h_samp_factor; |
308 | for (block_row = block_rows; block_row < compptr->v_samp_factor; |
309 | block_row++) { |
310 | thisblockrow = buffer[block_row]; |
311 | lastblockrow = buffer[block_row-1]; |
312 | jzero_far((void FAR *) thisblockrow, |
313 | (size_t) (blocks_across * SIZEOF(JBLOCK))); |
314 | for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) { |
315 | lastDC = lastblockrow[h_samp_factor-1][0]; |
316 | for (bi = 0; bi < h_samp_factor; bi++) { |
317 | thisblockrow[bi][0] = lastDC; |
318 | } |
319 | thisblockrow += h_samp_factor; /* advance to next MCU in row */ |
320 | lastblockrow += h_samp_factor; |
321 | } |
322 | } |
323 | } |
324 | } |
325 | /* NB: compress_output will increment iMCU_row_num if successful. |
326 | * A suspension return will result in redoing all the work above next time. |
327 | */ |
328 | |
329 | /* Emit data to the entropy encoder, sharing code with subsequent passes */ |
330 | return compress_output(cinfo, input_buf); |
331 | } |
332 | |
333 | |
334 | /* |
335 | * Process some data in subsequent passes of a multi-pass case. |
336 | * We process the equivalent of one fully interleaved MCU row ("iMCU" row) |
337 | * per call, ie, v_samp_factor block rows for each component in the scan. |
338 | * The data is obtained from the virtual arrays and fed to the entropy coder. |
339 | * Returns TRUE if the iMCU row is completed, FALSE if suspended. |
340 | * |
341 | * NB: input_buf is ignored; it is likely to be a NULL pointer. |
342 | */ |
343 | |
344 | METHODDEF(boolean) |
345 | compress_output (j_compress_ptr cinfo, JSAMPIMAGE input_buf) |
346 | { |
347 | my_coef_ptr coef = (my_coef_ptr) cinfo->coef; |
348 | JDIMENSION MCU_col_num; /* index of current MCU within row */ |
349 | int blkn, ci, xindex, yindex, yoffset; |
350 | JDIMENSION start_col; |
351 | JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; |
352 | JBLOCKROW buffer_ptr; |
353 | jpeg_component_info *compptr; |
354 | |
355 | /* Align the virtual buffers for the components used in this scan. |
356 | * NB: during first pass, this is safe only because the buffers will |
357 | * already be aligned properly, so jmemmgr.c won't need to do any I/O. |
358 | */ |
359 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
360 | compptr = cinfo->cur_comp_info[ci]; |
361 | buffer[ci] = (*cinfo->mem->access_virt_barray) |
362 | ((j_common_ptr) cinfo, coef->whole_image[compptr->component_index], |
363 | coef->iMCU_row_num * compptr->v_samp_factor, |
364 | (JDIMENSION) compptr->v_samp_factor, FALSE); |
365 | } |
366 | |
367 | /* Loop to process one whole iMCU row */ |
368 | for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; |
369 | yoffset++) { |
370 | for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row; |
371 | MCU_col_num++) { |
372 | /* Construct list of pointers to DCT blocks belonging to this MCU */ |
373 | blkn = 0; /* index of current DCT block within MCU */ |
374 | for (ci = 0; ci < cinfo->comps_in_scan; ci++) { |
375 | compptr = cinfo->cur_comp_info[ci]; |
376 | start_col = MCU_col_num * compptr->MCU_width; |
377 | for (yindex = 0; yindex < compptr->MCU_height; yindex++) { |
378 | buffer_ptr = buffer[ci][yindex+yoffset] + start_col; |
379 | for (xindex = 0; xindex < compptr->MCU_width; xindex++) { |
380 | coef->MCU_buffer[blkn++] = buffer_ptr++; |
381 | } |
382 | } |
383 | } |
384 | /* Try to write the MCU. */ |
385 | if (! (*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { |
386 | /* Suspension forced; update state counters and exit */ |
387 | coef->MCU_vert_offset = yoffset; |
388 | coef->mcu_ctr = MCU_col_num; |
389 | return FALSE; |
390 | } |
391 | } |
392 | /* Completed an MCU row, but perhaps not an iMCU row */ |
393 | coef->mcu_ctr = 0; |
394 | } |
395 | /* Completed the iMCU row, advance counters for next one */ |
396 | coef->iMCU_row_num++; |
397 | start_iMCU_row(cinfo); |
398 | return TRUE; |
399 | } |
400 | |
401 | #endif /* FULL_COEF_BUFFER_SUPPORTED */ |
402 | |
403 | |
404 | /* |
405 | * Initialize coefficient buffer controller. |
406 | */ |
407 | |
408 | GLOBAL(void) |
409 | jinit_c_coef_controller (j_compress_ptr cinfo, boolean need_full_buffer) |
410 | { |
411 | my_coef_ptr coef; |
412 | |
413 | coef = (my_coef_ptr) |
414 | (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_IMAGE, |
415 | SIZEOF(my_coef_controller)); |
416 | cinfo->coef = (struct jpeg_c_coef_controller *) coef; |
417 | coef->pub.start_pass = start_pass_coef; |
418 | |
419 | /* Create the coefficient buffer. */ |
420 | if (need_full_buffer) { |
421 | #ifdef FULL_COEF_BUFFER_SUPPORTED |
422 | /* Allocate a full-image virtual array for each component, */ |
423 | /* padded to a multiple of samp_factor DCT blocks in each direction. */ |
424 | int ci; |
425 | jpeg_component_info *compptr; |
426 | |
427 | for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; |
428 | ci++, compptr++) { |
429 | coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) |
430 | ((j_common_ptr) cinfo, JPOOL_IMAGE, FALSE, |
431 | (JDIMENSION) jround_up((long) compptr->width_in_blocks, |
432 | (long) compptr->h_samp_factor), |
433 | (JDIMENSION) jround_up((long) compptr->height_in_blocks, |
434 | (long) compptr->v_samp_factor), |
435 | (JDIMENSION) compptr->v_samp_factor); |
436 | } |
437 | #else |
438 | ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); |
439 | #endif |
440 | } else { |
441 | /* We only need a single-MCU buffer. */ |
442 | JBLOCKROW buffer; |
443 | int i; |
444 | |
445 | buffer = (JBLOCKROW) |
446 | (*cinfo->mem->alloc_large) ((j_common_ptr) cinfo, JPOOL_IMAGE, |
447 | C_MAX_BLOCKS_IN_MCU * SIZEOF(JBLOCK)); |
448 | for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) { |
449 | coef->MCU_buffer[i] = buffer + i; |
450 | } |
451 | coef->whole_image[0] = NULL; /* flag for no virtual arrays */ |
452 | } |
453 | } |
454 | |